id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0003/astro-ph0003347.html | ar5iv | text | # Untitled Document
Perspective of long baseline optical interferometry
S. K. Saha<sup>1</sup> e-mail: sks@iiap.ernet.in <sup>2</sup> e-mail: smorel@cfa.harvard.edu , and S. Morel<sup>2</sup>
<sup>1</sup>Indian Institute of Astrophysics, Bangalore 560 034, India.
<sup>2</sup>Infrared Optical Telescope Array, F. L. Whipple Observatory, 670 Mt-Hopkins Road,
Amado AZ 85645, USA.
Received 10. 1. 2000 ; Accepted 21. 2. 2000
Abstract. This article is a sort of sequel of the earlier extensive review by Saha (1999a) where emphasis was laid down on the ground based single aperture, as well as on the working long baseline optical interferometers (LBI) situated at the various observatories across the globe that are producing a large amount of astronomical results. Since the future of high resolution astronomy lies with the new generation of arrays, the numerous technical challenges of developing such systems are addressed indicating the current trends and the path to future progress in interferometry. The new generation interferometers such as Palomar testbed interferometer (PTI), Navy prototype optical interferometer (NPOI), Keck interferometer, Very large telescope interferometer (VLTI), Center for high angular resolution astronomy (CHARA) array, Optical very large array (OVLA), Mitaka optical infrared arrays (MIRA), etc., are being developed. A few of them, viz., PTI, NPOI, IOTA are producing results. Among the working interferometers that have been described earlier by Saha (1999a), the expansion of the Grand interféromètre à deux (two) télescopes (GI2T), Infrared and optical telescope array (IOTA) are in progress. The current status of all these interferometers stated above are enumerated. The data analysis being carried out using the working interferometers are also described. The space interferometry programmes are advancing very fast. Among the notable ones are the Space technology 3 (ST3), Space interferometry mission (SIM), and Darwin; they have already received funds. The technical details of these interferometers and their objectives are highlighted.
Key words: optical interferometry, arrays, space interferometry, astrometry.
Table of contents
1. Introduction
2. Historical development of ground-based interferometry
2.1. Intensity interferometry
2.2. Single aperture interferometry
2.3. Amplitude and phase interferometry
3. On-going ground-based interferometric projects
3.1. Palomar testbed interferometer (PTI)
3.2. Navy prototype optical interferometer (NPOI)
3.3. Keck interferometer
3.4. Very large telescopes interferometer (VLTI)
3.5. Center for high angular resolution astronomy (CHARA) arrays
3.6. Grand interféromètre à deux télescopes (GI2T) current status
3.7. Infrared-optical telescope array (IOTA) current status
3.8. Optical very large array (OVLA)
3.9. Mitaka optical infrared arrays (MIRA)
4. Data acquisition and processing in optical interferometry
4.1. Fringe acquisition and tracking
4.2. Data reduction
5. Image reconstruction
6. Space-borne interferometers
6.1. Astrometry from space
6.2. First space-borne interferometers
6.3. Searching for life on other planets
6.4. Long-term perspective
7. Conclusions
Acknowledgments
References
1. Introduction
The implementation of imaging by interferometry in optical astronomy is a challenging task. Though interferometry at optical wavelengths in astronomy began more than a century and a quarter ago (Fizeau, 1868), the progress in achieving high angular resolution has been modest. The first successful measurement of the angular diameter of $`\alpha `$ Orionis was performed in 1920 using stellar interferometer (Michelson and Pease, 1921), but the field lay dormant until it was revitalized by the development of intensity interferometry (Brown and Twiss, 1958). Over the last few decades, a marked progress has been witnessed in the development of this field, offering to realize the potential of the interferometric technique.
Single aperture speckle interferometry (Labeyrie, 1970) decodes the diffraction-limited spatial Fourier spectrum and image features of the object. A profound increase has been noticed in its contribution (Saha, 1999a and references therein) to measure fundamental stellar parameters, viz., (i) diameter of stars, (ii) separation of close binary stars, (iii) imaging of emission line of the active galactic nuclei (AGN), (iv) the spatial distribution of circumstellar matter surrounding objects, (v) the gravitationally lensed QSO’s, etc., (Saha, 1999a, 1999b and references therein).
Significant improvements in technological innovation over the past several years have brought the hardware to compensate in real-time for telescope image degradation induced by the atmospheric turbulence that distorts the characteristics of light traveling through it. The limitation is due to warping of iso-phase surfaces and intensity variation across the wavefront, thereby, distorting the shapes of the wavefront (Fried, 1966). The blurring suffered by such images is modeled as convolution with the point spread function (PSF). Wavefront sensing and adaptive optics (AO) are based on this hardware oriented correction (Babcock, 1953, Rousset et al., 1990).
Success in synthesizing images obtained from a pair of independent telescopes on a North-South baseline configuration (Labeyrie, 1975, Labeyrie et al., 1986, Shao et al., 1988), impelled astronomers to venture towards ground-based very large arrays (Davis et al., 1992). Potentials for progress in the direction of developing large interferometric arrays of telescopes (Labeyrie, 1996) are expected to provide images, spectra of quasar host galaxies, exo-planets that may be associated with stars outside the solar system (Labeyrie, 1995, 1998a, 1998b). Plans are also on to put an interferometer of a similar kind on the surface of the moon at the fall of this century. The technique of developing long baseline Fizeau-type interferometer for lunar operation consisting of 20 to 27 off-axis parabolic segments carried by robotic hexapodes that are movable during observing run has been suggested by Arnold et al., (1996). In a very recent article, Saha (1999a) has discussed at length about the interferometric techniques, that include the basic features of the working long baseline interferometers (LBI) with two or more optical telescopes. This review focuses on the current activities of the various groups across the globe to develop new ground-based, as well as space-borne interferometers in the optical domain, data processing techniques being adapted at the working LBIs; an account of historical development of high resolution astronomy is enunciated for the benefit of the readers as well. Some of the important results obtained with the new interferometers are also highlighted.
2. Historical development of ground-based interferometry
High angular resolution of an stellar object is an important aspect which astronomers are aspiring for. Ever since Fizeau (1868) had suggested to install a screen with two holes on top of the telescope that produce Young’s fringes at its focal plane as the fringes remain visible in presence of seeing, several attempts have been made with moderate sized telescopes to measure stellar diameters. Stéphan (1874) tried to resolve Sirius by using several masks with hole separation up to 65 cm on the 80 cm telescope of Observatoire de Marseille (France). No fringe contrast change was noticed and, therefore, only a maximum diameter of Sirius was deduced. One of the first significant results was the measurement of diameter of the satellites of Jupiter with a Fizeau interferometer on top of the Yerkes refractor by Michelson (1891). With the 100 inch telescope at Mt. Wilson (Anderson, 1920), the angular separation of spectroscopic binary star Capella was also determined.
To overcome the restrictions of the baseline, Michelson (1920) constructed the stellar interferometer equipped with 4 flat mirrors to fold the beams by installing a 7 m steel beam on top of the telescope afore-mentioned 100 inch telescope; the supergiant star $`\alpha `$ Orionis were resolved (Michelson and Pease, 1921). Due to the various difficulties, viz., (i) effect of atmospheric turbulence, (ii) variations of refractive index above small sub-apertures of the interferometer, and (iii) mechanical instability, the project was abandoned.
2.1. Intensity interferometry
The field of optical interferometry lay dormant until it was revitalized by the development of intensity interferometry (Brown and Twiss, 1958). Success in completing the intensity interferometer at radio wavelengths (Brown et al., 1952), in which the signals at the antennae are detected separately and the angular diameter of the source is obtained by measuring correlation of the intensity fluctuations of the signals as a function of antenna separation, Brown and Twiss (1958) demonstrated its potential at optical wavelengths by measuring the angular diameter of Sirius. Subsequent development of this interferometer with a pair of 6.5 meter light collector on a circular railway track spanning 188 meter (Brown et al., 1967), depicted the measurements of 32 southern binary stars with angular resolution limit of 0.5 milliarcseconds (Brown, 1974). The project was abandoned due to lack of photons beyond 2.5 magnitude stars.
2.2. Single aperture interferometry
Meanwhile, Labeyrie (1970) had invented speckle interferometric technique that retrieves the diffraction-limited information of an object. The diffraction-limited resolution of celestial objects viewed through the Earth’s turbulent atmosphere could be achieved with the large optical telescope, by post detection processing of a large data set of short-exposure images using Fourier-domain methods. Certain specialized moments of the Fourier transform of a short-exposure image contain diffraction-limited information about the object of interest. Owing to the turbulent phenomena associated with heat flow and winds in the atmosphere, the density of air fluctuates in space and time. The inhomogeneities of the refractive index of the air can have devastating effect on the resolution achieved by any large telescope. The disturbance takes the form of distortion of the shape of the wavefront and variations of the intensity across the wavefront. Due to the motion and temperature fluctuations in the air above the telescope aperture, inhomogeneities in the refractive index develop. These inhomogeneities have the effect of breaking the aperture into cells with different values of refractive index that are moved by the wind across the telescope aperture.
The power spectral density of refractive index fluctuations caused by the atmospheric turbulence follows a power law with large eddies having greater power (Tatarski, 1967). A plane wave propagating through the atmosphere of Earth is distorted by refractive index variation in the atmosphere (troposphere); it suffers phase fluctuations and reaches the entrance pupil of with patches of random excursions in phase (Fried, 1966). Therefore, the image of the star in the focal plane of a telescope is larger than its Airy disk (theoretical size 1.22$`\lambda `$/D is known as Rayleigh limit or diffraction limit, where, D is the diameter of the telescope). The size is equivalent to the atmospheric point spread function (point spread function is a modulus square of the Fourier transform of the aperture function). The resolution at the image plane of the telescope is determined by the width of the PSF which is of the order of (1.22$`\lambda /r_0`$), where, $`\lambda `$ is a wavelength of light and $`r_0`$ is the average size of the turbulence cell, which is of the order of 10 cm. The statistical properties of speckle pattern depend both on the coherence of the incident light and the properties of random medium. Mathematically, speckles are simply the result of adding many sine functions having different, random characteristics. Since the positive and negative values cannot cancel out everywhere, adding an infinite number of such sine functions would result in a function with 100 $`\%`$ constructed oscillations. Further details about the technique, way of recording speckles of any astronomical objects, image processing can be found in the most recent article by Saha (1999a, 1999b).
The afore-mentioned technique has been successful in obtaining spectacular results of a wide range of objects; a glimpse of such studies can be found in a recent review by Saha (1999a and references therein). Studies of the morphology of stellar atmospheres, the circumstellar environment of nova or supernova, YPN, long period variables (LPV), rapid variability of AGNs etc., are of paramount importance in astrophysics. Details of the structure of a wide range of stellar objects at scales of 0.015<sup>′′</sup> \- 0.03<sup>′′</sup> are routinely observed. The physical properties of red dwarfs in the vicinity of sun can also be looked into; some dwarfs may often be close binaries. Speckle interferometric technique has been extended to IR domain too. With the photon counting detector system (Saha, 1999a and references therein) which is an essential tool in the application of optical interferometric imaging that allows the accurate photon centroiding, as well as provides dynamic range needed for measurements of source characteristics, one can record the specklegrams of the object of faintest limiting magnitude. Further benefits have been witnessed when the atmospherically degraded images of these objects are applied to image restoration techniques (Liu and Lohmann, 1973, Rhodes and Goodman, 1973, Knox and Thomson, 1974, Lynds et al., 1976, Weigelt, 1977, Lohmann et al., 1983, Ayers and Dainty, 1988) for obtaining Fourier phase. Mapping the finer features of such objects would produce qualitative scientific results.
Development of various interferometric techniques, namely, (i) speckle spectroscopy (Grieger and Weigelt, 1992), (ii) speckle polarimetry (Falcke et al., 1996), (iii) pupil plane interferometry (Roddier and Roddier, 1988), (iv) Closure-phase method (Baldwin et al., 1986), (v) aperture synthesis using both partial redundant and non-redundant masking (Haniff et al., 1987, 1989, Nakajima et al., 1989, Busher et al., 1990, Bedding et al., 1992, 1994, Bedding, 1999), (vi) differential speckle interferometry (Petrov et al., 1986) too contribute in obtaining new results. Adaptive optics system introduces controllable counter wavefront distortion which both spatially and temporally follows that of the atmosphere. Adaptive optical systems may become standard tool for the new generation large telescopes. A considerable amount of new results have already been published. Detailed technique and the results can be seen in the recent reviews (Léna, 1997, Léna and Lai, 1999a, 1999b, Saha, 1999a).
2.3. Amplitude and phase interferometry
Subsequently, Labeyrie (1975) had developed a long baseline optical interferometer $``$ Interféromètre à Deux Télescopes (I2T) $``$ with a pair of 25 cm telescopes at Observatoire de Calern, France, exploiting the concept of merging speckles from both the telescopes. In other words, the fringed speckle can be visualized when a speckle from one telescope is merged with the speckle from the other telescope. His design combines features of the Michelson and of the radio interferometers. The use of independent telescopes increases the resolving capabilities. In this case, Coudé beams from both the telescopes arrive at central station and recombines them. Apart from the first measurements for a number of giant stars (Labeyrie, 1985), this interferometer also determined the effective temperatures of giant stars (Faucherre et al., 1983). In addition, resolving the gas envelope of the Be star $`\gamma `$ Cassiopeiae in the H$`\alpha `$ line (Thom et al., 1986) has a major achievement from I2T. In the infrared, diameters of cool bright giants and their effective temperature at 2.2 $`\mu `$m (DiBenedetto and Rabbia, 1987) have also been measured.
Following the success of its operation, Labeyrie (1978) undertook a project of building large interferometer known as Grand Interféromètre à Deux (two) Télescopes (GI2T) at the same observatory. This interferometer comprises of two 1.5 meter spherical telescopes on a North-South baseline, which are movable on a railway track (Labeyrie et al., 1986). The first scientific result came out of this interferometer in 1989 (Mourard et al., 1989) that had resolved the rotating envelope of hot star $`\gamma `$ Cassiopeiae. This object has been the favorite target to the GI2T (Stee et al., 1995, 1998). The emerging results on $`\beta `$ Lyrae, $`\delta `$ Cep (new and accurate distance estimate), P Cyg (the first discovery of an asymmetry in its wind) and $`\zeta `$ Tau (the first evidence for a one-armed oscillation in a Be star equatorial disk) are the most spectacular results from the GI2T too (Mourard et al., 1997, Harmanec et al., 1996, Vakili et al., 1997, 1998a, 1998b). The technical details of this kind of interferometers can be found in the recent article by Saha (1999a).
There are several long baseline interferometers, viz., (i) Mt. Wilson stellar interferometer, (ii) Sydney University stellar interferometer (SUSI), (iii) Cambridge Optical Aperture Synthesis Telescope (COAST), (iv) Infrared Optical Telescope Array (IOTA) are in operation. The technical details of these interferometers, as well as the results obtained so far can be found in the review article by Saha (1999a). However, a few salient objectives and programmes of these are reported in brief.
Measurements of precise stellar positions and motions of the stars are the major programmes being carried out with the Mark III interferometer at Mt. Wilson (Shao et al., 1990, Hummel, 1994). This set up has also been used to derive the fundamental stellar parameters, like the orbits for spectroscopic, as well as eclipsing binaries (Armstrong et al., 1992a, 1992b, Pan et al., 1992, Shao and Colavita, 1994), structure of circumstellar shells (Bester et al., 1991) etc.
The interferometers, namely, COAST, SUSI, IOTA etc., are relatively new. The expansion of a few of them are in progress. Nevertheless, they have produced several spectacular results to be mentioned. Among the notable results with COAST, mapping of the double-lined spectroscopic binary $`\alpha `$ Aurigae, resolving of $`\alpha `$ Tau are the important ones (Baldwin et al., 1996, 1998); detecting a circularly symmetric data with an unusual flat-topped and limb darkening profile of $`\alpha `$ Orionis (Burns et al., 1997), variations of the cycle of pulsation of Mira variable R Leonis (Burns et al., 1998) etc., have also been reported. With the SUSI interferometer, Davis et al., (1998, 1999) have determined the diameter of $`\delta `$ CMa with an accuracy of $`\pm `$1.8%. The results with the IOTA interferometer so far reported are from the near IR bands. They are in the form of measuring the angular diameters and effective temperatures of carbon stars (Dyck et al., 1996a), carbon Miras and S types (Van Belle et al., 1997), K and M giants and supergiants (Dyck et al., 1996b, 1998, Perrin et al., 1998), Mira variables (Van Belle et al., 1996), and cool giant stars (Dyck et al., 1995).
3. On-going ground-based interferometric projects
In view of the growing importance of high angular resolution interferometry, several projects of developing LBIs are in progress. Since the technical details of the well established interferometers have been already described in the recent article by Saha (1999a), the salient features of some of the on-going interferometric projects are enumerated.
3.1. Palomar testbed interferometer (PTI)
The Palomar testbed interferometer (PTI), is an infrared phase-tracking interferometer in operation situated at Palomar Observatory, California; it was developed by the Jet Propulsion Laboratory and California Institute of Technology for NASA as a test-bench for the Keck interferometer. The main thrust of this interferometer is to develop techniques and methodologies for doing narrow angle astrometry for the purpose of detecting extra-solar planets (Wallace et al., 1998) that measures the wobble in the position of a star caused by the transverse component of a companion’s motion.
Three 40 cm siderostats (steerable flat mirrors for sending starlight in a fixed direction) coupled to beam compressors (reducing the beam diameter) can be used pairwise to provide baselines up to 110 m (Colavita et al., 1999). This interferometer tracks the white light fringes using an array detector at 2.2 $`\mu `$m (K band) and active delay-lines with a range of $`\pm `$38 m. Among others, the notable feature of this interferometer is that of implementation of a dual-star astrometric ability; observation of fringes from 2 close stars simultaneously for phase referencing and narrow-angle astrometry. An end-to-end heterodyne laser metrology system is used to measure the optical path length of the starlight (Wallace et al., 1998). They also claimed the better performances after the recent upgradations of PTI, viz., a single mode fiber for spatial filtering, vacuum pipes to relay the beams, accelerometers on the siderostat mirror etc.
Malbet et al., (1998) have resolved the young stellar object FU Orionis using the said interferometer in the near infrared with a projected resolution better than 2 AU. Observations of the young binary stars, $`\iota `$ Peg have been also conducted by Pan et al., (1996) with this interferometer. They have determined its visual orbit with separation of 1 mas in R. A., having a circular orbit with a radii of 9.4 mas. Measurements of diameters and effective temperatures of G, K, and M giants and supergiants have been reported recently by Van Belle et al., (1999). The visual orbit for the spectroscopic binary $`\iota `$ Peg with interferometric visibility data recoded by PTI has also been derived (Boden et al., 1999).
3.2. Navy prototype optical interferometer (NPOI)
The astrometric array of NPOI, a joint project of the US Naval Research laboratory and the US Naval Observatory is designed to measure positions with precision comparable to that of Hipparcos (1997). This interferometer is located at the Lowell Observatory, Arizona and is capable of maintaining accuracy of the positions of the brightest Hipparcos stars while improving the precision of their proper motions. The anticipated wide-angle astrometric precision of the NPOI is about $``$2 mas. (Armstrong et al., 1998). Since the high precision astrometry is an important aspect to astronomy that helps in establishing cosmic distance scale, measurements of proper motion can confirm stars as members of cluster (known distance), may elucidate the dynamics of the Galaxy. NPOI plans to measure the positions of some radio stars that would help in matching radio sources with their optical counterparts.
This interferometer includes sub-arrays for imaging and for astrometry and is developed at Y-shaped (Very Large Array-like) baseline configuration. The light beams are passed through vacuum pipes to the central laboratory. For astrometric mode, 4 fixed siderostats (0.4 m diameter) are used with the baselines extendable from 19 m to 38 m (Armstrong et al., 1998). The shared back and covers 450-850 nm in 32 channels. The other notable features are being the delay system, active group-delay fringe tracking etc. The astrometric sub-array has a laser metrology system to measure the motions of the siderostats with respect to one another and to the bedrock. While for imaging mode, 6 transportable siderostats (0.12 m diameter) are used with the baselines from 2 m to 432 m. Three siderostat positions are kept with equal space for each arm of the Y. Coherence of imaging configuration is maintained by phase bootstrapping (see section 4.1). Observations in visible spectrum with 3-elements have been carried out using avalanche photo-diode as detector (Hummel et al., 1998). The dynamic range in the best of the NPOI images exceeds 100:1 (Armstrong et al., 1998).
A few examples of science with NPOI can be read from the following results. Pauls et al., (1998) have measured the limb darkened angular diameters of late-type giant stars using the said interferometer with three optical elements; measurement of non-zero closure phase has been performed on a single star. Hajian et al., (1998) have also observed the limb darkened diameters of two K giants, $`\alpha `$ Arietis and $`\alpha `$ Cassiopeiae with 20 spectral channels covering 520-850 nm. They were able to extend the spatial frequency coverage beyond the first zero of the stellar visibility function for these stars. Hummel et al., (1998) have determined the orbital parameters of two spectroscopic binaries, $`\zeta `$ Ursae Majoris (Mizar A), $`\eta `$ Pegasi (Matar) and derived masses and luminosities based on the data obtained with said interferometer; published radial velocities and Hipparcos trigonometrical parallaxes were used for the analysis.
3.3. Keck interferometer
The development of Keck interferometer consisting of 2 $`\times `$ 10 m apertures (main telescopes) with a fixed baseline of 85 m is in progress: the expected resolution of this is of the order of 5 mas at 2.2 $`\mu `$m. The baselines available with outrigger telescopes (4 $`\times `$ 1.8 m) will be between 25 m to 140 m (fixed baselines). For imaging the main telescope would be used with outriggers (Colavita et al., 1998). This project is funded by NASA and is being carried out by Jet Propulsion Laboratory (JPL) and California Association for Research in Astronomy (CARA). This large interferometer is located at Mauna Kea Observatory, Hawaii. It will combine phased pupils provided by adaptive optics for the main telescopes (up to V=9, 39 mas FWHM, Strehl ratio=30%) and fast tip/tilt correction on the outriggers. Beam recombination will be carried out by 5 two-way combiners at 1.5-2.4 $`\mu `$m for fringe tracking, astrometry, and imaging. Project for a 10 $`\mu `$m nulling-combiner for exo-zodiacal disk characterization is also undertaken.
The astrometric accuracy is expected to be of the order of 20 to 30 $`\mu `$as/$`\sqrt{\mathrm{hour}}`$. With the main telescopes observations for searching Jovian planets, as well as for characterizing the exo-zodiacal disks may commence by 2001 and with the addition of the outriggers, astrometric observation will be carried out by 2003.
3.4. Very large telescope interferometer (VLTI)
The VLTI, built by the European Southern Observatory and located at Cerro Paranal, Chile, will be a versatile facility consisting of four 8 m fixed telescopes (“unit telescopes”) and three 1.8 m mobile auxiliary telescopes which can be installed on any of the 30 stations built on the ground. The maximum baseline of VLTI is 200 m. Siderostats for first tests could be used as well (Derie et al., 2000). Coudé beams from these apertures are sent through delay-lines operating in rooms at atmospheric pressure but at a thoroughly controlled temperature in order to avoid turbulence. The beams reach an optical switch-yard to be directed to one of the four expected recombiners: VINCI (Kervella et al., 2000), MIDI (Leinert and Graser, 1998), AMBER (Petrov et al., 2000) or PRIMA (Quirrenbach et al., 1998).
VINCI (VLT INterferometer Commissioning Instrument) will be a single-mode fiber recombiner operating at 2.2 $`\mu `$m like FLUOR (see 3.7) and is intended to be used for debugging the upstream sub-systems of VLTI. MIDI (MID-Infrared) will be a beamsplitter-based recombiner that will be used for observations at 10 $`\mu `$m. AMBER (Astronomical Multiple BEam Recombiner) has been designed for observations between 1 $`\mu `$m and 2.5 $`\mu `$m. It will be able to perform recombination of three beams in order to use closure-phase techniques. PRIMA (Phase-Reference Imaging and Microarcsecond Astrometry) is a recombiner dedicated to narrow-angle astrometry (see 6.1). It should be able to reach a 10 $`\mu `$as resolution.
By the end of 1999, two unit telescopes were operating and the primary mirror has been installed on the third one. First fringes of the VLTI with two siderostats and VINCI are scheduled for the end of 2000. All the remaining combiners are scheduled to work either with the unit or the auxiliary telescopes by 2005.
3.5. Center for high angular resolution astronomy (CHARA) array
The Center for High Angular Resolution Astronomy (CHARA), Georgia State University, is currently building an interferometric array at Mt. Wilson, California, USA; it comprises six fixed 1 m telescopes arranged in a Y-shaped configuration with a maximum baseline of $``$350 m that would operate at optical and IR wave band (McAlister et al., 1998) with a limiting resolution of 0.2 mas. The main objective of this project is to measure the diameters, distances, masses and luminosities of stars, as well as to image features, viz., spots and flares on their surfaces. The aim of this project range from detecting other planetary systems to imaging the black hole driven central engines of quasars and active galaxies.
Light from the telescopes is sent through vacuum pipes to the centrally located Beam synthesis facility, a L-shaped long building (McAlister et al., 1994) that houses the optical path length equalizer (OPLE) and the beam combination laboratory (BCL). Constructions of piers for the five telescopes (McAlister et al., 1998) have been completed. Optical delay line carts and their metrology, similar versions that are used at NPOI and PTI, and control systems, are being developed at JPL, USA. The first fringes in K band from a star have been acquired with two telescopes of CHARA by the end of 1999. The other baselines and the visible spectrum recombiner should be operational in the next few years.
3.6. Grand interféromètre à deux télescopes (GI2T) current status
The detailed description of GI2T is available in a recent article by Saha (1999a). This interferometer has recently been upgraded with a new recombiner named REGAIN (REcombineur pour GrAnd Interféromètre). This recombiner, able to operate with three telescopes (Mourard et al., 1998) has the following features.
The 76 mm Coudé beams coming from the telescopes are first compressed to 5 mm in order to stabilize the pupil image in a fixed plane. Then, field rotators consisting of four plane mirrors are used for each beam to compensate the polarization difference affecting the visibility measured (Rousselet-Perraut et al., 1996). The different chromatic dispersion between the two beams (due to operation at atmospheric pressure) is compensated by using for each beam two prisms which can slide on their hypotenuse, forming therefore a plate with adjustable thickness. This thickness is modified every 4 minutes, following the variation of the altitude of the observed object. Figure 1 depicts the process performed by an arm of the REGAIN table prior to recombination. Unlike the previous recombiner which had to move to track fringes, REGAIN uses a delay-line named LAROCA (Ligne A Retard de l’Observatoire de la Côte-d’Azur) featuring a cat’s eye reflector with a variable curvature mirror. An adaptive optics for the 1.5 m telescopes of GI2T is in development (Vérinaud et al., 1998).
Figure 1. Optical processing of a beam from one telescope of GI2T by the REGAIN recombiner.
The focal instrument of GI2T/REGAIN will be a spectrometer working either in dispersed fringes mode or in Courtès mode. In dispersed fringes (see 4.1), the spectral range is 480 nm to 750 nm and the spectral resolution can reach $`R=30,000`$. Separated recombination of two orthogonal polarizations will be possible. The detectors used will be two CP40 photon cameras. The Courtès mode consists in forming images at different wavelengths of speckles with fringes given by the recombination. The spectrometer of REGAIN in Courtès mode will be able to give 16 images at the same time.
However, first fringes (at 2.2 $`\mu `$m) with GI2T/REGAIN have already been acquired in August 1999 (Weigelt et al., 2000) thanks to a different slit spectrometer and a PICNIC infrared camera built by the Max-Planck-Institute für Radioastronomie (Bonn, Germany).
3.7. Infrared optical telescope array (IOTA) current status
A description of IOTA has already been presented in this journal (Saha, 1999a). IOTA has recently been upgraded (Traub et al. 2000) with a supplementary optical path delaying system (consisting of a long-travel delay-line fixed while observing and a short-travel delay-line tracking the star), a new control system for the long delay-lines and new high-precision secondary mirror holders for the telescopes. The third collector similar to the two already existing (a 45 cm siderostat and a fixed Mersenne telescope compressing the beam by 10) will soon be operational. Identical reflections are applied to the beams from the collectors to the recombiner in order to avoid any loss of fringe contrast due to different polarization states between the two beams at the recombination point. Figure 2 depicts the overview of IOTA interferometer.
Figure 2. Overview of IOTA interferometer.
The focal instrumentation consists of two infrared recombiners. The first one uses a classical beamsplitter (see 4.1). The two recombined beams are focused on two pixels of a NICMOS III infrared camera (Millan-Gabet et al., 1999). This camera is able to read up to ten fringe frames per second. Each fringe frame, containing 256 samples, is made by scanning the optical path difference between the two beams with a mirror mounted on a 60-micron stroke piezo-electric transducer (PZT). Last scientific results obtained with this instrument include environment characterization of Herbig AeBe stars (Millan-Gabet et al., 1998) and dust shell diameter measurement of CI Cam (Traub et al., 1998).
The second recombiner named FLUOR (Fiber-Linked Unit for Optical Recombination, (Coudé du Foresto and Ridgway, 1992) consists of single-mode fiber optics interfering beams (Shaklan and Roddier, 1987). These fibers have been designed to propagate infrared light at K band (2.2 $`\mu `$m) in TEM mode only, like a coaxial cable. Therefore, only plane waves perpendicular to the axis of the fiber may propagate over long distances. Concretely, this results in a “spatial filtering”, smoothing the wavefronts that have been corrugated by atmospheric turbulence. Figure 3 depicts the principle of wavefront smoothing by spatial filtering with a pinhole and with a single-mode fiber. The advantage of such a technique for interferometry is a reduction of the uncertainty on the measured visibility.
Figure 3. Principle of wavefront smoothing by spatial filtering with a pinhole and with a single-mode fiber.
Drawbacks of spatial filtering are a loss of optical coupling efficiency and larger photometric variations due to the turbulence. The FLUOR experiment was originally set up at the McMath solar tower of the Kitt-Peak National Observatory (Arizona), with a 5 m baseline. It has been installed at IOTA since 1994. The current FLUOR bench use a 180-micron stroke PZT for scanning the fringes. The detector used is the NICMOS III of IOTA. Four pixels are read (two for the interferometric fiber outputs, two for the photometric fiber outputs). Accurate diameter measurements of Mira variable stars (Perrin, 1999) and cepheids (Kervella et al., 1999), effective temperature measurements of giant stars (Perrin et al., 1998), have recently been done with FLUOR on IOTA. Figure 4 depicts the schematic of the FLUOR recombiner.
Figure 4. Schematic of the FLUOR recombiner. P1 and P2 are the photometric output fibers. I1 and I2 are the interferometric output fibers. These outputs are imaged by a lens on a NICMOS infrared array detector.
Attempts to observe with single-mode fibers at longer wavelengths at IOTA (TISIS experiment) have been done in L band (3.5 $`\mu `$m) by Menesson et al., (1999), and in M band (5 $`\mu `$m) in March 1999 (Menesson et al., 2000). However, the thermal background of the instrument and the presence of an atmospheric $`\mathrm{H}_2\mathrm{O}`$ absorption line in M band barred fringe acquisition.
3.8. Optical very large array (OVLA)
The OVLA is a project that was initiated about 10 years ago by A. Labeyrie. It is proposed to build an interferometer using innovative concepts. First, the collectors will be radically different from what has already been imagined. Each telescope structure (Dejonghe et al., 1998) is a 2.8 m diameter fiberglass sphere which may be oriented in any direction thanks to three motors featuring specially designed “barrel-caster” cabestans, mounted on their shafts. The sphere rests on these three barrel-casters: when one of them is steering the sphere, no friction occurs from the two other ones. The mirror of each telescope (1.5 m diameter with f/1.7) is made of ordinary window glass. It is 24 mm thick and weights 180 kg. An active optics is, therefore, required in order to get high-quality wavefronts for interferometric purpose: the mirror rests on three hard points and 29 actuators able to accurately correct its shape. Correction of the spherical aberration may be done by applying an electric current through the mirror coating between two chosen points of the edge, thus heating the top side of the mirror to compensate the noticed temperature difference with the bottom side (about 0.5C). A secondary mirror makes the beam afocal and compressed, a third steerable flat mirror sends this beam out through a slit located on the sphere. A motorized shutter can partially close this slit when a barrel-caster and the slit are in coincidence. It has been projected to mount the sphere and its motorization on a six-leg robot (hexapode) able to move on the ground while fringes are acquired (in order to compensate the OPD between beams). Due to its original features, an OVLA telescope requires a significant amount of electronics and a control system (Lardière et al., 1998) able to manage in real-time motors, actuators, shutter, hexapode. The first OVLA telescope has been tested in October 1999 (Arnold et al., 2000): the point spread function of the optics was 4” FWHM. Improvements of the image quality by a better control of the actuators remains, therefore, to be done in order to obtain diffraction-limited images. Once this is done, the OVLA might be used as the third telescope of the GI2T interferometer becoming then the “GI3T”. Figure 5 depicts the first operating telescope of proposed OVLA.
Figure 5. One of the telescope of proposed OVLA (Courtesy: O. Lardière).
The OVLA has actually been thought for different possible aperture diameters including 12 to 25 m (Labeyrie, 1998c). A new telescope structure has been imagined for this class of very large collectors: the “cage telescope”, in which the sphere is replaced with an icosahedral truss steerable by a different mechanical system. There are several options for the configuration of the OVLA interferometer featuring 27 apertures (like the Very Large Array radio-interferometer). The first one is to build a Fizeau interferometer, i.e. a fragmented giant telescope. Each telescope is shaped as it was a segment of the paraboloid mirror of the synthesized giant telescope. The array is arranged on the ground to form an ellipse. Images are then directly obtained at a recombination station located at a focus of the ellipse. This Fizeau configuration was thought for LOVLI (Lunar Optical Very Large Interferometer), the moon-based version of OVLA. However, controlling the particular shape of each telescope will be very difficult. A recent new concept of interferometric recombination, especially imagined for OVLA, is called “densified pupil” (Labeyrie, 1996). In Fizeau mode, the ratio aperture diameter/separation is constant from light collection to recombination in the image plane (homothetic pupil). In Michelson mode, this ratio is not constant since the collimated beams have the same diameter from the output of the telescope to the recombination lens. The distance between pupils is equal to the baseline at the collection and to a much smaller value just before the recombining lens. The disadvantage of the Michelson mode is a very narrow field of view compared to the Fizeau’s. However, a densified pupil interferometer (“extreme Michelson mode”), where, in the recombination plane, the distance between two pupils corresponding to two telescopes is minimized to become about equal to their diameter, may be very interesting to get direct images without using the heavy procedure of aperture synthesis (visibility measurement, phase calibration, Fourier synthesis…). It can be demonstrated that the definition (i.e. number of pixels of the image) of a densified pupil interferometer is equal to the square of its number of apertures. One difficulty is cophasing all the beams. Since 27 ($`=3^3`$) telescopes are expected, the cophasing of the whole array may be done hierarchically (Pedretti and Labeyrie, 1999) by cophasing triplets of beams (yielding a honeycomb pattern in the image plane), then triplets of triplets, etc… The limitation of the cophasing procedure by the photon noise would not be very important. According to numerical simulations, the expected limiting magnitude of the densified pupil OVLA is 8.3 if 10 cm apertures are used and 20 for 10 m apertures.
3.9. Mitaka optical infrared arrays (MIRA)
The MIRA project (a collaboration between the University of Tokyo and the National Astronomical Observatory of Japan) does not consist of one but several interferometers built one-by-one, each instrument being an upgrade of the previous one. The first of the series was MIRA-I (Machida et al., 1998). It had 25 cm siderostats and a 4 m baseline. The fringe detector was designed for 800 nm wavelength. Its successor, MIRA-I.2 (Sato et al., 1998) has the same baseline and slightly larger siderostats (30 cm). It features the equipment encountered on many operating interferometers: beam compressors (yielding 30 mm beams), delay-line operating in vacuum, tip-tilt correction system and laser metrology. MIRA-I and MIRA-I.2 are instruments specially designed for practicing interferometry and testing devices. The experience acquired from these interferometers will be useful for building larger interferometers of the MIRA project like MIRA-II, MIRA-SG and MIRA-III, which will be instruments for astrophysical research.
4. Data acquisition and processing in optical interferometry
Operating a long-baseline interferometer (i.e. finding fringes, measuring their visibility, interpreting the result) is a long and difficult process. First, a correct determination of the baseline vector must be established. Once this has been done, one knows, for a given object to observe, how to set the position of the optical delay-line to get fringes within, usually, a few hundred micron interval around the expected null-OPD point. Then, optics must be adjusted to avoid various aberrations and vignetting, which may be difficult to avoid when light is fed through long and narrow pipes. Then, fringes are searched by adjusting the delay-line position. However, once they are found, mechanical constraints on the instrument, errors on the pointing model, thermal drifts, various vibrations and atmospheric turbulence make the null-OPD point changing. The position of the delay-line (or any other delaying device in the optical path) must be adjusted in order to keep the fringes within the “observation window”: usually, the error on the OPD must be less than the coherence length defined by:
$`L_c={\displaystyle \frac{\overline{\lambda }^2}{\mathrm{\Delta }\lambda }}.`$ (1)
Where $`\overline{\lambda }`$ is the mean wavelength observed and $`\mathrm{\Delta }\lambda `$ is the spectral interval. This real-time control is called “fringe-tracking”.
When enough fringe patterns have been recorded, the visibility may be extracted. Then, a set of measured visibilities obtained (i.e. samples in the Fourier plane $`(u,v)`$ corresponding to the image) allows to partially reconstruct the high-angular resolution image of the observed object.
4.1. Fringe acquisition and tracking
For visible spectrum, three possible set-ups for fringe acquisition exist. In the first one (white fringes), the OPD is temporally modulated by a sawtooth signal, using a fast and short-travel delaying device (usually, a reflector mounted on a PZT). The intensity of the recombined beams describes, therefore, over the time a fringe pattern that is recorded by one or several mono-pixel detectors (photo-multipliers, avalanche photo-diodes, InSb photometers). Natural OPD drift due to the Earth-rotation can also be used for acquiring fringes, as it was done by the SOIRDÉTÉ interferometer (Rabbia et al., 1990). The second method (channeled spectrum) consists of imaging the dispersed recombined beam on a linear detector (CCD or photon-counting camera). In the third one (dispersed fringes), beams are dispersed prior to be recombined. Unlike the two previous techniques, recombination is not done by overlapping the beams, but by focusing them with a common lens, like in the original Michelson stellar interferometer. The detector used is a 2-D photon-counting camera. In infrared, no photon-counting is possible with the current technology. It is, therefore, important to use as less pixels as possible in order to reduce the global readout noise. Hence, the “white” fringes set-up will be preferably used for infrared observations. Figure 6 depicts the various possible set-ups for beam recombination and fringe acquisition.
Figure 6. Three possible set-up for beam recombination and fringe acquisition: white fringes (a), channeled spectrum (b), dispersed fringes (c).
Techniques to compensate the OPD drift between the two beams of a standard optical interferometer may be classified into two categories: coherencing and cophasing. The aim of coherencing is to keep the OPD within the coherence area of the fringes. Cophasing is a more demanding technique because the OPD must remain much smaller than the wavelength: fast compensation of the OPD variations due to the differential “piston” mode of the turbulence is, therefore, done, in order to “freeze” the fringes.
In white light set-up, the coherencing, as for IOTA (Morel et al., 2000), is done by scanning the OPD while acquiring interferometric signal, and then finding the null-OPD point in the fringe pattern. This yields the OPD correction to apply to the delay-line, at a few Hz servo-loop rate. With a channeled spectrum, the OPD is proportional to the fringe frequency. The method used called “group-delay tracking” (Lawson, 1995) is based on the Fourier transform of each frame acquired. The integration of the moduli of all the computed Fourier Transforms yields a peak whose position is proportional to the OPD. Group-delay tracking has been used on SUSI (Lawson, 1994) and COAST (Lawson, 1998) interferometers. A similar technique (Koechlin et al., 1996) named “real-time active fringe-tracking” (RAFT) has been applied to dispersed fringes on GI2T using a 2-D Fourier Transform. The advantage of RAFT over group-delay tracking is the knowledge of the sign of the OPD to measure and the possibility to be used with apertures larger than $`r_0`$, where overlapping wavefronts that have been corrugated by the turbulence would blur the fringes. However, dispersed fringes with large apertures require a complex optical system to rearrange the speckles in the image plane before dispersion and recombination (Bosc, 1988). Both group-delay tracking and RAFT allow a slow servo-loop period (up to a few seconds) by multiplying the coherence length by the number of spectral channels used. It is important to notice that their common use of the Fourier Transform make them optimal in the sense that they yield the same OPD than a maximum likelihood estimator (Morel and Koechlin, 1998).
Cophasing is usually performed with white fringes, using the “synchronous detection” method, as it was used on the Mark III interferometer (Shao and Colavita, 1988): the OPD is quickly scanned over a wavelength. Signal acquired from the detector is then processed in order to yield the phase-shift to compensate and the visibility. This can be done easily (Shao and Staelin, 1977) by integrating signal over four $`\lambda /4`$ bins, named $`A`$, $`B`$, $`C`$ and $`D`$. Phase-shift and visibility modulus are then given by:
$`\mathrm{\Delta }\phi =\mathrm{arctan}\left({\displaystyle \frac{BD}{AC}}\right);V={\displaystyle \frac{\pi \sqrt{(AC)^2+(BD)^2}}{\sqrt{2}(A+B+C+D)}}.`$ (2)
The cophasing technique may be compared to adaptive optics, like coherencing with non-white fringes may be compared to active optics. We can notice that a compound method, based on the synchronous detection applied to signals from several spectral channels, has been used on the NPOI interferometer (Benson et al., 1998). Fringe-tracking is usually done from data acquired for scientific purpose (i.e. visibility extraction), in order to not “share” the photons between two instruments. Hence, the fringe signal-to-noise ratio (SNR) is optimal. This fringe SNR is given by the expression (Lawson, 1995):
$`\mathrm{SNR}{\displaystyle \frac{NV^2}{\sqrt{1+0.5\times NV^2}}}.`$ (3)
Where $`N`$ is the number of photons acquired and $`V`$ is the visibility modulus.
However, it may be optimal to have two recombiners, one for visibility measurement, the other one for fringe-tracking. For example, at long baselines, when the expected fringe visibility is too low for tracking, it is possible to use a longer wavelength where the fringe contrast, for a white observed object, is higher. Meanwhile, fringes for computing the visibility are acquired at shorter wavelength than for tracking. Another method where photons are shared is called “bootstrapping”. It consists in dividing the baseline into sub-baselines by adding apertures along. Fringe-tracking is performed on each sub-baseline, where the visibility is higher than with the whole baseline. Hence, fringes are tracked on the whole baseline as well. This method is used on the NPOI interferometer (Armstrong et al., 1998). Figure 7 depicts the principle of baseline bootstrapping.
Figure 7. Principle of baseline bootstrapping. Apertures are represented by circles. Fringe-tracking methods may be enhanced by introduction of a priori information, in order to allow observations at fainter $`V`$ or fainter magnitudes. Gorham (1998) has proposed to improve white light cophasing by filtering data with a function computed to reduce the photon noise. The gain for the tracking limit magnitude, at constant $`V`$, is between 0.5 or 0.7. Methods introducing a priori information for GDT or RAFT have been imagined as well (Padilla et al., 1998, Morel and Koechlin, 1998).
4.2. Data reduction
The optimal integration time required for measuring a visibility point is a trade-off between the number of photons to collect and the Earth rotation shifting the sampled point in the $`(u,v)`$ plane. Most of the interferometers use two apertures and are unable to recover the complex visibility. Therefore, the information to extract from a batch of fringes is the modulus of the visibility. Theoretically, using merely the Fourier Transform would give an optimal estimate of the visibility modulus, as demonstrated by Walkup and Goodman (1973). However, white-light fringes obtained from coherencing are flawed by the differential piston that modulates their frequency. Techniques used in radio-interferometry (where wavelengths are much longer), like fitting a sinewave through the fringe data, are, therefore, not suitable. Perrin (1997) has proposed a method to remove the piston from fringes. However, this method requires a high fringe signal-to-noise ratio and may only be applied when fringe SNR is important. Schloerb et al., (1999) use a model of the turbulence effects to extract the visibility modulus.
Due to the atmospheric turbulence affecting the wavefronts before recombination, measurements of $`V`$ are biased by a random factor depending on the seeing quality. Instrumental flaws leading to optical aberrations and non-balanced flux between the two beams modify the measured visibility modulus as well. It is, therefore, important to calibrate each measure on an object by measuring $`V`$ on an non-variable unresolved source (e.g., a farther star) in the neighborhood of the studied object and at the same turbulence condition, i.e. right away after data acquisition on the studied object. To reproduce the instrumental conditions, the calibrator must roughly be as bright as the object to calibrate.
5. Image reconstruction
Interferometers with two apertures have limited possibilities for image reconstruction due to the absence of phase visibility recovering. Objects assumed with circular symmetry (“standard” stars) may be reconstructed with two-aperture interferometers. However, a problem comes from the dark-limbening of the stars observed. The radial intensity profile of a star may be given (Hestroffer, 1997) by:
$`I(r)=I(0)\left(1{\displaystyle \frac{r^2}{R^2}}\right)^{\alpha /2}.`$ (4)
Where $`R`$ is the radius of the star and $`\alpha `$ is the dark-limbening factor depending on the stellar atmosphere.
Many interferometers cannot measure low visibilities existing at high angular frequency (i.e, when $`\sqrt{u^2+v^2}`$ is large), beyond the first zero of the visibility function. Reconstructions are, therefore, ambiguous and neither the diameter nor the dark-limbening factor may be accurately determined. Usually, $`\alpha `$ is an a priori information given by the stellar atmosphere model. The diameter is, therefore, deduced from $`\alpha `$ and the interferometric data.
Binary systems have already been characterized by optical stellar interferometry. The expression of two unresolved sources, i.e. a binary system, is $`O(𝐱)=a\delta (𝐱+𝐱_0)+b\delta (𝐱+𝐱_0+𝐱_s)`$, where $`|𝐱_s|`$ is the angular separation. The visibility modulus corresponding to this function at $`𝐮=(u,v)`$ is, therefore:
$`|\widehat{O}(𝐮)|=\sqrt{(ab)^2+4ab\mathrm{cos}^2(2\pi 𝐮.𝐱_s)}.`$ (5)
It is then useful to use a technique called “super-synthesis”: the $`(u,v)`$ plane is swept during an observation lasting several hours, due to Earth rotation. If we note $`\mathrm{B}_{EW}^{}`$ and $`\mathrm{B}_{NS}^{}`$ the orthogonal East-West and North-South components of the baseline vector at the ground of an interferometer located at the terrestrial latitude $`\theta _l`$, the $`(u,v)`$ point sampled from a star of declination $`\delta _{}`$, when its hour angle is $`H`$, is given by:
$`\{\begin{array}{cc}\hfill u=& (\mathrm{B}_{EW}^{}\mathrm{cos}H\mathrm{B}_{NS}^{}\mathrm{sin}\theta _l\mathrm{sin}H)/\lambda \hfill \\ \hfill v=& (\mathrm{B}_{EW}^{}\mathrm{sin}\delta _{}\mathrm{sin}H+\mathrm{B}_{NS}^{}\left(\mathrm{sin}\theta _l\mathrm{sin}\delta _{}\mathrm{cos}H+\mathrm{cos}\theta _l\mathrm{cos}\delta _{}\right))/\lambda \hfill \end{array}`$ (6)
After a large variation of $`H`$, several visibility moduli are therefore measured at different $`(u,v)`$ points and allow to determine the parameters ($`a`$,$`b`$ and $`𝐱_s`$) of the system by fitting the function described by Eq. 5.
Reconstruction of more complex images involves the knowledge of complex visibilities. The phase of a visibility may be deduced from closure-phase terms (Jennison, 1958) using three telescopes (i.e., wrapped sums of the phases of the three visibilities, including instrumental and atmospheric biases, sampled by the network at a given configuration). Closure-phase in optical interferometry has already been used (Baldwin et al., 1998, Hummel, 1998) for bright objects like the Capella binary system. Process after data acquisition consists of phase calibration and visibility phase reconstruction from closure-phase terms by techniques similar to bispectrum processing. From complex visibilities acquired from an array with a large number of aperture, it is possible to reconstruct the image by actually interpolating the function in the $`(u,v)`$ plane. This has been done in radio-interferometry for a few decades. The most popular algorithm for reconstructing image from $`(u,v)`$ plane samples is named CLEAN (Högbom, 1974). CLEAN subtracts iteratively, from the image given by the inverse Fourier Transform of the visibilities measured on the object (the “dirty map”), a fraction of the image given by the array from an unresolved source (the “dirty beam”) centered on the maximum value of the dirty map. However, this simple algorithm is subject to instability problems leading sometimes to wrong results. More robust versions of CLEAN have been designed (Cornwell, 1983, Dwarakanath et al., 1990). Among alternative algorithms for image reconstruction in aperture synthesis are maximum entropy method (MEM) and WIPE (Lannes et al., 1997). It can be proved (Maréchal et al., 1997) that both MEM and WIPE derivate from a common principle known as “maximum entropy on the mean”.
6. Space-borne interferometers
As classical astronomy at visible and infrared wavelengths already did it (NASA’s Hubble Space Telescope, ESA’s Infrared Space Observatory), long-baseline optical and infrared interferometry will, in the next years, take advantage of observing from outer space (absence of atmospheric turbulence, observation possible at any wavelength and for long periods, easy cooling of optics and detectors). However, the first projects for a space interferometer (FLUTE, TRIO) were proposed about two decades ago (Labeyrie et al., 1980, 1982a, 1982b). The main difficulty was to develop a technology featuring high-precision positioning as well as toughness required for space operation. Size and weight issues must also be addressed, depending on the chosen orbit. For example, if one wish to reach the Lagrangian point 2 (where the Sun and Earth gravitations are equal, stabilizing, therefore, any space-borne instrument), then the maximum payload mass of an Ariane V european launcher is 4998 kg.
6.1. Astrometry from space
The accurate determination of star angular positions will provide crucial data for astrophysics. For example, precise parallax distances of cepheids will help to establish a period/absolute magnitude relationship in order to calibrate distances of galaxies, thus reducing the uncertainty on the value of $`H_0`$. For some galaxies, distance from Earth could be computed by tracking the stars of the lower and upper sides of the observed galaxies, yielding its apparent transverse velocity $`v_t=v_0/D`$, where $`v_0`$ is the actual edge velocity and $`D`$ the distance from Earth. Using spectroscopic ground-based measures giving the radial velocity $`v_r=v_0.\mathrm{sin}i`$ (where $`i`$ is the disk inclination), one can deduce $`D`$:
$`D={\displaystyle \frac{v_r}{v_t\mathrm{sin}i}}.`$ (7)
The quest for extra-solar planets (exo-planets) is another challenge for high-precision astrometry. A planet orbiting around a star causes a revolution of the star around the center of gravity defined by the two masses. Like galaxy velocity, this periodical short motion known as “wobble” has a radial counterpart measurable from ground by spectrometry. Thus, if Doppler-Fizeau effect measurements have already led to detect from Earth jovian planets around stars, like for example 51-Peg (Mayor and Queloz, 1995) or 47-Uma (Marcy and Butler, 1996), smaller planets might be detected by measuring the stellar photocenter motion due to the wobble.
Very valuable astrometry results from space have already been obtained by the Hipparcos satellite (Perryman, 1989). Hipparcos used phase shift measurement of the temporal evolution of the photometric level of two stars seen drifting through a grid. The successor of Hipparcos, Gaia (Lindengren and Perryman, 1996), will probably use the same technique with improvements, yielding more accurate results on a larger number of objects. However, only space-borne interferometers will achieve very high precision angular measurements.
For an interferometer consisting of two apertures separated by a baseline B, the external optical delay $`d`$, while an object with altitude $`\theta `$ is observed in a broad spectral range (i.e. white light), is:
$`d=|𝐁|\times \mathrm{cos}\theta .`$ (8)
This delay can be deduced from the position of the optical delay-line of the instrument set up such that the central fringe of the interference pattern appears in a narrow observation window. The position, as well as $`|𝐁|`$, are measured by laser metrology. Hence, $`\theta `$ is deduced with a high precision. For a space-borne interferometer, the issue is to find a reference for the angle measured. Usually, a grid of far objects like quasars are used as a reference frame. Then, two modes of observation are possible: the “wide-angle” and the “narrow-angle” modes. In wide-angle mode, the large angle difference between the reference and the studied object usually requires collector motions. In narrow-angle, the two objects are in the field of view of the instrument, therefore, no motions are required and the accuracy of the measurement is improved. However, it is difficult to have always a correct reference star within the field of view for any studied object. Narrow-angle astrometry is, therefore, more suitable for wobble characterization. Figure 8 depicts the principle of an interferometer for astrometry.
Figure 8. Principle of an interferometer for astrometry.
6.2. First space-borne interferometers
The expected pioneer of this new generation of scientific spacecrafts is “Space Technology 3”, or ST3 (Gorham et al., 1999), formerly known as “Deep Space 3”. This NASA’s mission, scheduled for 2003, consists of two independent free-flying elements launched into an Earth-trailing heliocentric orbit. One is a collector sending light from the observed object to the second element featuring another collector, an optical delay-line and a beam recombiner. The aim of ST3 is the demonstration and validation of technologies that might be used for future space-borne interferometers like SIM or TPF (see further). Thus, the two elements of ST3 should be able to move up to 1 km from each other, thanks to ionic micro-engines, while being controlled by a laser metrology. However, the designed delay-line of ST3 can delay up to 20 m of optical pathlength only. A 200 m maximum projected baseline would, therefore, be possible with 1 km spacecraft separation. More than just a technology experiment, ST3 will be used as an imaging interferometer for studying Wolf-Rayet or Be stars (Linfield, 1999).
After ST3, SIM (Space Interferometry Mission) will be the next space-borne interferometer built and launched by NASA (Unwin et al., 1998). The main goal of SIM will be a collect of new high-precision astrometry results (see above), including the possibility of jovian planet detection around stars up to 1 kilo-parsec distant and terrestrial planet detection around nearby stars. The final design of SIM, known as “SIM Classic” has recently been decided (Unwin, 1999). It consists of one free-flyer with a 10 m boom supporting 30 cm collectors. The expected angular accuracy is 1 $`\mu `$as in narrow-angle mode (with a 1 field of view) and 4 $`\mu `$as in wide-angle mode. The sensitivity for astrometry is $`m_V=20`$ after four hour integration. SIM will work in the visible spectrum (0.4 to 0.9 $`\mu `$m). In order to get an accurate knowledge of the baseline vector B for wide-angle astrometry without collector motions, SIM will feature two auxiliary interferometers, aimed at reference stars (“grid-locking”). The schematic design of the SIM is depicted in figure 9.
Figure 9. The schematic design of the SIM (Courtesy: NASA/JPL/Caltech).
Besides its abilities for astrometry, SIM will feature a nulling mode. Like coronography, the nulling is a technique used for masking a bright central object in order to reveal its fainter environment (Bracewell and Macphie, 1979). Basically, a nulling system is an interferometer with a $`\pi `$ phase shift introduced in one beam. Therefore, the central fringe of the interference pattern is dark, allowing the fringe pattern from a faint object to appear. The quality of a nulling is defined by the “null depth” $`N`$:
$`N=(1V\mathrm{cos}\phi _e)/2(\pi \sigma _{\mathrm{OPD}}/\lambda )^2.`$ (9)
Where $`V`$ is the fringe visibility modulus, $`\phi _e`$ the phase error between the two recombined beams and $`\sigma _{\mathrm{OPD}}`$ the standard deviation of the optical path difference between the two beams. Nulling at $`N=25,000`$ has been obtained with laser light in laboratory (Serabyn, 1999). The nulling system of SIM is expected to reach $`N=10,000`$ with white light.
SIM should be launched in 2005 into an Earth-trailing heliocentric orbit and be operational from 2006 for a five-year mission.
6.3. Searching for life on other planets
The old, somehow philosophical, question “Is Earth the only planet sheltering life or not?”, might be answered in the next decades, thanks to interferometry. The knowledge of the chemical composition of any planetary atmosphere gives hints about the likeliness to find carbon-based life, as we know it, on this planet. Lovelock (1965) has suggested that the simultaneous presence on Earth of a highly oxidized gas, like $`\mathrm{O}_2`$, and a highly reduced gas, like $`\mathrm{CH}_4`$ and $`\mathrm{N}_2\mathrm{O}`$ is the result of the biochemical activity. However, finding spectral signatures of these gases on an exo-planet would be very difficult. An alternative life indicator would be ozone ($`\mathrm{O}_3`$), detectable as an absorption line at 9.6 $`\mu `$m. On Earth, ozone is photochemically produced from $`\mathrm{O}_2`$ and, as a component of the stratosphere, is not masked by other gases. Finding ozone would, therefore, indicate a significant quantity of $`\mathrm{O}_2`$ that should have likely been produced by photosynthesis (Léger et al., 1993). Moreover, for a star like the Sun, detecting ozone can be done 1000 times faster than detecting $`\mathrm{O}_2`$ at 0.76 $`\mu `$m: estimates made by Angel and Woolf (1997) show that the requirements for planet detection in the visible with an 8 m telescope are not matchable with current technology.
One of the imagined instruments for ozone search on exo-planets is “Darwin” (Penny et al., 1998), a.k.a. IRSI (InfraRed Space Interferometer), a project selected by ESA as a “cornerstone mission”. The aim is the discovery and characterization of terrestrial planet systems around nearby stars (closer than 15 pc) by direct detection (i.e. involving the detection of photons from the planet and not from the star as it is done with Doppler-Fizeau effect detection or wobble detection). The design of this instrument has not been established yet, but some features will likely be found in the final version of Darwin.
Basically, Darwin will have to overcome two major difficulties for achieving Earth-like planet detection. The first one concerns stellar light quenching. Interferometric nulling techniques will obviously be employed to address this issue. Severe requirements about the optical quality of the nulling device might involve spatial filtering (Ollivier and Mariotti, 1997), by pinhole or single-mode fibers, to smooth the beam wavefronts. The second difficulty is the expected presence of exo-zodiacal light (infrared emission from the dust surrounding the observed star).
Several solutions for the instrument design have, therefore, been imagined. The first one consists of five free-flying collectors (1 to 2 m telescopes) rotating around a central recombiner. The image of the exo-planetary system is constructed after a 2$`\pi `$ rotation of the system. The odd number of apertures enables a recovery of the signal from a planet (which is an asymmetrical object from the axis of rotation defined by the star) drowned in the signal from the exo-zodiacal disk (which is a symmetrical object around the star). An alternative configuration, recently imagined, consists of six collectors arranged to form a triangle (Mariotti and Menesson, 1998). In this configuration, no rotation of the system is required. Darwin is expected to be launched in 2009 into an orbit at 5 AU from the sun, in order to reduce the illumination by solar zodiacal light.
A project named TPF (Terrestrial Planet Finder), very similar to Darwin/IRSI is currently studied by NASA (Beichman, 1998). Like Darwin, the final design has not been decided yet. The current version (Lawson et al., 1999) features four aligned 3.5 m free-flying telescopes and a central recombiner. The baseline from the two most separated telescopes can span from 75 m to 1 km. Like the original design of Darwin, the collectors of TPF will rotate around the recombiner for planet detection. In this case the maximum baseline is 135 m. For planet imaging, telescopes will move along parallel straight lines and could be separated by 1 km. Instrumentation for spectroscopy on TPF will include a $`R=30`$ spectrometer working between 7 $`\mu `$m and 20 $`\mu `$m for planet detection, and a $`R=300`$ spectrometer working between 3 $`\mu `$m and 30 $`\mu `$m for imaging. The expected time to find a planet and then to determine whether ozone is present in its atmosphere should be 15 days for each star zeroed in on. The launch of TPF is expected in 2010. Its five-year mission will start one year later. Instead of a far orbit location as Darwin is supposed to reach, TPF will be placed on a Earth-trailing orbit or at the Lagrangian point 2. Despite the problem of solar zodiacal light (that is expected to be overcome by using telescopes larger than Darwin’s), such orbits provide easier radio-transmissions, a larger available solar power and a heavier payload possible for the launcher. Figure 10 depicts the conceptual design of TPF
Figure 10. The conceptual design of the TPF (Courtesy: NASA/JPL/Caltech).
6.4. Long-term perspective
Space-borne interferometry projects for years spanning from 2020 to 2050 already exist. However, the reader should be aware that such projects must be regarded as drafts for future instruments. No one can forecast, today, how future space-borne interferometers will actually look like.
For the post-TPF era, NASA has imagined an enhanced version featuring four 25 m telescopes and a $`R1000`$ spectrometer. This interferometer would be able to detect on an exo-planet lines of gases directly produced by biochemical activity. The next step proposed by NASA is an array of 25 telescopes, 40 m diameter each, that would yield 25 $`\times `$ 25 pixel images of an Earth-like planet at 10 pc, revealing its geography and eventually oceans or chlorophyll zones.
A comparable project has been proposed by Labeyrie (1999). It consists of 150 telescopes, 3 m diameter each, forming an interferometer with a 150 km maximum baseline. Such an instrument would give a 40 $`\times `$ 40 pixel image of an Earth-like planet at 3 pc, providing the same information as described previously.
8. Conclusions
The angular resolution of any stellar object in the visible wavelength can vastly be improved by using long baseline interferometry. The angular diameter for more than 50 stars have been measured (DiBenedetto and Rabbia, 1987, Mozurkewich et al., 1991, Dyck et al., 1993) with accuracy better than 1% in some cases with the ground-based long baseline amplitude interferometers at optical and IR wavelengths. Apart from the measurements of the diameters, distances, masses on the stellar surfaces, among others, this technique can detect the morphological details, viz., (i) spots and flares, (ii) granulations, (ii) oblateness etc., of giant stars. Eclipsing binaries (Algol type) which show evidence of detached gas rings around the primary are also good candidates for long baseline interferometry. The potential of this type of interferometer can be envisaged in determining the fundamental astrophysical informations of circumstellar envelopes such as, the diameter of inner envelope, colour, symmetry, radial profile etc. The objectives of very large array in optical/IR wave bands range from detecting other planetary systems to imaging the black hole driven central engines of quasars and active galaxies.
Several long baseline interferometers are either in operation or under development at various stages. Rapid increase in the scientific output at optical, as well as at infrared wave bands using these interferometers can be foreseen at the begining of this millennium. With improved technology, the long baseline interferometric arrays of large telescopes fitted with high level adaptive optics system that applies dark speckle coronograph (Boccaletti et al., 1998) may provide snap-shot images at their recombined focus using the concept of densified-pupil imaging (Pedretti and Labeyrie, 1999), and yield improved images and spectra of objects. One of the key areas where the new technology would make significant contributions is the astrometric detection and characterization of exo-planets.
However, the role of smaller interferometers should not be neglected, since such instruments could be useful for long observations of binary systems, testing new focal instrumentation designed for larger interferometers, or observing at short wavelengths (blue, UV). Moreover, this class of interferometers would be easily accessible to a broader community of astronomers and might be employed as education tools.
On the other hand, space-borne LBIs would provide the best ever spatial resolution of faint objects as the fringes can be recorded with longer integration time. The greatest advantage of such a project is being the absence of atmospheric turbulence. The bright prospects of LBI programmes can be witnessed from the future space interferometers like SIM, ST3 and Darwin which are effectively funded projects that will become an essential tool at the cutting edge of astronomical research in the new millennium.
Acknowledgments: The authors thank Dr P. R. Lawson at Jet Propulsion Laboratory, USA and Dr O. Lardière at Observatoire de Haute Provence, France, for photographs used in this paper. The service rendered by Mr. B. A. Varghese, Indian Institute of Astrophysics, Bangalore, India is also gratefully acknowledged. One of the authors Dr S. Morel is grateful to DGA-DRET (the scientific research office of the French Ministry of Defense) for having funded his post-doctoral stay at IOTA, and to JPL for its useful documentation and lectures.
References
Anderson J. A., 1920, Ap. J., 51, 263.
Angel R., Woolf N. J., 1997, Ap J, 475, 373.
Armstrong J. T., Hummel C. A., Mozurkewich D., 1992a, Proc. ESO-NOAO conf. ‘High Resolution Imaging Interferometry’, eds., J. M. Beckers & F. Merkle, Garching bei München, Germany, 673.
Armstrong J. T., Mozurkewich D., Pauls T. A., Hajian A. R., 1998, Proc. SPIE., conf. ‘Astronomical Interferometry’, 3350, 461.
Armstrong J. T., Mozurkewich D., Rickard L. J., Hutter D. J., Benson J. A., Bowers P. F., Elias N. M., Hummel C. A., Johnston K. J., Buscher D. F., Clark J. H., Ha L., Ling L. -C., White N. M., Simon R. S., 1998, Ap J, 496, 550.
Armstrong J. T., Mozurkewich D., Vivekanand M., Simon R. S., Denison C. S., Johnston K. J., Pan X. -P., Shao M., Colavita M. M., 1992b, A J, 104, 241.
Arnold L., Labeyrie A., Mourard D., 1996, Adv. Space Res., 18, 1149.
Arnold L., Lardière O., Dejonghe J., 2000, Proc. SPIE, conf. ‘Interferometry in Optical Astronomy’, 4006, (in preparation).
Ayers G. R., Dainty J. C., 1988, Opt. Lett., 13, 457.
Babcock H. W., 1953, PASP, 65, 229.
Baldwin J. E., Beckett R. C., Boysen R. C., Burns D., Buscher D. F., Cox G. C., Haniff C. A., Mackay C. D., Nightingale N. S., Rogers J., Scheuer P. A. G., Scott T. R., Tuthill P. G., Warner P. J., Wilson D. M. A., Wilson R. W., 1996, A & A, 306, L13.
Baldwin J. E., Boysen R. C., Haniff C. A., Lawson P. R., Mackay C. D., Rogers J., St-Jacques D., Warner P. J., Wilson D. M. A., Young J. S., 1998, Proc. SPIE., conf. ‘Astronomical Interferometry’, 3350, 736.
Baldwin J. E., Haniff C. A., Mackay C. D., Warner P. J., 1986, Nature, 320, 595.
Bedding T. R., 1999, astro-ph/9901225, PASP (to appear).
Bedding T. R., Robertson J. G., Marson R. G., 1994, A & A, 290, 340.
Bedding T. R., Robertson J. G., Marson R. G., Gillingham P. R., Frater R. H., O’Sullivan J. D., 1992, Proc. ESO-NOAO, conf. ‘High Resolution Imaging Interferometry’, eds. J. M. Beckers & F. Merkle, Garching bei München, Germany, 391.
Beichman C. A., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 719.
Benson J. A., Mozurkewich D., Jefferies S. M., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 493.
Bester M., Danchi W. C., Degiacomi C. G., Townes C. H., 1991, Ap J, 367, L27.
Boccaletti A., Moutou C., Labeyrie A., Kohler D., Vakili F., 1998, A & A, 340, 629.
Boden A. F., Koresko C. D., Van Belle G. T., Colavita M. M., Dumont P. J., Gubler J., Kulkarni S. R., Lane B. F., Mobley D., Shao M., Wallace J. K., 1999, Ap J, 515, 356.
Bosc I., 1988, Proc. ESO-NOAO, conf. ‘High-Resolution Imaging by Interferometry’, ed. F. Merkle, Garching bei München, Germany, 735.
Bracewell R. N., Macphie R. H., 1997, Icarus, 38, 136.
Brown R. H., 1974, ‘The Intensity Interferometry, its Applications to Astronomy’, Taylor & Francis, London.
Brown R. H. and Twiss R. Q., 1958, Proc. Roy. Soc. A., 248, 222.
Brown R. H., Davis J. and Allen L. R., 1967, MNRAS, 137, 375.
Brown R. H., Jennison R. C. and Das Gupta M. K., 1952, Nature, 170, 1061.
Burns D., Baldwin J. E., Boysen R. C., Haniff C. A., Lawson P. R., Mackay C. D., Rogers J., Scott T. R., Warner P. J., Wilson D. M. A., Young J. S., 1997, MNRAS, 290, L11.
Burns D., Baldwin J. E., Boysen R. C., Haniff C. A., Lawson P. R., Mackay C. D., Rogers J., Scott T. R., St-Jacques D., Warner P. J., Wilson D. M. A., Young J. S., 1998, MNRAS, 297, 467.
Busher D. F., Haniff C. A., Baldwin J. E., Warner P. J., 1990, MNRAS., 245, 7.
Butler P. R., Marcy G. W., 1996, Ap J, 464, L153.
Colavita M. M., Boden A. F., Crawford S. L., Meinel A. B., Shao M., Swanson P. N., Van Belle G. T., Vasist G., Walker J. M., Wallace J. P., Wizinowich P. L., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 776.
Colavita M. M., Wallace J. K., Hines B. E., Gursel Y., Malbet F., Palmer D. L., Pan X. P., Shao M., Yu J. W., Boden A. F., Dumont P. J., Gubler J., Koresko C. D., Kulkarni S. R., Lane B. F., Mobley D. W., Van Belle G. T., 1999, A J, 510, 505.
Cornwell T. J., 1983, A & A, 121, 281.
Coudé du Foresto V., Ridgway S.T., 1992, Proc. ESO-NOAO, conf. ‘High Resolution Imaging Interferometry’, eds. J. M. Beckers and F. Merkle, Garching bei München, Germany, 731.
Davis J., Tango W. J., Booth A. J., Minard R. A., Brummelaar t. T. A., Shobbrook R. R., 1992, Proc. ESO-NOAO, conf. ‘High Resolution Imaging Interferometry’, eds. J. M. Beckers and F. Merkle, Garching bei München, Germany, 741.
Davis J., Tango W. J., Booth A. J., O’Byrne J. W., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 726.
Davis J., Tango W. J., Booth A. J., Thorvaldson E. D., Giovannis J., 1999, MNRAS, 303, 783.
Dejonghe J., Arnold L., Lardière O., Berger J.-P., Cazalé C., Dutertre S., Kohler D., Vernet D., 1998, Proc. SPIE, conf. ‘Advanced Technology Optical/IR Telescopes’, 3352, 603.
Derie F., Ferrari M., Brunetto E., Duchateau M., Amestica R., Aniol P., 2000, Proc. SPIE, conf. ‘Interferometry in Optical Astronomy’, 4006, (in preparation).
DiBenedetto G. P., Conti G., 1983, Ap J, 268, 309.
DiBenedetto G. P., Rabbia Y., 1987, A & A, 188, 114.
Dwarakanath K. S., Deshpande, A. A., Udaya Shankar N., 1990, J. Astr. Astron, 11, 311.
Dyck H. M., Benson J. A., Carleton N. P., Coldwell C. M., Lacasse M. G., Nisenson P., Panasyuk A. V., Papaliolios C. D., Pearlman M. R., Reasenberg R. D., Traub W. A., Xu X., Predmore R., Schloerb F. P., Gibson D., 1995, A J, 109, 378.
Dyck H. M., Benson J. A., Ridgway S. T., 1993, PASP, 105, 610.
Dyck H. M., Benson J. A., Van Belle G. T., Ridgway S. T., 1996b, A J, 111, 1705.
Dyck H. M., Van Belle G. T., Benson J. A., 1996a, A J, 112, 294.
Dyck H. M., Van Belle G. T., Thomson R. R., 1998, A J, (to appear).
Falcke H., Davidson K., Hofmann K. -H., Weigelt G., 1996, A & A, 306, L17.
Faucherre M., Bonneau D., Koechlin L., Vakili F., 1983, A & A, 120, 263.
Fizeau H., 1868, C. R. Acad. Sci. Paris, 66, 934.
Fried D. C., 1966, J. Opt. Soc. Am., 56, 1972.
Goodman J. W., 1968, Introduction to Fourier optics, McGraw Hill Book Co., New-York.
Gorham P. W., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 116.
Gorham P. W., Folkner W. M., Blackwood G. H., 1999, conf. ‘Working on the fringe’, Dana Point, USA, to be published in ASP Conference Series, eds. S. Unwin and R. Stachnik.
Grieger F., Weigelt G., 1992, Proc. ESO-NOAO, conf. ‘High Resolution Imaging Interferometry’, eds. J. M. Beckers and F. Merkle, Garching bei München, Germany, 481.
Hajian A. R., Armstrong J. T., Hummel C. A., Benson J. A., Mozurkevich D., Pauls T. A., Hutter D. J., Elias N. M., Johnston K. J., Rickard L. J., White N. M., 1998, Ap J, 496, 484.
Haniff C. A., Busher D. F., Christou J. C., Ridgway S. T., 1989, MNRAS, 241, 694.
Haniff C. A., Mackay C. D., Titterington D. J., Sivia D., Baldwin J. E., Warner P. J., 1987, Nature, 328, 694.
Harmanec P., Morand F., Bonneau D., Jiang Y., Yang S., Guinan E. P., Hall D. S., Mourard D., Hadrava P., Bozic H., Sterken C., Tallon-Bosc I., Walker G. A. B., McCook P. M., Vakili F., Stee P., 1996, A & A, 312, 879.
Hestroffer D., 1997, A & A, 327, 199.
The Hipparcos catalogue, 1997, ESA, SP-1200.
Högbom J. A., 1974, A & AS, 15, 417.
Hummel C. A., 1994, IAU Symp. 158, ‘Very high resolution imaging’ ed., J. G. Robertson and W. J. Tango, 448.
Hummel C. A., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 483.
Hummel C. A., Mozurkevich D., Armstrong J. T., Hajian A. R., Elias N. M., Hutter D. J., 1998, A J, 116, 2536. Jennison R. C., 1958, MNRAS, 118, 276.
Kervella P., Traub W. A., Lacasse M. G., 1999, conf. ‘Working on the Fringe’, Dana Point, USA, to be published in ASP Conference Series, eds. S. Unwin and R. Stachnik.
Kervella P., Coudé du Foresto V., Glindemann A., 2000, Proc. SPIE, conf. ‘Astronomical Interferometry’, 4006 (in preparation).
Knox K. T., Thompson B. J., 1974, Ap J, 193, L45.
Koechlin L., Lawson P. R., Mourard D., Blazit A., Bonneau D., Morand F., Stee P., Tallon-Bosc I., Vakili F., 1996, Appl. Opt., 35, 3002.
Labeyrie A., 1970, A & A., 6, 85.
Labeyrie A., 1975, Ap. J., 196, L71.
Labeyrie A., 1978, Ann. Rev. A & A., 16, 77.
Labeyrie A., 1985, 15th. Advanced Course, Swiss Society of Astrophys. and Astron. ed.. A. Benz, M. Huber and M. Mayor, 170.
Labeyrie A., 1995, A & A, 298, 544.
Labeyrie A., 1996, A & AS, 118, 517.
Labeyrie A., 1998a, conf. ’Extrasolar planets: formation, detection and modeling’, Lisbon, Portugal (to appear).
Labeyrie A., 1998b, Proc. NATO-ASI, conf.‘Planets outside the solar system’, Cargèse, Corsica - France.
Labeyrie A., 1998c, Proc. SPIE, conf. ‘Astronomical interferometry’, 3350, 960.
Labeyrie A., 1999, conf. ‘Working on the fringe’, Dana Point, USA, to be published in ASP Conference Series, eds. S. Unwin and R. Stachnik.
Labeyrie A., Kibblewhite J., de Graauw T., Roussel Ph., Noordam J., Weigelt G., 1982b, Proc. CNES, conf. ‘Very long baseline interferometry’, 477.
Labeyrie A., Praderie F., Steinberg J., Vatoux S., Wouters F., 1980, Proc. KPNO, conf. ‘Optical and infrared telescopes for the 1990’s’, ed. A. Hewitt, 1020.
Labeyrie A., Schumacher G., Savaria E., 1982a, Adv. Space Res., 2, 11.
Labeyrie A., Schumacher G., Dugué M., Thom C., Bourlon P., Foy F., Bonneau D. and Foy R., 1986, A & A., 162, 359.
Lannes A., Anterrieu E., Maréchal P., 1997, A & AS, 123, 183.
Lardière O., Arnold L., Berger J.-P., Cazalé C., Dejonghe J., Labeyrie A., Mourard D., 1998, Proc. SPIE, conf. ‘Telescope Control Systems’, 3351, 107.
Lawson P. R., 1994, PASP, 106, 917.
Lawson P. R., 1995, J. Opt. Soc. Am A, 12, 366.
Lawson P. R., Baldwin J. E., Warner P. J., Boysen R. C., Haniff C. A., Rogers J., Saint-Jacques D., Wilson D. M. A., Young J. S., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 753.
Lawson P. R., Dumont P. J., Colavita M. M., 1999, AAS Meeting 194.
Léger A., Pirre M., Marceau F.J., 1993, A & A., 277, 309 (1993).
Leinert C., Graser U., 1998, Proc. SPIE, conf. ‘Astronomical interferometry’, 3350, 389.
Léna P., 1997, Experimental Astr., 7, 281.
Léna P., Lai O., 1999a, ‘Adaptive Optics in Astronomy’, ed. F. Roddier, Cambridge Univ. Press, 351.
Léna P., Lai O., 1999b, ‘Adaptive Optics in Astronomy’, ed. F. Roddier, Cambridge Univ. Press, 371.
Lindengren L., Perryman M. A. C., 1996, A & AS, 116, 579.
Linfield R., Gorham P. W., 1999, conf. ‘Working on the fringe’, Dana Point, USA, to be published in ASP Conference Series, eds. S. Unwin and R. Stachnik.
Liu Y. C., Lohmann A. W., 1973, Opt. Comm., 8, 372.
Lohmann A. W., Weigelt G. P., Wirnitzer B., 1983, App. Opt., 22, 4028.
Lovelock J.E., 1965, Nature, 207, 568.
Lynds C. R., Worden S. P., Harvey J. W., 1976, Ap J, 207, 174.
Machida Y., Nishikawa J., Sato K., Fukushima T., Yoshizawa M., Honma Y., Torii Y., Matsuda K., Kubo K., Ohashi M., Suzuki S., Iwashita H., Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 202.
Malbet F., Berger J. -P., Colavita M. M., Koresko C. D., Beichman C., Boden A. F., Kulkarni S. R., Lane B. F., Mobley D. W., Pan X. -P., Shao M., van Belle G. T., Wallace J. K., 1998, astro-ph/9808326, Ap. JL. (accepted).
Maréchal P., Anterrieu E., Lannes A., 1997, ASP Conf., 125, 158.
Mariotti J. -M., Menesson B., 1998, Internal ESA report.
Mayor M., Queloz D., Nature, 1995, 378, 355.
McAlister H. A., Bagnuolo W. G., ten Brummelaar, Hartkopf W. I., Shure M. A., Sturmann L., Turner N. H., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 947.
McAlister H. A., Bagnuolo W. G., ten Brummelaar, Hartkopf W. I., Turner N. H., Garrison A. K., Robinson W. G., Ridgway S. T., 1994, Proc. SPIE, 2200, 129.
Mennesson, B., Mariotti J. -M., Coudé du Foresto V., Perrin, G., Ridgway S. T., Ruilier C., Traub, W. A., Carleton N. P., Lacasse, M. G., Mazé G., 1999, A & A, 346, 181.
Mennesson B., Perrin G., Chagnon G., Coudé du Foresto V., Morel S., Ruilier C., Traub W. A., Carleton N. P., Lacasse M. G., 2000, Proc. SPIE, conf. ‘Interferometry in Optical Astronomy’, 4006 (in preparation).
Michelson A. A., 1891, Nature 45, 160.
Michelson A. A., 1920, Ap. J., 51, 257.
Michelson A. A., and Pease F. G., 1921, Ap. J., 53, 249.
Millan-Gabet R., Schloerb P. F., Traub W. A., Carleton N. P., 1999, PASP, 111, 238.
Millan-Gabet R., Schloerb P. F., Traub W. A., 1998, AAS Meeting 193.
Morel S., Koechlin L., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 1057.
Morel S., Traub W. A., Bregman J. D., Mah R., Wilson E., 2000, Proc. SPIE, conf. ‘Interferometry in Optical Astronomy’, 4006 (in preparation).
Mourard D., Bonneau D., Koechlin L., Labeyrie A., Morand F., Stee P., Tallon-Bosc I., Vakili F., 1997, A & A, 317, 789.
Mourard D., Bosc I., Labeyrie A., Koechlin A., Saha S., 1989, Nature, 342, 520.
Mourard D., Thureau N., Antonelli P., Bério P., Blanc, J.-C., Blazit A., Boit J.-L., Bonneau D., Chesneau O., Clausse, J.-M., Corneloup, J.-M., Dalla R., Dugué M., Glentzlin A., Hill L., Labeyrie A., Lemerrer J., Menardi S., Merlin G., Moreaux, G., Petrov R., Rebattu S., Rousselet-Perraut K., Stee P., Tallon-Bosc I., Trastour J., Vakili F.; Vérinaud C., Voet C., Waultier G., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 517.
Mozurkewich D., Johnston K. J., Simon R., Hutter D. J., Colavita M. M., Shao M., Pan X.-P., 1991, A J, 101, 2207.
Nakajima T., Kulkarni S. R., Gorham P. W., Ghez A. M., Neugebauer G., Oke J. B., Prince T. A., Readhead A. C. S., 1989, A J, 97, 1510.
Ollivier M., Mariotti J.-M., 1997, Appl. Opt., 36, 5340.
Padilla C. E., Karlov V. I., Matson L. K., Soosaar K., Brummelaar T. ten, 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 1045.
Pan X.-P., Shao M., Colavita M. M., 1992, IAU Colloq. 135., ASP Conf. Proc. 32, ‘Complementary Approaches to Double and Multiple Star Research’, eds. H. A. McAlister and W. I. Hartkopf, 502.
Pan X.-P., Kulkarni S. R., Colavita M. M., Shao M., 1996, Bull. Am. Astron. Soc., 28, 1312. Pauls T. A., Mozurkewich D., Armstrong J. T., Hummel C. A., Benson J. A., Hajian A. R., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 467.
Pedretti E., Labeyrie A., 1999, A & AS, 137, 543.
Penny A. J., Léger A., Mariotti J. -M., Schalinski C., Eiora C., Laurance R., Fridlund M., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 666.
Perrin G., 1997, A & AS, 121, 553.
Perrin G., Coudé du Foresto V., Ridgway S. T., Mariotti J.-M., Traub W. A., Carleton N. P., Lacasse M. G., 1998, A & A, 331, 619.
Perrin G., Coudé du Foresto V., Ridgway S. T., Menesson B., Ruilier C., Mariotti J -M., Traub W. A., Lacasse M. G., 1999, A & A, 345, 221.
Perryman M. A. C., 1998, Nature, 340, 111.
Petrov R., Roddier F., Aime C., 1986, J. Opt. Soc. Am. A, 3, 634.
Petrov R., Malbet F., Richichi A., Hofmann K. H., Agabi K., Antonelli P., Aristidi E., Baffa C., Beckmann U., Bério P., Bresson Y., Cassaing F., Chelli A., Dress A., Dugué M., Duvert G., Forveille T., Fossat E., Gennari S., Geng M., Glentzlin A., Kamm D., Lagarde S., Lecoarer E., Le Contel J.-M., Lisi F., Lopez B., Mars G., Martinot-Lagarde G., Monin J., Mouillet D., Mourard D., Rousselet-Perraut K., Perrier-Bellet C., Puget P., Rabbia Y., Rebattu S., Reynaud F., Robbe-Dubois S., Sacchettini M., I. Tallon-Bosc, Weigelt G., 2000, Proc. SPIE, conf. ‘Interferometry in Optical Astronomy’, 4006 (in preparation).
Quirrenbach A., Coudé du Foresto V., Daigne G., Hofmann K. H., Hofmann R., Lattanzi M., Osterbart R., Le Poole R. S., Queloz D., Vakili F., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 807.
Rabbia Y., Mekarnia D., Gay J., 1990, Proc. SPIE, conf. ‘Infrared Technology’, 1341, 172.
Rhodes W. T., Goodman J. W., 1973, J. Opt. Soc. Am., 63, 647.
Roddier C., Roddier F., 1988, Proc. NATO-ASI, conf. ‘Diffraction Limited Imaging with Very Large Telescopes’, eds. D. M. Alloin and J. -M. Mariotti, Cargèse, Corsica - France, 221.
Rousset G., Fontanella J. C., Kem P., Gigan P., Rigaut F., Léna P., Boyer P., Jagourel P., Gaffard J. P., Merkle F., 1990, A & A, 230, L29.
Rousselet-Perraut K. , Vakili F. , Mourard D., 1996, Opt. Engin., 35, 2943.
Saha S. K., 1999a, BASI., 27, 443.
Saha S. K., 1999b, Ind. J. Phys., 73B, 552.
Sato K., Nishikawa J., Yoshizawa M., Fukushima T., Machida Y., Honma Y., Kuwabara R., Suzuki S., Torii Y., Kubo K., Matsuda K., Iwashita H., Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 212.
Schloerb F. P., Millan-Gabet, R. S., Traub, W. A., 1999, AAS Meeting 194.
Serabyn E., 1999, Appl. Opt., 38, 4213.
Shaklan S. B., Roddier F., 1987, Appl. Opt., 26, 2159.
Shao M., Colavita M. M., 1988, A & A, 193, 357.
Shao M., Colavita M. M., 1994, IAU Symp. 158, ‘Very high resolution imaging’, eds. J. G. Robertson and W. J. Tango, 413.
Shao M., Colavita M. M., Hines B. E., Hershey J. L., Hughes J. A., Hutter D. J., Kaplan G. H., Johnston K. J., Mozurkewich D., Simon R. H., Pan X. -P., 1990, A J, 100, 1701.
Shao M., Colavita M. M., Hines B. E., Staelin D. H., Hutter D. J., Johnston K. J., Mozurkewich D., Simon R. H., Hershey J. L., Hughes J. A., Kaplan G. H., 1988, A & A, 193, 357.
Shao M., Staelin D. H., 1977, J. Opt. Soc. Am., 67, 81.
Stee P., de Araújo, Vakili F., Mourard D., Arnold I., Bonneau D., Morand F., Tallon-Bosc I., 1995, A & A, 300, 219.
Stee P., Vakili F., Bonneau D., Mourard D., 1998, A & A, 332, 268.
Stéphan H., 1874, C. R. Acad. Sci. Paris, 76, 1008.
Tatarski V. I., 1967, ‘Wave Propagation in a Turbulent Medium’, Dover, N. Y.
Thom C., Granes P., Vakili F., 1986, A & A, 165, L13.
Traub W. A., Millan-Gabet R., Garcia M. R., 1998, AAS Meeting 193.
Traub W. A., Carleton N. P., Brewer M. K., Lacasse M. G., Millan-Gabet R., Morel S., Papaliolios C., Porro I., 2000, Proc. SPIE, conf. ‘Interferometry in Optical Astronomy’, 4006, (in preparation).
Unwin S. C., Turyshev S. G., Shao M., 1998, Proc. SPIE, conf. ‘Astronomical Interferometry’, 3350, 551.
Unwin S. C., 1999, Private communication.
Vakili F., Bério P., Bonneau D., Chesneau O., Mourard D., Stee P., Thureau N., 1998a, conf. ‘Be stars’, ed. A. M. Hubert and C. Jaschek, 173.
Vakili F., Mourard D., Bonneau D., Morand F., Stee P., 1997, A & A, 323, 183.
Vakili F., Mourard D., Stee P., Bonneau D., Bério P., Chesneau O., Thureau N., Morand F., Labeyrie A., Tallon-Bosc I., 1998b, A & A, 335, 261.
Van Belle G. T., Dyck H. M., Benson J. A., Lacasse M. G., 1996, A J., 112, 2147.
Van Belle G. T., Dyck H. M., Thompson R. R., Benson J. A., Kannappan S. J., 1997, A J., 114, 2150.
Van Belle G. T., Lane B. F., Thompson R. R., Boden A. F., Colavita M. M., Dumont P. J., Mobley D. W., Palmer D;, Shao M., Vasisht G. X., Wallace J. K., Creech-Eakman M. J., Koresko C. D., Kulkarni S. R., Pan X.-P., Gubler J., 1999, A J, 117, 521.
Vérinaud C., Blazit A., de Bonnevie A., Bério P., 1998, conf. ‘Catching the Perfect Wave’, eds. S. R. Restaino, W. Junor and N. Duric., 131.
Walkup J. F., Goodman J. W., 1973, J. Opt. Soc. Am., 63, 399.
Wallace J. K., Boden A. F., Colavita M. M., Dumont P. J., Gursel Y., Hines B., Koresko C. D., Kulkarni S. R., Lane B. F., Malbet F., Palmer D., Pan X. P., Shao M., Vasisht G. X., Van Belle G. T., Yu J., 1998, Proc. SPIE, conf. ‘Astronomical interferometry’, 3350, 864.
Weigelt G., 1977, Opt. Communication, 21, 55.
Weigelt G., Mourard D., Abe L., Beckmann U., Blöecker T., Chesneau O., Hillemanns C., Hoffmann K. H., Ragland S., Schertl D., Scholz M., Stee P., Thureau N., Vakili F., 2000, Proc. SPIE, conf. ‘Interferometry in Optical Astronomy’, 4006, (in preparation). |
warning/0003/cs0003020.html | ar5iv | text | # ACLP: Integrating Abduction and Constraint SolvingThis system has been developed in collaboration with A. Michael and C. Mourlas. The system can be obtained from http://www.cs.ucy.ac.cy/aclp/.
## Introduction
The ACLP framework and system is an attempt to address the problem of providing a high-level declarative programming (or modeling) enviroment for problems of Artificial Intelligence which at the same time has an acceptable computational performance. Its key elements are (i) the support of abduction as a central inference of the system, to facilitate a high-level of expressivity for problem representation, and (ii) the use of constraint solving to enhance the efficiency of the computational process of abductive inference as this is applied on the high-level representation of the problem at hand.
It has been argued in (?) that declarative problem solving, where the proplem representation contains information about properties that hold true in the problem domain rather than information on methods of how we would solve the problem, and abduction are closely related to each other. In an (ideal) declarative setting problem solving consists of filling in missing information from the theory that represents the problem. In other words, the solution consists of an extension of the basic description of the problem so that the problem task (or goal) is satisfied in this extended description. This process of extending the theory is called abduction. For example, in a logical setting, abduction as a problem solving method, assumes that the general data structure for the solution to a problem (or solution carrier) is at the predicate level and hence a solution is described in the same terms and level as the problem itself.
Indeed, abduction allows a high-level representation of problems close to their natural specification suitable for addressing a variety of problems in AI, such as diagnosis, planning and scheduling, natural language understanding, assimilation of sensor data and user modeling. The main advantage of using abduction to solve these problems is the high-level representation or modeling environment that it offers. This in turn provides a high degree of modularity and flexibility which is useful for applications with complex and changing requirements. But although the utility of abduction for formulating such problems in AI is well proven there has been little work (see though (?)) to address the question of whether these abductive formulations can form the basis for computationally effective solutions to realistic problems.
ACLP tries to address this problem by a non-trivial integration of constraint solving within the abductive process. The general pattern of computation in ACLP consists of a cooperative interleaving between hypotheses and constraint generation, via abductive inference, with constraint satisfaction of the generated constraints. Abductive reasoning provides an incremental reduction of the high-level problem representation and goals to abductive hypotheses together with lower-level constraints whose form is problem independent. The integration of abductive reasoning with constraint solving in ACLP is cooperative, in the sense that the constraint solver not only solves the final constraint store generated by the abductive reduction but also affects dynamically this abductive search for a solution. It enables abductive reductions to be pruned early by setting new suitable constraints on the abducible assumptions into the constraint store, provided that this remains satisfiable. During the ACLP computation there is a non-trivial interaction between (i) reduction of goals and consistency checking of abducible assumptions, (ii) setting new constraints in the constraint store of reduction and (iii) generating further abductive hypotheses.
## General Information
Currently, the ACLP system is implemented as a meta-interpreter on top of the CLP language of ECLiPSe. As such the system is relatively compact comprising about 500 lines of code. It is based on the abductive proof procedure developed in (?) (which in turn follows a series of proof procedures (?), (?), (?)) and uses the CLP constraint solver of ECLiPSe to handle constraints over finite domains (integer and atomic elements). The architecture of the system is quite general and can be implemented in a similar way with other constraint solvers.
The ACLP system runs on any platform on which ECLiPSe runs. This includes all major platforms. It can be obtained, together with information on how to use it, from the following web address: http://www.cs.ucy.ac.cy/aclp/. ACLP programs (see section Applying the System below) are loaded into ECLiPSe together with the ACLP system file, aclp.pl, and executed by calling the top-level ECLiPSe query:
aclp-solve(+Goal, +Initial-hypothesis, ?output-variable).
The output-variable returns a list of abducible hypotheses, with their domain variables constrained according to the dynamic constraints that were generated through the unfolding of the “relevant” part, with respect to the Goal, of the program and the integrity constraints. A subsequent step of labelling on these variables is needed to give a ground solution of our query, +Goal. Various constraint predicates of ECLiPSe can be used at this stage e.g. min\_max/2 or minimize/2 to find an optimal ground solution. If we are not interested in such further optimization we can use the simpler queries:
aclp-solve(+Goal)
aclp-solve(+Goal, +Initial-hypothesis).
The initial-hypothesis variable is a list of ground abducible facts which we want the system to take as given when constructing a solution. It is used when we have partial information about the solution that we are looking for. (If no such information is known then this is given as the empty list.) Its typical use is when we want to recompute the solution to a goal under some new requirements by adapting the old solution as for example in the case of rescheduling. The old solution (or part of this) will then form the initial-hypothesis.
## Applying the System
The ACLP system is a programming environment on top of the ECLiPSe language. An ACLP program is an abductive theory consisting of a triple $`<P,A,IC>`$ where:
* $`P`$ is a finite set of user-defined ECLiPSe clauses,
* $`A`$ is a set of declarations of abducibles predicates in the form of ECLiPSe facts as: $`abducible\mathrm{\_}predicate(predicate\mathrm{\_}name/arity)`$,
* $`IC`$ is a set of integrity constraints written as ECLiPSe rules of the form: $`ic:B_1,\mathrm{},B_n.(n1),`$ where:
+ at least one of the goals $`B_1,\mathrm{},B_n`$ has an abducible predicate, and
+ the rest of the goals can be either positive or negative literals on user-defined predicates or constraint predicates of ECLiPSe.
The (lower-level) problem independent CLP constraint predicates that can be used in the body of a program rule or an integrity constraint can be (i) arithmetic constraint predicates (over the integers) or (ii) logical contraint predicates. The constraint predicates on finite domain variables of ECLiPSe that are supported by the current ACLP implementation are:
the value of variable T1 is not equal to that of variable T2.
the value of variable T1 is equal to that of variable T2.
the value of variable T1 is less than that of variable T2.
the value of variable T1 is less or equal to that of variable T2.
the value of variable T1 is greater than that of variable T2.
the value of variable T1 is greater or equal to that of variable T2.
The equality and inequality constraints are also supported over other non-arithmetic user-defined finite domains. The system also has a term equality constraint, T1##=T2 , where the terms $`T1`$ and $`T2`$ can contain variables one level deep inside a function sympol. In addition, ACLP supports logical constraints such as conjunction, #$``$, and disjunction, #$``$. These simple constraints can be combined to build complex logical constraint expressions. During the ACLP computation constraints maybe negated and their negation is set in the current constraint store. This negation of the constraints is the usual mathematical negation, e.g. the negation of the aritmetic constraint T1#\<T2 is T1#\>=T2. The negation of the logical constraint #$``$ is #$``$.
An ACLP program, $`<P,A,IC>`$, can contain negation as failure literals in $`P`$ and $`IC`$. Negation as failure is handled through abduction simply as another type of abducible in the theory. All occurences of $`not(p)`$ in the program $`P`$ are replaced by $`not\mathrm{\_}p`$ which is treated as an abducible with the canonical integrity constraint $`ic:not\mathrm{\_}p,p.`$ In the current implementation it is necessary for the user to specify explicitly both the fact that $`not\mathrm{\_}p`$ is abducible by adding a statement $`abducible\mathrm{\_}predicate(not\mathrm{\_}p/arity)`$ in the program as well as adding the above canonical constraint in the program. The semantics of negation as failure is that of (partial) Stable Models in the program $`P`$ and Generalised (partial) Stable Models when we consider the whole abductive theory with its integrity constraints $`IC`$. The details of the abductive semantics for ACLP programs and the particular treatment of NAF can be found in (?).
As an example, the ACLP program below is an implementation of the basic axioms (of persistence) of the Event Calculus (?) suitable for abductive planning. The program $`P`$ consists of the following clauses:
```
holds_at(P,E) :- initially(P,T),
not clipped(T,E,P).
holds_at(P,E) :- initiates(P,A),
time(T), T #< E,
act(T,A),
not clipped(T,E,P).
```
together with the auxiliary definitions:
```
between(A,B,C) :- A #<B, B#<C.
time(T) :- maximum_time(Max), T :: 1..Max.
```
The abducible predicates in $`A`$ are the action predicate $`act/2`$ and the NAF predicate $`not\mathrm{\_}clipped/3`$ declared by the following clauses:
```
abducible_predicate(act/2).
abducible_predicate(not_clipped/3).
```
The integrity constraints in $`IC`$ contain the negation as failure constraint as the clause:
```
ic :- not_clipped(T,E,P), terminates(P,A1),
act(C,A2), A1 ##= A2, between(T,C,E).
```
and contraints that encode the preconditions of actions written as clauses of the general form:
```
ic :- act(T,A), not preconditions(A,T).
```
In a specific planning domain, e.g. the trucks domain, this will be extended with clauses for the $`initiates`$ and $`terminates`$ predicates in $`P`$, for example:
```
initiates(in(Obj,Truck),
load_truck(Obj,Truck,Loc)).
terminates(at(Obj,Loc),
load_truck(Obj,Truck,Loc)).
```
and the definitions, again in $`P`$, of the preconditions of the specific actions in the domain, for example:
```
preconditions(load_truck(Obj,Truck,Loc),T):-
holds_at(at(Obj,Loc),T),
holds_at(at(Truck,Loc),T).
```
The initial state is defined by a set of facts of the form $`initially(Property,0)`$, e.g. $`initially(at(package1,city1\mathrm{\_}1),0).`$
### Methodology
ACLP has been applied to several different types of abductive problems such as planning, air-crew scheduling, optical music recognition, analysis of software requirements and intelligent information integration (see below sectionUsers and Useability). Although most of these applications are not of ”industrial scale” they indicate some methodological guidelines that can be followed when using ACLP. As ACLP is a general development framework with no specific application domain these guidelines can only be themselves of a general nature.
The central advantage of an abductive approach is the high-level declarative representation that it allows. This means that the development of the program can be done incrementally starting first with a ”pure” declarative representation based on a simple model of the problem and gradually refine this model to reflect more and more particular domain knowledge of the problem at hand. A central first decision to be taken is the choice of abducibles for the problem. These play the important role of the solution carriers or answers to the problem goals. Each problem has its own abducible answer predicates (c.f. the usual answer holder of a logical variable in LP and CLP ) which means that we can describe directly in our theory (the ACLP program) the desired properties of the solution.
An important methodological step is the distinction between strict validity requirements on the solution of our problem, which are separated in the integrity constraints $`IC`$ of the ACLP theory, and the basic model of our problem which is described in the program $`P`$ of the ACLP theory. A good such separation means that we can then incrementally refine this basic model to improve the quality of the solution without affecting its validity (which is always ensured by the integrity constraints in $`IC`$). As we refine our representation we include more domain specific information, exploiting any natural structures of the problem at hand, that can help to improve the computation of the solution.
At a final step of refinement of the problem representation we can develop the model in order to control the choice of abducible in the abductive reduction of the problem goals. This choice can be implemented to follow either some heuristics, priorities, or algorithm for optimality, to control both the computational efficiency and the quality of the solutions. We can then experiment with different design alternatives adopting different strategies to study how this would affect the quality of the solutions. In large scale problems the user can also experiment with different orders in which the integrity constraints are satisfied. Generally, the heuristic of trying first more specific integrity constraints gives better results.
An importnat characteristic of an ACLP representation of a problem is the flexibility it offers under new or dynamically changing requirements. Once we have one complete representation of the problem we can easily experiment with different requirements on the solution, by changing the integrity constraints which specialize the general model to the needs and preferences of a particular case. This can be done in a modular way by affecting only the integrity constraints to reflect the new requirements.
### Users and Useability
ACLP is a high-level knowledge representation environment which supports directly abduction. Its use requires some basic knowledge of logic programming, constraint logic programming (?) and abductive logic programming (?). As it is implemented on top of ECLiPSe knowledge of this particular CLP language can help. In some cases it is also useful to understand some of the basic search heuristics that ACLP and ECLiPSe underneath use in their computation (see below in section Evaluating the System). Details of how to use it with examples can be found at the web page of ACLP at: http://www.cs.ucy.ac.cy/aclp/.
The ACLP framework as a declarative problem solving paradigm can be used to address several different types of problems. Its developers have applied it initially to the problems of scheduling and planning (??) to test its computional effectiveness and its flexibility in problem representation. Also it has been used in an industrial application of crew-schedulling (?). Other groups have used ACLP for (i) optical music recognition (?) where ACLP was used to implement a system that can handle recognition under incomplete information, and (ii) resolving inconsistencies in software requirements (?) where (a simplified form of) ACLP was used to identify the causes of inconsistency and suggest changes that can restore consistency of the specification. Also the intelligent information integration work of (?) although it does not use ACLP in its implementation its approach to information integration is based on an ACLP representation.
Currently, we are considering two new applications of ACLP. One is that of the development of an information integration mediator for integrating information suitable for electronic commerce applications. The other application area concerns the further development of the problem of planning with emphasis on (i) the study of a systematic way to exploit domain specific information, and (ii) the problem of planning under incomplete information about the initial state of the problem.
We also mention that ACLP programs can be generated automatically from example data using a machine learning technique called abductive concept learning. For details of this method and a related system see http://www-lia.deis.unibo.it/Software/ACL/.
## Evaluating the System
At this initial stage of the development of the ACLP system the main aim of its evaluation is to understand the cost of the extra high-level expressivity layer that it gives (over for example CLP approaches) in comparison with the advantages of modularity that this may provide. The ACLP system has thus been evaluated mainly in two different directions: (i) computational efficiency, particularly in comparison with the underlying CLP language of ECLiPSe on which it is implemented, and (ii) flexibility under changes of the problem specification. The overall evaluation of an application under ACLP is a combination of these two factors together with the quality of the generated solutions under some optimization criteria when such criteria apply.
For example, the air-crew scheduling application in (?) produced solutions (for the small sized company of Cyprus Airways) that were judged to be of good quality, comparable to manually generated solutions by experts of many years on the particular problem, while at the same time it provided a flexible platform on which the company<sup>1</sup><sup>1</sup>1 Unfortunately, the company has decided not to use the system for reasons that relate more to their policy of adopting a global solution to the full computerization of their operation with direct compatibility between their different systems. could easily experiment with changes in policy and preferences. Also the re-scheduling module of the system was judged to be of high-value both as a tool for adjusting the initially generated solution and for handling unexpected changes on the day of operation.
The computational effectiveness of the ACLP system depends on two factors: (a) the effectiveness of the reduction of the high-level ACLP representation to lower-level finite domain constraints and (b) the effeciency of the underlying constraint solver in propagating (or solving) these constraints. In fact, these two factors are interrelated as in many cases the reduction in (a) depends on the completeness of the propagation in (b). For some problems where these are not strongly related e.g. in the case of job-shop sheduling we can see that in comparison with (b) the overhead for the reduction in (a) is small. Information on these evaluation experimenents can be found at the ACLP web pages. Further results of comparison on problems where factors (a) and (b) are loosesly coupled can be found in the recent work of (?) where experiments with various types of systems, including ACLP, on the constraint satisfaction problems of the N-queens and graph colouring have been perfomed.
Another, but limited, comparison that we have carried out in order to test the effectiveness of the current ACLP implementation was a comparison with the use of Constraint Handling Rules (CHRs) (?) on the same problems of job-shop scheduling. On the whole the ACLP system was at least as effecient as CHR. It should be noted though that these comparions were carried out before recent developments on CHRs.
In problems where the search space of the reduction of the high-level specification depends strongly on the fast detection that the contraint store of finite domain constraints is becoming unsatisfiable, the overall computational efficiency of ACLP can be sensitive to the particular way of modeling the problem and the amount of specific domain knowledge it contains pertaining to the computation aspects of the problem. Such a problem is that of planning. As an example, table 1 shows the execution time (on a SUN Ultra-1 with 64Mb RAM) and the number of moves required for an indicative set of blocks world planning problems. The representation of the problem that was used was purely declarative with the exception that the towers of the final state were to be build in a horizontal fashion from the bottom up. The number of available positions on the table was restricted to be one third of the total number of blocks in order to make the problems more computationally demanding. These times are comparable with the execution times of solving such problems directly in ECLiPSe as reported in (?).
The flexibility of the ACLP system as a knowledge representation framework is tested by examining how easy it is for a given ACLP representation to be adapted under changes in the requirements of the original problem. There are two factors to measure here: (1) the programming effort required to adapt an existing solution to the new problem, and (2) the computational robustness of the system under such changes. Experiments in the domains of job-shop scheduling, air-crew scheduling and planning show that the extra programming effort in ACLP is considerably smaller than the corresponding effort when the problem is represented directly in ECLiPSe. In many cases, the effort required in ACLP is simply the addition of some new integrity constraints written directly from the declarative specification of the new requirements. The same experiments show that on the whole the computational performance of ACLP remains within the same order of magnitute under changes which affect the problem only locally in one part, e.g. extra requirements on the moves allowed for a particular ”small” subset of blocks in the problem of blocks world planning.
Another feature of the flexibility of ACLP is the ability to use it to recompute the solution for a given goal, under some new information about the particular instance of the problem, so that the new solution remains ”close” to the old solution, e.g. it contains a minimal number of changes from the old solution. Experiments to test this feature have been performed on the problems of job-shop and air-crew scheduling. Table 2 shows an example of the results. For each one of these problems a new requirement of some resource unavailability was added (shown below in the first column). The rescheduling results are shown in the third column of the table, which gives the time together with the number of changes needed on the existing solution in order to satisfy the new requirements. The fourth column displays the analogous information for the control experiment of re-executing the goal with the extra requirement represented in the program but now without any initial solution.
### Future Development
ACLP is a general purpose declarative programming framework. Hence its evaluation must combine different aspects of its performance. At this initial stage the emphasis in the development of the first prototype was on its declarativeness together with an acceptable computational performance.
The search that ACLP performs in constructing a solution needs further study for improvement. Currently, the system employs a few simple heuristics to help in its search. As ACLP is parametric on the underlying finite domain constraint solver improvements on its performance will improve the ACLP performance. More important though, is the interaction between the abductive reduction and the satisfaction of the finite domain constaints that it generates. This is the major aspect of the search space of ACLP. Hence one way to improve the ACLP search is to develop further the interface between the abductive reduction and the constraint solver so that the propagation of the domain constraints varies according to some heuristic criteria on the point of the search space where the request to the constraint solver is made.
In particular, while the abductive process is reducing the high-level goal and integrity constraints there are choice points where we can either introduce a new abducible hypothesis in the solution or instead backtrack higher up in the search space. In many cases, this decision can be made to depend on the satisfaction of some of the lower-level domain constraints that are generated by the abductive reduction. We can then evaluate the significance of their satisfaction (currently the system adopts a very simple form of evaluation) and depending on this guide the search to introduce or not a new hypotheses in the solution. This has to be combined with the general heuristic of abductive search of prefering to reuse hypotheses and delay the specialisation of (non-ground) hypotheses or the generation of new ones. In developing though a better search for ACLP it maybe necessary to restrict our attention to separate classes of problems e.g. to develop separately an ACLP planner from an ACLP system for diagnosis.
These considerations of improving the general purpose search strategy of the system is an important next stage of development. On the other hand, it is clear that the general improvement of efficiency that can be achieved is limited as we are aiming to use the system for computational hard problems. Hence another line of development is to provide more facilities for problem specific information to be incorporated in the representation of the problem whose exploitation can improve the performance of the system on the particular problem at hand. This problem specific information could include information to directly control the search of the system on the particular problem in the same spirit of recent developments for controlling models in constraint programming (?). |
warning/0003/nlin0003037.html | ar5iv | text | # Fredholm methods for billiard eigenfunctions in the coherent state representation
## I Introduction
The precise test of semiclassical approximations in the presence of chaos is of grest interest to establish the limits of applicability of periodic orbit theory and its resummations. This test can be done in model systems both on approximations to the spectrum or to the stationary states. For the calculation of the spectrum the most efficient tool in this respect seems to be the spectral determinant and several calculations have demonstrated that, given enough periodic orbits, the spectrum can be accurately represented semiclassically. However, a more sensitive test - and still a great challenge - is the semiclassical representation of single eigenfunctions. This includes the study of the scar phenomena and the eventual deviations from uniformity of eigenfunctions in accordance to the Berry-Voros hypothesis and Schnirelman’s theorem .
Just as for spectral problems, the use of Fredholm methods allows for the most efficient encoding of classical information in the calculation for single eigenfunctions . In this paper we review these methods and apply them to the calculation of Husimi distributions of stadium eigenfunctions.
This paper is organized as follows. In Sec. II we review the Fredholm method for billiard eigenfunctions. Fredholm theory allows us to find the solution to certain type of integral or operator equations . For billiards these methods can be applied to the boundary integral equation. In Sec. III we make the semiclassical approximation that is based on the approximation of the traces and powers of the propagator as sums over the periodic points of the underlying classical system. The propagator itself is taken as Bogomolny’s $`𝐓`$ operator . We choose the coherent state representation and obtain an expression for the semiclassical Husimi representation of the eigenfunctions in terms of classical invariants: periodic points, their monodromy matrices and Maslov indeces. In Sec. IV we apply this scheme for the stadium billiard. Our conclusions and perspectives are presented in Sec. V.
## II Fredholm formulae for eigenfunctions
Fredholm theory gives the solution to a certain class of integral equations, which can also be written as operator equations . A Fredholm integral equation of second type is
$`\chi (q)=\chi _0(q)+\lambda {\displaystyle 𝑑q^{}𝐓(q^{},q)\chi (q^{})}.`$ (1)
All the functions are defined in a finite domain. If the known functions $`\chi _0(q)`$ and $`𝐓(q^{},q)`$ are well behaved, the Fredholm alternative holds: there is a unique solution $`\chi `$ with the same analytic properties or the homogeneous equation ($`\chi _0=0`$) has a solution. There is a set of complex parameters $`\lambda _i`$ for which the solution is not unique. In operator notation, the inverse of $`(1\lambda 𝐓)`$ exists if $`\lambda \lambda _i`$. In this case, this inverse can be written as
$`{\displaystyle \frac{1}{1\lambda 𝐓}}={\displaystyle \frac{𝐌(\lambda )}{D(\lambda )}},`$ (2)
where the operator $`𝐌(\lambda )`$ and the function $`D(\lambda )`$ are series in $`\lambda `$. If $`𝐓`$ is a compact operator, $`D(\lambda )`$ and $`𝐌(\lambda )`$ are entire in $`\lambda `$ and, thus, absolutely convergent. The explicit form for the series expansion in terms of powers of $`𝐓`$ is given below. In what follows we apply this general theory assuming $`𝐓`$ to be unitary and of finite dimension $`N`$. Both assumptions are justified in the semiclassical limit for the quantization of billiards .
### A Secular equation
The $`k`$’s eigenvalues are given by the secular equation $`P(k)=\text{det}(1𝐓(k))=0`$. We can expand this determinant as
$`P(k)={\displaystyle \underset{n=0}{\overset{N}{}}}\beta _n(k),`$ (3)
where the coefficients $`\beta _n(k)`$ are related to the traces of $`𝐓(k)`$, $`b_n(k)\text{tr}𝐓^n(k)`$, through
$`\beta _n(k)={\displaystyle \frac{1}{n}}{\displaystyle \underset{j=1}{\overset{n}{}}}\beta _{nj}(k)b_j(k).`$ (4)
Thus, knowledge of the traces up to a certain $`n_{max}`$ implies the knowledge of the coefficients $`\beta _n`$ up to the same $`n_{max}`$.
If $`𝐓(k)`$ is unitary, $`P(k)`$ is self reversive, meaning that its coefficients satisfy
$`\beta _{Nj}(k)=(1)^N\overline{\beta }_j(k)\text{det}𝐓(k).`$ (5)
This condition alone forces the eigenvalues of $`𝐓`$ at fixed $`k`$ to be symmetric with respect to the unit circle: if $`\lambda `$ is an eigenvalue, then $`1/\overline{\lambda }`$ is an eigenvalue too. Of course, if $`𝐓`$ is unitary, then this condition is automatically satisfied but we can use it in our semiclassical approach to partially restore unitarity.
The contributions from coefficients $`\beta _i`$ with $`i>[(N+1)/2]`$ can be expressed in terms of coefficients $`\beta _i`$ with $`i[(N+1)/2]`$. ($`[x]`$ is the integer part of $`x`$.) So, if $`N`$ is even:
$`P(k)=\eta (k)+det𝐓\overline{\eta }(k),`$ (6)
$`\eta (k)={\displaystyle \underset{j=0}{\overset{N/21}{}}}\beta _j(k)+{\displaystyle \frac{1}{2}}\beta _{N/2},det𝐓={\displaystyle \frac{\beta _{N/2}}{\overline{\beta }_{N/2}}}.`$ (7)
If $`N`$ is odd:
$`P(k)=\eta (k)det𝐓\overline{\eta }(k),`$ (8)
$`\eta (k)={\displaystyle \underset{j=0}{\overset{(N1)/2}{}}}\beta _j(k),det𝐓={\displaystyle \frac{\beta _{(N+1)/2}}{\overline{\beta }_{(N1)/2}}}.`$ (9)
As a consequence of the imposition of this symmetry on the operator $`𝐓`$ we obtain two advantages: only traces up to half the Heisenberg time $`t_H=N`$ are needed and the eigenvalues are constrained to lie on the unit circle or in symmetric pairs.
These formulae relate $`P(k)`$ with the traces of powers of $`𝐓(k)`$ which, in turn are related semiclassically to periodic orbits and to the smoothed density of states . They have been tested extensively for the hyperbola billiard by Keating and Sieber .
### B Green function
To extend these methods to the calculation of eigenfunctions, we define a generalized Green function $`𝐆(k)`$:
$`𝐆(k)={\displaystyle \frac{𝐓(k)}{1𝐓(k)}}.`$ (10)
This operator has poles at the billiard eigenvalues $`k=k_\nu `$ and its residues are the projectors onto the corresponding eigenfunctions. It has a Fredholm expression as
$`𝐆(k)={\displaystyle \frac{𝐓(k)𝐂^t(1𝐓(k))}{P(k)}},`$ (11)
where $`𝐂^𝐭(\mathrm{𝟏}𝐓(𝐤))`$ is the transpose of the cofactor matrix of $`1𝐓(k)`$ and, as in Eq. (2), has an expansion in powers of $`𝐓(k)`$.
It is then convenient to define a normalized Green operator as
$`𝐠(k)={\displaystyle \frac{𝐆(k)}{\text{tr}(𝐆(k))}},`$ (12)
where the singularities in the denominator have been eliminated. The normalized Green operator has the property $`𝐠(k_\nu )=|\psi _\nu \psi _\nu |`$, where $`|\psi _\nu `$ is the eigenvector corresponding to eigenvalue $`k_\nu `$. Then, we can write $`𝐠(k)`$ in the following way:
$`𝐠(k)={\displaystyle \frac{𝐓(k)C^t(1𝐓(k))}{\text{tr}(𝐓(k)C^t(1𝐓(k))).}}`$ (13)
As the cofactor matrix can be expanded in powers of the propagator and as the propagator itself is unitary we write the normalized Green operator in terms of the powers of the propagator and their traces up to $`N/2`$ (if $`N`$ is even):
$`𝐠(k)={\displaystyle \frac{\underset{i=0}{\overset{\frac{N}{2}1}{}}c_i(k)𝐓^{i+1}(k)\text{det}𝐓(k)\underset{i=0}{\overset{\frac{N}{2}1}{}}\overline{c}_i(k)𝐓^i(k)}{_{i=0}^{\frac{N}{2}1}c_i(k)\text{tr}(𝐓^{i+1}(k))\text{det}𝐓(k)_{i=0}^{\frac{N}{2}1}\overline{c}_i(k)\text{tr}(𝐓^i(k))}},`$ (14)
where the coefficients $`c_i(k)`$ are given by
$`c_i(k)={\displaystyle \underset{n=i}{\overset{\frac{N}{2}1}{}}}\beta _{ni}(k).`$ (15)
An analogous formula can be derived in case $`N`$ is odd.
The coefficients $`c_i(k)`$ are dependent on the traces of $`𝐓^n`$ through Eqs. (4) and (15) and thus are independent of the chosen representation. On the other hand, the expression for the powers of the propagator will depend on the representation chosen for the calculation of the eigenfunctions. If the coordinate representation $`|q`$ is chosen, Eq. (14) relates the probability density $`|\varphi _\nu (q)|^2`$ to the diagonal powers of the propagator $`q|𝐓^n(k)|q`$. If the Weyl representation is chosen, then Eq. (14) gives the Wigner distribution of $`|\varphi _\nu `$ in terms of the Weyl propagator . Here we choose the coherent state representation to find the equivalent Husimi distribution.
We remark that Eq. (14) is a very compact and representation independent derivation of formulae that were previously very laboriously derived for the Wigner case. It prepares in an optimal way the grounds for the semiclassical approximation because its ingredients are all dependent on classical elements, namely periodic orbits, phase space volume and generating function.
Using the fact that at $`k=k_\nu `$ the normalized Green function is the projector onto the corresponding eigenstate, we can obtain the Husimi distribution as
$`_{\psi _\nu }(z,\overline{z})={\displaystyle \frac{z|𝐠(k_\nu )|z}{z|z}}={\displaystyle \frac{1}{z|z}}{\displaystyle \frac{\underset{i=0}{\overset{\frac{N}{2}1}{}}c_i(k)z|𝐓^{i+1}(k)|z\text{det}𝐓(k)\underset{i=0}{\overset{\frac{N}{2}1}{}}\overline{c}_i(k)z|𝐓^i(k)|z}{_{i=0}^{\frac{N}{2}1}c_i(k)\text{tr}(𝐓^{i+1}(k))\text{det}𝐓(k)_{i=0}^{\frac{N}{2}1}\overline{c}_i(k)\text{tr}(𝐓^i(k))}}.`$ (16)
This scheme was successfuly applied in simple quantum maps .
## III Semiclassical approximation
Green’s theorem allows us to reduce the Schrödinger equation for the billiard with Dirichlet boundary conditions to the following linear homogeneous equation for the normal derivative on the border $`\varphi (s)`$
$`\varphi (s)=2{\displaystyle 𝑑s^{}\varphi (s^{})𝐊(s,s^{};k)},`$ (17)
where the kernel is
$`𝐊(s,s^{};k)={\displaystyle \frac{ik}{2}}\mathrm{cos}\psi (s)H_1^{(1)}(k|𝐫(s)𝐫^{}(s^{})|),`$ (18)
with $`k`$ the wave number, $`\psi (s)`$ the angle between the normal at $`s`$ and the line that connects $`𝐫(s)`$ with $`𝐫^{}(s^{})`$ (see Fig. 1) and $`H_1^{(1)}`$ the Hankel function of first type and order one.
We introduce the wave function $`\mu (s)`$
$`\varphi (p)={\displaystyle \frac{1}{ik}}\sqrt{1p^2}\mu (p),`$ (19)
with $`\varphi (p)`$ and $`\mu (p)`$ the momentum representations of $`\varphi (s)`$ and $`\mu (s)`$. This transformation makes the kernel symmetric and turns Eq. (17) to
$`\mu (s)={\displaystyle 𝐓(s^{},s;k)\mu (s^{})𝑑s^{}}.`$ (20)
The semiclassical theory of the kernel $`𝐓(s^{},s;k)`$ is based on two fundamental properties: $`𝐓`$ is semiclassically unitary and has an effective dimension $`N(k)=Lk/\pi `$, where $`L`$ is the length of the billiard. Moreover, the kernel is given by the generating function of the classical Birkhoff map (see Eq.(21)). These properties have been extensively tested and will be assumed in what follows.
Thus, we make the semiclassical approximation by taking $`𝐓`$ as Bogomolny’s operator (21) and by evaluating all integrals by stationary phase approximation. The $`𝐓`$ operator for convex billiards in the plane, taking its border as Poincaré section and using Birkhoff coordinates is
$`𝐓(s^{},s;k)=\left({\displaystyle \frac{k}{2\pi i}}\right)^{\frac{1}{2}}\left|{\displaystyle \frac{^2l(s^{},s)}{ss^{}}}\right|^{1/2}\mathrm{exp}\left(ikl(s^{},s)i{\displaystyle \frac{\pi }{2}}\nu \right),`$ (21)
where the bounce map generated by $`l(s^{}s)`$, the arc length between $`s`$ and $`s^{}`$; $`\nu `$ is the Maslov index. The quantization condition is $`\text{det}(1𝐓(k))=0`$.
In the semiclassical theory for the spectral determinant
$`P(k)=\text{det}(1𝐓(k))`$ (22)
the approximate unitarity of $`𝐓`$ can be used efficiently to reduce the number of periodic orbits needed for the computation of the spectrum. Similar manipulations of the Fredholm formulae allow for the same reduction in the semiclassical calculation of single eigenfunctions of the billiard. We write down a formula giving the projector on single eigenfucntions as a finite sum involving powers of the map $`𝐓`$ and its traces.
### A Semiclassical traces and determinant
First we need the traces. It is a well known fact that they adopt the following semiclassical expression :
$`[b_n]_{scl}={\displaystyle \underset{PO,n=n_pr}{}}{\displaystyle \frac{n_p}{|\text{det}(IM_p^r)|^{(1/2)}}}\mathrm{exp}\left(ir(kl_p\nu _p\pi /2)\right),`$ (23)
where the sum goes over all the primitive PO’s of the billiard with period $`n_p`$, which must be a divisor of $`n`$, Maslov index $`\nu _p`$, length $`l_p`$ and monodromy matrix $`M_p`$. The Maslov index can be interpreted geometrically: $`\pi \nu _p`$ is the angle swept by the unstable manifold of $`M_p`$ along the PO.
The determinant of $`𝐓`$ can be obtained as
$`[\text{det}𝐓(k)]_{scl}=(1)^N\text{exp}(2\pi i𝒩(k)),`$ (24)
where $`𝒩(k)`$, the number of states between 0 y $`k`$, is
$`𝒩(k)={\displaystyle \frac{1}{4\pi }}𝒜k^2{\displaystyle \frac{1}{4\pi }}Lk,`$ (25)
with $`𝒜`$ the area of billiard and $`L`$ its length.
These are the semiclassical ingredients needed for the calculation of the spectrum and of the coefficients $`c_i(k)`$.
### B Semiclassical propagator in coherent state representation
We first obtain the coherent state representation for one iteration of the bounce map:
$`z^{}|𝐓|z={\displaystyle 𝑑s𝑑s^{}z^{}|s^{}s^{}|𝐓|ss|z}.`$ (26)
That is to say:
$`z^{}|𝐓|z=\left({\displaystyle \frac{k}{\pi \sigma ^2}}\right)^{1/2}\left({\displaystyle \frac{k}{2\pi i}}\right)^{1/2}\text{e}^{(i\pi \nu /2)}\times {\displaystyle 𝑑s𝑑s^{}\left|\frac{^2l}{ss^{}}\right|^{1/2}\mathrm{exp}\left(ik\mathrm{\Phi }(s^{},s)\right)}`$ (27)
where
$`\mathrm{\Phi }(s^{},s)={\displaystyle \frac{i}{2}}z^2+{\displaystyle \frac{i}{2}}z^2{\displaystyle \frac{i}{2\sigma ^2}}s^2{\displaystyle \frac{i\sqrt{2}}{\sigma }}z^{}s^{}+{\displaystyle \frac{i}{2\sigma ^2}}s^2{\displaystyle \frac{i\sqrt{2}}{\sigma }}\overline{z}s+l(s^{},s),`$ (28)
and $`z^{}=(q^{}/\sigma i\sigma p^{})/\sqrt{2}`$ and $`z=(q/\sigma i\sigma p)/\sqrt{2}`$. The most important contributions come from those points $`s^{}`$ and $`s^{}`$ that make stationary the phase $`\mathrm{\Phi }`$:
$`{\displaystyle \frac{\mathrm{\Phi }}{s}}(s^{},s^{})={\displaystyle \frac{i}{\sigma ^2}}s^{}{\displaystyle \frac{i\sqrt{2}}{\sigma }}\overline{z}+{\displaystyle \frac{l}{s}}(s^{},s^{})=0`$ (29)
$`{\displaystyle \frac{\mathrm{\Phi }}{s^{}}}(s^{},s^{})={\displaystyle \frac{i}{\sigma ^2}}s^{}{\displaystyle \frac{i\sqrt{2}}{\sigma }}z^{}+{\displaystyle \frac{l}{s^{}}}(s^{},s^{})=0.`$ (30)
The solution to (29) satisfying the reality conditions is
$`s^{}=q{\displaystyle \frac{l}{s}}(s^{},s^{})=p`$ (31)
$`s^{}=q^{}{\displaystyle \frac{l}{s^{}}}(s^{},s^{})=p^{}.`$ (32)
This is a classical trajectory from $`(q,p)`$ to $`(q^{},p^{})`$. Thus, the matrix element $`z^{}|𝐓|z`$ will be non zero only if $`z`$ and $`z^{}`$ are connected by the classical dynamics. Let us call these points $`z_c`$ and $`z_c^{}`$ and let us calculate the matrix element to next order in their neighbourhoods, $`z_c^{}+\delta z^{}|𝐓|z_c+\delta z`$. To this effect we expand $`\mathrm{\Phi }(s^{},s)`$ in Eq. (27) to second order and, after some algebra:
$`\mathrm{\Phi }(\delta s^{},\delta s)l(q_c^{},q_c)+\left[i\delta z^{}\overline{z}_{}^{}{}_{c}{}^{}i\delta \overline{z}z_c{\displaystyle \frac{i}{2}}\overline{z}_cz_c{\displaystyle \frac{i}{2}}\overline{z}_{}^{}{}_{c}{}^{}z_c^{}\right]+\left[{\displaystyle \frac{i}{4}}(\overline{z}_c^{}_{}{}^{}2z_c^{}_{}{}^{}2\overline{z}_c^2+z_c^2)\right]+`$ (33)
$`+\left[{\displaystyle \frac{i}{2}}\delta z^2+{\displaystyle \frac{i}{2\sigma ^2}}\delta s^2{\displaystyle \frac{i\sqrt{2}}{\sigma }}\delta z^{}\delta s^{}+{\displaystyle \frac{i}{2}}\delta \overline{z}^2+{\displaystyle \frac{i}{2\sigma ^2}}\delta s^2{\displaystyle \frac{i\sqrt{2}}{\sigma }}\delta \overline{z}\delta s+{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{^2l}{s^2}}\delta s^2+2{\displaystyle \frac{^2l}{s^{}s}}\delta s^{}\delta s+{\displaystyle \frac{^2l}{s^2}}\delta s^2\right)\right],`$ (34)
where $`\delta s=sq_c`$ and $`\delta s^{}=s^{}q_c^{}`$. We change to new integration variables $`\delta s`$ and $`\delta s^{}`$ and obtain:
$`z_c^{}+\delta z^{}|𝐓|z_c+\delta z\left({\displaystyle \frac{k}{\pi \sigma ^2}}\right)^{1/2}\left({\displaystyle \frac{k}{2\pi i}}\right)^{1/2}\mathrm{exp}(i\pi \nu /2)\times {\displaystyle 𝑑\delta s𝑑\delta s^{}\left|\frac{^2l}{ss^{}}\right|^{1/2}\mathrm{exp}\left(ik\mathrm{\Phi }(\delta s^{},\delta s)\right)}`$ (35)
We now insert Eq. (33) in Eq. (35). All terms are constant with respect to integration except the last one in square brackets. The resulting integral is the coherent state representation, with respect to $`|\delta z`$, of the linearized map, whose generating function is quadratic, which we have introduced in Ec. (A19):
$`\delta z^{}|𝐓|\delta z={\displaystyle \frac{1}{\sqrt{\overline{s}_c}}}\mathrm{exp}\left[{\displaystyle \frac{k}{2\overline{s}_c}}\left(\overline{r}_c\delta z^2+2\delta z^{}\delta \overline{z}+r_c\delta \overline{z}^2\right)\right],`$ (36)
where $`r_c`$ y $`s_c`$ are the matrix elements of the linearized map in complex coordinates. Finally we arrive at:
$`z_c^{}+\delta z^{}|𝐓|z_c+\delta z\mathrm{exp}\left[{\displaystyle \frac{k}{4}}(\overline{z}_c^{}_{}{}^{}2z_c^{}_{}{}^{}2\overline{z}_c^2+z_c^2)\right]\times `$ (37)
$`\times \mathrm{exp}\left[k\left(\delta z^{}\overline{z}_{}^{}{}_{c}{}^{}+\delta \overline{z}z_c+{\displaystyle \frac{1}{2}}\overline{z}_cz_c+{\displaystyle \frac{1}{2}}\overline{z}_{}^{}{}_{c}{}^{}z_c^{}\right)\right]\times \mathrm{exp}\left(ikli{\displaystyle \frac{\pi }{2}}\nu \right)\delta z^{}|𝐓|\delta z.`$ (38)
This result lets us evaluate the matrix element we were looking for, $`z|𝐓^n|z/z|z`$, that will be a sum of contributions of periodic points $`z_{pp}`$ of period $`n`$ in the semiclassical limit:
$`{\displaystyle \frac{z|𝐓^n|z}{z|z}}{\displaystyle \underset{pp,n}{}}{\displaystyle \frac{z_{pp}+\delta z|𝐓^n|z_{pp}+\delta z}{z_{pp}+\delta z|z_{pp}+\delta z}}.`$ (39)
To obtain the composition $`z_{pp}+\delta z|𝐓^n|z_{pp}+\delta z`$ we use the expression (37) and the composition rule of Eq. (A20). Then
$`{\displaystyle \frac{z|𝐓^n|z}{z|z}}{\displaystyle \underset{pp,n}{}}{\displaystyle \frac{1}{\sqrt{\overline{s}_{pp}}}}\lambda _{pp}\mathrm{exp}(ikl_{pp}i{\displaystyle \frac{\pi }{2}}\nu _{pp})\times \mathrm{exp}\left[{\displaystyle \frac{k}{2\overline{s}_{pp}}}\left(\overline{r}_{pp}\delta z^2+2\delta z\delta \overline{z}+r_{pp}\delta \overline{z}^2\right)k\delta z\delta \overline{z}\right],`$ (40)
where $`\lambda _{pp}`$ can be calculated by Eq. (A21), $`\nu _{pp}=n`$ (because of the Dirichlet boundary conditions) and $`l_{pp}`$ is the length of the PO starting from $`(q_{pp},p_{pp})`$. As we can see, the matrix element behaves as a gaussian in the vecinities of the periodic point. This allows us to write the Husimi representation of the $`n`$-th power of the propagator as a sum of contributions from periodic points of period $`n`$. Each term of the sum is a gaussian packet in phase space whose parameters are related to the monodromy matrix in complex coordinates. A PO composed by $`n`$ points will give $`n`$ different contributions to this sum, due to the fact that the monodromy matrices at each point differ. However, the invariant properties of these matrices are the same and the usual Gutzwiller-Tabor trace formula can be recovered by integration. Of course, a periodic point of period $`n`$ will contribute also to the $`rn`$ ($`r`$ natural) powers of the propagator.
We should remark at this point that the different semiclassical representations of the propagator in terms od the corresponding generating function are only semiclassically equivalent and thus can give different results at finite $`N`$. This is not true for the calculation fo the spectral determinant, whose semiclassical expression in terms of periodic orbits is the same in all representations. It is because of this that the different ways of computing eigenfunctions are not equivalent. For the calculation of $`|\varphi _\nu (s)|^2`$ the closed (but not necesarily periodic) orbits are needed . For the Wignaer function calculation only periodic points are needed but each contribution is extended in phase space. In the present formalism we will obtain the Husimi distributions of eigenfunctions in terms of deformed localized gaussians centered in the periodic points, constructed solely in terms of classical information.
Symmetries
Our system, the stadium billiard, has two discrete spatial symmetries: $`R_x`$ and $`R_y`$, the two reflections with respect of the coordinate axes. These spatial symmetries in the domain reflect in the border and, thus, in the classical and quantum map on it. Their action on the Birkhoff coordinates of phase space $`(q,p)`$ is
$`R_x(q,p)(Lq,p),R_y(q,p)({\displaystyle \frac{L}{2}}q,p),R_xR_y(q,p)({\displaystyle \frac{L}{2}}+q,+p).`$ (41)
In order to have coherent states on the border with correct symmetries we need to project them using $`𝐑_x`$ and $`𝐑_y`$, the unitary representations of the symmetries $`𝐑_x|x,y=|x,y`$ and $`𝐑_y|x,y=|x,y`$. Then we define
$`|z_{\sigma _x\sigma _y}=\left({\displaystyle \frac{1+\sigma _x𝐑_x}{2}}\right)\left({\displaystyle \frac{1+\sigma _y𝐑_y}{2}}\right){\displaystyle \frac{|z}{\sqrt{z|z}}},`$ (42)
where $`\sigma _x,\sigma _y=\pm 1`$ and $`𝐑_x`$ and $`𝐑_y`$ move the center of the coherent state according to Ec. (41).
In this way, the diagonal matrix elements of the propagator in symmetrized coherent state representation are
$`{\displaystyle \frac{1}{z|z}}z|𝐓\left({\displaystyle \frac{1+\sigma _x𝐑_x}{2}}\right)\left({\displaystyle \frac{1+\sigma _y𝐑_y}{2}}\right)|z={\displaystyle \frac{1}{4z|z}}(z|𝐓|z+\sigma _xz|\mathrm{𝐓𝐑}_x|z+\sigma _yz|\mathrm{𝐓𝐑}_y|z+\sigma _x\sigma _yz|\mathrm{𝐓𝐑}_x𝐑_y|z).`$ (43)
We have already calculated $`z|𝐓^n|z`$. We still have to calculate the other three contributions, $`z|𝐓^n𝐑_x|z`$, $`z|𝐓^n𝐑_y|z`$ and $`z|𝐓^n𝐑_x𝐑_y|z`$. We can conclude using the results we have already obtained that each of them will be a sum of gaussians centered in those points $`z`$ that the dynamics connects with their symmetric partners, $`R_xz`$, $`R_yz`$, $`R_xR_yz`$, respectively. These points belong to POs whose periods are $`2n`$ which are symmetric under the operations $`R_x`$, $`R_y`$, $`R_xR_y`$, respectively. The increment $`R\delta z`$ with respect to $`Rz`$ ($`RR_x,R_y,R_xR_y`$) is related to the increment $`\delta z`$ with respect to $`z`$ through:
$`R\delta z=t_R\delta z,t_R=\{\begin{array}{cc}1\text{ if }R=R_x\hfill & \\ 1\text{ if }R=R_y\hfill & \\ 1\text{ if }R=R_xR_y.\hfill & \end{array}`$ (47)
Thus we arrive at:
$`{\displaystyle \frac{z|𝐓^n𝐑|z}{z|z}}{\displaystyle \underset{pp,2n}{}}{\displaystyle \frac{1}{\sqrt{\overline{s}_{pp}}}}\lambda _{pp}\mathrm{exp}(ikl_{pp}i{\displaystyle \frac{\pi }{2}}\nu _{pp})\times \mathrm{exp}\left[{\displaystyle \frac{k}{2\overline{s}_{pp}}}\left(\overline{r}_{pp}\delta z^2+2t_R\delta z\delta \overline{z}+r_{pp}\delta \overline{z}^2\right)k\delta z\delta \overline{z}\right],`$ (48)
where the sum goes over the periodic points of period $`2n`$ that belong to POs symmetric under $`R`$. The quantities $`s_{pp}`$, $`r_{pp}`$, $`l_{pp}`$, $`\nu _{pp}`$ and $`\lambda _{pp}`$ are calculated along the trajectory that connect $`z`$ to $`Rz`$, i.e., half PO.
## IV Semiclassical eigenfunctions for the stadium
We use the semiclassical approach we introduced above for the stadium billiard. We choose odd-odd symmetries ($`\sigma _x=\sigma _y=1`$) and $`\sigma =2`$. We have periodic points with desymmetrized period up to 8 (around 800). We have used the symbolic dynamics developed by Biham and Kvale to obtain them. The wave number $`k`$ is related to the maximum period used in the expansion (16) by $`P(k)=\frac{L}{2\pi }k0.4k`$. In this way we can obtain semiclassical approximations of eigenfunctions of wave number $`k20`$.
In Fig. 2 we show the phase space representations of the first six powers of Bogomolny’s $`𝐓`$ operator for $`k=20`$; in Fig. 3, the semiclassical approximations. We see that the exact representations show global maxima in the bouncing ball region that cannot be reproduced semiclassically for the lowest powers. However, the overall semiclassical behaviour is very close to that of the exact representations. (Because of the symmetries we chose, we have no semiclassical approximation to the first power of the operator because the contribution of the only periodic point of period 1 is zero.)
We select two energy ranges: $`k[19.1,20.0]`$ and $`k[20.5,21.3]`$. There are 4 eigenenergies in each of these ranges. We show the absolute value of the secular determinant, $`|P(k)|`$, for each of them in Figs. 4 and 6. The full line is the semiclassical approximation, the dashed line is the secular determinant for Bogomolny’s operator. The vertical lines are the exact quantum $`k`$ eigenvalues calculated by the scaling method . We see a good approximation when we use the periodic point expansion. The agreement shows that in this region the spectrum is well represented semiclassically. (The discontinuities come from the change in dimension of the operator).
To keep the method consistent we evaluated the semiclassical Husimi expansion in those values of $`k`$ that minimize the semiclassical secular determinant. We see in Figs. 5 and 7 the exact eigenfunctions (first column) and their corresponding semiclassical Husimi representations (second column) obtained as the real part of Eq. (16). The global behaviour is well reproduced; however, the finer details are hard to mimic. The bouncing ball region is problematic: in some functions (e.g., $`k=21.16`$) some probability leaks to this region. Probably POs with longer periods that approximate the bouncing ball orbits could make a better picture for this region.
One of the advantages of formula (14) for the projector is that it has no singularities between eigenvalues. It is possible to study continualy its behaviour as a function of $`k`$ in order to see its sensitivity to changes in $`k`$. Some properties of the exact distribution as a function of $`k`$ are:
* The distribution is positive at the eigenvalues $`k_n`$.
* The distribution has $`N(k)`$ zeros at eigenvalues $`k_n`$.
These two properties follow from the fact that $`_{\psi _\nu }(z,\overline{z})`$ is the modulus of an analytic function. The distributions between eigenvalues can become negative. In particular, it can be shown that at the value of $`k`$ that maximizes $`P(k)`$, the distribution is constant. This properties can be used to control the semiclassical approximations.
In Figs. 8 and 9 we show the behaviour of the distribution $`z|𝐠(k)|z/z|z`$ between the semiclassical eigenvalues $`k=19.18`$ and $`k=19.38`$. When $`k=19.18`$ the distribution is positive and has well defined minima that approach zero. As we move away from the eigenvalue, the distribution changes smoothly. Initialy it moves away from the plane $`g=0`$ in the positive direction, then it comes back and turns negative. During this “evolution” it flattens visibly and we can’t discern its features. At $`k=19.32`$, aproximately the maximum of the secular determinant, see Fig. 4, the distribution is constant. At the semiclassical eigenvalue $`k=19.38`$ the distribution is positive again with well defined minima.
We can see from Figs. 5 and 7 that the semiclassical approximation is relatively good. It is not trivial to obtain a positive defined distribution with $`N`$ zeroes adding several hundreds of gaussians, each with its phase and deformation.
## V Conclusions
Using Fredholm theory we have given a very compact and representation independent derivation of the projector on a single eigenfunction for unitary quantum maps. Expressing the projector in the coherent state basis we wrote a semiclassical expression for the Husimi distributions of the billiard’s eigenfunctions. Each periodic point contributes with a gaussian centered in it whose parameters are calculated only with classical information. We should not underestimate the difficulties and complexities inherent to this method. Hundreds of gaussian contributions have to conspire to make a positive definite distribution with $`N`$ that approximate the quantum Husimi distributions.
The projector (14) can be represented in coordinate space. We obtain $`q|\psi \psi |q`$, whose semiclassical approximation can be directly compared to the probability density in the section. This representation has an additional difficulty, since the semiclassical approximation is written as a sum over closed trajectories, periodic or not, in configuration space. Those that are not periodic are more in number and more difficult to find. Anyway, we can apply our scheme for Bogomolny’s $`𝐓(q^{},q)`$ operator and compare the results with the exact quantum calculation. In Fig. 10 we see that the approximation is excelent at this level.
The maximum period $`P`$ in the expansions is related to the energy in the way $`P0.4k`$. Due to the exponential proliferation of orbits in chaotic systems, the method cannot be applied for arbitrarily high energies. The measure of this proliferation is the topologic entropy $`W`$ which relates the number $`N_P`$ of POs of a given period $`P`$ with the period itself, $`N_P=\mathrm{exp}(WP)`$ . For the stadium, $`W0.94`$. Then, for $`k100`$ we need POs of periods up to $`P=40`$, whose number is $`N_P\mathrm{exp}(0.94\times 40)10^{17}`$!
The Fredholm method we developed is a first step and shows that the eigenfunctions can be described as expansions in terms of the periodic points of the underlying classical system. It eliminates the divergencies associated that the schemes based on smoothings in energy have. However, the exponential divergence of periodic orbits poses a serious practical problem, as discussed in the previous paragraph. This method can only become practical for large $`k_\nu `$ if some way of selecting a few “important” orbits at each value of $`k`$ can be developed. Some results in this direction have been obtained by Vergini and Carlo .
## A Complex phase space
We introduce the following symplectic transformation $`Z`$, depending upon parameter $`\sigma `$, acting on a point of classical phase space $`(q,p)`$ :
$`\left(\begin{array}{cc}z\hfill & \\ p_z\hfill & \end{array}\right)=Z\left(\begin{array}{cc}q\hfill & \\ p\hfill & \end{array}\right)=\left(\begin{array}{cccc}1/\sqrt{2}\sigma \hfill & i\sigma /\sqrt{2}& & \\ i/\sqrt{2}\sigma \hfill & \sigma /\sqrt{2}& & \end{array}\right)\left(\begin{array}{cc}q\hfill & \\ p\hfill & \end{array}\right).`$ (A9)
Imposing reality conditions on the inverse transformation we see that $`\overline{z}=ip_z`$. A linear transformation $`M=\left(\begin{array}{cccc}a\hfill & b& & \\ c\hfill & d& & \end{array}\right)`$ in $`(q,p)`$ phase space has a representation $`M_z`$ in $`(z,p_z)`$ phase space by conjugation with $`Z`$
$`M_z=ZMZ^1=\left(\begin{array}{cccc}\overline{s}\hfill & ir& & \\ i\overline{r}\hfill & s& & \end{array}\right)\text{with}\{\begin{array}{cc}s=\frac{1}{2}\left[(a+d)i\left(\frac{b}{\sigma ^2}\sigma ^2c\right)\right]\hfill & \\ r=\frac{1}{2}\left[(da)+i\left(\frac{b}{\sigma ^2}+\sigma ^2c\right)\right]\hfill & \end{array}.`$ (A14)
This $`(z,p_z)`$ phase space allows a passage to quantum mechanics. This is done in a Hilbert Bargmann space by introducing operators $`𝐳`$ and $`𝐩_𝐳`$ that satisfy the conmutator relations
$`[𝐳,𝐩_𝐳]=i\mathrm{},[𝐳,𝐳]=[𝐩_𝐳,𝐩_𝐳]=0.`$ (A15)
Any vector $`|\psi `$ in Hilbert space can be represented in this new space as $`z|\psi =𝑑qz|qq|\psi `$, where the coherent states are $`z|q=(1/(\pi \mathrm{}\sigma ^2))^{1/4}\mathrm{exp}\left((1/\mathrm{})\left(z^2/2+q^2/(2\sigma ^2)\sqrt{2}zq/\sigma \right)\right)`$. The scalar product is $`\psi _1|\psi _2=\overline{\psi }_1(z)\psi _2(z)𝑑\mu (z)`$ with norm $`d\mu (z)=\frac{1}{\pi }\mathrm{exp}\left(z\overline{z}/\mathrm{}\right)d\text{Re}(z)d\text{Im}(z)`$.
We now define the Husimi representation of a vector $`\psi `$ as
$`_\psi (z){\displaystyle \frac{|z|\psi |^2}{z|z}}.`$ (A16)
It is a real positive function for every $`z`$ in the complex plane.
The representation of a linear symplectic transformation $`M`$ in phase space in terms of a unitary operator of Hilbert Bargmann space is
$`M=\left(\begin{array}{cccc}a\hfill & b& & \\ c\hfill & d& & \end{array}\right)z^{}|𝐔(M)|z={\displaystyle \frac{1}{\sqrt{|\overline{s}}|}}\mathrm{exp}\left({\displaystyle \frac{i}{2}}\text{arg}(\overline{s})\right)\mathrm{exp}\left[{\displaystyle \frac{k}{2\overline{s}}}\left(\overline{r}z^2+2z^{}\overline{z}+r\overline{z}^2\right)\right].`$ (A19)
This representation is up to a phase and its composition law is
$`{\displaystyle z^{}|𝐔(M_1)|zz|𝐔(M_2)|z^{\prime \prime }𝑑\mu (z)}=\lambda (M_1,M_2,M_1M_2)z^{}|𝐔(M_1M_2)|z^{\prime \prime },`$ (A20)
with
$`\lambda (M_1,M_2,M_1M_2)=\mathrm{exp}\left[{\displaystyle \frac{i}{2}}\left(\text{arg}(\overline{s})\text{arg}(\overline{s}_1)\text{arg}(\overline{s}_2)\text{arg}\left({\displaystyle \frac{\overline{s}}{\overline{s}_1\overline{s}_2}}\right)\right)\right]=\pm 1.`$ (A21)
The accumulated phase due to succesive transformations leads to the Maslov index of the trajectory.
In case the phase space shows periodicity in coordinate or momentum, we have to periodize the coherent states as in .
## Figure captions |
warning/0003/hep-th0003047.html | ar5iv | text | # 1.Introduction
## 1.Introduction
The discovery of the $`AdS_{d+1}/CFT_d`$ correspondence brings the role of conformal field theory to a special stage. The type $`II`$B string theory on the $`AdS_5S^5`$ are equivalent to $`𝒩=4`$ super Yang-Mills in Minkowski space-time. However, the calculation of the correlation functions of physical quantities are limited by our knowledge, except for the $`CFT_2`$, saying 2d conformal field theory (CFT) case. With the help of infinitely dimensional symmetries of 2d CFT, much more information can be obtained. For example, Seiberg et al. considered the duality between $`AdS_3`$ and $`CFT_2`$ . As a special case of this duality, a type $`II`$B string theory on $`AdS_3S^3T^4`$ is equivalent to a certain 2d superconformal field theory (SCFT), which corresponds to the IR limit of the dynamics of parallel $`D1`$-branes and $`D5`$-branes. Light-cone gauge quantization of string theories on $`AdS_3`$ are given in .
For 2d CFT, when the central charge of the theory is greater than one, the Virasoro symmetry must be enlarged , or more primary fields should be added, this extended structure is called $`W`$ symmetry. If there are a finite number of primary fields, then the values of central charge $`c`$ and conformal weight ( or spin) $`h`$ take on rational values. there are called rational conformal field theories (RCFT), see for review.
The $`Z_k`$ parafermion (PF) algebra is proposed by Zamolodchikov and Fateev for describing a two-dimensional statistical system with $`Z_k`$ symmetry associating ”spin” variables $`\sigma _r`$ to each node $`r`$ in a (square) lattice $`L`$, the $`\sigma _r`$ take the $`N`$ values $`\omega ^q(q=0,\mathrm{\hspace{0.33em}1},\mathrm{},k1)`$, where $`\omega =exp(2i\pi /k)`$. This generalizes the fermion of the Ising model, which corresponds to the node of $`Z_2`$. It is also known that there are various of statistical models, which can be described by this extended theory, such as the $`3`$-state Potts model ($`k=3`$) , Ashkin-Teller model ($`k=4`$) . The $`Z_4`$ parafermion also gives a consistent 6d string theory .
In fact, parafermion field is important in fractional superstring theory , $`W`$-string theory , furthermore in the compactification of a type $`II`$ string theories on a Calabi-Yau (CY) manifolds and the construction of $`𝒩=2`$ SCFT . Gepner model is the tensor products of $`𝒩=2`$ minimal models with the internal central charge $`c=9`$, which is exactly a solvable models for strings compactified on a CY manifolds. And its applications in $`Dp`$-brane theory are also presented in recently. For example, $`D0`$-branes, the wrapping of $`Dp`$-branes on $`p`$-dimensional supersymmetric cycles leads to BPS saturated. Its dynamics can be analyzed by a Ishibashi boundary states . The Ishibashi boundary state is the RCFT extension of a boundary state of open sting theories. In open sting theories, the boundary must be chosen such that the 2d CFT symmetry is not broken .
$$(L_n\overline{L}_n)|B>=0,$$
(1.1)
here $`|B>`$ is a boundary state. When the extension structure of the CFT forms a RCFT, the RCFT symmetry on the boundary must be hold also. So that the bulk left- and right-moving primary currents $`W`$, $`\overline{W}`$ have to satisfy certain relations on the boundary. On the construction of the boundary state, the Ishibashi states $`|i>>`$ have the following relation,
$$(W_n(1)^{h_W}\overline{W}_n)|i>>=0,$$
(1.2)
where the $`h_W`$ is the conformal dimension of $`W`$. So the explicit expression of $`W`$ algebra is important and helpful for solving this problem. For the intrinsic relations between the Gepner model and the parafermion , the construction of $`W`$ algebra without introducing the free boson has his own advantage.
It is well known that conformal algebra may be obtained from current algebra via Sugawara construction. Similar ways of constructing $`W`$-algebra from current algebras were found by Bais et al. , through simple third order Casimir in level one case and $`GKO`$ coset model in $`SU(N)_1SU(N)_k/SU(N)_{k+1}`$ case. The construction of $`W`$-algebra directly (not by free field realization ) from $`SU(2)_k`$ parafermion was also proposed , in which the so called $`Z`$-algebra technique was used. In a sense of that the generating PFs can be defined through the current algebras by projecting out the Cartan subalgebraic valued components, the $`Z`$-algebra construction may have the most similarity to the Sugawara construction. On the other hand, parafermion is a coset valued field. Thus the parafermion realization of $`W_k`$ algebra for specific level may unify the Sugawara and the coset construction. It has been conjectured that all RCFT can be represented as cosets, and that any CFT can be arbitrary well approximated by a rational theory. So the studying of rational theory has his own interesting. As we know, the bosonization representation of a conformal model provide a much bigger Fock space. We have to use the BRST operator ($`Q_{BRST}^2=0`$ or other restrictions) to project it onto the physical space. Hence there are much more complications if we use the free field realization. Therefore the advantage of our approach is that it avoids the ambiguity and complexity of the bosonization. It is also worthy to find $`W_n`$ algebras for PFs of higher rank group. The reason is the follows. The central charges of $`PFs`$
$$c=\frac{kD}{k+g}r,$$
(1.3)
where $`D`$, $`r`$ are dimension, rank of Lie algebra $`𝒢`$ respectively, and level $`k`$ of $`\widehat{𝒢}`$ is also an integer defining the cyclic symmetry of PFs. In $`SU(2)_k`$ case $`c=2(k1)/(k+2)`$ agrees with a special case in the Fateev-Lykyanov’s $`W`$-algebra series $`c=(k1)[1k(k+1)/p(p+1)]_{p=k+1}`$ . However, there is no known $`W`$-algebra, which is constructed from boson, current algebra or coset model, has the central charge coinciding with the PFs construction of groups with higher rank.
In we gave a construction of Virasoro algebra by using non-local fields (parafermions) which take values on coset space $`G/U(1)^r`$, where $`G`$ is a simply connected compact Lie group manifold, its Lie algebra $`𝒢`$ is a simple one with rank $`r`$. There the so called $`Z`$-algebra technique was used. We also extended this approach to construct the $`W`$-symmetries, $`W_3`$ algebra and $`W_5`$ algebra were obtained from $`SU(2)`$ and $`SU(3)`$, respectively (part results in $`SU(2)`$ case was reconsidered recently in ). While in ref. the construction of $`W_3`$ algebra from the $`SU(3)`$ parafermion was based on a special choice of the root set for summation, and turned out that the $`W`$ algebra were magical closed, while for other choice of root set, the construction was not correct.
As known that in ref. we only obtained the PFs construction of RCFT for $`SU(n)(n3)`$ cases, and further extension of this construction was not succeed. In fact, from $`SU(4)`$ PFs the next and direct goal this construction is failed for spin three primary field. However, it seems that the possibility was not removed at any extent, and the reasons what make these problems arising were unclear at that time. In this paper we will discuss these problems.
The layout of this paper is as follows. In section $`2`$, we recall some basic aspects of extended CFT and the parafermion field, establish our notations and obtain the identities which will be used at a later stage. In section $`3`$, using the relations obtained in section $`2`$, the Virasoro algebra constructed from arbitrary Lie algebra $`𝒢`$ PFs is given, the approach presented here greatly simplifies the calculation in . If we hope that the extension structure of RCFT is nontrivial, certain restriction must be put for the Lie algebra root set $`\mathrm{\Phi }`$ on which the parafermion fields take values. They coincide with the known results for the $`SU(2)`$ and $`SU(3)`$ cases, and get rid of the possibility by this construction from $`SU(4)`$ PFs for spin three primary field. Assuming the condition of the root set is satisfied, in section $`4`$ we obtain a spin $`3`$ primary field very general in simple-laced case, more detailed discussion is given for $`A_l`$ algebra case.
## 2.Brief review of RCFT
In this section we first review some basic aspects of Virasoro algebra, $`W`$ algebra and parafermion field. Then notations which will be used in the sequel are introduced.
The OPE of the stress momentum tensor is
$$T(z)T(w)=\frac{c/2}{(zw)^4}+\frac{2T(w)}{(zw)^2}+\frac{T(w)}{zw}+\mathrm{},$$
(2.1)
the commutator of its modes generate a chiral algebra which is just the Virasoro algebra. To simplify the the expression of OPE, we denote the last equation as,
$$[TT]_4=c/2,[TT]_3=0,[TT]_2=2T,[TT]_2=T,.$$
(2.2)
As mentioned previously in the introduction, in the case of the central charge greater than one, it is necessary to enlarge the symmetry of the CFT by adding primary field with spin great than two . The first nontrivial chiral primary field $`W(z)`$ of conformal dimension $`3`$ (For the left chiral field its conformal dimension is identical with its spin, so we use them without difference.) with the OPE,
$`W(z)W(w)=`$ $`{\displaystyle \frac{c/3}{(zw)^6}}+{\displaystyle \frac{2T(w)}{(zw)^4}}+{\displaystyle \frac{T(w)}{(zw)^3}}`$ (2.3)
$`+{\displaystyle \frac{1}{(zw)^2}}(2b^2\mathrm{\Lambda }(w)+{\displaystyle \frac{3}{10}}^2T)`$
$`+{\displaystyle \frac{1}{zw}}(b^2\mathrm{\Lambda }(w)+{\displaystyle \frac{1}{15}}^3T)+\mathrm{},`$
$`T(z)W(w)=`$ $`{\displaystyle \frac{3W(w)}{(zw)^2}}+{\displaystyle \frac{W(w)}{zw}}+\mathrm{},`$ (2.4)
where $`\mathrm{\Lambda }(z)=[TT]_0(z)\frac{3}{10}^2T(z)`$, and constant $`b^2=16/(22+5c)`$. The primary feature of $`W`$ field is governed by the equation (2.4). Identically, we express the above equation as
$$[WW]_6=c/3,[WW]_5=0,[WW]_4=2T,[WW]_3=T,$$
$$[WW]_2=(2b^2\mathrm{\Lambda }(w)+\frac{3}{10}^2T),[WW]_1=(b^2\mathrm{\Lambda }(w)+\frac{1}{15}^3T),$$
(2.5)
$$[TW]_{6,5,4,3}=0,[TW]_2=3W,[TW]_1=W.$$
(2.6)
Parafermionic currents are primary fields of the 2d CFT. The general parafermion defined for root lattices are proposed in . For a semi-simple Lie algebra $`𝒢`$ there are $`Dr`$ generating parafermion operators $`\psi _\alpha `$, where $`D=dim𝒢`$ and $`r=rank𝒢`$ are dimension and rank of $`𝒢`$, respectively. $`\alpha `$ is a root of $`𝒢`$ (analogy to their counter part in the antiholomorphic sector $`\overline{\psi }_\alpha `$, which will be left out for simplicity). For general parafermion we denote them by a vector in the root lattices $`M`$ mod a lattice $`kML`$, where $`ML`$ is the long root lattices and $`k`$ is a constant identified with the level in the corresponding affine Lie algebra $`\widehat{𝒢}`$ . The generating parafermions are defined through their relationship with current algebra. Thus define the fields
$`\chi _\alpha (z)=\sqrt{{\displaystyle \frac{2k}{\alpha ^2}}}:\psi _\alpha (z)exp(i\alpha \varphi (z)/\sqrt{k}):,`$
$`h_j(z)h_{\alpha _j}(z)={\displaystyle \frac{2i\sqrt{k}}{\alpha _j^2}}\alpha _j_z\varphi (z),`$ (2.7)
for any root $`\alpha `$ and simple root $`\alpha _j`$. We require that the currents $`\chi (z)`$ and $`h(z)`$ (Cartan subgroup valued components) obey the OPE of the current algebra. Because of the mutually semi-local property between the two parafermions, the radial ordering product is a multivalued functions, so we can define the radial order product of (generating) parafermions (PFs) $`\psi _\alpha (z),\psi _\beta (w)`$ ($`\alpha ,\beta `$ are roots of the underlying Lie group)
$`R(\psi _\alpha (z)\psi _\beta (w))`$
$`=\{\begin{array}{cc}\psi _\alpha (z)\psi _\beta (w),\mathrm{},\omega ^{(k1)\alpha \beta }\psi _\alpha (z)\psi _\beta (w),|z|>|w|;\hfill & \\ \omega ^{\alpha \beta /2}\psi _\beta (w)\psi _\alpha (z),\mathrm{}\omega ^{(k1/2)\alpha \beta }\psi _\beta (w)\psi _\alpha (z),|z|<|w|,\hfill & \end{array}`$ (2.10)
where $`\omega =exp(2\pi i/k)`$. The RHS of (2.10) is the requirement of analysis for the field. By using (2.10) one therefore have the following relation for the parafermion fields
$$R\left(\psi _\alpha (z)\psi _\beta (w)\right)(zw)^{\alpha \beta /k}=R\left(\psi _\beta (w)\psi _\alpha (z)\right))(wz)^{\alpha \beta /k}.$$
(2.11)
which is an extension of that for fermion (i.e. $`\alpha \beta =1,k=2`$), and boson (i.e. $`k\mathrm{}`$). We will drop the $`R`$ symbol in the following without confusion. The OPE of the parafermion fields defined by
$`\psi _\alpha (z)\psi _\beta (w)(zw)^{\alpha \beta /k}`$ $`={\displaystyle \frac{\delta _{\alpha ,\beta }}{(zw)^2}}+{\displaystyle \frac{\epsilon _{\alpha ,\beta }/\sqrt{k}}{zw}}\psi _{\alpha +\beta }(w)`$ (2.12)
$`+{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}(zw)^n[\psi _\alpha \psi _\beta ]_n,`$
$`{\displaystyle \underset{n=2}{\overset{\mathrm{}}{}}}(zw)^n[\psi _\alpha \psi _\beta ]_n,`$
Which means that we have
$$[\psi _\alpha \psi _\beta ]_l=0,(l3)[\psi _\alpha \psi _\beta ]_2=\delta _{\alpha ,\beta },[\psi _\alpha \psi _\beta ]_1=\frac{\epsilon _{\alpha ,\beta }}{\sqrt{k}}\psi _{\alpha +\beta }$$
(2.13)
where $`\epsilon _{\alpha ,\beta }`$ is the structure constant of Lie algebra $`𝒢`$ (see more details in the Appendix).
For every field in the parafermion theory there is a pair of charges $`(\lambda ,\overline{\lambda })`$, which take values in the weight lattice. So we denote such field by $`\varphi _{\lambda ,\overline{\lambda }}(z,\overline{z})`$ . The OPE of the $`\psi _\alpha `$ and $`\varphi _{\lambda ,\overline{\lambda }}(z,\overline{z})`$ is given by
$$\psi _\alpha (z)\varphi _{\lambda ,\overline{\lambda }}(w,\overline{w})=\underset{\mathrm{}}{\overset{\mathrm{}}{}}(zw)^{m1\alpha \lambda }A_m^{\alpha ,\lambda }\varphi _{\lambda ,\overline{\lambda }}(w,\overline{w})$$
(2.14)
which means that we define the action of the operator (mode) $`A_m^{\alpha ,\lambda }`$ on $`\varphi _{\lambda ,\overline{\lambda }}(z)`$ by the integration
$$A_m^{\alpha ,\lambda }\varphi _{\lambda ,\overline{\lambda }}(w,\overline{w})=_{c_w}𝑑z(zw)^{m+\alpha \lambda }\psi _\alpha (z)\varphi _{\lambda ,\overline{\lambda }}(w,\overline{w})$$
(2.15)
where $`c_w`$ is the contour around $`w`$, and for simplicity the notation $`𝑑z\frac{dz}{2\pi i}`$ is implied.
Assuming that fields $`A_\alpha `$ and $`B_\beta `$ are arbitrary function of parafermions with parafermion charges $`\alpha `$ and $`\beta `$. The fields are local ($`\alpha `$,or $`\beta =0`$) or semilocal ($`\alpha =\beta `$=root of the underlying Lie algebra). The OPE of them can be written as
$$R(A_\alpha (z)B_\beta (w))(zw)^{\alpha \beta /k}=\underset{n=[h_A+h_B]}{\overset{\mathrm{}}{}}[AB]_n(w)(zw)^n,$$
(2.16)
in which $`[h_A]`$ means the integral part of dimension $`A`$. Hence we have
$$[A_\alpha (z)B_\beta ]_n(w))=_wdzA_\alpha (z)B_\beta (w)(zw)^{n1+\alpha \beta /k}$$
(2.17)
and some relations
$$[A_\alpha (z)B_\beta ]_n(w)=(n+1\alpha \beta /k)[A_\alpha (z)B_\beta ]_{(n+1)}(w)$$
(2.18)
$$[A_\alpha B_\beta ]_n(w)=(n1+\alpha \beta /k)[A_\alpha (z)B_\beta ]_{n1}(w)+[A_\alpha (z)B_\beta ]_n(w)$$
(2.19)
$$[A_\alpha (z)B_\beta ]_n(w)+[A_\alpha B_\beta ]_n(w)=[A_\alpha (z)B_\beta ]_n(w)$$
(2.20)
$`[^nA_\alpha (z)B_\beta ]_0(w)=(n\alpha \beta /k)\mathrm{}(1\alpha \beta /k)[A_\alpha (z)B_\beta ]_n(w)`$
$`{\displaystyle \frac{\mathrm{\Gamma }(n+1\alpha \beta /k)}{\mathrm{\Gamma }(1\alpha \beta /k)}}[A_\alpha (z)B_\beta ]_n(w)`$ (2.21)
in which the $`\mathrm{\Gamma }`$ is the usual $`\mathrm{\Gamma }`$\- unction. It is easy to find a relation between three-fold radial ordering products
$`\{{\displaystyle _w}du{\displaystyle _w}dzR(A(u)R(B(z)C(w)))`$
$`{\displaystyle _w}𝑑z{\displaystyle _w}𝑑u()^{\alpha \beta /k}R(B(z)R(A(u)C(w)))`$
$`{\displaystyle _w}dz{\displaystyle _z}duR(R(A(u)B(z))C(w))\}`$
$`(zw)^{p1+\beta \gamma /k}(uw)^{q1+\gamma \alpha /k}(uz)^{r1+\alpha \beta /k}=0,`$ (2.22)
where the integers $`p,q,r`$ are in the region $`\mathrm{}<p[h_B+h_C],\mathrm{}<q[h_C+h_A],\mathrm{}<r[h_A+h_B]`$, and $`\alpha ,\beta ,\gamma `$ are parafermionic charges of the fields $`A,B,`$ and $`C`$ respectively. This equation is an extension of the identity $`A(BC)B(AC)[A,B]C=0`$. The contours are self evident. Performing the binomial expansion, we can rewrite the last equation as
$`{\displaystyle _w}𝑑u{\displaystyle _w}𝑑zR(A(u)R(B(z)C(w))){\displaystyle \underset{i=p}{\overset{\mathrm{}}{}}}C_{r1+\alpha \beta /k}^{(ip)}`$
$`\times (zw)^{i1+\beta \gamma /k}(uw)^{Q1+(\beta +\gamma )\alpha /k}`$
$`+(1)^r{\displaystyle _w}𝑑z{\displaystyle _w}𝑑uR(B(u)R(A(z)C(w))){\displaystyle \underset{j=q}{\overset{\mathrm{}}{}}}C_{r1+\alpha \beta /k}^{(jq)}`$
$`\times (zw)^{Qj1+\beta (\alpha +\gamma )/k}(uw)^{j1+\gamma )\alpha /k}`$
$`={\displaystyle _w}𝑑z{\displaystyle _z}𝑑uR(R(A(u)B(z))C(w)){\displaystyle \underset{l=r}{\overset{\mathrm{}}{}}}C_{q1+\alpha \gamma /k}^{(lr)}`$
$`\times (zw)^{Ql+(\alpha +\beta )\gamma /k}(uz)^{lr+\beta \alpha /k},`$ (2.23)
From the two equations above we obtain the following Jacobi-like identity relations
$`{\displaystyle \underset{i=p}{\overset{[h_B+h_C]}{}}}C_{r1+\alpha \beta /k}^{(ip)}[A[BC]_i]_{Qi}(w)`$
$`+()^r{\displaystyle \underset{j=q}{\overset{[h_C+h_A]}{}}}C_{r1+\alpha \beta /k}^{(jq)}[B[AC]_j]_{Qj}(w)`$
$`={\displaystyle \underset{k=r}{\overset{[h_B+h_A]}{}}}()^{(kr)}C_{q1+\alpha \gamma /k}^{(kr)}[[AB]_kC]_{Qk}(w),`$ (2.24)
in which $`Q=p+q+r1,C_x^{(l)}=\frac{()^lx(x1)\mathrm{}(xl+1)}{l!}`$, and $`C_0^{(0)}=C_n^{(0)}=C_1^{(l)}=1,C_p^{(l)}=0`$, for $`p,l>0,l>p`$. This identity is important for our usage, we will use it extensively. Performing analytic continuation one more equation is obtained
$$[BA]_r(w)=\underset{i=r}{\overset{[h_A+h_B]}{}}\frac{()^t}{(tr)!}^{tr}[AB]_t(w),$$
(2.25)
and two special cases should be mentioned $`(n0)`$
$$[Aconst.]_n(w)=const.\frac{1}{n!}^nA(w),[const.A]_n(w)=const.\delta _{n,0}B(w).$$
(2.26)
In all of the previous equations $`A,B,C`$ can be compound operators. We can calculate any coefficient in OPE from fundamental equation (2.12).
## 3.PFs constructions CFT
In this section, we present an another approach beside the $`Z`$-algebra technique used in. This approach greatly simplifies the calculation of , and the restriction on the root set arises naturally from the definition of the primary parafermion field. The detailed derivation of OPE of the stress momentum tensor is given.
For the notation conveniences, define $`N_{\alpha ,\beta }\epsilon _{\alpha ,\beta }/\sqrt{k}`$, and the following identities are hold by $`N_{\alpha ,\beta }`$,
$$N_{\alpha ,\beta }=N_{\beta ,\alpha }=N_{\alpha ,\beta }=\frac{(\alpha +\beta )^2}{\beta ^2}N_{\alpha ,\alpha +\beta }.$$
(3.1)
If we only consider the simple-laced case, the results are
$$N_{\alpha ,\beta }=N_{\beta ,\alpha }=N_{\alpha ,\beta }=N_{\alpha ,\alpha +\beta }.$$
(3.2)
Further more, we define $`\tau _\alpha =[\psi _\alpha \psi _\alpha ]_0`$, $`\eta _\alpha =[\psi _\alpha \psi _\alpha ]_1`$, $`\mathrm{\Omega }_\alpha =[\psi _\alpha \psi _\alpha ]_2`$. We calculate the OPE of $`\tau _\alpha `$ with $`\psi _\alpha `$, and $`\tau _\beta `$. The results coincide with the OPE of stress momentum tensor, and the modes of $`\tau _\beta `$ give the Virasoro algebras.
From the definition of $`\tau _\beta `$, and the (2.24), obviously we have,
$`\tau _\alpha =\tau _\alpha ,`$ (3.3)
$`\left[\tau _\alpha \psi _\beta \right]_l=0,(l3),`$ (3.4)
setting $`Q=p=2,q=1,r=0`$ in the (2.24), we have
$`[\tau _\alpha \psi _\beta ]_2`$ (3.5)
$`=[\tau _\alpha \psi _\beta ]_2`$
$`=[\psi _\alpha [\psi _\alpha \psi _\beta ]_2]_0+[\psi _\alpha [\psi _\alpha \psi _\beta ]_1]_1+(1+\alpha ^2/k)[\psi _\alpha [\psi _\alpha \psi _\beta ]_2]_0`$
$`{\displaystyle \frac{\alpha \beta }{k}}[[\psi _\alpha \psi _\alpha ]_1\psi _\beta ]_1+{\displaystyle \frac{\alpha \beta }{2k}}\left(1{\displaystyle \frac{\alpha \beta }{k}}\right)[[\psi _\alpha \psi _\alpha ]_2\psi _\beta ]_0`$
$`=\delta _{\alpha ,\beta }\psi _\alpha +N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }\psi _\beta +(1+\alpha ^2/k)\delta _{\alpha ,\beta }\psi _\alpha `$
$`+{\displaystyle \frac{\alpha \beta }{2k}}(1{\displaystyle \frac{\alpha \beta }{k}})\psi _\beta `$
$`=\delta _{\alpha ,\beta }\psi _\alpha +N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }\psi _\beta `$
$`+(1+\alpha ^2/k)\delta _{\alpha ,\beta }\psi _\alpha {\displaystyle \frac{\alpha \beta }{2k}}(1+{\displaystyle \frac{\alpha \beta }{k}})\psi _\beta .`$ (3.6)
We denote $`\tau =_{\alpha \mathrm{\Phi }}\tau _\alpha `$, where the $`\mathrm{\Phi }`$ is the root set for summation. We require that the set satisfy the conditions
$$\{\mathrm{\Phi }\}\{\mathrm{\Phi }\}=\mathrm{},\{\mathrm{\Phi }\}\{\mathrm{\Phi }\}=\mathrm{\Delta },$$
Obviously the number of roots in $`\mathrm{\Phi }`$ equals the number of ones in $`P`$. In fact, the $`\mathrm{\Phi }`$ can be obtained from $`P`$ by some Weyl reflections. From the equation (3.6) we obtain
$$[\tau \psi _\beta ]_2=\left(1+\underset{\alpha \mathrm{\Phi }}{}(\alpha \beta /2k(\alpha \beta )^2/2k^2+N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta })\right)\psi _\beta ,(\beta \mathrm{\Phi }),$$
(3.7)
where without loosing generality we choose $`\beta \mathrm{\Phi }`$ for convenience, and we will not mention it in the later stage. From the general theory of the conformal fields , we know that the conformal dimension of the parafermion $`\psi _\alpha `$ is $`(1\alpha ^2/2k)`$. We normalize the $`\tau `$ to
$$T=\frac{k}{k+g}\tau ,$$
(3.8)
in which the $`g`$ is the dual Coxeter number, and $`k`$ is the level of the representation of $`\widehat{𝒢}`$, which is the affinization of the classical Lie algebra $`𝒢`$. For a consistent theory, we require
$$[T\psi _\beta ]_2=\left(1\frac{\beta ^2}{2k}\right)\psi _\beta $$
(3.9)
or, equivalently,
$$[\tau \psi _\beta ]_2=\left(1+\frac{2g\beta ^2}{2k}\frac{g\beta ^2}{2k^2}\right)\psi _\beta $$
(3.10)
Comparing the last equation with (3.7), we get the following conditions for set $`\mathrm{\Phi }`$.
$$\underset{\alpha \mathrm{\Phi }}{}\left(\frac{\alpha \beta }{2k}+N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }\right)=\frac{2g\beta ^2}{2k},$$
(3.11)
$$\underset{\alpha \mathrm{\Phi }}{}(\alpha \beta )^2=\underset{\alpha P}{}(\alpha \beta )^2=g\beta ^2.$$
(3.12)
The last two equations are just the consistent condition for PFs construction of CFT. From the definition of $`g`$ we know that the condition (3.12) is satisfied for any given Lie algebra ($`\psi ^2=2`$). While the condition (3.11) brings a constraint on root system of $`𝒢`$. Therefore we get a necessary condition for the root set on which the summation is defined for a consistent theory. From (3.6) we obtain:
$$k\underset{\alpha \mathrm{\Phi }}{}\left(N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }\right)=\beta ^2,$$
(3.13)
while on the other hand, we have
$$k\underset{\alpha \mathrm{\Phi }}{}\left(N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }+N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }\right)=2g2\beta ^2.$$
(3.14)
So we get the solution
$$k\underset{\alpha \mathrm{\Phi }}{}N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }=\frac{2g\beta ^2}{2},$$
(3.15)
$$k\underset{\alpha \mathrm{\Phi }}{}N_{\alpha ,\beta }N_{\alpha ,\alpha \beta }=\frac{2g3\beta ^2}{2},$$
(3.16)
and we have (if $`N_{\alpha ,\beta }0`$, or, $`N_{\alpha ,\beta }0`$)
$$\underset{\alpha \mathrm{\Phi }}{}(\alpha \beta )=0,$$
(3.17)
in which $`\beta `$ is an arbitrary element of the $`\mathrm{\Delta }`$, so we can re-express the last identity as,
$$\underset{\alpha \mathrm{\Phi }}{}\alpha =0,$$
(3.18)
This is the condition for root system $`𝒢`$ on which the summation will be taken over. Which says that for very simple root $`\alpha _i`$ the sum of his height in $`\mathrm{\Phi }`$ must be zero. For $`SU(3)_k`$ as an example $`\mathrm{\Phi }=\{\alpha _1,\alpha _2,\alpha _3=(\alpha _1+\alpha _2,)\}`$, this coincides with the result in . In this paper we only consider the simple-laced cases for simplicity, saying $`\alpha ^2=2`$, and $`\alpha \beta =1`$, if $`\alpha +\beta \mathrm{\Delta }`$. When $`g=2,N_{\alpha ,\beta }=0`$, this is a special case, and we have $`\alpha =\beta `$, this is the $`SU(2)_k`$ ($`Z_k`$ symmetry). please see for more details. While for $`g2`$, we have the following identities:
$$k\underset{\alpha \mathrm{\Phi }}{}N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }=g1,k\underset{\alpha \mathrm{\Phi }}{}N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }=g3,$$
(3.19)
$$k\underset{\alpha \mathrm{\Phi }}{}\alpha \beta N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }=(g1),k\underset{\alpha \mathrm{\Phi }}{}\alpha \beta N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }=g3,$$
(3.20)
$$\underset{\alpha \mathrm{\Phi }}{}(\alpha \beta )^2=2g,\underset{\alpha \mathrm{\Phi }}{}(\alpha \beta )^3=6,\underset{\alpha \mathrm{\Phi }}{}(\alpha \beta )^4=2g+12.$$
(3.21)
The proof of the above identities is very simple. Using these identity (we will not mention them separately), we have
$$[T\psi _\beta ]_2=(11/k)\psi _\beta $$
(3.22)
repeating the same procedure, we have
$$[T\psi _\beta ]_1=\psi _\beta $$
(3.23)
in the process of deriving the last equation, the identity,
$`{\displaystyle \underset{\alpha \mathrm{\Phi }}{}}(N_{\alpha ,\beta }[\psi _\alpha \psi _{\alpha +\beta }]_0+N_{\alpha ,\beta }[\psi _\alpha \psi _{\alpha +\beta }]_0)`$
$`={\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha \mathrm{\Phi }}{}}\left(N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }+N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }\right)\psi _\beta ,`$ (3.24)
is used. We can express the results as the OPE
$$T(z)\psi _\beta (w)=\frac{11/k}{(zw)^2}+\frac{1}{zw}\psi _\beta (w)+\mathrm{}.$$
(3.25)
Repeating the same process for $`T`$, one can get the OPE of the $`T(z)T(w)`$. We leave out the detail of the all, and just give one example of them, saying
$`[\tau _\alpha \tau _\beta ]_2`$ $`=[[\tau _\alpha \psi _\beta ]_1\psi _\beta ]_1+[[\tau _\alpha \psi _\beta ]_2\psi _\beta ]_0+[\psi _\beta [\tau _\alpha \psi _\beta ]_2]_0`$ (3.26)
$`=\delta _{\alpha ,\beta }[\psi _\alpha \psi _\beta ]_0+(1+\alpha ^2/k)\delta _{\alpha ,\beta }[\psi _\beta \psi _\alpha ]_0`$
$`+\delta _{\alpha ,\beta }[\psi _\beta \psi _\alpha ]_0+(1+\alpha ^2/k)\delta _{\alpha ,\beta }[\psi _\alpha \psi _\beta ]_0`$
$`+(N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }+N_{\alpha ,\beta }N_{\alpha ,\alpha +\beta }{\displaystyle \frac{(\alpha \beta )^2}{k^2}})\tau _\beta `$
$`+N_{\alpha ,\beta }[[\psi _\alpha \psi _{\alpha +\beta }]_0\psi _\beta ]_1+N_{\alpha ,\beta }[[\psi _\alpha \psi _{\alpha +\beta }]_0\psi _\beta ]_1`$
$`+{\displaystyle \frac{\alpha \beta }{k}}(1+\alpha ^2/k)(\delta [\psi _\alpha \psi _\beta ]_0\delta _{\alpha ,\beta }[\psi _\alpha \psi _\beta ]_0),`$
consider the summation for $`\alpha `$ and $`\beta `$, we have
$$[TT]_2=2T,$$
(3.27)
the other terms can be derived in the same manner, they read
$$[TT]_4=c/2,[TT]_3=0,[TT]_1=T,$$
(3.28)
equivalently, we can re-express these results as
$$T(z)T(w)=\frac{c/2}{(zw)^4}+\frac{2T}{(zw)^2}+\frac{T(w)}{zw}+\mathrm{},$$
(3.29)
in which the central charge $`c`$ is given by formula (1.3). The last equation is the OPE of the stress momentum tensor. In fact, notice that $`\tau _\alpha =\tau _\alpha `$, and there is only single index $`\alpha `$ needs for summation, we can extend the summation over $`\mathrm{\Phi }`$ to $`\mathrm{\Delta }`$ without difficulty for $`\tau _\alpha `$. And if we choose the root set $`\mathrm{\Delta }`$ for summation, no essential differences will arise, for $`_{\alpha \mathrm{\Delta }}\alpha =0`$. The only difference is the normal constant, which are two times of the original one. So the PFs construction for Virasoro can be extended to any given Lie algebras. Therefore we obtain parafermion representation of the Virasoro algebras underlying any given Lie algebra with arbitrary level $`k`$. It is trivial to extend this construction to semi-simple Lie algebra case, if the condition is satisfied by every copy of the simple Lie algebra.
## 4.PFs realization of RCFT with spin $`3`$
In this section we will construct a spin three field, calculate the OPE of the field with $`T`$. It turns out that the field is a primary field. This is one of the main features of $`W_3`$ algebra. We conjecture that this field is the first primary field in $`W`$ algebras. We then calculate the OPE of the spin $`3`$ field with itself, and in which a spin $`4`$ primary field emerges.
Using the notation introduced in the previous, i.e. $`\eta _\beta [\psi _\beta \psi _\beta ]_1`$, further we define $`w_\beta \eta _\beta \eta _\beta `$, and $`w_3=_{\beta \mathrm{\Phi }}w_\beta `$. Obviously, $`w_\beta =w_\beta `$, $`_{\beta \mathrm{\Phi }}w_\beta =_{\beta \mathrm{\Phi }}w_\beta `$.
Perform the same calculation by properly choosing of $`Q,p,q`$ in the Jacobi-like identity (the final result is the same for different choice, but for certain choice the calculation becomes simpler), and we have,
$$[Tw_3]_5=[Tw_3]_4=[Tw_3]_3=0,$$
(4.1)
while for example, setting $`Q=r=1,q=2`$, we have
$`[T\eta _\beta ]_2`$ $`=[\psi _\beta [T\psi _\beta ]_2]_1+[[T\psi _\beta ]_1\psi _\beta ]_0+[[T\psi _\beta ]_2\psi _\beta ]_1`$ (4.2)
$`=(22/k)\eta _\beta +[\psi _\beta \psi _\beta ]_0`$
$`=3\eta _\beta ,`$
therefore we have,
$$[Tw_\alpha ]_2=3w_\alpha ,or,[Tw_3]_2=3w_3.$$
(4.3)
In the process for deriving those results, some known identities are used without mention. Similarly, we obtain,
$$[Tw_3]_1=w_3.$$
(4.4)
Equivalently, the OPE expression is
$$T(z)w_3(w)=\frac{3w_3(w)}{(zw)^3}+\frac{w_3(w)}{zw}+\mathrm{}$$
(4.5)
and we complete the proof for the $`w_3`$ that it is a spin three primary field.
However, from $`_{\beta \mathrm{\Phi }}w_\beta =_{\beta \mathrm{\Phi }}w_\beta `$, we know that the summation of $`w_\alpha `$ defined on the root set $`\mathrm{\Delta }`$ is identical to zero. It means that we cannot find any extension of the Virasoro algebras ($`W`$-algebra) on the total root system $`\mathrm{\Delta }`$ for any Lie algebras $`𝒢`$. If we expect that such extension to be existence, one part of the roots (we denote it by $`\mathrm{\Phi }`$), on which the spin three field are defined, must be separated out. While the restriction of the central charge require that the number of the elements in $`\mathrm{\Phi }`$ is the half of the ones in $`\mathrm{\Delta }`$, or is the same as in $`P`$. On the other hand, the consistence of the theory brings more constraints on the roots set $`\mathrm{\Phi }`$. We can express them as
Proposition: The parafermion representation of spin three primary field $`w_3`$ is invariant under the Weyl reflection up to a minus one.
$`s(w_3)=\pm w_3,`$ (4.6)
$`s_\beta (w_\alpha )=w_\alpha {\displaystyle \frac{2(\alpha \beta )}{\beta ^2}}w_\beta ,\alpha ,\beta \mathrm{\Phi }.`$ (4.7)
Starting from this, we have the following relation for the height of $`\mathrm{\Phi }`$, it reads,
$$h_\mathrm{\Phi }=h_\mathrm{\Phi }\pm h_\mathrm{\Phi },$$
(4.8)
then the solution of the equation is $`h_\mathrm{\Phi }=0`$, which recover the consistent condition. We used the $`Z`$-algebra technique to construct $`W_3`$-algebra for $`SU(3)`$ PFs in, for simplicity we choose the symmetric roots $`\mathrm{\Phi }=\{\alpha _1,\alpha _2,\alpha _3=(\alpha _1+\alpha _2)\}`$ for summation there. At now we see that, this choice is essential and unique. Here, we have no enough space to list out the calculation in detail. The OPE of $`W_3(z)W_3(w)(g>2)`$ are
$`W_3(z)W_3(w)`$ $`={\displaystyle \frac{c/3}{(zw)^6}}+{\displaystyle \frac{2T}{(zw)^4}}+{\displaystyle \frac{T}{(zw)^3}}`$ (4.9)
$`+{\displaystyle \frac{1}{(zw)^2}}\left(2b^2\mathrm{\Lambda }(w)+{\displaystyle \frac{3}{10}}^2T(w)+V(w)\right)`$
$`+{\displaystyle \frac{1}{zw}}\left(b^2\mathrm{\Lambda }(w)+{\displaystyle \frac{1}{15}}^3T(w)+V(w)\right),`$
$`W_3`$ $`=\left({\displaystyle \frac{k^3}{6(k2)(k+1)(k+g)}}\right)^{1/2}w_3,`$ (4.10)
in which
$`V(z)=`$ $`{\displaystyle \frac{2(4k+3)k^2}{3(k2)(k+1)(k+g)}}{\displaystyle \underset{\alpha \mathrm{\Phi }}{}}(\mathrm{\Omega }_\alpha +\mathrm{\Omega }_\alpha )`$ (4.11)
$`{\displaystyle \frac{2k^2}{(k2)(k+1)(k+g)}}{\displaystyle \underset{\alpha ,\beta \mathrm{\Phi }}{}}\alpha \beta [\tau _\alpha \tau _\beta ]_02b^2[TT]_0`$
$`+\left({\displaystyle \frac{3}{10}}(2b^21){\displaystyle \frac{k}{2(k2)}}\right)^2T.`$
is a spin four primary field, which can be proven from general CFT , or by directly calculation. It is very obviously that $`_{\alpha ,\beta \mathrm{\Delta }}\alpha \beta [\tau _\alpha \tau _\beta ]_0=0`$, and the left parts of $`V`$ is not a primary field anymore. So the definition of $`V`$ cannot be extended to the root set $`\mathrm{\Delta }`$, For $`g=3`$, saying the $`SU(3)_k`$ case, its expression reduce to
$`V(z)=`$ $`{\displaystyle \frac{2(4k+3)k^2}{3(k2)(k+1)(k+3)}}{\displaystyle \underset{\alpha \mathrm{\Phi }}{}}(\mathrm{\Omega }_\alpha +\mathrm{\Omega }_\alpha )`$ (4.12)
$`+{\displaystyle \frac{2k^2}{(k2)(k+1)(k+3)}}{\displaystyle \underset{\alpha \mathrm{\Phi }}{}}[\tau _\alpha \tau _\alpha ]_0`$
$`+\left({\displaystyle \frac{2(k+3)}{3(k2)(k+1)}}2b^2\right)[TT]_0`$
$`+\left({\displaystyle \frac{3}{10}}(2b^21){\displaystyle \frac{k}{2(k2)}}\right)^2T.`$
which is null at $`k=3`$ , so the algebra is closed. The detailed calculation can be fund . However, for $`g4`$ cases, the full solution of this problem is still open for their complexity. In the scene that a spin $`4`$ primary field emerges, the algebra is not closed. However, for higher rank Lie algebra, the central charge, which reflects the character of symmetry is larger. From the general theory of CFT, we know that the more independent primary fields are needed for the larger central charge. From this point view, it is natural that the field is not closed at spin $`3`$. Unfortunately, we do not find an effective approach to calculate the number of the independent primary fields at now. For the well known method to enumerate the independent generating fields is the so called ”character technique” . By that technique, the independent generating fields is the same as the number of independent Casimirs of $`𝒢`$. From this point view, no spin three primary field will emerge in $`SU(2)`$ case, and this is indeed the fact from other approaches. But we have been obtained a $`W_5`$ algebra from the $`SU(2)`$ PFs.
From the above discussion that the root set $`\mathrm{\Phi }`$ forms a closed cycle in the root space. However, for a lot of Lie algebras, this condition cannot be satisfied. In fact, we know that the set $`\mathrm{\Phi }`$ can be obtained from the positive system $`P`$ by some appropriate Weyl reflection. Obviously, this requires the height of the $`P`$ to be:
$$h_p=2\underset{\alpha P}{}n_\alpha ,h_\rho =\underset{\alpha P}{}n_\alpha ,$$
(4.13)
where $`n_\alpha N`$, is the times of the $`\alpha `$ emerging in $`P`$. (4.13) says that, for every simple root $`\alpha _i`$, his times appearing in $`P`$ must be an even number. Obviously, one cannot find such set $`\mathrm{\Phi }`$ for many Lie algebras. For $`A_l`$ algebra, the positive system is $`P=\{\alpha _1,\alpha _2,\mathrm{}\alpha _l,\alpha _1+\alpha _2,\mathrm{},\alpha _{l1}+\alpha _l,\alpha _1+\alpha _2+\alpha _3,\mathrm{},\alpha _{l2}+\alpha _{l1}+\alpha _l,\mathrm{}\mathrm{},\alpha _1+\alpha _2+\mathrm{}+\alpha _{l1}+\alpha _l\}`$, the last one in it is the highest root. It is obviously that, the height of simple roots $`\alpha _j`$ in $`P`$ (the sum of the multiplicities of $`\alpha _j`$ as an element of a root in $`P`$) is $`h_{\alpha _j}(A_l)=j(lj+1)`$. So in general, the set $`\mathrm{\Phi }`$ does not exist for any algebras $`A_{2n+1},(n1)(h_{\alpha _1}=2n+1,h_{\alpha _2}=4n,\mathrm{})`$, and $`\mathrm{\Phi }`$ exists for algebras $`A_{2n}`$ $`(n1)`$ (The height labeled by arbitrary simple root $`\alpha _i`$ is an even number. For example, the height $`h_{\alpha _1}=2n`$ and $`h_{\alpha _2}=2(2n1)`$ ). For $`D_1=A_1`$, $`D_2=A_1A_1`$, so there are no problem in these two cases; while for $`D_3=A_3`$, AND NO solution can be found in this case. For simplicity we leave the discussion of the algebras $`D_l(4)`$, $`E_6,E_7,E_8`$ and non-simple laced algebras to other place.
## 5. Discussion
In this paper, we consider a construction of $`W`$-symmetries (algebras) through a special kind of coset currents, the non-local currents (parafermion), which take values on the coset space $`G/U(1)^r`$, where rank $`r`$ Lie group $`G`$ is limited to a simply connected compact one. It turns out that, the restriction given by the $`W`$-symmetry on the PFs underlying Lie algebra is very stringent. Our extension is very general, all of the previously known results of PFs construction are very simple example of the present discussion. In fact, the present discussion can be generalized to semi-simple Lie algebra case, if the condition of the root set is hold for every copy of its simple Lie algebra.
The important property of OPE is that the singular parts of it can be governed by certain algebra. In PFs construction we use part of the regular terms of the parafermionic OPE , and they satisfy certain algebras also. The property and relation of their higher order terms need further studying.
Because of the the definition of RCFT invariant boundary state is directly relevant to $`W`$-algebra, and the important role played by boundary state in $`Dp`$-brane theory, it is reasonable to expect that $`W`$-algebra (geometry) would be found his position in the $`Dp`$-brane theory studying. The number of independent primary fields are relevant to fusion rule and character. So the problems need to further explore.
For $`\psi _\alpha `$ is primary currents, from the general discussion of the boundary state of open string, it seems that, a kind of new boundary state
$$(\psi _n^\alpha \pm (1)^{\alpha ^2/2k}\psi _n^\alpha )|B>=0.$$
(5.1)
should exist. By adding appropriately chosen $`U(1)`$ current, spin $`1`$ currents or spin $`3/2`$ supercurrents, and their corresponding boundary states can be obtained.
Acknowledgments: One of the authors (Ding) would like to thanks H. Fan, K.J. Shi, Y. K. Lau, S. K. Wang and L. Zhao for fruitful discussion. The work was supported in part by the ”Natural Science Foundation of China” and the ”Project of 973”.
## 6. Appendix
For the convenience of usage, here we recall some basic data of the root system of the Lie algebras.
Any simple Lie algebra can be classified to the classical series $`A_l`$, $`B_l`$, $`C_l`$, $`D_l`$, and the except ones, $`E_6`$, $`E_7`$, $`E_8`$, $`F_4`$ and $`G_2`$. There are at most two different length roots for any of them. For our purposes we choose the basis as
$$[E_\alpha ,E_\beta ]=\epsilon _{\alpha ,\beta }E_{\alpha +\beta }$$
(6.1)
Denote the root set of $`𝒢`$ by $`\mathrm{\Delta }`$, We have $`\epsilon _{\alpha ,\beta }0`$, if $`\alpha +\beta \mathrm{\Delta }`$, $`\epsilon _{\alpha ,\beta }=0`$, if $`\alpha +\beta \mathrm{\Delta }`$, and
$$\epsilon _{\alpha ,\beta }=\epsilon _{\beta ,\alpha }=\epsilon _{\alpha ,\beta }=\frac{(\alpha +\beta )^2}{\beta ^2}\epsilon _{\alpha ,\alpha +\beta }.$$
(6.2)
Let $`\mathrm{\Pi }=\{\alpha _1,\alpha _2,\mathrm{},\alpha _r\}`$ be a simple system of roots of $`𝒢`$, $`r`$ be the rank of $`𝒢`$, and Let $`P`$ be the corresponding positive system, then we have:
(i)$`\{P\}\{P\}=\mathrm{},\{P\}\{P\}=\mathrm{\Delta },\alpha ,\beta P,\alpha +\beta \mathrm{\Delta },\alpha +\beta P`$;
(ii)If, $`1i,jr`$, then $`\alpha _i\beta _jP`$, $`\alpha _i\beta _j0`$;
(iii)If $`\alpha P,\alpha =_i^rm_i\alpha _i,m_i0,`$ while $`\rho `$ is half the sum of the $`P`$, $`\rho =\frac{1}{2}_{\alpha P}\alpha `$. Then the height of the root $`\alpha `$ is $`h_\alpha =_i^rm_i`$; further more we define the total height of the root system $`P`$, $`h_t=_{\alpha P}h_\alpha `$, obviously, $`h_t=2h_\rho `$;
(iv)Weyl reflection $`s_\beta (\alpha )=\alpha \frac{2(\alpha ,\beta )}{(\beta ,\beta )}\beta \alpha \frac{2\alpha \beta }{\beta ^2}\beta `$.
If $`M`$ generate an irreducible representation of $`𝒢`$, Let $`Q_m`$ be the quadratic Casimir operator of the representation, it has an unique highest weight $`\lambda `$, then,
$$Q_m=\lambda (\lambda +2\rho ).$$
(6.3)
For a adjoint representation of the Lie algebra, denoting the accompanying quadratic Casimir operator as $`Q_\psi ,\psi `$ is the highest weight of the adjoint representation. Then we introduce
$$g=Q_\psi /\psi ^2=1+2\rho \psi /\psi ^2,$$
(6.4)
using the data given previous, we have
$$\psi /\psi ^2=\underset{i=1}{\overset{r}{}}m_i\alpha _i/\alpha _{i}^{}{}_{}{}^{2},$$
(6.5)
and so,
$$g=1+\underset{i=1}{\overset{r}{}}m_i=\underset{i=0}{\overset{r}{}}m_i,$$
(6.6)
where $`g`$ is the so-called the dual Coxeter number of the affine Lie algebra. More directly we can use the Freudenthal-de Vries Strange formula:
$$\frac{|\rho |^2}{g}=\frac{Dim𝒢}{12}\frac{D}{12},$$
(6.7)
In the $`ADE`$ cases, $`g=1+h=1+r`$, where the $`h`$ is the height of the highest root. |
warning/0003/math0003079.html | ar5iv | text | # Loops of Lagrangian submanifolds and pseudoholomorphic discs
## 1 Introduction
In this paper we study the Hofer geometry for exact loops of Lagrangian submanifolds of a symplectic manifold $`(M,\omega )`$. Think of such a loop as a submanifold $`\mathrm{\Lambda }S^1\times M`$ such that the projection $`\mathrm{\Lambda }S^1`$ is a submersion and
$$\mathrm{\Lambda }_t:=\{zM|(e^{2\pi it},z)\mathrm{\Lambda }\}$$
is a Lagrangian submanifold of $`M`$ for every $`t`$. The loop is called exact if there exists a Hamiltonian isotopy $`\psi _t`$ of $`M`$ such that $`\psi _t(\mathrm{\Lambda }_0)=\mathrm{\Lambda }_t`$ for every $`t`$. The Hofer length of an exact Lagrangian loop $`\mathrm{\Lambda }`$ is defined by
$$\mathrm{}(\mathrm{\Lambda }):=_0^1\left(\underset{\mathrm{\Lambda }_t}{\mathrm{max}}H_t\underset{\mathrm{\Lambda }_t}{\mathrm{min}}H_t\right)𝑑t,$$
where the Hamiltonian functions $`H_t:M`$ are chosen such that the corresponding Hamiltonian isotopy $`\psi _t:MM`$ satisfies $`\psi _t(\mathrm{\Lambda }_0)=\mathrm{\Lambda }_t`$. It is interesting to minimize the Hofer length over the Hamiltonian isotopy class of $`\mathrm{\Lambda }`$. This infimum will be denoted by
$$\nu (\mathrm{\Lambda })=\nu (\mathrm{\Lambda };M,\omega ):=\underset{\mathrm{\Lambda }\mathrm{\Lambda }^{}}{inf}\mathrm{}(\mathrm{\Lambda }^{}).$$
As an explicit example consider the space $`=(P^n,P^n)`$ of Lagrangian submanifolds of $`P^n`$ that are diffeomorphic to $`P^n`$. It contains the finite dimensional manifold $`\mathrm{PL}(n+1)`$ of projective Lagrangian planes. The space $`\mathrm{PL}(n+1)`$ is the orbit of $`P^n`$ under the action of $`\mathrm{PU}(n+1)`$ and its fundamental group is isomorphic to $`_{n+1}`$. Consider the loop $`\mathrm{\Lambda }^kS^1\times P^n`$ defined by
$$\mathrm{\Lambda }^k:=\underset{t}{}\{e^{2\pi it}\}\times \varphi _{kt}(P^n),$$
(1)
where $`\varphi _t([z_0:\mathrm{}:z_n]):=[e^{\pi it}z_0:z_1:\mathrm{}:z_n]`$ and $`k`$. The loops $`\mathrm{\Lambda }^j`$ and $`\mathrm{\Lambda }^k`$ are homotopic in $`\mathrm{PL}(n+1)`$ (as based loops) if and only if they are Hamiltonian isotopic (as free loops) if and only if $`kj`$ is divisible by $`n+1`$. If $`kj`$ is not divisible by $`n+1`$ then $`\mathrm{\Lambda }^j`$ and $`\mathrm{\Lambda }^k`$ can be distinguished by the Maslov index. More precisely, every Lagrangian loop $`\mathrm{\Lambda }S^1\times P^n`$, with fibres $`\mathrm{\Lambda }_t`$ Lagrangian isotopic to $`P^n`$, has a well defined Maslov index $`\mu (\mathrm{\Lambda })_{n+1}`$. It is defined as the Maslov index of a smooth map $`u:D=\{z||z|1\}M`$ such that $`u(e^{2\pi it})\mathrm{\Lambda }_t`$. Such maps $`u`$ always exist and the Maslov indices of any two such maps differ by an integer multiple of $`n+1`$. It turns out that
$$\mu (\mathrm{\Lambda }^k)k\text{ mod }n+1.$$
(2)
In the case $`n=1`$ the loop $`\mathrm{\Lambda }^1`$ is obtained by rotating a great circle on the 2-sphere through 180 degrees around an axis that passes through the circle. The result is an embedding of the Klein bottle into $`S^1\times S^2`$. The image of this embedding is a Lagrangian submanifold of $`D\times S^2`$ with respect to a suitable symplectic form. In contrast $`\mathrm{\Lambda }^0`$ is a Lagrangian torus in $`D\times S^2`$. In general, the cases where $`n`$ is even and where $`n`$ is odd are topologically different. If $`n`$ is even, then $`\mathrm{\Lambda }^k`$ is diffeomorphic to $`S^1\times P^n`$ for every $`k`$. If $`n`$ is odd then $`\mathrm{\Lambda }^j`$ is diffeomorphic to $`\mathrm{\Lambda }^k`$ if and only if $`kj`$ is even, and $`\mathrm{\Lambda }^k`$ is orientable if and only if $`k`$ is even. In particular, $`\mathrm{\Lambda }^k`$ is diffeomorphic to $`\mathrm{\Lambda }^0=S^1\times P^n`$ whenever $`k`$ is even.
Fix $`k\{1,\mathrm{},n\}`$ and consider the exact Lagrangian loop
$$\mathrm{\Lambda }:=\underset{t}{}\{e^{2\pi it}\}\times \psi _t(P^n),$$
where
$$\psi _t([z_0:\mathrm{}:z_n]):=([z_0:e^{\pi it}z_1:\mathrm{}:e^{\pi it}z_k:z_{k+1}:\mathrm{}:z_n]).$$
This loop is Hamiltonian isotopic to $`\mathrm{\Lambda }^k`$ and it has Hofer length $`1/2`$, whereas $`\mathrm{\Lambda }^k`$ has Hofer length $`k/2`$. The next theorem asserts that $`\mathrm{\Lambda }`$ minimizes the Hofer length in its Hamiltonian isotopy class and hence is a geodesic for the Hofer metric.
Theorem A Let $`\omega \mathrm{\Omega }^2(P^n)`$ denote the Fubini-Study form that satisfies the normalization condition $`_{P^n}\omega ^n=1.`$ Then
$$\nu (\mathrm{\Lambda }^k;P^n,\omega )=\frac{1}{2}$$
for $`k=1,\mathrm{},n`$ and $`\nu (\mathrm{\Lambda }^0)=0`$.
This is a Lagrangian analogue of a theorem by Polterovich about loops of Hamiltonian symplectomorphisms of complex projective space. Following we introduce two other invariants of exact Lagrangian loops $`\mathrm{\Lambda }S^1\times M`$ that can be expressed in terms of Hamiltonian connection $`2`$-forms $`\tau `$ on the trivial bundle $`D\times M`$ that vanish over $`\mathrm{\Lambda }`$. Let $`𝒯(\mathrm{\Lambda })\mathrm{\Omega }^2(D\times M)`$ denote the space of such connection $`2`$-forms. The relative K-area $`\chi (\mathrm{\Lambda })`$ is obtained by minimizing the Hofer norm of the curvature $`\mathrm{\Omega }_\tau `$ over $`𝒯(\mathrm{\Lambda })`$. The third invariant is related to the relative cohomology classes $`[\tau ]H^2(D\times M,\mathrm{\Lambda };)`$ of $`\tau 𝒯(\mathrm{\Lambda })`$. These form a $`1`$-dimensional affine space parallel to the subspace generated by the integral cohomology class $`\sigma :=[dxdy/\pi ]`$. For $`\tau _0,\tau _1𝒯(\mathrm{\Lambda })`$ define $`s(\tau _1,\tau _0)`$ by $`s(\tau _1,\tau _0)\sigma =[\tau _1][\tau _0].`$ The invariant $`\epsilon (\mathrm{\Lambda })`$ is defined by
$$\epsilon (\mathrm{\Lambda }):=\epsilon ^+(\tau _0,\mathrm{\Lambda })\epsilon ^{}(\tau _0,\mathrm{\Lambda }),$$
for $`\tau _0𝒯(\mathrm{\Lambda })`$, where
$$\epsilon ^+(\tau _0,\mathrm{\Lambda }):=inf\{s(\tau ,\tau _0)|\tau 𝒯(\mathrm{\Lambda }),\tau ^{n+1}>0\},$$
$$\epsilon ^{}(\tau _0,\mathrm{\Lambda }):=sup\{s(\tau ,\tau _0)|\tau 𝒯(\mathrm{\Lambda }),\tau ^{n+1}<0\}.$$
Theorem B For every exact Lagrangian loop $`\mathrm{\Lambda }S^1\times M`$
$$\epsilon (\mathrm{\Lambda })\chi (\mathrm{\Lambda })=\nu (\mathrm{\Lambda }).$$
A lower bound for $`\epsilon (\mathrm{\Lambda })`$ can sometimes be obtained by studying pseudoholomorphic sections of $`D\times M`$ with boundary values in $`\mathrm{\Lambda }`$. We assume that the pair $`(M,\mathrm{\Lambda }_0)`$ is monotone and fix a class $`AH_2(D\times M,\mathrm{\Lambda };)`$ that satisfies
$$n\pm \mu _\mathrm{\Lambda }(A)N2,$$
where $`n=dim\mathrm{\Lambda }_0=dimM/2`$, $`N`$ denotes the minimal Maslov number of the pair $`(M,\mathrm{\Lambda }_0)`$, and $`\mu _\mathrm{\Lambda }`$ denotes the Maslov class. Under these assumptions we define Gromov invariants
$$\mathrm{Gr}_A^\pm (\mathrm{\Lambda })H_{n\pm \mu _\mathrm{\Lambda }(A)}(\mathrm{\Lambda }_0;_2).$$
A connection $`2`$-form $`\tau 𝒯(\mathrm{\Lambda })`$ and an $`\omega `$-compatible almost complex structure $`J`$ on $`M`$ determine an almost complex structure $`\stackrel{~}{J}=\stackrel{~}{J}(\tau ,J)`$ on $`D\times M`$. Under our assumptions the moduli space of $`\stackrel{~}{J}(\tau ,\pm J)`$-holomorphic sections of $`D\times M`$ is, for a generic $`\tau `$, a compact smooth manifold of dimension $`n\pm \mu _\mathrm{\Lambda }(A)`$. The Gromov invariant is defined as the image of the mod-2 fundamental class under the evaluation map $`uu(1)`$. Now let $`\mathrm{\Lambda }^kS^1\times P^n`$ be given by (1) with $`1kn`$. Let $`A^\pm H_2(D\times P^n,\mathrm{\Lambda }^k;)`$ be the homology classes of the constant sections $`u^+(x,y)[1:0:\mathrm{}:0]`$ and $`u^{}(x,y)[0:\mathrm{}:0:1]`$.
Theorem C $`\mathrm{Gr}_{A^\pm }^\pm (\mathrm{\Lambda }^k)0`$.
Theorem C can be interpreted as an existence result for pseudoholomorphic sections and we shall use this to prove that $`\epsilon (\mathrm{\Lambda }^k)1/2`$. On the other hand the Hamiltonian isotopy class of $`\mathrm{\Lambda }^k`$ contains a loop of length equal to $`1/2`$. Hence Theorem A follows from Theorem B.
We expect that the same techniques can be used to obtain similar results for general symplectic quotients of $`^n`$ by subgroups of $`\mathrm{U}(n)`$. These quotients will not, in general, satisfy our assumption of monotonicity for the definition of the Gromov invariants. However, it should be possible to derive the same conclusions by using the invariants introduced in Cieliebak–Gaio–Salamon instead. This programme will be carried out elsewhere.
In Polterovich studied the Hofer length of loops $`\psi _t=\psi _{t+1}:MM`$ of Hamiltonian symplectomorphisms. Let $`PS^2`$ denote the Hamiltonian fibration associated to the Hamiltonian loop. Poltervich introduced invariants $`\nu ^\pm (P)`$, $`\chi ^\pm (P)`$, and $`\epsilon ^\pm (P)`$ on which our invariants are modelled. Here $`\nu ^+(P)`$ is obtained by minimizing the positive part of the Hofer length in a given Hamiltonian isotopy class, the K-area $`\chi ^+(P)`$ is a symplectic analogue of an invariant introduced by Gromov , and the invariant $`\epsilon ^+(P)`$ is based on the coupling construction of Guillemin–Lerman–Sternberg . In Polterovich proves that these invariants are equal:
$$\epsilon ^\pm (P)=\chi ^\pm (P)=\nu ^\pm (P).$$
We adopt the convention $`\pm \nu ^\pm (P)0`$. Let us denote by $`\nu (P)`$, $`\chi (P)`$, and $`\epsilon (P)`$ the Hamiltonian analogues of our invariants of Lagrangian loops. These were also considered by Polterovich and he noted that
$$\epsilon (P)=\epsilon ^+(P)\epsilon ^{}(P)=\nu ^+(P)\nu ^{}(P)\nu (P).$$
This is the Hamiltonian analogue of Theorem B. Now consider the Lagrangian loop $`\mathrm{\Lambda }S^1\times \overline{M}\times M`$ given by
$$\mathrm{\Lambda }_t=\mathrm{graph}(\psi _t).$$
The invariants introduced by Polterovich are related to our invariants by
$$\nu (\mathrm{\Lambda })\nu (P),\epsilon (\mathrm{\Lambda })\epsilon (P).$$
The Gromov invariants of the fibration $`P`$ associated to a Hamiltonian loop were independently studied by Seidel and his results were used by Lalonde–McDuff–Polterovich to prove that Hamiltonian loops act trivially on homology. Our results on the Gromov invariants can be viewed as Lagrangian analogues of results in on the Gromov invariants of symplectic fibrations.
The present paper is organized as follows. In Section 2 we discuss background material about the Hofer metric. The space of Lagrangian submanifolds is naturally foliated by Hamiltonian isotopy classes and the Hofer metric is defined on each leaf of this foliation. In Section 3 we introduce the invariants $`\nu (\mathrm{\Lambda })`$, $`\chi (\mathrm{\Lambda })`$, and $`\epsilon (\mathrm{\Lambda })`$ of exact Lagrangian loops and give a proof of Theorem B. In the $`2`$-dimensional case the invariant $`\nu (\mathrm{\Lambda })`$ can sometimes be computed explicitly. This is done in Section 4 for the $`2`$-torus. In Section 5 we introduce the Gromov invariants and in Section 6 we prove Theorems A and C. In Appendix A we prove a result about Hamiltonian isotopy on Riemann surfaces which is used in Section 4.
Acknowledgement: We would like to thank Leonid Polterovich for suggesting the topic and for many helpful discussions.
## 2 The Hofer metric for Lagrangian submanifolds
Let $`(M,\omega )`$ be a $`2n`$-dimensional symplectic manifold and $`L`$ be a compact connected $`n`$-manifold without boundary. Denote by
$$𝒳=\left\{\iota \mathrm{Emb}(L,M)\right|\iota ^{}\omega =0\}$$
the space of Lagrangian embeddings of $`L`$ into $`M`$. The group $`𝒢=\mathrm{Diff}(L)`$ acts on this space by $`\iota \iota \varphi `$ for $`\varphi 𝒢`$. Two Lagrangian embeddings $`\iota _0,\iota _1𝒳`$ lie in the same $`𝒢`$-orbit if and only if they have the same image $`\mathrm{\Lambda }=\iota _0(L)=\iota _1(L)`$. Hence the quotient space
$$:=𝒳/𝒢$$
can be naturally identified with the set of Lagrangian submanifolds of $`M`$ that are diffeomorphic to $`L`$. A function $`:t\mathrm{\Lambda }_t`$ is called smooth if there exists a smooth function $`\times LM:(t,q)\iota _t(q)`$ such that $`\iota _t(L)=\mathrm{\Lambda }_t`$ for all $`t`$. One can think of $``$ as an infinite dimensional manifold.
###### Lemma 2.1
The tangent space of $``$ at a point $`\mathrm{\Lambda }`$ can be naturally identified with the space of closed $`1`$-forms on $`\mathrm{\Lambda }`$:
$$T_\mathrm{\Lambda }=\left\{\beta \mathrm{\Omega }^1(\mathrm{\Lambda })\right|d\beta =0\}$$
Proof: Let $`\times LM:(t,q)\iota _t(q)`$ be a smooth function such that $`\iota _t𝒳`$ for all $`t`$ and define
$$\alpha _t:=\omega (v_t,d\iota _t)\mathrm{\Omega }^1(L),v_t:=_t\iota _t𝒞^{\mathrm{}}(L,\iota _{t}^{}{}_{}{}^{}TM).$$
(3)
Then
$$0=_t\iota _{t}^{}{}_{}{}^{}\omega =d\alpha _t$$
and hence the tangent space of $`𝒳`$ at $`\iota `$ is given by
$$T_\iota 𝒳=\{v𝒞^{\mathrm{}}(L,\iota ^{}TM)|\omega (v,d\iota )\mathrm{\Omega }^1(L)\text{ is closed}\}.$$
The tangent space to the $`𝒢`$-orbit consists of all vector fields of the form $`v=d\iota \xi `$, where $`\xi \mathrm{Vect}(L)`$. The map $`v\omega (v,d\iota )`$ identifies the quotient space $`T_\iota 𝒳/T_\iota (\iota 𝒢)`$ with the space of closed $`1`$-forms on $`L`$.
If $`\iota _t,\iota _t^{}𝒳`$ are two smooth paths in $`𝒳`$ that satisfy $`\iota _t^{}=\iota _t\varphi _t`$ for some path $`\varphi _t𝒢`$ then the vector fields $`v_t:=_t\iota _t`$ and $`v_t^{}:=_t\iota _t^{}`$ are related by
$$v_t^{}=v_t\varphi _t+d\iota _t\xi _t\varphi _t$$
where $`\xi _t\mathrm{Vect}(L)`$ generates the diffeomorphism $`\varphi _t`$ via $`_t\varphi _t=\xi _t\varphi _t.`$ Hence the $`1`$-forms $`\alpha _t:=\omega (v_t,d\iota _t)`$ and $`\alpha _t^{}:=\omega (v_t^{},d\iota _t^{})`$ are related by
$$\alpha _t^{}=\varphi _{t}^{}{}_{}{}^{}\alpha _t.$$
Hence two closed $`1`$-forms $`\alpha ,\alpha ^{}\mathrm{\Omega }^1(L)`$ corresponding to two Lagrangian embeddings $`\iota `$ and $`\iota ^{}=\iota \varphi `$ represent the same tangent vector of $``$ if and only if $`\alpha ^{}=\varphi ^{}\alpha `$ or, equivalently, $`\iota _{}\alpha =\iota _{}^{}{}_{}{}^{}\alpha ^{}`$. This proves the lemma. $`\mathrm{}`$
Let $`:t\mathrm{\Lambda }_t`$ be a smooth path of Lagrangian submanifolds. We define the derivative of this path at time $`t`$ by
$$_t\mathrm{\Lambda }_t:=\iota _{t}^{}{}_{}{}^{}\alpha _t,$$
where the path $`𝒳:t\iota _t`$ is chosen such that $`\iota _t(L)=\mathrm{\Lambda }_t`$ for every $`t`$ and $`\alpha _t`$ is defined by (3). The proof of Lemma 2.1 shows that the $`1`$-form $`\beta _t=\iota _{t}^{}{}_{}{}^{}\alpha _t\mathrm{\Omega }^1(\mathrm{\Lambda }_t)`$ is closed and is independent of the choice of the lift $`t\iota _t`$ used to define it.
We wish to study Hamiltonian isotopies of Lagrangian submanifolds. This corresponds to paths in $``$ that are tangent to the subbundle
$$=\left\{(\mathrm{\Lambda },\beta )T\right|\mathrm{\Lambda },\beta \mathrm{\Omega }^1(\mathrm{\Lambda })\text{ is exact}\}.$$
Abstractly, one can think of $``$ as a distribution on $``$. It follows from Weinstein’s Lagrangian neighbourhood theorem that this distribution is integrable. We shall see that the leaf through $`\mathrm{\Lambda }_0`$ consists of all Lagrangian submanifolds of $`M`$ that are Hamiltonian isotopic to $`\mathrm{\Lambda }_0`$ . To be more precise, let $`\times M:(t,z)H_t(z)`$ be a smooth Hamiltonian function and denote by $`\times MM:(t,z)\psi _t(z)`$ the Hamiltonian isotopy generated by $`H`$ via
$$\frac{d}{dt}\psi _t=X_t\psi _t,\iota (X_t)\omega =dH_t,\psi _0=\mathrm{id}.$$
(4)
###### Lemma 2.2
Let $`:t\mathrm{\Lambda }_t`$ be a smooth path of Lagrangian submanifolds and $`\psi _t`$ be a Hamiltonian isotopy on $`M`$ generated by the Hamiltonian functions $`H_t:M`$ via (4). Then $`\mathrm{\Lambda }_t=\psi _t(\mathrm{\Lambda }_0)`$ for every $`t`$ if and only if
$$_t\mathrm{\Lambda }_t=dH_t|_{\mathrm{\Lambda }_t}$$
for every $`t`$.
Proof: Choose a smooth path $`𝒳:t\iota _t`$ such that $`\iota _t(L)=\mathrm{\Lambda }_t`$ for every $`t`$ and let $`\alpha _t\mathrm{\Omega }^1(L)`$ be defined by (3). Then $`_t\mathrm{\Lambda }_t=dH_t|_{\mathrm{\Lambda }_t}`$ if and only if $`d(H_t\iota _t)=\alpha _t`$. It follows from the definitions that this is equivalent to
$$X_t(\iota _t(q))_t\iota _t(q)\mathrm{im}d\iota _t(q)$$
for all $`t`$ and all $`q`$. This means that there exists a smooth family of vector fields $`\xi _t\mathrm{Vect}(L)`$ such that
$$X_t\iota _t=_t\iota _t+d\iota _t\xi _t.$$
Equivalently, $`\psi _t\iota _0=\iota _t\varphi _t,`$ where the isotopy $`\varphi _t\mathrm{Diff}(L)`$ is generated by $`\xi _t`$ via $`_t\varphi _t=\xi _t\varphi _t`$ and $`\varphi _0=\mathrm{id}.`$ This proves the lemma. $`\mathrm{}`$
The previous lemma shows that every path in $``$ that is generated by a Hamiltonian isotopy is tangent to $``$. The converse is proved next.
###### Lemma 2.3
A smooth path $`[0,1]:t\mathrm{\Lambda }_t`$ is tangent to $``$ if and only if there exists a Hamiltonian isotopy $`t\psi _t`$ such that $`\psi _t(\mathrm{\Lambda }_0)=\mathrm{\Lambda }_t`$ for every $`t`$.
Proof: The “if” part was proved in Lemma 2.2. Suppose that the path $`t\mathrm{\Lambda }_t`$ is tangent to $``$. Choose a smooth function $`[0,1]𝒳:t\iota _t`$ such that $`\iota _t(L)=\mathrm{\Lambda }_t`$ for every $`t`$ and let $`\alpha _t\mathrm{\Omega }^1(L)`$ be defined by (3). By assumption, $`\alpha _t`$ is exact for every $`t`$. Fix a smooth path $`q_tL`$ and, for every $`t`$, choose $`h_t:L`$ such that
$$dh_t=\alpha _t,h_t(q_t)=0.$$
Then the function $`\times L:(t,q)h_t(q)`$ is smooth. We construct a smooth function $`[0,1]\times M:(t,z)H_t(z)`$ such that
$$H_t\iota _t=h_t.$$
(5)
Choose an almost complex structure $`J`$ on $`M`$ that is compatible with $`\omega `$. Let $`\epsilon >0`$ be so small that, for every $`t[0,1]`$, the map
$$T\mathrm{\Lambda }_tM:(z,v)\mathrm{exp}_z(Jv)$$
restricts to a diffeomorphism from the $`\epsilon `$-neighbourhood of the zero section in $`T\mathrm{\Lambda }_t`$ onto the open neighbourhood
$$U_t:=\left\{\mathrm{exp}_z(Jv)\right|z\mathrm{\Lambda }_t,vT_z\mathrm{\Lambda }_t,|v|<\epsilon \}$$
of $`\mathrm{\Lambda }_t`$ in $`M`$. Choose a cutoff function $`\rho :[0,\epsilon ][0,1]`$ such that $`\rho (r)=1`$ for $`r<\epsilon /3`$ and $`\rho (r)=0`$ for $`r>2\epsilon /3`$. Define $`H_t:M`$ by
$$H_t(\mathrm{exp}_z(Jv)):=\rho (|v|)h_t\iota _{t}^{}{}_{}{}^{1}(z)$$
for $`z\mathrm{\Lambda }_t`$ and $`vT_z\mathrm{\Lambda }_t`$ with $`|v|<\epsilon `$, and by $`H_t(z):=0`$ for $`zMU_t`$. Then $`H_t`$ satisfies (5) and hence
$$dH_t|_{\mathrm{\Lambda }_t}=\iota _{t}^{}{}_{}{}^{}dh_t=\iota _{t}^{}{}_{}{}^{}\alpha _t=_t\mathrm{\Lambda }_t.$$
By Lemma 2.2, the Hamiltonian isotopy $`\psi _t`$ generated by $`H_t`$ satisfies $`\psi _t(\mathrm{\Lambda }_0)=\mathrm{\Lambda }_t`$ for every $`t`$. This proves the lemma. $`\mathrm{}`$
###### Remark 2.4
The Hamiltonian functions constructed in Lemma 2.3 satisfy
$$\mathrm{max}H_t=\mathrm{max}h_t,\mathrm{min}H_t=\mathrm{min}h_t$$
(6)
for every $`t`$. With a slightly more sophisticated argument one can show that the Hamiltonian functions can be chosen such that the Hamiltonian vector fields $`X_t`$ satisfy $`_t\iota _t=X_t\iota _t`$ and hence the resulting Hamiltonian isotopy satisfies
$$\psi _t\iota _0=\iota _t.$$
(7)
However, in general there does not exist a Hamiltonian isotopy that satisfies both (6) and (7).
###### Lemma 2.5
Let $`:t\mathrm{\Lambda }_t`$ be a smooth path of Lagrangian submanifolds. Let $`\mathrm{Diff}(M,\omega ):t\psi _t`$ be a symplectic isotopy and define $`\beta _t\mathrm{\Omega }^1(M)`$ by $`\beta _t:=\iota (Y_t)\omega ,`$ where $`_t\psi _t=Y_t\psi _t.`$ Then $`\beta _t`$ is closed and the path $`\mathrm{\Lambda }_t^{}:=\psi _{t}^{}{}_{}{}^{1}(\mathrm{\Lambda }_t)`$ satisfies
$$_t\mathrm{\Lambda }_t^{}=\psi _{t}^{}{}_{}{}^{}\left(_t\mathrm{\Lambda }_t\beta _t|_{\mathrm{\Lambda }_t}\right).$$
Proof: Choose a lift $`𝒳:t\iota _t`$ of $`t\mathrm{\Lambda }_t`$ and denote
$$\iota _t^{}:=\psi _{t}^{}{}_{}{}^{1}\iota _t,\alpha _t:=\omega (_t\iota _t,d\iota _t),\alpha _t^{}:=\omega (_t\iota _t^{},d\iota _t^{}).$$
Then $`\alpha _t^{}=\alpha _t\iota _{t}^{}{}_{}{}^{}\beta _t`$ and hence
$$_t\mathrm{\Lambda }_t^{}=\iota _{t}^{}{}_{}{}^{}\alpha _t^{}=\psi _{t}^{}{}_{}{}^{}\iota _{t}^{}{}_{}{}^{}\alpha _t^{}=\psi _{t}^{}{}_{}{}^{}\left(\iota _{t}^{}{}_{}{}^{}\alpha _t\beta _t\right)=\psi _{t}^{}{}_{}{}^{}\left(_t\mathrm{\Lambda }_t\beta _t\right)$$
as claimed. $`\mathrm{}`$
The subbundle $`T`$ carries a natural norm. Following Hofer we define the norm of an exact $`1`$-form $`\alpha =dh\mathrm{\Omega }^1(\mathrm{\Lambda })`$ by
$$dh:=\mathrm{max}h\mathrm{min}h.$$
This norm gives rise to a distance function on each leaf of the foliation determined by $``$. Let $`_0`$ be such a leaf. By Lemma 2.3, $`_0`$ is the Hamiltonian isotopy class of any Lagrangian submanifold $`\mathrm{\Lambda }_0`$. Let $`[0,1]_0:t\mathrm{\Lambda }_t`$ be a smooth path in $`_0`$. The length of this path is defined by
$$\mathrm{}(\{\mathrm{\Lambda }_t\}):=_0^1_t\mathrm{\Lambda }_t𝑑t.$$
Lemma 2.3 and Remark 2.4 show that
$$\mathrm{}(\{\mathrm{\Lambda }_t\})=\underset{\psi _t(\mathrm{\Lambda }_0)=\mathrm{\Lambda }_t}{inf}\mathrm{}(\{\psi _t\}),$$
(8)
where the infimum runs over all Hamiltonian isotopies $`t\psi _t`$ that satisfy $`\psi _t(\mathrm{\Lambda }_0)=\mathrm{\Lambda }_t`$ for all $`t`$ and $`\mathrm{}(\{\psi _t\})`$ denotes the Hofer length (cf. ).
Now let $`\mathrm{\Lambda },\mathrm{\Lambda }^{}_0`$ and denote by $`𝒫(\mathrm{\Lambda },\mathrm{\Lambda }^{})`$ the space of all smooth paths $`[0,1]_0:t\mathrm{\Lambda }_t`$ that connect $`\mathrm{\Lambda }_0=\mathrm{\Lambda }`$ to $`\mathrm{\Lambda }_1=\mathrm{\Lambda }^{}`$. The distance between $`\mathrm{\Lambda }`$ and $`\mathrm{\Lambda }^{}`$ is defined by
$$d(\mathrm{\Lambda },\mathrm{\Lambda }^{}):=\underset{\{\mathrm{\Lambda }_t\}𝒫(\mathrm{\Lambda },\mathrm{\Lambda }^{})}{inf}\mathrm{}(\{\mathrm{\Lambda }_t\}).$$
(9)
It follows immediately from (8) that
$$d(\mathrm{\Lambda },\mathrm{\Lambda }^{})=\underset{\psi (\mathrm{\Lambda })=\mathrm{\Lambda }^{}}{inf}d(\mathrm{id},\psi )$$
(10)
where the infimum runs over all Hamiltonian symplectomorphisms $`\psi `$ of $`M`$ that satisfy $`\psi (\mathrm{\Lambda })=\mathrm{\Lambda }^{}`$ and $`d(\mathrm{id},\psi )`$ denotes the Hofer distance (cf. ). The function (9) is obviously nonnegative, symmetric, and satisfies the triangle inequality. That it defines a metric is a deep theorem due to Chekanov .
###### Theorem 2.6 (Chekanov)
If $`\mathrm{\Lambda }\mathrm{\Lambda }^{}`$ then $`d(\mathrm{\Lambda },\mathrm{\Lambda }^{})>0`$.
###### Remark 2.7
In Milinković studied geodesics in the space of Lagrangian submanifolds. Generalizing a result by Bialy and Polterovich , he proved that the distance of two exact Lagrangian submanifolds $`\mathrm{\Lambda }=\mathrm{graph}(dS)`$ and $`\mathrm{\Lambda }^{}=\mathrm{graph}(dS^{})`$ of the cotangent bundle $`T^{}L`$ is given by
$$d(\mathrm{\Lambda },\mathrm{\Lambda }^{})=d(SS^{}).$$
## 3 Invariants of Lagrangian loops
In this section we shall consider exact loops of Lagrangian submanifolds. In the terminology of the previous section this corresponds to loops inside a leaf of the foliation of $``$ determined by $``$. We shall construct three invariants of Hamiltonian isotopy classes of such loops and study the relations between them.
### 3.1 The minimal length
Continue the notation of Section 2. A Lagrangian loop in $`M`$ is a smooth function $`:t\mathrm{\Lambda }_t`$ such that
$$\mathrm{\Lambda }_{t+1}=\mathrm{\Lambda }_t$$
for all $`t`$. Such a loop determines a subset $`\mathrm{\Lambda }S^1\times M`$ defined by
$$\mathrm{\Lambda }:=\left\{(e^{2\pi it},z)\right|t,z\mathrm{\Lambda }_t\}.$$
(11)
Note that a loop $`:t\mathrm{\Lambda }_t`$ is smooth if and only if this set $`\mathrm{\Lambda }`$ is a smooth submanifold of $`S^1\times M`$. We shall frequently identify the loop $`:t\mathrm{\Lambda }_t`$ with the corresponding submanifold $`\mathrm{\Lambda }S^1\times M`$.
A Lagrangian loop $`t\mathrm{\Lambda }_t`$ is called exact if it is tangent to $``$, i.e. $`_t\mathrm{\Lambda }_t\mathrm{\Omega }^1(\mathrm{\Lambda }_t)`$ is exact for every $`t`$. Two exact Lagrangian loops $`t\mathrm{\Lambda }_t`$ and $`t\mathrm{\Lambda }_t^{}`$ are called Hamiltonian isotopic if there exists a smooth function $`[0,1]\times :(s,t)\mathrm{\Lambda }_{s,t}`$ such that
$$\mathrm{\Lambda }_{0,t}=\mathrm{\Lambda }_t,\mathrm{\Lambda }_{1,t}=\mathrm{\Lambda }_t^{},$$
the map $`t\mathrm{\Lambda }_{s,t}`$ is an exact Lagrangian loop for every $`s`$, and $`_s\mathrm{\Lambda }_{s,t}\mathrm{\Omega }^1(\mathrm{\Lambda }_{s,t})`$ is exact for all $`s`$ and $`t`$. Here the function $`[0,1]\times :(s,t)\mathrm{\Lambda }_{s,t}`$ is called smooth if there exists a smooth function $`[0,1]\times \times LM:(s,t,q)\iota _{s,t}(q)`$ such that $`\iota _{s,t}(L)=\mathrm{\Lambda }_{s,t}`$ for all $`s`$ and $`t`$. Let $`\mathrm{\Lambda },\mathrm{\Lambda }^{}S^1\times M`$ be two exact Lagrangian loops. We write $`\mathrm{\Lambda }\mathrm{\Lambda }^{}`$ iff $`\mathrm{\Lambda }`$ is Hamiltonian isotopic to $`\mathrm{\Lambda }^{}`$. A Hamiltonian isotopy class corresponds to a component in the free loop space of a leaf $`_0`$ of the foliation determined by $``$. To every such Hamiltonian isotopy class we assign the real number
$$\nu (\mathrm{\Lambda }):=\underset{\mathrm{\Lambda }^{}\mathrm{\Lambda }}{inf}\mathrm{}(\mathrm{\Lambda }^{}).$$
So $`\nu (\mathrm{\Lambda })`$ is obtained by minimizing the Hofer length over all exact Lagrangian loops that are Hamiltonian isotopic to $`\mathrm{\Lambda }`$.
### 3.2 The relative K-area
Following Polterovich we introduce the notion of relative K-area. This invariant is defined in terms of Hamiltonian connections on the symplectic fibre bundle $`D\times MD`$ that preserve the subbundle $`\mathrm{\Lambda }D\times M`$ defined by (11). Here $`D`$ denotes the closed unit disc. We begin by recalling the basic notions of symplectic connections and curvature (cf. ). Think of a connection on $`D\times M`$ as a horizontal distribution. Any such connection is determined by a connection $`2`$-form on $`D\times M`$ of the form
$$\tau =\omega +\alpha dx+\beta dy+fdxdy$$
where $`\alpha =\alpha _{x,y}\mathrm{\Omega }^1(M)`$, $`\beta =\beta _{x,y}\mathrm{\Omega }^1(M)`$, and $`f=f_{x,y}\mathrm{\Omega }^0(M)`$ depend smoothly on $`x+iyD`$. The horizontal subspace is the $`\tau `$-orthogonal complement of the vertical subspace. Explicitly, the horizontal lifts of $`/x`$ and $`/y`$ at $`(x+iy,z)D\times M`$ are the vectors $`(1,X_{x,y}(z))`$ and $`(i,Y_{x,y}(z))`$, respectively, where the vector fields $`X=X_{x,y},Y=Y_{x,y}\mathrm{Vect}(M)`$ are defined by
$$\iota (X)\omega =\alpha ,\iota (Y)\omega =\beta .$$
Thus the connection associated to $`\tau `$ is independent of $`f`$. It is called symplectic if $`\alpha _{x,y}`$ and $`\beta _{x,y}`$ are closed for all $`x+iyD`$, and Hamiltonian if $`\alpha _{x,y}`$ and $`\beta _{x,y}`$ are exact for all $`x+iyD`$ and $`\tau `$ is closed.<sup>1</sup><sup>1</sup>1 In a connection is called Hamiltonian if parallel transport along every loop in the base is a Hamiltonian symplectomorphism. In the case of a simply connected base this is equivalent to the existence of a closed $`2`$-form $`\tau `$ that represents this connection. In contrast, we call a connection Hamiltonian if parallel transport along every path is a Hamiltonian symplectomorphism. This notion only makes sense when the bundle in question is equipped with a trivialization. Thus a Hamiltonian connection $`2`$-form has the form
$$\tau =\omega +dFdx+dGdy+(_xG_yF+c)dxdy,$$
(12)
where $`F,G:D\times M`$ and $`c:D`$ are smooth maps such that the functions $`F_{x,y}=F(x+iy,)`$ and $`G_{x,y}=G(x+iy,)`$ have mean value zero:
$$_MF_{x,y}\omega ^n=_MG_{x,y}\omega ^n=0.$$
In (12) the $`d`$ in $`dF`$ denotes the differential on $`M`$, i.e. $`dF`$ denotes the smooth family $`x+iydF_{x,y}`$ of $`1`$-forms on $`M`$, and similarly for $`dG`$. We shall only consider Hamiltonian connections with the property that parallel transport along the boundary preserves $`\mathrm{\Lambda }`$.
###### Lemma 3.1
Let $`\tau `$ be a Hamiltonian connection $`2`$-form on $`D\times M`$ of the form (12) and denote
$$H_t:=2\pi \mathrm{sin}(2\pi t)F_{\mathrm{cos}(2\pi t),\mathrm{sin}(2\pi t)}+2\pi \mathrm{cos}(2\pi t)G_{\mathrm{cos}(2\pi t),\mathrm{sin}(2\pi t)}.$$
(13)
Let $`:t\mathrm{\Lambda }_t`$ be an exact Lagrangian loop, let $`\mathrm{\Lambda }D\times M`$ be defined by (11), and choose a smooth function $`\iota :\times LM`$ such that $`\iota _t(L)=\mathrm{\Lambda }_t`$, where $`\iota _t:=\iota (t,)`$. Then the following are equivalent.
Parallel transport of $`\tau `$ along the boundary preserves $`\mathrm{\Lambda }`$.
$`\iota ^{}\tau =0`$.
$`dH_t|_{\mathrm{\Lambda }_t}=_t\mathrm{\Lambda }_t`$ for every $`t`$.
Proof: The parallel transport of $`\tau `$ along a curve $`tx(t)+iy(t)`$ is determined by the Hamiltonian functions
$$H_t=\dot{x}(t)F_{x(t),y(t)}+\dot{y}(t)G_{x(t),y(t)}$$
via (4). The functions $`H_t`$ in (13) correspond to the path $`te^{2\pi it}`$. By Lemma 2.2, the Hamiltonian isotopy determined by $`H_t`$ preserves $`\mathrm{\Lambda }`$ if and only if $`dH_t|_{\mathrm{\Lambda }_t}=_t\mathrm{\Lambda }_t`$ for every $`t`$. This shows that (i) is equivalent to (iii).
To prove the equivalence of (ii) and (iii) note that
$$\tau (_t\iota _t,d\iota _t)=\omega (_t\iota _tX_t\iota _t,d\iota _t),$$
where $`X_t\mathrm{Vect}(M)`$ denotes the Hamiltonian vector field of $`H_t`$ as in (4). The right hand side vanishes if and only if $`dH_t|_{\mathrm{\Lambda }_t}=_t\mathrm{\Lambda }_t`$ and the left hand side vanishes if and only if $`\iota ^{}\tau =0`$. This proves the lemma. $`\mathrm{}`$
For every exact Lagrangian loop $`:t\mathrm{\Lambda }_t`$ let us denote the set of Hamiltonian connections that preserve $`\mathrm{\Lambda }`$ by
$$𝒯(\mathrm{\Lambda })=\left\{\tau \mathrm{\Omega }^2(D\times M)\right|\tau \text{ has the form }(\text{12}),\tau |_{T\mathrm{\Lambda }}=0\}.$$
We shall prove in Lemma 3.2 below that this set is nonempty. Let $`:t\mathrm{\Lambda }_t^{}`$ be another exact Lagrangian loop. A diffeomorphism
$$\mathrm{\Psi }:(D\times M,\mathrm{\Lambda })(D\times M,\mathrm{\Lambda }^{})$$
is called a fibrewise (Hamiltonian) symplectomorphism if it has the form $`\mathrm{\Psi }(x+iy,z)=(x+iy,\psi _{x,y}(z)),`$ where $`\psi _{x,y}:MM`$ is a (Hamiltonian) symplectomorphism for all $`x,y`$. In the case $`\mathrm{\Lambda }=\mathrm{\Lambda }^{}`$ we denote by $`𝒢(\mathrm{\Lambda })`$ the group of fibrewise Hamiltonian symplectomorphisms of $`(D\times M,\mathrm{\Lambda })`$. This group acts on $`𝒯(\mathrm{\Lambda })`$ by $`\tau \mathrm{\Psi }^{}\tau `$. The curvature of a connection $`2`$-form $`\tau `$ of the form (12) is the function $`\mathrm{\Omega }_\tau :D\times M`$ defined by
$$\mathrm{\Omega }_\tau (x,y,z):=\{F_{x,y},G_{x,y}\}(z)+_yF_{x,y}(z)_xG_{x,y}(z)$$
(14)
for $`x+iyD`$ and $`zM`$. It is sometimes useful to think of the curvature as a $`2`$-form $`\mathrm{\Omega }_\tau dxdy`$ on $`D\times M`$ rather than a function.
###### Lemma 3.2
For every exact Lagrangian loop $`:t\mathrm{\Lambda }_t`$ the set $`𝒯(\mathrm{\Lambda })`$ is nonempty.
Two exact Lagrangian loops $`\mathrm{\Lambda }`$ and $`\mathrm{\Lambda }^{}`$ are Hamiltonian isotopic if and only if the corresponding pairs $`(D\times M,\mathrm{\Lambda })`$ and $`(D\times M,\mathrm{\Lambda }^{})`$ are fibrewise Hamiltonian symplectomorphic.
If $`\tau `$ is a Hamiltonian connection $`2`$-form on $`D\times M`$ and $`\mathrm{\Psi }`$ is a fibrewise Hamiltonian symplectomorphism of $`D\times M`$ then
$$\mathrm{\Omega }_{\mathrm{\Psi }^{}\tau }=\mathrm{\Omega }_\tau \mathrm{\Psi }.$$
Proof: Let $`\varphi _t\mathrm{Diff}(L)`$ be defined by
$$\iota _{t+1}\varphi _t=\iota _t.$$
Since $`L`$ is connected there exists a smooth path $`L:tq_t`$ such that, for every $`t`$,
$$q_{t+1}=\varphi _t(q_t).$$
(15)
For example choose $`q_t`$ in the interval $`0t1`$ such that $`q_t=q_1`$ for $`1\epsilon t1`$ and $`q_t=\varphi _{t}^{}{}_{}{}^{1}(q_1)`$ for $`0t\epsilon `$. Then define $`q_t`$ for $`t`$ such that (15) is satisfied. Let $`h_t:L`$ be defined by
$$dh_t=\alpha _t:=\omega (_t\iota _t,d\iota _t),h_t(q_t)=0.$$
By (15), the function $`t\iota _t(q_t)`$ is $`1`$-periodic in $`t`$ and the proof of Lemma 2.1 shows that the $`1`$-forms $`\iota _{t}^{}{}_{}{}^{}\alpha _t`$ are $`1`$-periodic in $`t`$. Hence the functions $`h_t\iota _{t}^{}{}_{}{}^{1}`$ are $`1`$-periodic in $`t`$ and hence, so are the functions $`H_t`$ defined in the proof of Lemma 2.3. Now define
$$\stackrel{~}{H}_t(z):=H_t(z)\frac{_MH_t\omega ^n}{_M\omega ^n}.$$
Let $`\rho :[0,1][0,1]`$ be a smooth cutoff function such that $`\rho (r)=0`$ for $`r<\epsilon `$ and $`\rho (r)=1`$ for $`r>1\epsilon `$ and define $`\tau `$ by
$$\mathrm{\Phi }^{}\tau =\omega +\rho (r)d\stackrel{~}{H}_tdt+\dot{\rho }(r)\stackrel{~}{H}_tdrdt,$$
(16)
where $`\mathrm{\Phi }:[0,1]\times [0,1]\times MD\times M`$ is given by $`\mathrm{\Phi }(r,t,z)=(re^{2\pi it},z).`$ Explicitly, $`\tau `$ has the form (12) where $`F,G:D\times M`$ are given by
$$F_{x,y}=\frac{\mathrm{sin}(2\pi t)\rho (r)}{2\pi r}\stackrel{~}{H}_t,G_{x,y}=\frac{\mathrm{cos}(2\pi t)\rho (r)}{2\pi r}\stackrel{~}{H}_t,$$
(17)
for $`x+iy=re^{2\pi it}`$. These functions have mean value zero and satisfy (13) with $`H_t`$ replaced by $`\stackrel{~}{H}_t`$. Since $`H_t\iota _t=h_t`$ it follows as in the proof of Lemma 2.3 that
$$d\stackrel{~}{H}_t|_{\mathrm{\Lambda }_t}=dH_t|_{\mathrm{\Lambda }_t}=_t\mathrm{\Lambda }_t.$$
By Lemma 3.1, the parallel transport of $`\tau `$ along the boundary preserves $`\mathrm{\Lambda }`$. Hence $`\tau `$ is an element of $`𝒯(\mathrm{\Lambda })`$. This proves (i).
We prove (ii). Assume first that there exists a fibrewise Hamiltonian symplectomorphism of the form $`\mathrm{\Psi }(x+iy,z)=(x+iy,\psi _{x+iy}(z))`$ such that
$$\psi _{e^{2\pi it}}(\mathrm{\Lambda }_t)=\mathrm{\Lambda }_t^{}$$
for every $`t`$. Define
$$\psi _{s,t}:=\psi _{se^{2\pi it}},\mathrm{\Lambda }_{s,t}:=\psi _{s,t}(\mathrm{\Lambda }_t)$$
for $`0s1`$ and $`t`$. Then $`t\mathrm{\Lambda }_{s,t}`$ is an exact Lagrangian loop for every $`s`$ and $`_s\mathrm{\Lambda }_{s,t}\mathrm{\Omega }^1(\mathrm{\Lambda }_{s,t})`$ is exact for all $`s`$ and $`t`$. Hence the Lagrangian loop $`\mathrm{\Lambda }_{1,t}=\mathrm{\Lambda }_t^{}`$ is Hamiltonian isotopic to $`\mathrm{\Lambda }_{0,t}=\psi _0(\mathrm{\Lambda }_t)`$. Since $`\psi _0`$ is a Hamiltonian symplectomorphism, the loop $`t\psi _0(\mathrm{\Lambda }_t)`$ is Hamiltonian isotopic to $`t\mathrm{\Lambda }_t`$. Conversely, suppose that $`t\mathrm{\Lambda }_t`$ and $`t\mathrm{\Lambda }_t^{}`$ are two exact Lagrangian loops that are Hamiltonian isotopic. Choose an exact isotopy $`(s,t)\mathrm{\Lambda }_{s,t}`$ such that $`\mathrm{\Lambda }_{0,t}=\mathrm{\Lambda }_t,`$ $`\mathrm{\Lambda }_{1,t}=\mathrm{\Lambda }_t^{},`$ and $`_s\mathrm{\Lambda }_{s,t}=0`$ for $`s1/2`$. As in the proof of (i), one can construct a smooth family of Hamiltonian functions $`H_{s,t}:M`$ such that
$$H_{s,t+1}=H_{s,t},dH_{s,t}|_{\mathrm{\Lambda }_{s,t}}=_s\mathrm{\Lambda }_{s,t}.$$
Define the Hamiltonian symplectomorphisms $`\psi _{s,t}:MM`$ by
$$_s\psi _{s,t}=X_{s,t}\psi _{s,t},\iota (X_{s,t})\omega =dH_{s,t},\psi _{0,t}=\mathrm{id}.$$
Then $`\psi _{s,t}=\mathrm{id}`$ for $`s1/2`$ and the required fibrewise Hamiltonian symplectomorphism is given by $`\mathrm{\Psi }(se^{2\pi it},z):=(se^{2\pi it},\psi _{s,t}(z))`$.
We prove (iii). Let $`\tau `$ be given by (12) and suppose that
$$\mathrm{\Psi }(x+iy,z)=(x+iy,\psi _{x,y}(z))$$
is a fibrewise Hamiltonian symplectomorphism. Choose smooth functions $`A,B:D\times M`$ such that the functions $`A_{x,y}:=A(x+iy,)`$ and $`B_{x,y}:=B(x+iy,)`$ have mean value zero and the Hamiltonian vector fields $`X_A=X_{A_{x,y}}`$ and $`X_B=X_{B_{x,y}}`$ satisfy
$$_x\psi =X_A\psi ,_y\psi =X_B\psi .$$
(18)
Then
$$\mathrm{\Psi }^{}\tau =\omega +d\stackrel{~}{F}dx+d\stackrel{~}{G}dy+(_x\stackrel{~}{G}_y\stackrel{~}{F}+c)dxdy,$$
where
$$\stackrel{~}{F}=(FA)\mathrm{\Psi },\stackrel{~}{G}=(GB)\mathrm{\Psi }.$$
Hence
$`\mathrm{\Omega }_{\mathrm{\Psi }^{}\tau }`$ $`=`$ $`_x\stackrel{~}{G}_y\stackrel{~}{F}\{\stackrel{~}{F},\stackrel{~}{G}\}`$
$`=`$ $`_x(GB)\mathrm{\Psi }+d(GB)X_A\mathrm{\Psi }`$
$`_y(FA)\mathrm{\Psi }d(FA)X_B\mathrm{\Psi }`$
$`\{(FA),(GB)\}\mathrm{\Psi }`$
$`=`$ $`(_xG_yF\{F,G\})\mathrm{\Psi }`$
$`(_xB_yA\{A,B\})\mathrm{\Psi }`$
$`=`$ $`\mathrm{\Omega }_\tau \mathrm{\Psi }.`$
The last equality follows from the definition of $`A`$ and $`B`$ in (18). This proves the lemma. $`\mathrm{}`$
The relative K-area of an exact Lagrangian loop $`\mathrm{\Lambda }`$ is defined by
$$\chi (\mathrm{\Lambda }):=\underset{\tau 𝒯(\mathrm{\Lambda })}{inf}\mathrm{\Omega }_\tau ,$$
where
$$\mathrm{\Omega }_\tau :=_D\left(\underset{zM}{\mathrm{max}}\mathrm{\Omega }_\tau (x,y,z)\underset{zM}{\mathrm{min}}\mathrm{\Omega }_\tau (x,y,z)\right)𝑑x𝑑y.$$
###### Theorem 3.3
For every exact Lagrangian loop $`\mathrm{\Lambda }S^1\times M`$
$$\chi (\mathrm{\Lambda })=\nu (\mathrm{\Lambda }).$$
Proof: Let $`:t\mathrm{\Lambda }_t`$ be an exact Lagrangian loop. Let $`\tau \mathrm{\Omega }^2(D\times M)`$ be the connection $`2`$-form defined by (16) in the proof of Lemma 3.2, where the cutoff function $`\rho :[0,1][0,1]`$ is chosen to be nondecreasing. Then
$$\mathrm{\Phi }^{}(Fdx+Gdy)=\rho \stackrel{~}{H}_tdt,$$
where $`F,G:D\times M`$ are given by (17) and $`\mathrm{\Phi }(r,t,z)=(re^{2\pi it},z)`$. Taking the differential of this $`1`$-form on $`[0,1]^2\times M`$ we find
$$\mathrm{\Phi }^{}((_xG_yF)dxdy)=\dot{\rho }\stackrel{~}{H}_tdrdt.$$
Since $`\{F,G\}=0`$ and $`\mathrm{\Phi }^{}(dxdy)=2\pi rdrdt`$ we obtain
$$\mathrm{\Omega }_\tau (re^{2\pi it},z)=\frac{\dot{\rho }(r)}{2\pi r}\stackrel{~}{H}_t(z).$$
Moreover,
$$\stackrel{~}{H}_t=\underset{M}{\mathrm{max}}\stackrel{~}{H}_t\underset{M}{\mathrm{min}}\stackrel{~}{H}_t=\underset{\mathrm{\Lambda }_t}{\mathrm{max}}\stackrel{~}{H}_t\underset{\mathrm{\Lambda }_t}{\mathrm{min}}\stackrel{~}{H}_t,$$
and hence
$$\mathrm{\Omega }_\tau =_0^1_0^1\dot{\rho }(r)\stackrel{~}{H}_t𝑑r𝑑t=_0^1\stackrel{~}{H}_t𝑑t=\mathrm{}(\mathrm{\Lambda }).$$
This implies $`\chi (\mathrm{\Lambda })\mathrm{}(\mathrm{\Lambda }).`$ If $`\mathrm{\Lambda }`$ and $`\mathrm{\Lambda }^{}`$ are Hamiltonian isotopic then, by Lemma 3.2 (ii), there exists a fibrewise Hamiltonian symplectomorphism $`\mathrm{\Psi }`$ of $`D\times M`$ such that $`\mathrm{\Psi }(\mathrm{\Lambda })=\mathrm{\Lambda }^{}`$. Hence $`\tau 𝒯(\mathrm{\Lambda }^{})`$ if and only if $`\mathrm{\Psi }^{}\tau 𝒯(\mathrm{\Lambda })`$. By Lemma 3.2 (iii), $`\chi (\mathrm{\Lambda })=\chi (\mathrm{\Lambda }^{})\mathrm{}(\mathrm{\Lambda }^{}).`$ Hence $`\chi (\mathrm{\Lambda })\nu (\mathrm{\Lambda }).`$
We prove that $`\nu (\mathrm{\Lambda })\chi (\mathrm{\Lambda })`$. Let $`\tau 𝒯(\mathrm{\Lambda })`$. We shall construct an exact Lagrangian loop $`\mathrm{\Lambda }^{}`$ that is Hamiltonian isotopic to $`\mathrm{\Lambda }`$ and satisfies
$$\mathrm{}(\mathrm{\Lambda }^{})\mathrm{\Omega }_\tau .$$
(19)
Suppose that $`\tau `$ has the form (12). Since the function $`c`$ in (12) has no effect on the curvature we may assume, without loss of generality, that $`c0`$. Define $`H=H_{r,t}:M`$ and $`K=K_{r,t}:M`$ by the formula
$$\mathrm{\Phi }^{}\tau =\omega +dKdr+dHdt+(_rH_tK)drdt.$$
Explicitly,
$$K_{r,t}=\mathrm{cos}(2\pi t)F_{re^{2\pi it}}+\mathrm{sin}(2\pi t)G_{re^{2\pi it}},$$
$$H_{r,t}=2\pi r\mathrm{cos}(2\pi t)G_{re^{2\pi it}}2\pi r\mathrm{sin}(2\pi t)F_{re^{2\pi it}}.$$
Define the Hamiltonian symplectomorphisms $`\psi _{r,t}:MM`$ by
$$_r\psi _{r,t}=X_{K_{r,t}}\psi _{r,t},\psi _{0,t}=\mathrm{id}.$$
Then the loop
$$\mathrm{\Lambda }_t^{}=\psi _{1,t}^{}{}_{}{}^{1}(\mathrm{\Lambda }_t)$$
is evidently Hamiltonian isotopic to $`\mathrm{\Lambda }`$. We shall prove that it satisfies (19). To see this, denote by $`\mathrm{\Psi }`$ the fibrewise Hamiltonian symplectomorphism of $`[0,1]^2\times M`$ given by
$$\mathrm{\Psi }(r,t,z)=(r,t,\psi _{r,t}(z)).$$
Then, as in the proof of Lemma 3.2, we obtain
$$\mathrm{\Psi }^{}\mathrm{\Phi }^{}\tau =\omega +dH^{}dt+_rH^{}drdt,$$
where $`H_{r,t}^{}=(H_{r,t}B_{r,t})\psi _{r,t}`$ and $`B_{r,t}:M`$ is defined by $`_t\psi _{r,t}=X_{B_{r,t}}\psi _{r,t}`$. These functions satisfy
$$\mathrm{\Omega }_\tau =_0^1_0^1_rH_{r,t}^{}𝑑r𝑑t,H_{0,t}^{}=0.$$
Moreover, by Lemma 2.5, we have
$`_t\mathrm{\Lambda }_t^{}`$ $`=`$ $`\psi _{1,t}^{}{}_{}{}^{}\left(_t\mathrm{\Lambda }_tdB_{1,t}|_{\mathrm{\Lambda }_t}\right)`$
$`=`$ $`\psi _{1,t}^{}{}_{}{}^{}(dH_{1,t}|_{\mathrm{\Lambda }_t})d(B_{1,t}\psi _{1,t})|_{\mathrm{\Lambda }_t^{}}`$
$`=`$ $`dH_{1,t}^{}|_{\mathrm{\Lambda }_t^{}}.`$
Hence the length of $`\mathrm{\Lambda }^{}`$ is given by
$`\mathrm{}(\mathrm{\Lambda }^{})`$ $`=`$ $`{\displaystyle _0^1}\left(\underset{\mathrm{\Lambda }_t^{}}{\mathrm{max}}H_{1,t}^{}\underset{\mathrm{\Lambda }_t^{}}{\mathrm{min}}H_{1,t}^{}\right)𝑑t`$
$``$ $`{\displaystyle _0^1}\left(\underset{M}{\mathrm{max}}H_{1,t}^{}\underset{M}{\mathrm{min}}H_{1,t}^{}\right)𝑑t`$
$`=`$ $`{\displaystyle _0^1}\left(\underset{M}{\mathrm{max}}\left({\displaystyle _0^1}_rH_{r,t}^{}dr\right)\underset{M}{\mathrm{min}}\left({\displaystyle _0^1}_rH_{r,t}^{}dr\right)\right)𝑑t`$
$``$ $`{\displaystyle _0^1}{\displaystyle _0^1}\left(\underset{M}{\mathrm{max}}_rH_{r,t}^{}\underset{M}{\mathrm{min}}_rH_{r,t}^{}\right)𝑑r𝑑t`$
$`=`$ $`{\displaystyle _0^1}{\displaystyle _0^1}_rH_{r,t}^{}𝑑r𝑑t`$
$`=`$ $`\mathrm{\Omega }_\tau .`$
Thus we have proved that for every $`\tau 𝒯(\mathrm{\Lambda })`$ there exists an exact Lagrangian loop $`\mathrm{\Lambda }^{}`$ that is Hamiltonian isotopic to $`\mathrm{\Lambda }`$ and satisfies $`\mathrm{}(\mathrm{\Lambda }^{})\mathrm{\Omega }_\tau `$. Hence $`\chi (\mathrm{\Lambda })\nu (\mathrm{\Lambda })`$ and this proves the theorem. $`\mathrm{}`$
### 3.3 The non-symplectic interval
Let $`\mathrm{\Lambda }D\times M`$ be an exact Lagrangian loop and $`\tau 𝒯(\mathrm{\Lambda })`$ be a Hamiltonian connection $`2`$-form. Since $`\tau `$ is closed and vanishes on $`\mathrm{\Lambda }`$ (see Lemma 3.1) it determines a relative cohomology class
$$[\tau ]H^2(D\times M,\mathrm{\Lambda };).$$
Let $`\mathrm{\Sigma }`$ be a compact oriented Riemann surface with (possibly empty) boundary $`\mathrm{\Sigma }`$. A smooth map $`v:(\mathrm{\Sigma },\mathrm{\Sigma })(D\times M,\mathrm{\Lambda })`$ determines a $`2`$-dimensional relative homology class
$$[v]:=v_{}[\mathrm{\Sigma }]H_2(D\times M,\mathrm{\Lambda };).$$
The pairing of this class with $`[\tau ]`$ is given by
$$[\tau ],[v]=_\mathrm{\Sigma }v^{}\tau .$$
Since every 2-dimensional integral homology class of the pair $`(D\times M,\mathrm{\Lambda })`$ can be represented by a smooth map $`v`$ as above, the cohomology class $`[\tau ]`$ is uniquely determined by these pairings. Define $`\sigma H^2(D\times M,\mathrm{\Lambda };)`$ by
$$\sigma ,[v]=\mathrm{deg}(\pi v)$$
(20)
for every $`v:(\mathrm{\Sigma },\mathrm{\Sigma })(D\times M,\mathrm{\Lambda })`$, where
$$\pi :(D\times M,\mathrm{\Lambda })(D,D)$$
denotes the obvious projection. In (20) the degree of a smooth map $`v_0:(\mathrm{\Sigma },\mathrm{\Sigma })(D,D)`$ is understood as the degree of its restriction to the boundary. It agrees with the number of preimages of an interior regular value, counted with appropriate signs (cf. Milnor ). Note that
$$\sigma =\frac{1}{\pi }[dxdy]$$
and hence $`\sigma `$ agrees with the pullback of the positive integral generator of $`H^2(D,D;)`$ under the projection $`\pi `$.
###### Lemma 3.4
Let $`\tau _0,\tau _1𝒯(\mathrm{\Lambda })`$. Then there exists a constant $`s=s(\tau _1,\tau _0)`$ such that
$$[\tau _1][\tau _0]=s\sigma .$$
Proof: Let $`\tau _i`$ be given by (12) with $`F,G,c`$ replaced by $`F_i,G_i,c_i`$ for $`i=0,1`$. Denote
$$F:=F_1F_0,G:=G_1G_0,c:=c_1c_0,$$
and let $`H_t:M`$ be defined by (13). Since $`\tau _0,\tau _1𝒯(\mathrm{\Lambda })`$ it follows from Lemma 3.1 that there exists a function $`h:/`$ such that
$$H_t|_{\mathrm{\Lambda }_t}h(t)$$
for every $`t`$. We shall prove that the required identity holds with
$$s:=_0^1h(t)𝑑t+_Dc𝑑x𝑑y.$$
To see this note that, by (13),
$$(Fdx+Gdy)|_\mathrm{\Lambda }=\pi ^{}\alpha _h.$$
(21)
where $`\alpha _h\mathrm{\Omega }^1(S^1)`$ denotes the pushforward of the $`1`$-form $`hdt\mathrm{\Omega }^1(/)`$ under the diffeomorphisms $`/S^1:[t]e^{2\pi it}`$. Let $`\mathrm{\Sigma }`$ be a compact oriented Riemann surface and $`v:\mathrm{\Sigma }D\times M`$ be a smooth function such that $`v(\mathrm{\Sigma })\mathrm{\Lambda }`$. Denote $`v_0:=\pi v:(\mathrm{\Sigma },\mathrm{\Sigma })(D,D)`$. Then
$`{\displaystyle _\mathrm{\Sigma }}v^{}(\tau _1\tau _0)`$ $`=`$ $`{\displaystyle _\mathrm{\Sigma }}v^{}\left(dFdx+dGdy+(_xG_yF+c)dxdy\right)`$
$`=`$ $`{\displaystyle _\mathrm{\Sigma }}v^{}(Fdx+Gdy)+{\displaystyle _\mathrm{\Sigma }}v_{0}^{}{}_{}{}^{}(cdxdy)`$
$`=`$ $`{\displaystyle _\mathrm{\Sigma }}v_{0}^{}{}_{}{}^{}\alpha _h+{\displaystyle _\mathrm{\Sigma }}v_{0}^{}{}_{}{}^{}(cdxdy)`$
$`=`$ $`s\mathrm{deg}(v_0).`$
The penultimate equality follows from (21) and the last from the identities
$$_\mathrm{\Sigma }v_{0}^{}{}_{}{}^{}\alpha _h=\mathrm{deg}(v_0)_{S^1}\alpha _h$$
(22)
and
$$_\mathrm{\Sigma }v_{0}^{}{}_{}{}^{}(cdxdy)=\mathrm{deg}(v_0)_Dc𝑑xdy.$$
(23)
Here (22) is the degree theorem for maps between compact $`1`$-manifolds and (23) is the degree theorem for maps between $`2`$-manifolds with boundary. More precisely, if the function $`c:D`$ has mean value zero then there exists a $`1`$-form $`\alpha \mathrm{\Omega }^1(D)`$ such that $`d\alpha =cdxdy`$ and $`\alpha |_{TS^1}=0.`$ This implies that the left hand side of (23) vanishes. Hence it suffices to establish (23) for constant functions $`c`$ and this reduces to (22). This proves the lemma. $`\mathrm{}`$
Let $`\tau _0𝒯(\mathrm{\Lambda })`$. We shall now address the question which cohomology classes $`[\tau _0]+s\sigma `$ can be represented by nondegenerate Hamiltonian connection $`2`$-forms. Such a $`2`$-form is a symplectic form on $`D\times M`$ with respect to which $`\mathrm{\Lambda }`$ is a Lagrangian submanifold. Denote
$$𝒯^\pm (\mathrm{\Lambda }):=\{\tau 𝒯(\mathrm{\Lambda })|\pm \tau ^{n+1}>0\}.$$
Here the inequality $`\tau ^{n+1}>0`$ means that $`\tau ^{n+1}=fdxdy\omega ^n`$, where $`f:D\times M`$ is a positive function. For $`\tau _0𝒯(\mathrm{\Lambda })`$ we define
$$\epsilon ^+(\tau _0,\mathrm{\Lambda }):=inf\left\{s(\tau ,\tau _0)\right|\tau 𝒯^+(\mathrm{\Lambda })\},$$
$$\epsilon ^{}(\tau _0,\mathrm{\Lambda }):=sup\left\{s(\tau ,\tau _0)\right|\tau 𝒯^{}(\mathrm{\Lambda })\}.$$
The proof of Theorem 3.5 below shows that the class $`[\tau _0]+s\sigma `$ can be represented by a symplectic form $`\tau 𝒯^\pm (\mathrm{\Lambda })`$ for $`\pm s`$ sufficiently large and hence $`\pm \epsilon ^\pm (\tau _0,\mathrm{\Lambda })<\mathrm{}`$. Evidently, $`\epsilon ^\pm (\tau _1,\mathrm{\Lambda })\epsilon ^\pm (\tau _0,\mathrm{\Lambda })=s(\tau _1,\tau _0)`$. Hence the number
$$\epsilon (\mathrm{\Lambda }):=\epsilon ^+(\tau _0,\mathrm{\Lambda })\epsilon ^{}(\tau _0,\mathrm{\Lambda })$$
is independent of the connection $`2`$-form $`\tau _0𝒯(\mathrm{\Lambda })`$ used to define it. This number is called the width of the nonsymplectic interval.
###### Theorem 3.5
For every exact Lagrangian loop $`\mathrm{\Lambda }D\times M`$
$$\epsilon (\mathrm{\Lambda })\chi (\mathrm{\Lambda }).$$
Proof: Let $`:t\mathrm{\Lambda }_t`$ be an exact Lagrangian loop and $`F,G:D\times M`$ be smooth functions such that the functions $`H_t:M`$ defined by (13) satisfy $`dH_t|_{\mathrm{\Lambda }_t}=_t\mathrm{\Lambda }_t`$ for every $`t`$. For every smooth function $`c:D`$ let $`\tau _c𝒯(\mathrm{\Lambda })`$ be given by (12). In particular, $`\tau _0`$ is given by (12) with $`c=0`$. We shall prove that
$$\epsilon ^+(\tau _0,\mathrm{\Lambda })_D\underset{zM}{\mathrm{max}}\mathrm{\Omega }_{\tau _0}(x,y,z)𝑑x𝑑y,$$
(24)
$$\epsilon ^{}(\tau _0,\mathrm{\Lambda })_D\underset{zM}{\mathrm{min}}\mathrm{\Omega }_{\tau _0}(x,y,z)𝑑x𝑑y.$$
(25)
To see this, note that
$$ndFdG\omega ^{n1}=\{F,G\}\omega ^n$$
and hence
$$\begin{array}{ccc}\hfill \tau ^{n+1}& =& (n+1)(_xG_yF+c)dxdy\omega ^n\hfill \\ & & +n(n+1)dFdxdGdy\omega ^{n1}\hfill \\ & =& (n+1)(_xG_yF\{F,G\}+c)dxdy\omega ^n\hfill \\ & =& (n+1)(c\mathrm{\Omega }_{\tau _0})dxdy\omega ^n.\hfill \end{array}$$
(26)
This shows that $`\tau _c`$ is nondegenerate if and only if $`c(x,y)\mathrm{\Omega }_{\tau _0}(x,y,z)`$ for all $`(x+iy,z)D\times M`$. Fix a number
$$s>_D\underset{zM}{\mathrm{max}}\mathrm{\Omega }_{\tau _0}(x,y,z)𝑑x𝑑y.$$
Choose a smooth function $`c:D`$ such that
$$c(x,y)>\underset{zM}{\mathrm{max}}\mathrm{\Omega }_{\tau _0}(x,y,z)$$
for all $`x+iyD`$ and
$$_Dc𝑑x𝑑y=s.$$
Then $`\tau _c`$ is nondegenerate and represents the class $`[\tau _c]=[\tau _0]+s\sigma .`$ This proves (24) and (25) follows from a similar argument. It follows from (24) and (25) that
$`\epsilon (\mathrm{\Lambda })`$ $`=`$ $`\epsilon ^+(\tau _0,\mathrm{\Lambda })\epsilon ^{}(\tau _0,\mathrm{\Lambda })`$
$``$ $`{\displaystyle _D}\left(\underset{zM}{\mathrm{max}}\mathrm{\Omega }_{\tau _0}(x,y,z)\underset{zM}{\mathrm{min}}\mathrm{\Omega }_{\tau _0}(x,y,z)\right)𝑑x𝑑y`$
$`=`$ $`\mathrm{\Omega }_{\tau _0}.`$
Since the curvature of $`\tau _0`$ is equal to the curvature of $`\tau _c`$ for every $`c`$ it follows that $`\epsilon (\mathrm{\Lambda })\mathrm{\Omega }_\tau `$ for every $`\tau 𝒯(\mathrm{\Lambda })`$ and hence $`\epsilon (\mathrm{\Lambda })\chi (\mathrm{\Lambda })`$. This proves the theorem. $`\mathrm{}`$
###### Remark 3.6
Let us denote
$$T(\mathrm{\Lambda }):=\left\{[\tau ]H^2(D\times M,\mathrm{\Lambda };)\right|\tau 𝒯(\mathrm{\Lambda })\}.$$
(27)
By Lemma 3.4, this set is a $`1`$-dimensional affine subspace of $`H^2(D\times M,\mathrm{\Lambda };)`$. Denote
$$T^\pm (\mathrm{\Lambda }):=\left\{[\tau ]\right|\tau 𝒯^\pm (\mathrm{\Lambda })\}.$$
These sets are open and connected. To prove connectedness, let $`\tau _i𝒯^+(\mathrm{\Lambda })`$ be given by (12) with $`F,G,c`$ replaced by $`F_i,G_i,c_i`$ for $`i=0,1`$. By (26), $`c_i>\mathrm{\Omega }_{\tau _i}`$. Assume without loss of generality that $`s(\tau _1,\tau _0)0`$. Then the path $`[0,1]T^+(\mathrm{\Lambda }):t[\tau _0]+ts(\tau _1,\tau _0)\sigma `$ connects $`[\tau _0]`$ with $`[\tau _1]`$. This shows that the sets $`T^\pm (\mathrm{\Lambda })`$ are connected. The complement $`T(\mathrm{\Lambda })(T^{}(\mathrm{\Lambda })T^+(\mathrm{\Lambda }))`$ is compact and connected. It can be expressed in the form
$$T(\mathrm{\Lambda })(T^{}(\mathrm{\Lambda })T^+(\mathrm{\Lambda }))=\left\{[\tau _0]+s\sigma \right|\epsilon ^{}(\tau _0,\mathrm{\Lambda })s\epsilon ^+(\tau _0,\mathrm{\Lambda })\}$$
for every $`\tau _0𝒯(\mathrm{\Lambda })`$. We do not know if this complement is always nonempty or, equivalently, if $`\epsilon (\mathrm{\Lambda })`$ is always nonnegative.
## 4 Loops on the 2-torus
Consider the torus $`M=𝕋^2=^2/^2`$ with the standard symplectic form
$$\omega =dxdy$$
and let $`\pi :^2𝕋^2`$ denote the projection. Let
$$B_r=\{(s,t)^2|s^2+t^2r^2\}$$
and suppose that $`S𝕋^2`$ is the image of an embedding $`B_1𝕋^2`$. Define
$$\mathrm{\Lambda }_t:=\mathrm{\Lambda }_t(S):=\left\{[x,y+t]\right|[x,y]S\}$$
(28)
(see Figure 1).
###### Theorem 4.1
Let $`S𝕋^2`$ be a closed embedded disc and $`t\mathrm{\Lambda }_t`$ be the exact Lagrangian loop defined by (28). Then
$$\nu (\mathrm{\Lambda })=\mathrm{area}(S).$$
Proof: We prove that $`\nu (\mathrm{\Lambda })\mathrm{area}(S)`$. To see this, choose smooth functions $`x,y:`$ such that
$$x(\theta +1)=x(\theta ),y(\theta +1)=y(\theta ),$$
and the map $`\iota _t:/𝕋^2`$ defined by
$$\iota _t(\theta ):=[x(\theta ),y(\theta )+t]$$
is an embedding with $`\iota _t(/)=\mathrm{\Lambda }_t`$. Then
$$\alpha _t:=\omega (_t\iota _t,d\iota _t)=\dot{x}d\theta \mathrm{\Omega }^1(/).$$
Hence $`\alpha _t=dh_t`$ where $`h_t=x:/`$. Hence
$$_t\mathrm{\Lambda }_t=dh_t=\mathrm{max}x\mathrm{min}x$$
and this implies
$$\mathrm{}(\mathrm{\Lambda })=\mathrm{max}x\mathrm{min}x.$$
By Proposition A.1 in the appendix, two loops $`t\mathrm{\Lambda }_t(S)`$ and $`t\mathrm{\Lambda }_t(S^{})`$, associated to two embedded discs $`S,S^{}𝕋^2`$ via (28), are Hamiltonian isotopic if and only if $`S`$ and $`S^{}`$ have the same area. Now for every $`\delta >0`$ there exists an embedded disc $`S^{}`$ (as illustrated in Figure 2) such that
$$\mathrm{area}(S)=\mathrm{area}(S^{}),\mathrm{max}x^{}\mathrm{min}x^{}<\mathrm{area}(S)+\delta ,$$
where $`x^{},y^{}:/`$ are chosen such that the map $`\iota ^{}(\theta )=[x^{}(\theta ),y^{}(\theta )]`$ defines an embedding $`/𝕋^2`$ whose image is $`S^{}`$. Hence the length of the loop $`t\mathrm{\Lambda }_t(S^{})`$ is bounded above by $`\mathrm{area}(S)+\delta `$. Thus we have proved that
$$\nu (\mathrm{\Lambda })\mathrm{area}(S).$$
To show the reverse inequality let $`t\mathrm{\Lambda }_t^{}`$ be an exact Lagrangian loop that is Hamiltonian isotopic to $`\mathrm{\Lambda }`$. Then
$$\mathrm{\Lambda }_0^{}=S^{},$$
where $`S^{}𝕋^2`$ is a smoothly embedded closed disc of the same area as $`S`$. Let $`\psi _t:𝕋^2𝕋^2`$ be a Hamiltonian isotopy such that
$$\psi _t(\mathrm{\Lambda }_0^{})=\mathrm{\Lambda }_t^{}.$$
We shall prove that
$$\mathrm{area}(S)\mathrm{}\left(\{\psi _t\}_{0t1}\right).$$
(29)
To see this, choose an embedded closed discs $`\stackrel{~}{S}^2`$ such that $`\pi (\stackrel{~}{S})=S^{}`$ and let $`\stackrel{~}{\psi }_t:^2^2`$ be a lift of $`\psi _t`$. Since $`\mathrm{\Lambda }^{}`$ is Hamiltonian isotopic to $`\mathrm{\Lambda }`$ we have $`\stackrel{~}{\psi }_{t+1}(\stackrel{~}{S})=\stackrel{~}{\psi }_t(\stackrel{~}{S})+(0,1)`$ and hence
$$\stackrel{~}{\psi }_1(\stackrel{~}{S})\stackrel{~}{S}=\mathrm{}.$$
Let $`\stackrel{~}{H}_t:^2`$ be the Hamiltonian functions that generate $`\stackrel{~}{\psi }_t`$ and have mean value zero over the fundamental domain $`[0,1]^2`$. Choose $`R>1`$ such that $`\stackrel{~}{\psi }_t(\stackrel{~}{S})B_R`$ for every $`t[0,1]`$ and let $`\beta :^2[0,1]`$ be a compactly supported cutoff function such that $`\beta |_{B_R}1`$. Then the functions
$$\widehat{H}_t:=\beta \stackrel{~}{H}_t$$
generate a compactly supported Hamiltonian isotopy $`\widehat{\psi }_t`$ of $`^2`$ that satisfies
$$\widehat{\psi }_1(\stackrel{~}{S})\stackrel{~}{S}=\mathrm{}.$$
Now it follows from the energy-capacity inequality in Hofer that the displacement energy of $`\stackrel{~}{S}`$ is bounded below by the area. Hence
$$\mathrm{area}(\stackrel{~}{S})d(\mathrm{id},\widehat{\psi }_1)\mathrm{}(\{\widehat{\psi }_t\}_{0t1})=\mathrm{}(\{\psi _t\}_{0t1}).$$
Since
$$\mathrm{area}(\stackrel{~}{S})=\mathrm{area}(S^{})=\mathrm{area}(S),$$
this proves (29). It follows from (29) and (8) that
$$\mathrm{area}(S)\mathrm{}(\mathrm{\Lambda }^{})$$
for every exact Lagrangian loop $`\mathrm{\Lambda }^{}`$ that is Hamiltonian isotopic to $`\mathrm{\Lambda }`$. Hence $`\mathrm{area}(S)\nu (\mathrm{\Lambda })`$. $`\mathrm{}`$
Theorem 4.1 shows that the invariant $`\nu (\mathrm{\Lambda })`$ is not necessarily invariant under Lagrangian isotopy, but only under exact Lagrangian isotopy. The techniques of proof are specific to the 2-dimensional case. To establish lower bounds for our invariants in higher dimensions we shall use existence theorems for pseudoholomorphic discs.
## 5 Relative Gromov invariants
Throughout we assume that our symplectic manifold $`(M,\omega )`$ is compact. The relative Gromov invariants of an exact Lagrangian loop $`\mathrm{\Lambda }D\times M`$ are defined in terms of holomorphic sections of the bundle $`D\times MD`$ with boundary values in $`\mathrm{\Lambda }`$. Let us denote by $`\mathrm{Map}_\mathrm{\Lambda }(D,M)`$ the space of smooth functions $`u:DM`$ that satisfy $`u(e^{2\pi it})\mathrm{\Lambda }_t`$ for every $`t`$. The Maslov class is a function
$$\mu _\mathrm{\Lambda }:\mathrm{Map}_\mathrm{\Lambda }(D,M)$$
defined as follows. Given $`u\mathrm{Map}_\mathrm{\Lambda }(D,M)`$ choose a trivialization of the tangent bundle $`u^{}TM`$. Then the tangent spaces $`T_{u(e^{2\pi it})}\mathrm{\Lambda }_t`$ define a loop of Lagrangian subspaces in $`(^{2n},\omega _0)`$ and $`\mu _\mathrm{\Lambda }(u)`$ is defined as the Maslov index of this loop (cf. ). This integer is independent of the choice of the trivialization used to define it, and it depends only on the homology class of $`u`$ in $`H_2(D\times M,\mathrm{\Lambda };)`$. We shall assume throughout that the pair $`(M,\mathrm{\Lambda }_0)`$ is monotone, i.e. there exists a $`\lambda >0`$ such that, for every smooth map $`v\mathrm{Map}_{\mathrm{\Lambda }_0}(D,M)`$,
$$_Dv^{}\omega =\lambda \mu _{\mathrm{\Lambda }_0}(v).$$
Here $`\mu _{\mathrm{\Lambda }_0}`$ denotes the Maslov class corresponding to the constant loop $`t\mathrm{\Lambda }_0`$. The minimal Maslov number of the pair $`(M,\mathrm{\Lambda }_0)`$ is defined by
$$N:=inf\left\{\mu _{\mathrm{\Lambda }_0}(v)\right|v:(D,D)(M,\mathrm{\Lambda }_0),\mu _{\mathrm{\Lambda }_0}(v)>0\}.$$
We shall define relative Gromov invariants for every tuple $`𝐭=(t_1,\mathrm{},t_k)^k`$ with $`0t_1<\mathrm{}<t_k<1`$ and every class $`AH_2(D\times M,\mathrm{\Lambda };)`$ that satisfies $`n\pm \mu _\mathrm{\Lambda }(A)N2.`$ The invariants are homology classes
$$\mathrm{Gr}_{A,𝐭}^\pm (\mathrm{\Lambda })H_{n\pm \mu _\mathrm{\Lambda }(A)}(\mathrm{\Lambda }_𝐭;_2),$$
where $`\mathrm{\Lambda }_𝐭:=\mathrm{\Lambda }_{t_1}\times \mathrm{}\times \mathrm{\Lambda }_{t_k}`$. These homology classes arise from certain moduli spaces $`_A(\tau ,\pm J)`$ of (anti-)holomorphic sections of the bundle $`D\times M`$ with boundary values in $`\mathrm{\Lambda }`$ that represent the class $`A`$. The points $`(e^{2\pi it_1},\mathrm{},e^{2\pi it_k})`$ determine an evaluation map
$$\mathrm{ev}_𝐭:_A(\tau ,\pm J)\mathrm{\Lambda }_𝐭$$
and $`\mathrm{Gr}_{A,𝐭}^\pm (\mathrm{\Lambda })`$ is defined as the image of the fundamental cycle of $`_A(\tau ,\pm J)`$ under the induced homomorphism on homology. We shall work with almost complex structures on $`D\times M`$ that are compatible with the fibration. Every such structure is determined by a family of almost complex structures on $`M`$ and a connection $`2`$-form $`\tau 𝒯(\mathrm{\Lambda })`$.
### 5.1 J-holomorphic discs
Let $`\mathrm{\Lambda }S^1\times M`$ be an exact Lagrangian loop and $`\tau 𝒯(\mathrm{\Lambda })`$ be a Hamiltonian connection $`2`$-form that preserves $`\mathrm{\Lambda }`$. Throughout we shall denote by $`𝒥(M,\omega )`$ the space of almost complex structures on $`TM`$ that are compatible with $`\omega `$. Let $`D𝒥(M,\omega ):(x,y)J_{x,y}`$ be a smooth family of such almost complex structures. Associated to the triple $`(\tau ,J,\mathrm{\Lambda })`$ there is a natural boundary value problem for smooth functions $`u:DM`$:
$$_xuX_F(u)+J(_yuX_G(u))=0,$$
(30)
$$u(e^{2\pi it})\mathrm{\Lambda }_t,t.$$
(31)
Here we abbreviate $`J=J_{x,y}`$, $`\tau `$ is given by (12), $`X_F=X_F(x,y,)\mathrm{Vect}(M)`$ denotes the Hamiltonian vector field of the function $`F=F(x,y,):M`$, and similarly for $`X_G`$. Following Gromov we observe that the solutions of (30) can be thought of as pseudo-holomorphic curves in $`D\times M`$.
###### Remark 5.1
Consider the almost complex structure $`\stackrel{~}{J}`$ on $`D\times M`$ given by
$$\stackrel{~}{J}=\stackrel{~}{J}(\tau ,J):=\left(\begin{array}{ccc}0& 1& 0\\ 1& 0& 0\\ JX_F+X_G& X_FJX_G& J\end{array}\right).$$
Then $`u:DM`$ is a solution of (30) if and only if the function
$$\stackrel{~}{u}(x,y)=(x,y,u(x,y))$$
(32)
is a $`\stackrel{~}{J}`$-holomorphic curve in $`D\times M`$, i.e.
$$_x\stackrel{~}{u}+\stackrel{~}{J}_y\stackrel{~}{u}=0.$$
If $`\tau `$ is given by (12) then, for every $`\stackrel{~}{\zeta }=(\xi ,\eta ,\zeta )T_{x,y,z}(D\times M)`$,
$$\tau (\stackrel{~}{\zeta },\stackrel{~}{J}\stackrel{~}{\zeta })=\left|\zeta \xi X_F\eta X_G\right|^2+(c\mathrm{\Omega }_\tau )(\xi ^2+\eta ^2).$$
Hence $`\stackrel{~}{J}`$ is tamed by $`\tau `$ whenever $`\tau 𝒯^+(\mathrm{\Lambda })`$ (see (26)). If $`\tau 𝒯^{}(\mathrm{\Lambda })`$ then $`\stackrel{~}{J}(\tau ,J)`$ is tamed by $`\tau `$.
The energy of a solution $`u`$ of (30) is defined by
$$E(u):=_D\left|_xuX_F(u)\right|^2𝑑x𝑑y.$$
The next lemma shows that the solutions of (30) and (31) that represent a given homology class $`AH_2(D\times M,\mathrm{\Lambda };)`$ satisfy a uniform energy bound.
###### Lemma 5.2
Let $`u:DM`$ be a smooth solution of (30) and (31) and denote by $`AH_2(D\times M,\mathrm{\Lambda };)`$ the homology class represented by the map $`\stackrel{~}{u}:DD\times M`$ defined by (32). Let $`c:D`$ be the function in (12). Then
$$E(u)=[\tau ],A+_D\left(\mathrm{\Omega }_\tau (x,y,u)c(x,y)\right)𝑑x𝑑y.$$
Proof: We compute
$`E(u)`$ $`=`$ $`{\displaystyle _D}\omega (_xuX_F(u),_yuX_G(u))𝑑x𝑑y`$
$`=`$ $`{\displaystyle _D}\left(\omega (_xu,_yu)dF(u)_yu+dG(u)_xu+\{F,G\}(u)\right)𝑑x𝑑y`$
$`=`$ $`{\displaystyle _D}\left(\omega (_xu,_yu)dF(u)_yu+dG(u)_xu\right)𝑑x𝑑y`$
$`+{\displaystyle _D}\left(\mathrm{\Omega }_\tau (x,y,u)(_yF)(u)+(_xG)(u)\right)𝑑x𝑑y`$
$`=`$ $`{\displaystyle _D}\left(\tau (_x\stackrel{~}{u},_y\stackrel{~}{u})c(x,y)\right)𝑑x𝑑y+{\displaystyle _D}\mathrm{\Omega }_\tau (x,y,u)𝑑x𝑑y.`$
This proves the lemma. $`\mathrm{}`$
Let us denote the moduli space of solutions of (30) and (31) that represent a given homology class $`AH_2(D\times M,\mathrm{\Lambda };)`$ by
$$_A(\tau ,J):=\{u:DM|u\text{ satisfies }(\text{30})\text{ and }(\text{31}),[\stackrel{~}{u}]=A\}.$$
We shall prove that, for a generic pair $`(\tau ,J)`$, this space is a smooth manifold of dimension $`n+\mu _\mathrm{\Lambda }(A).`$ Moreover, if the pair $`(M,\mathrm{\Lambda }_0)`$ is monotone with minimal Maslov number $`N`$ and $`n+\mu _\mathrm{\Lambda }(A)<N`$, we shall prove that $`_A(\tau ,J)`$ is compact, again for a generic pair $`(\tau ,J)`$. The key tool for establishing compactness is the energy bound of Lemma 5.2. Under these asumptions the moduli spaces can be used to define Gromov invariants of $`\mathrm{\Lambda }`$. The significance of these invariants for exact Lagrangian loops lies in the following observation.
###### Lemma 5.3
Let $`\mathrm{\Lambda }`$ be an exact Lagrangian loop and $`AH_2(D\times M,\mathrm{\Lambda };)`$. Suppose that for every $`\tau 𝒯^+(\mathrm{\Lambda })`$ there exists a $`J`$ such that $`_A(\tau ,J)\mathrm{}.`$ Then
$$\epsilon ^+(\tau _0,\mathrm{\Lambda })+[\tau _0],A0$$
for every $`\tau _0𝒯(\mathrm{\Lambda })`$.
Proof: Let $`\tau 𝒯^+(\mathrm{\Lambda })`$ and $`u_A(\tau ,J)`$. Let $`\stackrel{~}{u}:DD\times M`$ be given by (32). Then $`\stackrel{~}{u}`$ is a $`\stackrel{~}{J}(\tau ,J)`$-holomorphic curve. By Remark 5.1, $`\stackrel{~}{J}(\tau ,J)`$ is tamed by $`\tau `$. Hence
$$0<_D\stackrel{~}{u}^{}\tau =[\tau ],A=[\tau _0],A+s(\tau ,\tau _0).$$
The infimum of the numbers on the right is $`[\tau _0],A+\epsilon ^+(\tau _0,\mathrm{\Lambda })`$. This proves the lemma. $`\mathrm{}`$
A similar estimate for $`\epsilon ^{}(\tau _0,\mathrm{\Lambda })`$ can be obtained by studying anti-holomorphic curves. These are solutions of the equation
$$_xuX_F(u)J(_yuX_G(u))=0,$$
(33)
that satisfy the same boundary condition (31). Let us denote the moduli space of solutions by $`_A(\tau ,J)`$
###### Lemma 5.4
Let $`\mathrm{\Lambda }`$ be an exact Lagrangian loop and $`AH_2(D\times M,\mathrm{\Lambda };)`$. Suppose that for every $`\tau 𝒯^{}(\mathrm{\Lambda })`$ there exists a $`J`$ such that $`_A(\tau ,J)\mathrm{}.`$ Then
$$\epsilon ^{}(\tau _0,\mathrm{\Lambda })+[\tau _0],A0$$
for every $`\tau _0𝒯(\mathrm{\Lambda })`$.
Proof: Let $`\tau 𝒯^{}(\mathrm{\Lambda })`$ and $`u_A(\tau ,J)`$. Let $`\stackrel{~}{u}:DD\times M`$ be given by (32). Then $`\stackrel{~}{u}`$ is a $`\stackrel{~}{J}(\tau ,J)`$-holomorphic curve. By Remark 5.1, $`\stackrel{~}{J}(\tau ,J)`$ is tamed by $`\tau `$. Hence
$$0>_D\stackrel{~}{u}^{}\tau =[\tau ],A=[\tau _0],A+s(\tau ,\tau _0).$$
The supremum of the numbers on the right is $`[\tau _0],A+\epsilon ^{}(\tau _0,\mathrm{\Lambda })`$. This proves the lemma. $`\mathrm{}`$
### 5.2 Fredholm theory
In this subsection we examine the moduli spaces $`_A^\pm (\tau ,J)`$ in more detail and show that, for a generic $`J`$, these spaces are smooth manifolds of the predicted dimensions $`n\pm \mu _\mathrm{\Lambda }(A)`$. The arguments are standard (cf. ) and we shall only outline the main points. Fix an exact Lagrangian loop $`\mathrm{\Lambda }S^1\times M`$, a homology class $`AH_2(D\times M,\mathrm{\Lambda };)`$, and a constant $`p>2`$. Consider the Banach manifold
$$=W_{\mathrm{\Lambda },A}^{1,p}(D,M)$$
of all functions $`u:DM`$ of class $`W^{1,p}`$ that satisfy the boundary condition (31) and represent the class $`A`$. There is a natural vector bundle $``$ with fibres
$$_u=L^p(D,u^{}TM)$$
and the left hand sides of (30) and (33) define Fredholm sections $`^\pm :`$ given by
$$^\pm (u):=(u;\tau ,\pm J):=_xuX_F(u)\pm J(_yuX_G(u)).$$
The moduli spaces $`_A(\tau ,\pm J)`$ are the zero sets of these sections. The tangent space
$$T_u=W_\mathrm{\Lambda }^{1,p}(D,u^{}TM)$$
consists of all vector fields $`\xi W^{1,p}(D,u^{}TM)`$ along $`u`$ which are of class $`W^{1,p}`$ and satisfy the boundary condition $`\xi (e^{2\pi it})T_{u(e^{2\pi it})}\mathrm{\Lambda }_t.`$ The vertical differential of $`^\pm `$ at a zero $`u_A^\pm (\tau ,J)`$ is the linear operator
$$D_u^\pm =D^\pm (u):W_\mathrm{\Lambda }^{1,p}(D,u^{}TM)L^p(D,u^{}TM)$$
given by
$$D_u^\pm \xi ={}_{x}{}^{}\xi {}_{\xi }{}^{}X_{F}^{}(u)\pm J({}_{y}{}^{}\xi {}_{\xi }{}^{}X_{G}^{}(u))\pm ({}_{\xi }{}^{}J)(_yuX_G(u)).$$
(34)
Here $``$ denotes the Levi-Civita connection of the Riemannian metric
$$,_{x,y}=\omega (,J_{x,y})$$
and thus depends on $`x+iyD`$. The expression $`X_F`$ denotes the covariant derivative of $`X_F=X_{F_{x,y}}`$ with respect to the Levi-Civita connection at the point $`x+iy`$. The next theorem follows from the Riemann-Roch theorem for discs (see for example for a recent exposition) and the infinite dimensional implicit function theorem (see for example \[26, Appendix B\]). The proof is standard (see for example ) and will be omitted.
###### Theorem 5.5
For every $`uW_{\mathrm{\Lambda },A}^{1,p}(D,M)`$ the operators $`D_u^\pm `$ defined by (34) are Fredholm and their indices are
$$\mathrm{index}D_u^\pm =n\pm \mu _\mathrm{\Lambda }(u).$$
If $`D_u^\pm `$ is surjective for every $`u_A(\tau ,\pm J)`$ then $`_A(\tau ,\pm J)`$ is a smooth manifold of dimension
$$dim_A(\tau ,\pm J)=n\pm \mu _\mathrm{\Lambda }(A).$$
Fix an exact Lagrangian loop $`\mathrm{\Lambda }`$ and a connection $`2`$-form $`\tau 𝒯(\mathrm{\Lambda })`$. Let us denote by $`𝒥(D;M,\omega )`$ the space of all smooth families of almost complex structures $`J:D𝒥(M,\omega )`$. Such a family $`J𝒥(D;M,\omega )`$ is called regular for (30) if $`D_u^+`$ is surjective for every $`AH_2(D\times M,\mathrm{\Lambda };)`$ and every $`u_A(\tau ,J)`$. Similarly, $`J𝒥(D;M,\omega )`$ is called regular for (33) if $`D_u^{}`$ is surjective for every $`AH_2(D\times M,\mathrm{\Lambda };)`$ and every $`u_A(\tau ,J)`$. We shall denote set of all families of almost complex structures that are regular for (30), respectively (33), by
$$𝒥_{\mathrm{reg}}^\pm (\tau ,\mathrm{\Lambda })𝒥(D;M,\omega ).$$
The proof of the next theorem is a standard application of the Sard-Smale theorem (cf. ) and will be omitted.
###### Theorem 5.6
The sets $`𝒥_{\mathrm{reg}}^\pm (\tau ,\mathrm{\Lambda })`$ are of the second category in $`𝒥(D;M,\omega )`$ in the sense of Baire, i.e. they are countable intersections of open and dense subsets of $`𝒥(D;M,\omega )`$. In particular, they are dense.
Let $`\tau _0,\tau _1𝒯(\mathrm{\Lambda })`$ and choose regular families of almost complex structures
$$J_0𝒥_{\mathrm{reg}}^\pm (\tau _0,\mathrm{\Lambda }),J_1𝒥_{\mathrm{reg}}^\pm (\tau _1,\mathrm{\Lambda }).$$
By Theorem 5.5, the spaces $`_A(\tau _0,\pm J_0)`$ and $`_A(\tau _1,\pm J_1)`$ are smooth manifolds of the same dimension. These manifolds are cobordant. To construct a cobordism choose a smooth path $`[0,1]𝒯(\mathrm{\Lambda }):\lambda \tau _\lambda `$ that connects $`\tau _0`$ to $`\tau _1`$. Let us denote by
$$𝒥=𝒥([0,1]\times D,J_0,J_1;M,\omega )$$
the space of smooth homotopies $`[0,1]𝒥(D;M,\omega ):\lambda J_\lambda `$ that connect $`J_0`$ to $`J_1`$. Given $`\{J_\lambda \}𝒥`$ denote
$$𝒲_A(\{\tau _\lambda \},\{\pm J_\lambda \})=\left\{(\lambda ,u)\right|\mathrm{\hspace{0.17em}0}\lambda 1,u_A(\tau _\lambda ,\pm J_\lambda )\}.$$
A homotopy $`\{J_\lambda \}𝒥`$ is called regular if, for every $`AH_2(D\times M,\mathrm{\Lambda };)`$ and every pair $`(\lambda ,u)𝒲_A(\{\tau _\lambda \},\{\pm J_\lambda \})`$,
$$\mathrm{im}D_{\lambda ,u}^\pm +\xi _{\lambda ,u}^\pm =L^p(D,u^{}TM).$$
Here $`D_{\lambda ,u}^\pm `$ is defined by (34) with $`\tau `$ and $`J`$ replaced by $`\tau _\lambda `$ and $`J_\lambda `$, respectively, and $`\xi _{\lambda ,u}^\pm L^p(D,u^{}TM)`$ is given by
$$\xi _{\lambda ,u}^\pm :=X_{_\lambda F_\lambda }(u)\pm J_\lambda (u)X_{_\lambda G_\lambda }(u)_\lambda J_\lambda (u)(_yuX_{G_\lambda }(u)).$$
The set of all regular homotopies will be denoted by
$$𝒥_{\mathrm{reg}}^\pm (\{\tau _\lambda \},J_0,J_1,\mathrm{\Lambda })𝒥.$$
The proof of the next theorem is again standard and will be omitted.
###### Theorem 5.7
Let $`[0,1]𝒯(\mathrm{\Lambda }):\lambda \tau _\lambda `$ be a smooth family of connection $`2`$-forms and suppose that $`J_0𝒥_{\mathrm{reg}}^\pm (\tau _0,\mathrm{\Lambda })`$ and $`J_1𝒥_{\mathrm{reg}}^\pm (\tau _1,\mathrm{\Lambda }).`$ Then the sets $`𝒥_{\mathrm{reg}}^\pm (\{\tau _\lambda \},J_0,J_1,\mathrm{\Lambda })𝒥`$ are of the second category in the sense of Baire. Moreover, if $`\{J_\lambda \}𝒥_{\mathrm{reg}}^\pm (\{\tau _\lambda \},J_0,J_1,\mathrm{\Lambda })`$ then $`𝒲_A(\{\tau _\lambda \},\{\pm J_\lambda \})`$ is a smooth manifold of dimension
$$dim𝒲_A(\{\tau _\lambda \},\{\pm J_\lambda \})=n\pm \mu _\mathrm{\Lambda }(A)+1$$
with boundary
$$𝒲_A(\{\tau _\lambda \},\{\pm J_\lambda \})=_A(\tau _0,\pm J_0)_A(\tau _1,\pm J_1).$$
### 5.3 Compactness
###### Theorem 5.8
Let $`\mathrm{\Lambda }S^1\times M`$ be an exact Lagrangian loop and suppose that the pair $`(M,\mathrm{\Lambda }_0)`$ is monotone. Let $`AH_2(D\times M,\mathrm{\Lambda };)`$ and denote by $`N`$ the minimal Maslov number of the pair $`(M,\mathrm{\Lambda }_0)`$.
(i) If
$$n\pm \mu _\mathrm{\Lambda }(A)N1$$
then $`_A(\tau ,\pm J)`$ is compact for every $`\tau 𝒯(\mathrm{\Lambda })`$ and every $`J𝒥_{\mathrm{reg}}^\pm (\tau ,\mathrm{\Lambda })`$.
(ii) If
$$n\pm \mu _\mathrm{\Lambda }(A)N2$$
then $`𝒲_A(\{\tau _\lambda \},\{\pm J_\lambda \})`$ is compact for every smooth path $`[0,1]𝒯(\mathrm{\Lambda }):\lambda \tau _\lambda `$, every $`J_0𝒥_{\mathrm{reg}}^\pm (\tau _0,\mathrm{\Lambda })`$, every $`J_1𝒥_{\mathrm{reg}}^\pm (\tau _1,\mathrm{\Lambda })`$, and every regular homotopy $`\{J_\lambda \}𝒥_{\mathrm{reg}}^\pm (\{\tau _\lambda \},J_0,J_1,\mathrm{\Lambda })`$.
The proof of Theorem 5.8 relies on the following theorem about Gromov compactness for $`J`$-holomorphic discs. This result is implicitly contained in Gromov’s original paper and has been folk knowledge since then. However, the full details of the proof have not so far appeared in the literature. They were recently carried out by Frauenfelder in his Diploma thesis. In his thesis Frauenfelder also discusses the corresponding notion of stable maps for pseudoholomorphic discs.
###### Theorem 5.9 (Gromov)
Let $`(\tau ^\nu ,J^\nu )𝒯(\mathrm{\Lambda })\times 𝒥(D;M,\omega )`$ be a sequence that converges in the $`𝒞^{\mathrm{}}`$-topology to $`(\tau ,J)𝒯(\mathrm{\Lambda })\times 𝒥(D;M,\omega )`$. Let $`AH_2(D\times M,\mathrm{\Lambda };)`$ and $`u^\nu _A(\tau ^\nu ,\pm J^\nu )`$. If $`u^\nu `$ has no $`𝒞^{\mathrm{}}`$-convergent subsequence then there exist
finitely many points $`(x_i,y_i)D`$ and maps $`v_i:S^2M`$, $`i=1,\mathrm{},k`$,
finitely many points $`t_j`$ and maps $`w_j:DM`$, $`j=1\mathrm{},\mathrm{}`$,
a map $`u_0:DM`$,
such that $`v_i`$ is a nonconstant $`J_{x_i,y_i}`$-(anti)holomorphic sphere for $`i=1,\mathrm{},k`$, $`w_j`$ is a nonconstant $`J_{e^{2\pi it_j}}`$-(anti)holomorphic disc with $`w_j(D)\mathrm{\Lambda }_{t_j}`$ for $`j=1,\mathrm{},\mathrm{}`$, $`u_0_{A_0}(\tau ,\pm J)`$ for some $`A_0H_2(D\times M,\mathrm{\Lambda };)`$, and
$$A=A_0+\underset{i=1}{\overset{k}{}}[v_i]+\underset{j=1}{\overset{\mathrm{}}{}}[w_j].$$
(35)
Here $`[v_i]`$ and $`[w_j]`$ denote the induced homology classes in $`H_2(D\times M,\mathrm{\Lambda };)`$ and one of the integers $`k`$ and $`\mathrm{}`$ is nonzero.
###### Remark 5.10
(i) Let $`\stackrel{~}{M}`$ be a compact manifold and $`\stackrel{~}{L}\stackrel{~}{M}`$ be a compact submanifold of half the dimension. Suppose that $`\stackrel{~}{\omega }^\nu `$ is a sequence of symplectic forms on $`M`$ that converges to $`\stackrel{~}{\omega }`$ in the $`𝒞^{\mathrm{}}`$-topology such that $`\stackrel{~}{L}`$ is a Lagrangian submanifold of $`(\stackrel{~}{M},\stackrel{~}{\omega }^\nu )`$ for every $`\nu `$. Suppose that $`\stackrel{~}{J}^\nu `$ is a sequence of $`\stackrel{~}{\omega }^\nu `$-tame almost complex structures on $`\stackrel{~}{M}`$ that converges in the $`𝒞^{\mathrm{}}`$-topology to $`\stackrel{~}{J}`$. In Frauenfelder proves, in particular, that a sequence of $`\stackrel{~}{J}^\nu `$-holomorphic discs $`\stackrel{~}{u}^\nu :(D,D)(\stackrel{~}{M},\stackrel{~}{L})`$ that represent a fixed homology class $`AH_2(\stackrel{~}{M},\stackrel{~}{L};)`$ has a subsequence that converges (in a precisely defined sense) to a tree consisting of $`\stackrel{~}{J}`$-holomorphic spheres in $`\stackrel{~}{M}`$ and $`\stackrel{~}{J}`$-holomorphic discs in $`\stackrel{~}{M}`$ with boundary in $`\stackrel{~}{L}`$ such that the sum of their homology classes in $`H_2(\stackrel{~}{M},\stackrel{~}{L};)`$ is equal to $`A`$. The techniques in are an adaptation of those in Hofer–Salamon for holomorphic spheres to the case of holomorphic discs.
(ii) The moduli space $`_A(\tau ,\pm J)`$ does not depend on the function $`c:DM`$ in (12). Hence we may assume without loss of generality that the connection forms $`\tau ^\nu `$ in Theorem 5.9 lie in $`𝒯^\pm (\mathrm{\Lambda })`$. Under this assumption the manifold $`\stackrel{~}{M}=D\times M`$, the submanifold $`\stackrel{~}{L}=\mathrm{\Lambda }`$, the symplectic forms $`\stackrel{~}{\omega }^\nu =\pm \tau ^\nu `$, the almost complex structures $`\stackrel{~}{J}^\nu =\stackrel{~}{J}(\tau ^\nu ,\pm J^\nu )`$ defined in Remark 5.1, and the functions $`\stackrel{~}{u}^\nu `$ given by (32) satisfy the requirements of (i).
(iii) Theorem 5.9 follows from (i) and (ii) since each bubble in the limit curve is contained in a fibre of the (trivial) fibration $`D\times M`$. To see this, note that each curve $`v_i`$ appears as the limit of a sequence
$$v_i^\nu (x,y)=u^\nu (x_i^\nu +\epsilon ^\nu x,y_i^\nu +\epsilon ^\nu y),$$
where $`x_i^\nu x_i`$, $`y_i^\nu y_i`$, $`\epsilon ^\nu 0`$, and
$$\underset{\nu \mathrm{}}{lim}\frac{\epsilon ^\nu }{1\sqrt{(x_i^\nu )^2+(y_i^\nu )^2}}=0.$$
One can show that, after passing to a suitable subsequence, the sequence $`v_i^\nu `$ converges to $`v_i`$ in the $`𝒞^{\mathrm{}}`$-topology on the complement of some finite set. The functions $`v_i^\nu `$ satisfy
$$_xv_i^\nu \epsilon ^\nu X_{F^\nu }+J^\nu (_yv_i^\nu \epsilon ^\nu X_{G^\nu })=0,$$
where $`X_{F^\nu }`$, $`X_{G^\nu }`$, and $`J^\nu `$ are evaluated at the point $`(x_i^\nu +\epsilon ^\nu x,y_i^\nu +\epsilon ^\nu y,v_i^\nu )`$. It follows that the limit curve $`v_i`$ extends to a $`J_{x_i,y_i}`$-holomorphic sphere. The holomorphic discs $`w_j`$ appear as similar limits with $`x_j+iy_j=e^{2\pi it_j}`$ and
$$\underset{\nu \mathrm{}}{lim}\frac{\epsilon ^\nu }{1\sqrt{(x_j^\nu )^2+(y_j^\nu )^2}}>0.$$
A similar argument as above, with coordinates on the upper halfplane, then shows that the limit curve $`w_j`$ is a $`J_{e^{2\pi it_j}}`$-holomorphic disc with boundary values in $`\mathrm{\Lambda }_{t_j}`$.
(iv) The limit curve in (i) is a stable map consisting of $`\stackrel{~}{J}`$-holomorphic discs and spheres. For closed curves this concept is due to Kontsevich . Some of the components of the stable map may be constant. However, these do not contribute to the homology class and can be neglected for our purposes. If the original sequence $`\stackrel{~}{u}^\nu `$ does not have a $`𝒞^{\mathrm{}}`$-convergent subsequence, then the limit curve has more than one nonconstant component. This shows that in Theorem 5.9 either $`k`$ or $`\mathrm{}`$ is nonzero.
Proof of Theorem 5.8: We prove (ii). Suppose, by contradiction, that $`𝒲_A(\{\tau _\lambda \},\{\pm J_\lambda \})`$ is not compact. Then there exists a sequence
$$(\lambda ^\nu ,u^\nu )𝒲_A(\{\tau _\lambda \},\{\pm J_\lambda \})$$
that has no convergent subsequence. We may assume without loss of generality that $`\lambda ^\nu `$ converges to $`\lambda _0`$. Then, by Theorem 5.9, there exist nonconstant $`J_{\lambda _0;x_i,y_i}`$-(anti)holomorphic spheres $`v_i:S^2M`$ for $`i=1,\mathrm{},k`$, nonconstant $`J_{\lambda _0;e^{2\pi it_j}}`$-(anti)holomorphic discs $`w_j:(D,D)(M,L_{t_j})`$ for $`j=1,\mathrm{},\mathrm{}`$, and an element $`u_0_{A_0}(\tau _{\lambda _0},\pm J_{\lambda _0})`$ for some $`A_0H_2(D\times M,\mathrm{\Lambda };)`$ such that (35) is satisfied. Since the pair $`(M,\mathrm{\Lambda }_t)`$ is monotone with minimal Maslov number $`N`$ for every $`t`$ we have
$$\pm \mu _\mathrm{\Lambda }(v_i)N,\pm \mu _\mathrm{\Lambda }(w_j)N$$
for $`i=1,\mathrm{},k`$ and $`j=1,\mathrm{},\mathrm{}`$. Since either $`k`$ or $`\mathrm{}`$ is nonzero this implies
$`n\pm \mu _\mathrm{\Lambda }(A)`$ $`=`$ $`n\pm \mu _\mathrm{\Lambda }(A_0)\pm {\displaystyle \underset{i=1}{\overset{k}{}}}\mu _\mathrm{\Lambda }(v_i)\pm {\displaystyle \underset{j=1}{\overset{\mathrm{}}{}}}\mu _\mathrm{\Lambda }(w_j)`$
$``$ $`n\pm \mu _\mathrm{\Lambda }(A_0)+N.`$
Since $`\{J_\lambda \}𝒥_{\mathrm{reg}}^\pm (\{\tau _\lambda \},J_0,J_1,\mathrm{\Lambda })`$, the moduli space $`𝒲_{A_0}(\{\tau _\lambda \},\{\pm J_\lambda \})`$ is a smooth manifolds of dimension
$`dim𝒲_{A_0}(\{\tau _\lambda \},\{\pm J_\lambda \})`$ $`=`$ $`n\pm \mu _\mathrm{\Lambda }(A_0)+1`$
$``$ $`n\pm \mu _\mathrm{\Lambda }(A)+1N`$
$`<`$ $`0.`$
Hence
$$𝒲_{A_0}(\{\tau _\lambda \},\{\pm J_\lambda \})=\mathrm{},$$
in contradiction to the fact that
$$(\lambda _0,u_0)𝒲_{A_0}(\{\tau _\lambda \},\{\pm J_\lambda \}).$$
Thus we have proved (ii). The proof of (i) is almost word by word the same and will be left to the reader. $`\mathrm{}`$
### 5.4 Gromov invariants
Fix an exact Lagrangian loop $`\mathrm{\Lambda }S^1\times M`$ and a class $`AH_2(D\times M,\mathrm{\Lambda };)`$. Throughout we shall assume that the pair $`(M,\mathrm{\Lambda }_0)`$ is monotone and
$$n\pm \mu _\mathrm{\Lambda }(A)N2,$$
(36)
where $`N`$ denotes the minimal Maslov number of the pair $`(M,\mathrm{\Lambda }_0)`$. Fix a tuple $`𝐭=(t_1,\mathrm{},t_k)^k`$ such that $`0t_1<\mathrm{}<t_k<1`$ and denote
$$\mathrm{\Lambda }_𝐭:=\mathrm{\Lambda }_{t_1}\times \mathrm{}\times \mathrm{\Lambda }_{t_k}.$$
For $`\tau 𝒯(\mathrm{\Lambda })`$ and $`J𝒥(D;M,\omega )`$ we define $`\mathrm{ev}_𝐭:_A(\tau ,\pm J)\mathrm{\Lambda }_𝐭`$ by
$$\mathrm{ev}_𝐭(u):=(u(e^{2\pi it_1}),\mathrm{},u(e^{2\pi it_k})).$$
If $`J𝒥_{\mathrm{reg}}^\pm (\tau ,\mathrm{\Lambda })`$ then, by Theorems 5.5 and 5.8, the moduli space $`_A^\pm (\tau ,J)`$ is a compact smooth manifolds(without boundary) of dimension $`n\pm \mu _\mathrm{\Lambda }(A)`$. It is not necessarily orientable. Let
$$[_A(\tau ,\pm J)]H_{n\pm \mu _\mathrm{\Lambda }(A)}(_A(\tau ,\pm J);_2)$$
denote the fundamental cycle. The Gromov invariants are defined by
$$\mathrm{Gr}_{A,𝐭}^\pm (\mathrm{\Lambda }):=\mathrm{ev}_{𝐭}^{}{}_{}{}^{}[_A^\pm (\tau ,J)]H_{n\pm \mu _\mathrm{\Lambda }(A)}(\mathrm{\Lambda }_𝐭;_2).$$
(37)
###### Lemma 5.11
The homology classes $`\mathrm{Gr}_{A,𝐭}^\pm (\mathrm{\Lambda })H_{n\pm \mu _\mathrm{\Lambda }(A)}(\mathrm{\Lambda }_𝐭;_2)`$ are independent of the choices of the connection $`2`$-form $`\tau 𝒯(\mathrm{\Lambda })`$ and the almost complex structure $`J𝒥_{\mathrm{reg}}^\pm (\tau ,\mathrm{\Lambda })`$ used to define them.
Proof: Theorems 5.7 and 5.8 (ii). $`\mathrm{}`$
###### Corollary 5.12
Let $`A^\pm H_2(D\times M,\mathrm{\Lambda };)`$ satisfy (36) and suppose that
$$\mathrm{Gr}_{A^\pm ,𝐭^\pm }^\pm (\mathrm{\Lambda })0$$
for some $`𝐭^\pm `$. Then
$$\epsilon ^+(\tau ,\mathrm{\Lambda })[\tau ],A^+,\epsilon ^{}(\tau ,\mathrm{\Lambda })[\tau ],A^{}$$
for every $`\tau 𝒯(\mathrm{\Lambda })`$.
Proof: Lemmata 5.3 and 5.4. $`\mathrm{}`$
## 6 Complex projective space
In this section we shall use the Gromov invariants to compute the K-area of certain exact Lagrangian loops in $`P^n`$. The archetypal example is the half turn of a great circle in the 2-sphere. An explicit computation shows that the Hofer length of this loop is $`1/2`$. We shall use Corollary 5.12 and Theorems 3.3 and 3.5 to show that this loop minimizes the Hofer length in its Hamiltonian isotopy class.
### 6.1 Rotations of real projective space
Consider the complex projective space
$$M=P^n$$
equipped with symplectic form $`\omega `$ that is induced by the Fubini-Study metric and satisfies the normalization condition
$$_{P^n}\omega ^n=1.$$
Let $`L=P^n`$ and fix an integer $`k\{1,\mathrm{},n\}`$. As in the introduction, we consider the exact Lagrangian loop
$$\mathrm{\Lambda }:=\underset{t}{}\{e^{2\pi it}\}\times \psi _t(P^n),$$
(38)
where
$$\psi _t([z_0:\mathrm{}:z_n])=([z_0:e^{\pi it}z_1:\mathrm{}:e^{\pi it}z_k:z_{k+1}:\mathrm{}:z_n]).$$
The Hamiltonian isotopy $`\psi _t`$ is generated, via (4), by the the time independent Hamiltonian function $`H_t=H:P^n`$ given by
$$H([z_0:\mathrm{}:z_n])=\frac{k}{2n+2}\frac{|z_1|^2+\mathrm{}+|z_k|^2}{2(|z_0|^2+\mathrm{}+|z_n|^2)}.$$
(39)
This function has mean value zero and Hofer norm
$$H=\mathrm{max}H\mathrm{min}H=\frac{1}{2}.$$
Since $`H`$ attains its maximum and its minimum on $`\mathrm{\Lambda }_t=\psi _t(P^n)`$ it follows that $`\mathrm{}(\mathrm{\Lambda })=1/2.`$
### 6.2 The Maslov index
We prove that the minimal Maslov number of the pair $`(P^n,P^n)`$ is
$$N=n+1.$$
(40)
For $`n=1`$ this is obvious. For $`n>1`$ consider the homology exact sequence of the pair $`(P^n,P^n)`$. It has the form
$$0H_2(P^n;)H_2(P^n,P^n;)H_1(P^n;)0.$$
Now $`P^n`$ decomposes the line $`P^1P^n`$ into two discs that represent the same homotopy class in $`\pi _2(P^n,P^n)`$. Hence there is an element $`AH_2(P^n,P^n;)`$ such that $`2A`$ is equal to the image of the generator under the homomorphism
$$H_2(P^n;)H_2(P^n,P^n;).$$
This implies that $`A`$ is the generator of $`H_2(P^n,P^n;)`$. Since the Maslov class of $`2A\pi _2(P^n,P^n)`$ is equal to $`2c_1(TP^n),[P^1]=2n+2`$ we have proved (40).
###### Lemma 6.1
Let $`(M,\omega )`$ be a symplectic manifold and $`\mathrm{\Lambda }S^1\times M`$ be an exact Lagrangian loop such that $`(M,\mathrm{\Lambda }_0)`$ is a monotone pair with minimal Maslov number $`N`$. Then
$$\mu _\mathrm{\Lambda }(u_1)\mu _\mathrm{\Lambda }(u_0)\text{ }\text{mod}\text{ }N$$
for all $`u_0,u_1\mathrm{Map}_\mathrm{\Lambda }(D,M)`$.
Proof: If $`u_0(e^{2\pi it})=u_1(e^{2\pi it})`$ for every $`t`$ then $`u_0`$ (with reversed orientation) and $`u_1`$ form a sphere and the difference $`\mu _\mathrm{\Lambda }(u_1)\mu _\mathrm{\Lambda }(u_0)`$ is equal twice the first Chern number of this sphere. Hence the difference of the Maslov numbers is an even multiple of $`N`$. This continues to hold whenever $`u_0|_D`$ is homotopic to $`u_1|_D`$ as a section of the bundle $`\mathrm{\Lambda }S^1`$. For any two maps $`u_0,u_1\mathrm{Map}_\mathrm{\Lambda }(D,M)`$ there exists a smooth function $`v:(D,D)(M,\mathrm{\Lambda }_0)`$ such that $`v(1)=u_0(1)`$ and the connected sum $`u_0\mathrm{\#}v`$ is homotopic to $`u_1`$ along the boundary. Hence, by what we have just proved,
$$\mu _\mathrm{\Lambda }(u_1)\mu _\mathrm{\Lambda }(u_0)\mu _{\mathrm{\Lambda }_0}(v)2N.$$
Since $`\mu _{\mathrm{\Lambda }_0}(v)`$ is an integer multiple of $`N`$, the lemma is proved. $`\mathrm{}`$
Returning to the loop $`\mathrm{\Lambda }S^1\times P^n`$ we observe that (39) is a Morse-Bott function with critical manifolds
$$C^+:=\{[0:z_1:\mathrm{}:z_k:0:\mathrm{}:0]|(z_1,\mathrm{},z_k)^k\{0\}\},$$
$$C^{}:=\{[z_0:0:\mathrm{}:0:z_{k+1}:\mathrm{}:z_n]|(z_0,z_{k+1},\mathrm{},z_n)^{nk+1}\{0\}\}.$$
Note that $`H`$ attains its minimum $`(kn1)/(2n+2)`$ on $`C^+`$ and its maximum $`k/(2n+2)`$ on $`C^{}`$. Moreover, $`C^\pm P^n\mathrm{\Lambda }_t`$ for every $`t`$. Let us denote by
$$A^\pm H_2(D\times P^n,\mathrm{\Lambda };)$$
the homology classes represented by the constant functions $`DP^n`$ with values in $`C^\pm P^n`$. The next lemma shows that $`\mathrm{\Lambda }`$ has Maslov index $`k_{n+1}`$ as claimed in the introduction (see (2)). It also shows that the homology classes $`A^\pm H_2(D\times P^n,\mathrm{\Lambda };)`$ satisfy the condition (36) for the definition of the Gromov invariants.
###### Lemma 6.2
$$\mu _\mathrm{\Lambda }(A^+)=k1n,\mu _\mathrm{\Lambda }(A^{})=k.$$
(41)
Proof: In the case of $`A^{}`$, consider the constant function $`u(x,y)p:=[1:0:\mathrm{}:0].`$ Then a trivialization of the pullback tangent bundle $`u^{}TP^n`$ is determined by the coordinate chart $`[z_0:\mathrm{}:z_n](z_1/z_0,\mathrm{},z_n/z_0).`$ In these coordinates the Hamiltonian flow is $`\zeta (e^{\pi it}\zeta _1,\mathrm{},e^{\pi it}\zeta _k,\zeta _{k+1},\mathrm{},\zeta _n).`$ Since $`T_p\mathrm{\Lambda }_0^n^nT_pP^n,`$ we see that the Maslov index of the loop $`tT_p\mathrm{\Lambda }_t`$ is equal to $`k`$. This proves the second equation in (41) and the first follows from a similar argument. $`\mathrm{}`$
### 6.3 Computation of the Gromov invariants
Since $`N=n+1`$ it follows from Lemma 6.2 that the classes $`A^\pm `$ satisfy (36) and hence the requirements of Theorem 5.8. The next theorem shows that the Gromov invariants $`\mathrm{Gr}_{A^\pm ,0}^\pm (\mathrm{\Lambda })`$ are nonzero. Here the subscript $`0`$ corresponds to the choice $`𝐭=t_1=0`$ for the evaluation map.
###### Theorem 6.3
$$\mathrm{Gr}_{A^+,0}^+(\mathrm{\Lambda })=[P^{k1}]H_{k1}(P^n;_2),$$
$$\mathrm{Gr}_{A^{},0}^{}(\mathrm{\Lambda })=[P^{nk}]H_{nk}(P^n;_2).$$
Proof: Let $`\tau 𝒯(\mathrm{\Lambda })`$ be given by (12) with $`c=0`$ and
$$F_{x,y}=\frac{\mathrm{sin}(2\pi t)\rho (r)}{2\pi r}H,G_{x,y}=\frac{\mathrm{cos}(2\pi t)\rho (r)}{2\pi r}H,$$
where $`re^{2\pi it}=x+iy`$ and $`H`$ is given by (39). As in (17), $`\rho :[0,1][0,1]`$ is a smooth nondecreasing cutoff function such that $`\rho (r)=0`$ for $`r`$ near $`0`$ and $`\rho (r)=1`$ for $`r`$ near $`1`$. The formula
$$\mathrm{\Omega }_\tau (re^{2\pi it},z)=\frac{\dot{\rho }(r)}{2\pi r}H(z)$$
(42)
for $`zP^n`$ shows that $`\mathrm{\Omega }_\tau (x,y,z)0`$ for $`zC^+`$ and $`\mathrm{\Omega }_\tau (x,y,z)0`$ for $`zC^{}`$. By (42) and Lemma 5.2 with $`c=0`$ and $`E(u)=0`$,
$$[\tau ],A^+=\frac{k1n}{2n+2},[\tau ],A^{}=\frac{k}{2n+2}.$$
(43)
The explicit formulae for $`F`$ and $`G`$ show that $`C^\pm `$ consist entirely of critical points of $`F_{x,y}`$ and $`G_{x,y}`$ for all $`x+iyD`$. This shows that the constant functions $`u:DP^n`$ with values in $`C^+C^{}`$ are horizontal for the symplectic connection determined by $`\tau `$. In explicit terms $`_xu=X_F(u)`$ and $`_yu=X_G(u)`$. Hence these constant functions satisfy both equations (30) and (33) for every $`J𝒥(D;P^n,\omega )`$. The constant functions with values in $`(C^+C^{})P^n`$ satisfy in addition the boundary condition (31). The formula (41) shows that the constant solutions with values in $`C^+P^n`$ and those with values in $`C^{}P^n`$ represent different homology classes.
We prove that, for every $`J𝒥(D;P^n,\omega )`$,
$$_{A^+}(\tau ,J)=\{u:DC^+P^n|du=0\}.$$
(44)
To see this, let $`u_{A^+}(\tau ,J)`$. Then, by Lemma 5.2 and (42),
$`0`$ $``$ $`E(u)`$
$`=`$ $`[\tau ],A^++{\displaystyle _D}\mathrm{\Omega }_\tau (x,y,u(x,y))𝑑x𝑑y`$
$`=`$ $`{\displaystyle \frac{kn1}{2n+2}}{\displaystyle _0^1}{\displaystyle _0^1}\dot{\rho }(r)H(u(re^{2\pi it}))𝑑r𝑑t`$
$``$ $`{\displaystyle \frac{kn1}{2n+2}}{\displaystyle _0^1}{\displaystyle _0^1}\dot{\rho }(r)\mathrm{min}Hdrdt`$
$`=`$ $`0.`$
Hence every $`u_{A^+}(\tau ,J)`$ satisfies $`E(u)=0`$ and
$$\dot{\rho }(r)0H(u(re^{2\pi it}))=\mathrm{min}H.$$
The latter implies that $`u(x_0,y_0)C^+`$ for some point $`x_0+iy_0D`$ and the former implies that $`u`$ is a horizontal section of $`D\times M`$ with respect to $`\tau `$. Now let $`x_1+iy_1D`$, choose a path $`[0,1]D:tx(t)+iy(t)`$ that connects $`x_0+iy_0`$ to $`x_1+iy_1`$, and define $`z:[0,1]M`$ by
$$z(t):=u(x(t),y(t)).$$
Then $`z(0)C^+`$ and
$$\dot{z}(t)=\dot{x}(t)X_{F_{x(t),y(t)}}(z(t))+\dot{y}(t)X_{G_{x(t),y(t)}}(z(t)).$$
Since $`C^+`$ consists of critical points of $`F_{x,y}`$ and $`G_{x,y}`$ for all $`x+iyD`$ it follows that $`z(t)=z(0)`$ for all $`t[0,1]`$. Hence $`u`$ is constant. The boundary condition shows that this constant lies in $`C^+P^n`$. This proves (44). Hence $`_{A^+}(\tau ,J)`$ is diffeomorphic to $`P^{k1}`$ for every $`J`$ and, in particular, for every $`J𝒥_{\mathrm{reg}}^+(\tau ,\mathrm{\Lambda })`$. The evaluation map $`uu(1)`$ is obviously an embedding of $`_{A^+}(\tau ,J)P^{k1}`$ into $`P^n`$. A similar assertion holds for $`_A^{}(\tau ,J)`$ and this proves the theorem. $`\mathrm{}`$
### 6.4 Invariants of projective Lagrangian loops
Let $`\mathrm{PL}(n+1)`$ denote the manifold of projective Lagrangian planes in $`P^n`$. There is a fibration
$$S^1/\{\pm 1\}\mathrm{L}(n+1)\mathrm{PL}(n+1),$$
where $`\mathrm{L}(n+1)`$ denotes the manifold of Lagrangian subspaces of $`^{n+1}`$ and $`S^1`$ acts by multiplication. The generator $`te^{\pi it}`$ of $`\pi _1(S^1/\{\pm 1\})`$ gives rise to a loop of Lagrangian subspaces of Maslov index $`n+1`$. Hence the homotopy exact sequence of the fibration shows that the fundamental group of $`\mathrm{PL}(n+1)`$ is isomorphic to $`_{n+1}`$. For $`k`$ we denote by $`\mathrm{\Lambda }^kS^1\times P^n`$ the exact Lagrangian loop defined by (1) in the introduction, i.e $`\mathrm{\Lambda }_t^k:=\varphi _{kt}(P^n),`$ where $`\varphi _t([z_0:\mathrm{}:z_n])=[e^{\pi it}z_0:z_1:\mathrm{}:z_n].`$ If $`k`$ is divisible by $`n+1`$ then this loop is contractible. If $`k\{1,\mathrm{},n\}`$ and $`kk^{}\mathrm{mod}n+1`$ then $`\mathrm{\Lambda }^k^{}`$ is Hamiltonian isotopic to $`\mathrm{\Lambda }`$.
###### Corollary 6.4
If $`k`$ is not divisible by $`n+1`$ then
$$\nu (\mathrm{\Lambda }^k)=\chi (\mathrm{\Lambda }^k)=\epsilon (\mathrm{\Lambda }^k)=\frac{1}{2}.$$
If $`k`$ is divisible by $`n+1`$ then $`\nu (\mathrm{\Lambda }^k)=\chi (\mathrm{\Lambda }^k)=0.`$
Proof: The loop $`\mathrm{\Lambda }`$, given by (38), is Hamiltonian isotopic to $`\mathrm{\Lambda }^k`$ and hence
$$\epsilon (\mathrm{\Lambda }^k)=\epsilon (\mathrm{\Lambda }),\chi (\mathrm{\Lambda }^k)=\chi (\mathrm{\Lambda }),\nu (\mathrm{\Lambda }^k)=\nu (\mathrm{\Lambda }).$$
By Theorem 6.3, $`\mathrm{Gr}_{A^+,0}^+(\mathrm{\Lambda })0`$ and $`\mathrm{Gr}_{A^{},0}^{}(\mathrm{\Lambda })0.`$ Hence, by Corollary 5.12 and (43),
$$\epsilon ^+(\tau ,\mathrm{\Lambda })[\tau ],A^+=\frac{n+1k}{2n+2},\epsilon ^{}(\tau ,\mathrm{\Lambda })[\tau ],A^{}=\frac{k}{2n+2}.$$
Here $`\tau 𝒯(\mathrm{\Lambda })`$ denotes the connection $`2`$-form introduced in the proof of Theorem 6.3. Hence
$$\epsilon (\mathrm{\Lambda })=\epsilon ^+(\tau ,\mathrm{\Lambda })\epsilon ^{}(\tau ,\mathrm{\Lambda })\frac{1}{2}.$$
Since $`\nu (\mathrm{\Lambda })1/2`$ the result follows from Theorems 3.3 and 3.5. $`\mathrm{}`$
###### Remark 6.5
Our invariants do not distinguish between $`\mathrm{\Lambda }^j`$ and $`\mathrm{\Lambda }^k`$ unless one of the numbers is divisible by $`n+1`$ and the other is not. However, if
$$\mathrm{gcd}(j,n+1)\mathrm{gcd}(k,n+1)$$
then the iterated loops $`\mathrm{\Lambda }^{mj}`$ and $`\mathrm{\Lambda }^{mk}`$ have different invariants for some $`m`$. To see this suppose, without loss of generality, that $`\mathrm{gcd}(j,n+1)<\mathrm{gcd}(k,n+1)`$ and denote
$$m:=\frac{n+1}{\mathrm{gcd}(k,n+1)}<\frac{n+1}{\mathrm{gcd}(j,n+1)}.$$
Then $`mk`$ is divisible by $`n+1`$ whereas $`mj`$ is not. By Corollary 6.4,
$$\nu (\mathrm{\Lambda }^{mj})\nu (\mathrm{\Lambda }^{mk}).$$
In the case of Hamiltonian loops the analogue of the line $`𝒯(\mathrm{\Lambda })`$ has a natural basepoint and in that case there are separate invariants $`\epsilon ^+(P)`$ and $`\epsilon ^{}(P)`$ that contain finer information than their difference.
###### Remark 6.6
We conjecture that the constant loop $`\mathrm{\Lambda }^0=S^1\times P^n`$ satisfies $`\epsilon (\mathrm{\Lambda }^0)=0`$. This does not follow from the techniques of this paper. The homology class $`A^0H^2(D\times P^n,S^1\times P^n;),`$ represented by the constant maps $`DP^n`$, satisfies $`\mu _{\mathrm{\Lambda }^0}(A^0)=0.`$ Hence $`A^0`$ does not satisfy our condition (36) for the definition of the Gromov invariants, although the arguments of Theorem 6.3 carry over to the constant loop $`\mathrm{\Lambda }^0`$ with $`A^+=A^{}=A^0`$. It should be possible to circumvent the problems arising from Gromov compactness by using the invariants introduced in Cieliebak–Gaio–Salamon . We expect that these techniques apply to the constant loop $`\mathrm{\Lambda }^0`$ in $`P^n`$.
###### Remark 6.7
We expect that the techniques of also apply to symplectic quotients of $`^n`$ that do not satisfy our monotonicity hypothesis. This should give rise to results similar to the ones in this section for general toric varieties.
###### Remark 6.8
Let $`(M,\omega )`$ be a symplectic $`2n`$-manifold and $`L`$ be a closed $`n`$-manifold with $`H^1(L;)=0`$. In Weinstein considers the space of all pairs $`(\mathrm{\Lambda },\rho )`$ where $`\mathrm{\Lambda }M`$ is a Lagrangian submanifold diffeomorphic to $`L`$ and $`\rho `$ is a volume form on $`\mathrm{\Lambda }`$ (or a smooth measure in the nonorientable case). He interpretes this space as the cotangent bundle of $`=(M,\omega ,L)`$ and examines the symplectic action functional on the loop space of $`T^{}`$. In Donaldson interpretes this cotangent bundle as a symplectic quotient of the space of all embeddings $`\iota :LM`$ with vanishing cohomology class $`\iota ^{}[\omega ]`$ by the group of volume preserving diffeomorphisms of $`L`$ (with respect to any given smooth measure). The group action is Hamiltonian and the zero set of the moment map is the space of Lagrangian embeddings of $`L`$ into $`(M,\omega )`$. It would be interesting to examine analogues of the invariants studied in the present paper for loops in $`T^{}`$ and relate these to the work of Weinstein and Donaldson. This will be investigated in .
## Appendix A Symplectic isotopy on Riemann surfaces
The following results are known. However, we could not find proofs in the literature and present them here for the sake of completeness.
###### Proposition A.1
Let $`\mathrm{\Sigma }`$ be a compact connected oriented Riemann surface with area form $`\omega `$ and $`S,S^{}\mathrm{\Sigma }`$ be two closed embedded discs with the same area. Then there exists a Hamiltonian symplectomorphism $`\psi :\mathrm{\Sigma }\mathrm{\Sigma }`$ such that $`\psi (S)=S^{}.`$
The proof relies on the following three lemmata. The first asserts that, in dimension $`2`$, a symplectomorphism is smoothly isotopic to the identity if and only if it is symplectically isotopic to the identity. For the $`2`$-torus this follows from the characterization of Hamiltonian symplectomorphisms in Conley–Zehnder \[5, Theorem 6\]. In general the proof is a parametrized version of Moser isotopy. The work of Seidel shows that the result has no analogue in higher dimensions.
###### Lemma A.2
Let $`\mathrm{\Sigma }`$ be a compact oriented Riemann surface with area form $`\omega `$ and $`\psi :\mathrm{\Sigma }\mathrm{\Sigma }`$ be a symplectomorphism. Then $`\psi `$ is smoothly isotopic to the identity if and only if it is symplectically isotopic to the identity.
Proof: Let $`[0,1]\mathrm{Diff}(\mathrm{\Sigma }):t\psi _t`$ be a smooth isotopy such that $`\psi _0=\mathrm{id}`$ and $`\psi _1=\psi `$. Define
$$\omega _t:=\psi _{t}^{}{}_{}{}^{}\omega ,\omega _{s,t}:=s\omega +(1s)\omega _t$$
for $`0s,t1`$. Then $`\omega _{s,0}=\omega _{s,1}=\omega _{1,t}=\omega `$ and $`\omega _{0,t}=\omega _t`$ for all $`s`$ and $`t`$. Fix a Riemannian metric on $`\mathrm{\Sigma }`$ with volume form $`\omega `$ and let $`\alpha _t\mathrm{\Omega }^1(\mathrm{\Sigma })`$ be defined by
$$d\alpha _t=\omega _t\omega ,\alpha _t\mathrm{im}d^{}.$$
Choose $`X_{s,t}\mathrm{Vect}(\mathrm{\Sigma })`$ such that $`\iota (X_{s,t})\omega _{s,t}=\alpha _t`$ and define $`\psi _{s,t}\mathrm{Diff}(\mathrm{\Sigma })`$ by
$$_s\psi _{s,t}=X_{s,t}\psi _{s,t},\psi _{0,t}=\psi _t.$$
Then $`_s(\psi _{s,t}^{}{}_{}{}^{}\omega _{s,t})=0`$ and $`\psi _{0,t}^{}{}_{}{}^{}\omega _{0,t}=\omega `$. Hence $`\psi _{s,t}^{}{}_{}{}^{}\omega _{s,t}=\omega `$ for all $`s`$ and $`t`$. Moreover, $`\psi _{s,0}=\mathrm{id}`$ and $`\psi _{s,1}=\psi `$ for all $`s`$. Hence $`t\psi _{1,t}`$ is the required symplectic isotopy from $`\mathrm{id}`$ to $`\psi `$. $`\mathrm{}`$
###### Lemma A.3
Let $`\mathrm{\Sigma }`$ be a compact oriented Riemann surface, $`S\mathrm{\Sigma }`$ be an embedded closed disc, and $`\omega _0,\omega _1\mathrm{\Omega }^2(\mathrm{\Sigma })`$ be two area forms such that
$$_\mathrm{\Sigma }(\omega _1\omega _0)=_S(\omega _1\omega _0)=0.$$
Then there exists a smooth isotopy $`\psi _t:\mathrm{\Sigma }\mathrm{\Sigma }`$ such that
$$\psi _0=\mathrm{id},\psi _{1}^{}{}_{}{}^{}\omega _1=\omega _0,\psi _t(S)=S$$
for every $`t[0,1]`$.
Proof: The result follows again from Moser isotopy. We prove that there exists a $`1`$-form $`\alpha \mathrm{\Omega }^1(\mathrm{\Sigma })`$ such that
$$d\alpha +\omega _1\omega _0=0,\alpha |_{TS}=0.$$
(45)
To see this, choose any $`1`$-form $`\beta \mathrm{\Omega }^1(\mathrm{\Sigma })`$ such that $`d\beta +\omega _1\omega _0=0`$. Then the integral of $`\beta `$ over $`S`$ vanishes and so $`\beta |_S`$ is exact. Hence there exists a smooth function $`f:\mathrm{\Sigma }`$ such that $`(\beta df)|_{TS}=0`$ and the $`1`$-form $`\alpha :=\beta df`$ satisfies (45). Now let $`\omega _t:=t\omega _1+(1t)\omega _0`$ and define $`X_t\mathrm{Vect}(\mathrm{\Sigma })`$ and $`\psi _t\mathrm{Diff}(\mathrm{\Sigma })`$ by
$$_t\psi _t=X_t\psi _t,\iota (X_t)\omega _t=\alpha ,\psi _0=\mathrm{id}.$$
Then $`X_t`$ is tangent to $`S`$ for every $`t`$. Hence $`\psi _t`$ preserves $`S`$ and $`\psi _{t}^{}{}_{}{}^{}\omega _t=\omega _0`$ for every $`t`$. This proves the lemma. $`\mathrm{}`$
###### Lemma A.4
Let $`\mathrm{\Sigma }`$ be a compact connected Riemann surface and $`S,S^{}\mathrm{\Sigma }`$ be two embedded discs. Then there exists a diffeomorphism $`f:\mathrm{\Sigma }\mathrm{\Sigma }`$ such that $`f`$ is isotopic to the identity and $`f(S)=S^{}`$.
Proof: Choose orientation preserving embeddings $`\varphi ,\varphi ^{}:B_1\mathrm{\Sigma }`$ such that $`\varphi (B_1)=S`$ and $`\varphi ^{}(B_1)=S^{}`$. We prove the result in four steps.
Step 1: There exists a diffeomorphism $`g:\mathrm{\Sigma }\mathrm{\Sigma }`$ that is isotopic to the identity and satisfies $`g\varphi (0)=\varphi ^{}(0)`$.
Choose a path $`\gamma :[0,1]\mathrm{\Sigma }`$ such that $`\gamma (0)=\varphi (0)`$ and $`\gamma (1)=\varphi ^{}(0)`$. Next choose a smooth family of vector fields $`X_t\mathrm{Vect}(\mathrm{\Sigma })`$ such that $`X_t(\gamma (t))=\dot{\gamma }(t)`$ for every $`t`$. Then the diffeomorphisms $`g_t:\mathrm{\Sigma }\mathrm{\Sigma }`$, defined by $`_tg_t=X_tg_t`$ and $`g_0=\mathrm{id}`$, satisfy $`g_t(\gamma (0))=\gamma (t)`$ for every $`t`$. Hence $`g_1`$ satisfies the requirements of Step 1.
Step 2: $`\varphi `$ can be chosen such that $`d(g\varphi )(0)=d\varphi ^{}(0)`$.
We prove that, for every matrix $`\mathrm{\Psi }^{2\times 2}`$ such that $`det(\mathrm{\Psi })>0`$, there exists a diffeomorphism $`\psi :B_1B_1`$ such that $`d\psi (0)=\mathrm{\Psi }`$. To see this, let
$$P:=(\mathrm{\Psi }^T\mathrm{\Psi })^{1/2}$$
and choose $`Q\mathrm{SO}(2)`$ and $`a,b>0`$ such that $`QPQ^T=\mathrm{diag}(a,b)`$. Next choose smooth functions $`\alpha ,\beta :[0,1][0,1]`$ such that $`\dot{\alpha }(r)>0`$ and $`\dot{\beta }(r)>0`$ for all $`r`$ and
$$\alpha (r)=\{\begin{array}{cc}\hfill ar,& \text{for }r\text{ near }0,\hfill \\ \hfill r,& \text{for }r\text{ near }1,\hfill \end{array}\beta (r)=\{\begin{array}{cc}\hfill br,& \text{for }r\text{ near }0,\hfill \\ \hfill r,& \text{for }r\text{ near }1.\hfill \end{array}$$
Then the diffeomorphism $`\psi _0:B_1B_1`$ defined by
$$\psi _0(x,y)=(\alpha (|x|)x/|x|,\beta (|y|)y/|y|)$$
satisfies $`d\psi _0(0)=\mathrm{diag}(a,b)`$. Hence the function
$$\psi (z):=\mathrm{\Psi }(\mathrm{\Psi }^T\mathrm{\Psi })^{1/2}Q^T\psi _0(Qz)$$
is a diffeomorphism of $`B_1`$ and satisfies $`d\psi (0)=\mathrm{\Psi }`$. To prove Step 2, let $`\mathrm{\Psi }`$ be defined by
$$d(g\varphi )(0)\mathrm{\Psi }=d\varphi ^{}(0),$$
choose a diffeomorphism $`\psi :B_1B_1`$ such that $`d\psi (0)=\mathrm{\Psi }`$, and replace $`\varphi `$ by $`\varphi \psi `$.
Step 3: $`\varphi `$ can be chosen such that $`g\varphi (z)=\varphi ^{}(z)`$ for $`|z|`$ sufficiently small.
By Step 2, we may assume that $`d(g\varphi )(0)=d\varphi ^{}(0)`$. Choose $`\delta >0`$ such that $`\varphi ^{}(B_\delta )g(S)`$ and consider the function
$$h:=\varphi ^1g^1\varphi ^{}:B_\delta B_1.$$
This function is an embedding and satisfies $`dh(0)=1\mathrm{l}`$. Choose a smooth cutoff function $`\beta :[0,1][0,1]`$ such that $`\beta (r)=1`$ for $`r1/3`$ and $`\beta (r)=0`$ for $`r2/3`$. For $`0<\epsilon <\delta `$ define $`h_\epsilon :B_1B_1`$ by
$$h_\epsilon (z):=\beta (|z|/\epsilon )h(z)+(1\beta (|z|/\epsilon ))z.$$
Then $`h_\epsilon `$ is a diffeomorphism for $`\epsilon >0`$ sufficiently small and $`g\varphi h_\epsilon (z)=\varphi ^{}(z)`$ for $`|z|<\epsilon /3`$. Hence the embedding $`\varphi h_\epsilon `$ satisfies the requirements of Step 3 for $`\epsilon >0`$ sufficiently small.
Step 4: We prove the lemma.
By Step 3, there exist embeddings $`\varphi ,\varphi ^{}:B_1\mathrm{\Sigma }`$, a constant $`\epsilon >0`$, and a diffeomorphism $`g:\mathrm{\Sigma }\mathrm{\Sigma }`$ such that $`g`$ is isotopic to the identity and
$$|z|<\epsilon g\varphi (z)=\varphi ^{}(z).$$
Choose $`\delta >0`$ such that $`\varphi `$ and $`\varphi ^{}`$ extend to embeddings of $`B_{1+\delta }`$ into $`\mathrm{\Sigma }`$. Choose a smooth function $`\rho :[0,1+\delta ][0,1+\delta ]`$ such that $`\dot{\rho }(r)>0`$ for every $`r`$ and
$$\rho (r)=\{\begin{array}{cc}\hfill r,& \text{for }r\epsilon /2,\hfill \\ \hfill 1,& \text{for }r=\epsilon ,\hfill \\ \hfill r,& \text{for }r1+\delta /2.\hfill \end{array}$$
Let $`f:\mathrm{\Sigma }\mathrm{\Sigma }`$ be given by
$$f(\varphi (z)):=\varphi (\rho (|z|)z/|z|)$$
for $`zB_{1+\delta }`$ and by $`f=\mathrm{id}`$ in $`\mathrm{\Sigma }\varphi (B_{1+\delta })`$. Then $`f`$ is isotopic to the identity and $`f\varphi (B_\epsilon )=S`$. Similarly, there exists a diffeomorphism $`f^{}:\mathrm{\Sigma }\mathrm{\Sigma }`$ that is isotopic to the identity and satisfies $`f^{}\varphi ^{}(B_\epsilon )=S^{}`$. The diffeomorphism $`f^{}gf^1`$ is isotopic to the identity and maps $`S`$ to $`S^{}`$. This proves the lemma. $`\mathrm{}`$
Proof of Proposition A.1: By Lemma A.4, there exists a diffeomorphism $`f:\mathrm{\Sigma }\mathrm{\Sigma }`$ that is isotopic to the identity and satisfies $`f(S)=S^{}.`$ Since $`S`$ and $`S^{}`$ have the same area, we obtain
$$_\mathrm{\Sigma }(f^{}\omega \omega )=_S(f^{}\omega \omega )=0.$$
By Lemma A.3, there exists a diffeomorphism $`\psi :\mathrm{\Sigma }\mathrm{\Sigma }`$ that is isotopic to the identity and satisfies
$$\psi ^{}f^{}\omega =\omega ,\psi (S)=S.$$
Hence $`\varphi :=f\psi `$ is isotopic to the identity and
$$\varphi ^{}\omega =\omega ,\varphi (S)=S^{}.$$
By Lemma A.2, $`\varphi `$ is symplectically isotopic to the identity. Let $`t\varphi _t`$ be a symplectic isotopy such that $`\varphi _0=\mathrm{id}`$ and $`\varphi _1=\varphi `$. Then the embedded discs $`S_t:=\varphi _t(S)`$ all have the same area and $`S_0=S`$, $`S_1=S^{}`$. Hence $`tS_t`$ is an exact Lagrangian path. By Lemma 2.3, there exists a Hamiltonian isotopy $`t\psi _t`$ of $`\mathrm{\Sigma }`$ such that $`\psi _t(S_0)=S_t`$ for all $`t`$. Hence $`\psi _1(S)=S^{}`$ and this proves the proposition. $`\mathrm{}`$ |
warning/0003/gr-qc0003073.html | ar5iv | text | # Closed Lightlike Curves in Non-linear Electrodynamics
## I Introduction
### A Introductory Remarks
One of the most elegant characteristics that singles out the gravitational interaction consists in the possibility –turned into an actual formulation of gravity by Einstein — of associating gravitational phenomena to the metric structure of the spacetime. Such a geometrical view is impossible to be mimic by processes envolving other kinds of forces. The main reason for this is that all other known forces do not show the universal property that is typical of gravity. Indeed, this was the main drawback that induced Einstein unified program to its failure. However an interesting analogy with the Einstein way of looking into some processes appeared recently, carrying a less ambitious program but having a deep connection with it. It consists in a fresh method that —taking into account such limitation of non-universality — looks for special physical situations that allow an equivalent description in terms of an effective modification of the geometry of spacetime. In other words: even if a given kind of force cannot be geometrized, one can discover some special (and non-trivial) situations in which a restricted geometrization is possible. It is clearly understood that it is by no means a geometrization scheme in a broad sense but only a very limited one, although very useful, as we shall see. The interest on this arises, of course, from the possibility of applying such geometrization in a sufficiently large and meaningfull set of examples. This is precisely the situation that one encounters in some interesting and diverse circumstances which has led to the claim that nongravitational processes can indeed simulate modifications of the geometry of spacetime.
Just to consider two cases that have been presented in a coherent and self-contained way we refer to and that deal respectivelly with:
* The propagation of photons in non-linear electrodynamics;
* Processes envolving certain properties of superfluid $`{}_{}{}^{3}He.`$
In this paper we will deal only with the first case. Before going into this let us make another rather general comment to clarify our work here. We will be concerned with the propagation of photons in a nonlinear electrodynamics in terms of a modification of the metric of the spacetime. Such modification is nothing but an effective structure that yields an equivalent description of the photon paths. This should not be taken as an universal modification of the geometry of the spacetime. We shall see however that such geometric tool is very powerful and allows the analysis of light propagation to be accomplished in a very simple way. Besides, with this method we can transpose part of the behavior of photons from the well-known combined Maxwell-Einstein framework to the nonlinear case of electrodynamics. An example of this analogy will be presented in this paper. It concerns the possibility of the presence of photons closed paths in spacetime. We will see that the remarkable Gödel analysis of the existence of closed timelike curves (CTC) in a rotating universe can be transposed to the case of photons in non-linear electrodynamics. The electromagnetic field generated by a charged string yields the possibility of the existence of closed lightlike curves (CLC) for the photons.
### B Synopsis
In the last years there has been an increasing interest on properties of the gravitational field that describe, within the context of General Relativity(GR), geometries that allow the appearance of closed timelike curves (CTC) (see, for instance , , , , ). Such unusual geometries, exact solutions of the classical Einstein equations of GR, pose a thorough problem of compatibility in the realm of field theory — just to quote one difficulty — and it is a real challenge to deal with them. As an example, we can point out the case of traversable wormholes that allow the existence of two nonequivalent paths for the possible travel of a real observer to go from one point $`P`$ of spacetime to another point $`Q`$ inducing thus the existence of a closed path in spacetime. Gödel cosmological solution is another case which also provides such kind of undesirable paths. The attraction for such geometries rests on the deep understanding of the theory allowed by their analysis. In the present paper we show that similar paths can be generated in configurations of pure electromagnetic field in a non-linear regime.
In order to achieve such a proof we must first review some recent papers that show this hidden geometrical character of the photon propagation in nonlinear electrodynamics , , . We will recover such results in a very simple way in the next section.
### C Definitions and notations
We call the electromagnetic tensor $`F_{\mu \nu }`$, while its dual $`F_{\mu \nu }^{}`$ is
$$F_{\alpha \beta }^{}\frac{1}{2}\eta _{\alpha \beta }^{\mu \nu }F_{\mu \nu },$$
(1)
where $`\eta _{\alpha \beta \mu \nu }`$ is the completely antisymmetric Levi-Civita tensor; the Minkowski metric tensor is represented by its standard form $`\eta ^{\mu \nu }.`$ The two invariants constructed with these tensors are defined as
$$FF^{\mu \nu }F_{\mu \nu },$$
(2)
$$GF^{\mu \nu }F_{\mu \nu }^{}.$$
(3)
Once the modifications of the linearity of electrodynamics which will be dealt here with do not break the gauge invariance of the theory, we can restrict our analysis to the general form of the modified Lagrangian for electrodynamics, written as a functional of the above invariants. In the present paper we limit our analysis to the case in which the Lagrangian depends only on $`F:`$
$$L=L(F).$$
(4)
We denote by $`L_F`$ the derivative of the Lagrangian $`L`$ with respect to the invariant $`F`$; and similarly for the higher order derivatives. We are particularly interested in the derivation of the characteristic surfaces which guide the propagation of the field discontinuities.
## II The Method of the Effective Geometry
We will make a short review of the Hadamard method in order to obtain the propagation equations for the discontinuities of the electromagnetic field. We limit our analysis to the case in which all modifications on the linear electrodynamics can be described by a Lagrangian $`L`$ as in Eq. (4)
Let $`\mathrm{\Sigma }`$ be a surface of discontinuity for the electromagnetic field. Following Hadamard we assume that the field itself is continuous when crossing $`\mathrm{\Sigma }`$, while its first derivative presents a finite discontinuity. We accordingly set
$$[F_{\mu \nu }]_\mathrm{\Sigma }=0,$$
(5)
and
$$[_\lambda F_{\mu \nu }]_\mathrm{\Sigma }=f_{\mu \nu }k_\lambda ,$$
(6)
in which the symbol
$$[J]_\mathrm{\Sigma }\underset{\delta 0^+}{lim}\left(J|_{\mathrm{\Sigma }+\delta }J|_{\mathrm{\Sigma }\delta }\right)$$
represents the discontinuity of the arbitrary function $`J`$ through the surface $`\mathrm{\Sigma }`$ characterized by the equation $`\mathrm{\Sigma }(x^\mu )=constant`$. The tensor $`f_{\mu \nu }`$ is called the discontinuity of the field, and
$$k_\lambda =_\lambda \mathrm{\Sigma }$$
(7)
is the propagation vector.
The equations of motion are
$$_\nu \left(L_FF^{\mu \nu }\right)=0.$$
(8)
Following the definitions and procedure presented above one gets from the discontinuity of the equation of motion Eq. (8):
$$L_Ff^{\mu \nu }k_\nu +2L_{FF}\xi F^{\mu \nu }k_\nu =0,$$
(9)
where $`\xi `$ is defined by
$$\xi F^{\alpha \beta }f_{\alpha \beta }.$$
(10)
The cyclic identity yields
$$f_{\mu \nu }k_\lambda +f_{\nu \lambda }k_\mu +f_{\lambda \mu }k_\nu =0.$$
(11)
Multiplying this equation by $`k^\lambda F^{\mu \nu }`$ gives
$$\xi k_\nu k_\mu \gamma ^{\mu \nu }+2F^{\mu \nu }f_{\nu \lambda }k^\lambda k_\mu =0,$$
(12)
in which $`\gamma _{\mu \nu }`$ is the Minkowski metric tensor written in an arbitrary coordinate system. From the Eq. (9) it results:
$$f_{\mu \nu }k^\nu =\mathrm{\hspace{0.17em}2}\frac{L_{FF}}{L_F}\xi F_{\mu \nu }k^\nu .$$
(13)
After some algebraic manipulations the equation of propagation of the disturbances is obtained:
$$\left\{\gamma ^{\mu \nu }+\mathrm{\Lambda }^{\mu \nu }\right\}k_\mu k_\nu =0$$
(14)
in which the quantity $`\mathrm{\Lambda }^{\mu \nu }`$ is
$$\mathrm{\Lambda }^{\mu \nu }4\frac{L_{FF}}{L_F}F^{\mu \alpha }F_\alpha ^\nu .$$
(15)
It then follows that the photon path is kinematically described by
$$g^{\mu \nu }k_\mu k_\nu =0,$$
(16)
where the effective metric $`g^{\mu \nu }`$ is given by
$$g^{\mu \nu }=L_F\gamma ^{\mu \nu }4L_{FF}F^\mu _\lambda F^{\lambda \nu .}$$
(17)
Furthermore, once the wave vector $`k_\alpha `$ is a gradient, the photon path is a geodesic in the effective geometry . We re-obtained then the remarkable result that the discontinuities of the electromagnetic field in a nonlinear electrodynamics propagates along null geodesics of an effective geometry which depends on the properties of the background field<sup>1</sup><sup>1</sup>1The proof that the path is indeed a geodesics is given in the appendix B..
From the general expression of the energy-momentum tensor for an electromagnetic theory $`L=L(F)`$ we have
$$T_{\mu \nu }=4L_FF_{\mu }^{}{}_{}{}^{\alpha }F_{\alpha \nu }L\gamma _{\mu \nu }.$$
(18)
We can then re-write the effective geometry in a more appealing form in terms of the energy momentum tensor. We obtain, using Eq. (18) into Eq. (17)
$$g^{\mu \nu }=\gamma ^{\mu \nu }+𝒩T^{\mu \nu },$$
(19)
where the functions $``$ and $`𝒩`$ are given by
$$=L_F+\frac{LL_{FF}}{L_F},$$
(20)
$$𝒩=\frac{L_{FF}}{L_F}$$
(21)
As a consequence of this, the Minkowskian norm of the propagation vector $`k_\mu `$ reads
$$\gamma ^{\mu \nu }k_\mu k_\nu =\frac{𝒩}{}T^{\mu \nu }k_\mu k_\nu .$$
(22)
## III Charged String
The physical system we will analyse consists in a (idealized infinitely long) thin charged cylinder<sup>2</sup><sup>2</sup>2We neglect all gravitational effects once there is no substantial difference introduced by gravity, as far as the phenomenon we are interested to exhibit here is concerned.. The flat Minkowskian background geometry written in a $`(t,r,\phi ,z)`$ coordinate system takes the form
$$ds^2=dt^2dr^2+2h_0d\phi dt+g(r)d\phi ^2dz^2$$
(23)
where $`g(r)=h_0^2\omega ^2r^2.`$ Although locally such geometry is Minkowskian, it can be associated to a spinning string, due to its global properties. We shall see that this has no effect on our analysis, once CLC’s do not exist in the Maxwell-Einstein theory but only in the case of nonlinear electrodynamics. This shows that our example of CLC is not a gravitational effect. Furthermore, one could set $`\omega =1`$ in all calculations without any qualitative changing in our result. In order to avoid any artificial trouble with causality we will limit the range of the coordinate $`r`$ to be strictly larger than $`r_0,`$ that is, $`r>r_0`$ where $`r_0^2=h_0^2/\omega ^2.`$ In this domain of validity, this system is regular and well defined .
From the symmetry properties of this system, the unique non-null component of the electric field is $`F_{01}=E(r)`$. In this case<sup>3</sup><sup>3</sup>3In order to avoid any difficulty with the coordinates we take the radius of the charged string to be greater that the value $`r_0.`$, the equation of motion reduces to
$$L_FE=\frac{Q}{r}$$
(24)
where $`Q`$ is a constant. We are interested here in the analysis of the propagation of electromagnetic waves in such background. Following our previous treatment the photons propagate as if the metric structure of spacetime were changed into an effective Riemannian geometry. From Eq.(17) we obtain the components of the effective metric. The non-vanishing covariant components<sup>4</sup><sup>4</sup>4See the appendix A. are:
$$g_{tt}=\frac{\omega ^2r^2}{h_0^2+\mathrm{\Psi }\omega ^2r^2}$$
(25)
$$g_{t\phi }=h_0g_{tt}$$
(26)
$$g_{rr}=\frac{1}{\mathrm{\Lambda }}$$
(27)
$$g_{\phi \phi }=\mathrm{\Psi }\omega ^2r^2g_{tt}$$
(28)
$$g_{zz}=\mathrm{\hspace{0.17em}1},$$
(29)
where $`\mathrm{\Psi }`$ and $`\mathrm{\Lambda }`$ are
$$\mathrm{\Psi }=1\left(\frac{h_0}{\omega r}\right)^24\frac{L_{FF}E^2}{L_F}$$
(30)
$$\mathrm{\Lambda }=\mathrm{\hspace{0.17em}1}4\frac{L_{FF}E^2}{L_F}.$$
(31)
The photon paths are null geodesics in such modified geometry. Let us consider the curve defined by the equations $`t=constant,`$ $`r=constant,`$ and $`z=constant.`$ Along such a curve the element of length reduces to
$$ds_{eff}^2=\mathrm{\Psi }\omega ^4r^4\left(\frac{1}{h_0^2+\mathrm{\Psi }\omega ^2r^2}\right)d\phi ^2.$$
(32)
Thus, for the photon to follow such a path the radius $`r=r_c`$ must be such that $`\mathrm{\Psi }(r_c)=0.`$
Let us emphasize that the possibility of the presence of CLC’s depends crucially on the non-linearity of the electromagnetic field. Indeed, it is a direct consequence of the form of the above geometry that $`\mathrm{\Psi }`$ can vanish only if $`L_{FF}`$ is different from zero. In the linear Maxwell electrodynamics this phenomenon is forbidden<sup>5</sup><sup>5</sup>5Let us remark that a similar analysis on the photon path can be made for Maxwell theory in a non-linear dielectric medium. This is a direct consequence of the propagation equations in a non-linear dielectric medium as it was shown in . . Thus we are allowed to claim that it is a new property which depends on the non-linearity of the electromagnetic process. This is a general formalism valid for arbitrary form of the Lagrangian. We now turn to a specific example in which such situation occurs.
### A A toy model
We set for the nonlinear Lagrangian the form<sup>6</sup><sup>6</sup>6This form is very similar to Born-Infeld (BI) model. However, the sign in the field term inside the square-root is opposite to that used in BI Lagrangian. This makes a crucial difference in what concerns the appearance of CLC, as shown in the text.:
$$L=\frac{b^2}{2}\left(\sqrt{1\frac{F}{b^2}}1\right)$$
(33)
in which $`b`$ is an arbitrary constant.
A solution of the equation for the electric field yields
$$E=4bQ\left(b^2r^232Q^2\right)^{\frac{1}{2}}.$$
(34)
The field must be defined for any value of $`r`$ larger than $`r_0.`$ Thus, the minimum value of the radius $`r_{min}`$ that follows from this expression must be larger than $`\left(\frac{h_o}{\omega }\right)^2,`$ yielding a compromise between the constants. Using this value on the expression of the effective metric gives
$`ds_{eff}^2={\displaystyle \frac{r^2}{r^2l^2}}dt^2+{\displaystyle \frac{2h_0r^2}{r^2l^2}}dtd\phi `$ $``$ (35)
$`{\displaystyle \frac{r^2}{r^2+l^2}}dr^2\omega ^2r^2{\displaystyle \frac{r^2l^2\frac{h_0^2}{\omega ^2}}{r^2l^2}}d\phi ^2dz^2,`$ (36)
where $`l^2\frac{32Q^2}{b^2}.`$ The value for which $`\mathrm{\Psi }`$ vanishes is given by
$$r_c^2=\left(\frac{h_0}{\omega }\right)^2+\frac{32Q^2}{b^2}.$$
For $`r=r_c`$ the photon follows a closed spacetime path.
## IV Final Comments
It has been known from more than half a century that gravitational processes allows the existence of closed paths in spacetime. This led to the belief that this strange situation occurs uniquely under the effect of gravity. In the present paper we have shown that this is not the case. Indeed, we show here that photons can follow closed paths (CLC) due to electromagnetic forces in a non-linear regime. We presented an specific example of a theory in which CLC exists. In the limit case where the non-linearities are neglected the presence of CLC is no more possible. Thus we are allowed to claim that this new property depends crucially on the non-linearity of the electromagnetic process and it is not possible to exist in Maxwell theory.
This shows that the existence of CLC is not an exclusive property of gravitational interaction: it can exists also in pure electromagnetic processes depending on the non-linearities of the background field. The existence of such CLC in both gravitational and electromagnetic processes asks for a deep review of the causal structure displayed by the photon path.
## V Appendix A: The inverse metric
In the case in which the Lagrangian depends only on the invariant $`F`$ the effective geometry takes the form
$$g^{\mu \nu }=L_F\gamma ^{\mu \nu }4L_{FF}F^\mu _\lambda F^{\lambda \nu }$$
(37)
The inverse metric, the covariant tensor $`g_{\mu \nu }`$ defined by
$$g^{\mu \nu }g_{\nu \alpha }=\delta _\alpha ^\mu $$
can be easily evaluated by taking into account the identities
$$F^\mu _\lambda F^{\lambda \nu }F^\mu _\lambda F^{\lambda \nu }=\frac{1}{2}F\gamma ^{\mu \nu },$$
and
$$F^\mu _\lambda F^{\lambda \nu }=\frac{G}{4}\gamma ^{\mu \nu }.$$
A direct manipulation yields the result
$$g_{\mu \nu }=A\gamma _{\mu \nu }+BF_\mu ^\lambda F_{\lambda \nu }$$
(38)
where $`A`$ and $`B`$ are
$$A=\frac{1}{R}\left(L_F+2FL_{FF}\right),$$
$$B=\frac{4L_{FF}}{R},$$
and $`R`$ is defined by
$$R=L_F{}_{}{}^{2}+L_{FF}(2FL_FG^2).$$
## VI Appendix B: The Effective Null Geodesics
The geometrical relevance of the effective geometry (17) goes beyond its immediate definition. Indeed, we will demonstrate here that the integral curves of the vector $`k_\nu `$ (i.e., the trajectories of such nonlinear photons) are in fact geodesics. In order to achieve this result it will be required an underlying Riemannian structure for the manifold associated with the effective geometry. In other words, this means a set of Levi-Civita connection coefficients $`\mathrm{\Gamma }^\alpha _{\mu \nu }=\mathrm{\Gamma }^\alpha _{\nu \mu }`$, by means of which there exists a covariant differential operator $`_\lambda `$ (the covariant derivative) such that
$$_\lambda g^{\mu \nu }g^{\mu \nu }_{;\lambda }g^{\mu \nu }_{,\lambda }+\mathrm{\Gamma }^\mu _{\sigma \lambda }g^{\sigma \nu }+\mathrm{\Gamma }^\nu _{\sigma \lambda }g^{\sigma \mu }=0.$$
(39)
From (39) it follows that the effective connection coefficients are completely determined from the effective geometry by the usual Christoffel formula.
Contracting (39) with $`k_\mu k_\nu `$ results
$$k_\mu k_\nu g^{\mu \nu }_{,\lambda }=2k_\mu k_\nu \mathrm{\Gamma }^\mu _{\sigma \lambda }g^{\sigma \nu }.$$
(40)
Differentiating (16) and remembering $`g^{\mu \nu }=g^{\nu \mu }`$ one gets
$$2k_{\mu ,\lambda }k_\nu g^{\mu \nu }+k_\mu k_\nu g^{\mu \nu }_{,\lambda }=0.$$
(41)
Inserting (40) for the last term on the left hand side of (41) we obtain
$$g^{\mu \nu }k_{\mu ,\lambda }k_\nu g^{\sigma \nu }\mathrm{\Gamma }^\mu _{\sigma \lambda }k_\mu k_\nu =0.$$
(42)
Relabeling contracted indices we can rewrite (42) as
$$g^{\mu \nu }k_{\mu ;\lambda }k_\nu g^{\mu \nu }\left[k_{\mu ,\lambda }\mathrm{\Gamma }^\sigma _{\mu \lambda }k_\sigma \right]k_\nu =0.$$
(43)
Now, as the propagation vector $`k_\mu =\mathrm{\Sigma }_{,\mu }`$ is an exact gradient one can write $`k_{\mu ;\lambda }=k_{\lambda ;\mu }`$. With this identity and defining $`k^\nu g^{\mu \nu }k_\mu `$ equation (43) reads
$$k_{\mu ;\lambda }k^\lambda =0,$$
(44)
which states that $`k_\mu `$ is a geodesic vector. Remembering it is also a null vector (with respect to the effective geometry $`g^{\mu \nu }`$) its integral curves are thus null geodesics. We can restate our previous demonstration as:
> The discontinuities of the electromagnetic field in a nonlinear electrodynamics propagates along null geodesics of an effective geometry which depends on the properties of the background field.
Thus, photons follow geodesics in an effective geometry. Since the photon does not have electric charge, the force that acts on it in a non-linear electromagnetic field has a distinct character than that of the Lorentz force. Indeed, from the geodesic equation, the electromagnetic field acts on the photon by means of a force given by
$$f^\alpha =\mathrm{\Delta }^\alpha {}_{\mu \nu }{}^{}k_{}^{\mu }k^\nu $$
(45)
in which the quantity $`\mathrm{\Delta }^\alpha _{\mu \nu }`$ can be displayed in terms of the effective geometry $`g^{\mu \nu }`$ and its inverse $`g_{\mu \nu }`$ given above by the Christoffel form.
$$\mathrm{\Delta }^\alpha {}_{\mu \nu }{}^{}=\frac{1}{2}(L_F\gamma ^{\alpha \beta }+\mathrm{\Phi }^{\alpha \beta })(_\nu g_{\beta \mu }+_\mu g_{\beta \nu }_\beta g_{\mu \nu })$$
where $`\mathrm{\Phi }^{\alpha \beta }4L_{FF}F^\alpha _\lambda F^{\lambda \beta }.`$ Hence, it follows that the net effect of the force that the field exerts on the photon has very similar properties as the gravitational force. This is precisely the reason that allows us to interpret the action of the electromagnetic field on the photon to be a mimic of the behavior of a massless particle in a gravitational field. |
warning/0003/hep-th0003177.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The $`(1+1)`$ dimensional sigma model describes the motion of a string on a manifold. The sigma model is specified by giving a triplet of data $`(M,g,B)`$ where $`M`$ is the target $`n`$-dimensional manifold, $`g`$ is a metric on $`M`$, and $`B`$ is a $`2`$-form on $`M`$. The lagrangian for this model is
$$=\frac{1}{2}g_{ij}(x)\left(\frac{x^i}{\tau }\frac{x^j}{\tau }\frac{x^i}{\sigma }\frac{x^j}{\sigma }\right)+B_{ij}(x)\frac{x^i}{\tau }\frac{x^j}{\sigma }$$
(1.1)
with canonical momentum density
$$\pi _i=\frac{}{\dot{x}^i}=g_{ij}\dot{x}^j+B_{ij}x^j,$$
(1.2)
where an overdot denotes the time derivative ($`/\tau `$) and a prime denotes the space derivative ($`/\sigma `$) on the worldsheet. What is remarkable is that it possible for two completely different sigma models, $`(M,g,B)`$ and $`(\stackrel{~}{M},\stackrel{~}{g},\stackrel{~}{B})`$, to describe the same physics. By this we mean that there is a canonical transformation between the space of paths on $`M`$ and the corresponding one on $`\stackrel{~}{M}`$ that preserves the respective hamiltonians. This phenomenon is known as *target space duality*.
This is the first of two articles where we develop a systematic framework for studying target space duality at the classical level. We do not consider quantum aspects of target space duality nor do we consider examples involving mirror symmetry. Most of our considerations are local but phrased in a manner that is amenable to globalization. We analyze the local geometric requirements necessary for target space duality. The study of target space duality has developed by discovering a succession of more and more complicated examples (see below). We show that the known examples of abelian duality, nonabelian duality and Poisson-Lie duality are all derivable as special cases of the framework. We show that target space duality boils down to the study of some very special symplectic manifolds that allow the reduction of the structure group of the frame bundle to $`\mathrm{SO}(n)`$. In article I we develop the general theory and apply it so some very simple examples. In article II we systematically apply the theory to a variety of scenarios and we reproduce nonabelian duality and Poisson-Lie duality. The theory is applied to other geometric situations that lead us deep into unknown questions in Lie algebra theory. We try to make article I self contained. References to equations and sections in article II are preceded by II, *e.g.*, (II-8.3).
What is the value in developing a general framework for studying classical target space duality? The framework may say something about the what is string theory. We believe that there is some parameter space that describes string theory. For special values of the parameters we get the familiar Type I, Type II-A, Type II-B, *etc*. theories and that these are related by various dualities. If we can get a handle on the class of symplectic manifolds that lead to target space duality we may be able to get a better idea about the parameter space of string theory.
The simplest target space duality is abelian duality. Here a theory with target space $`S^1`$ or $``$ is dual to a theory with target space $`S^1`$ or $``$. For a comprehensive review and history of abelian duality look in . It should also be mentioned that it has been known for a long time, see *e.g.* , that the abelian duality transformation is a canonical transformation. A first attempt to generalize abelian duality to groups led to the pseudochiral model of Zakharov and Mikhailov as a dual to the nonlinear sigma model. Nappi showed that these models were not equivalent at the quantum level. The correct dual model was first found by Fridling and Jevicki and Fradkin and Tseytlin using path integral methods. String theory motivated a renewed interest in abelian and nonabelian duality . It was shown that the duality transformation was canonical and these ideas were generalized in a variety of ways . The form of the generating functions for duality transformation gave hints that nonabelian duality was associated with the geometry of the cotangent bundle of the group.
The most intricate target space duality discovered thus far is the Poisson-Lie duality of Klimcik and Severa . In this example we see a very nontrivial geometrical structure playing a central role. A Poisson-Lie group $`G`$ is a Lie group with a Poisson bracket that is compatible with the group multiplication law. Drinfeld showed that Poisson-Lie groups are determined by a Lie bialgebra $`𝔤_D=𝔤\stackrel{~}{𝔤}`$ where $`𝔤`$ is the Lie algebra of $`G`$ and $`\stackrel{~}{𝔤}`$ is the Lie algebra of a Lie group $`\stackrel{~}{G}`$, See Appendix II-B.1. The two Lie algebras are coupled together in a very symmetric way. A Lie group $`G_D`$ with Lie algebra $`𝔤_D`$ is called a Drinfeld double. It should be pointed out that $`\stackrel{~}{G}`$ is also a Poisson-Lie group. By using a clever argument, Klimcik and Several discovered that if the metric $`g`$ and $`B`$ field on a Poisson-Lie group $`G`$ was of a special form then there would be a corresponding metric $`\stackrel{~}{g}`$ and $`\stackrel{~}{B}`$-field on the group $`\stackrel{~}{G}`$. Their observations follow from the symmetric way that $`G`$ and $`\stackrel{~}{G}`$ enter into the Drinfeld double $`G_D`$. They showed that that by writing down a “first order” sigma model on $`G_D`$ they could derive either the model on $`G`$ or the model on $`\stackrel{~}{G}`$ by taking an appropriate slice. Here one explicitly sees that the the target manifold and the target dual manifold are carefully glued together into a larger space. Klimcik and Severa do not explicitly write down the duality transformation but they are totally explicit about the metric and $`B`$ field. It was Sfetsos who wrote down the duality transformation, verified that it was a canonical transformation, and constructed the generating function for the canonical transformation, see also .
At the time of the work by Klimcik and Severa, the author had been working on a program to develop a general theory of target space duality, see . In that article I advocated the use of generating functions of the type (2.2) because they would lead to a linear relationship<sup>3</sup><sup>3</sup>3For nonpolynomial generating functions look at . between $`(dx/d\sigma ,\pi )`$ and $`(d\stackrel{~}{x}/d\sigma ,\stackrel{~}{\pi })`$ that preserved the quadratic nature of the sigma model hamiltonians. I discussed the geometry which was involved and explained the role played in this geometry by the hamiltonian density $``$ and the momentum density $`𝒫`$. Explicit formulas relating the geometries of the two manifolds were not given in that article for the following reason. The formulation I had at the time involved variables $`(x,p)`$ where essentially $`\pi =dp/d\sigma `$. This gave a certain symmetry to some of the equations but at a major price. The $`B`$ field gauge symmetry $`BB+dA`$ became a nonlocal symmetry in $`(x,p)`$ space and the gauge symmetry was no longer manifest. Only for special choices of $`A`$ was the gauge transformation local. The formulas I had derived respected the special gauge transformations but I could not verify general gauge invariance. Sfetsos exploited some of the geometric constraints I had proposed and he was able to explicitly construct the duality transformation for Poisson-Lie duality. Sfetsos’ work is very interesting. He conjectures the form of the duality transformation and he knows the geometric data $`(M,g,B)`$ and $`(\stackrel{~}{M},\stackrel{~}{g},\stackrel{~}{B})`$ from the work of Klimcik and Severa. He now uses this information and certain integrability constraints to explicitly work out the generating function for the canonical transformation. Sfetsos’ computation may be reinterpreted as the construction of a known symplectic structure on the Drinfeld double, see Section II-3.
In this article I present a general theory for target space duality that is manifestly gauge invariant with respect to $`B`$ field gauge transformations. I consider what could be called *irreducible duality* where there are no spectator fields. All the fields participate actively in the duality transformation. I show that the duality transformation arises because of the existence of a special symplectic manifold $`P`$ that locally looks like $`M\times \stackrel{~}{M}`$ and admits a double fibration. The duality transformation exists only when there exists a compatible confluence of several distinct geometric structures associated to the manifold $`P`$: an $`\mathrm{O}(2n)`$ structure related to the hamiltonian density (3.1), an $`\mathrm{O}(n,n)`$ structure related to the momentum density (3.2), an $`\mathrm{O}(n)\times \mathrm{O}(n)`$ structure associated with the sigma model metrics, and a $`\mathrm{Sp}(2n)`$ structure related to the symplectic form. This is why these symplectic manifolds are very special and rare. I develop the general theory and then show how the known examples of abelian duality, nonabelian duality and Poisson-Lie duality follow. The general theory indicates that there are probably many more examples. For example, in Section 8.3 I write down families of nonlinear duality transformations that map a theory with target space $`^n`$ into one with target space $`^n`$. I also investigate a variety of scenarios and pose open mathematical questions deeply related to the theory of Lie algebras.
This work differs from the work of Sfetsos in a variety of ways. There are two types of constraints on the canonical transformation: algebraic constraints having to do with quadratic form of the hamiltonian density and differential constraints having to do integrability conditions. Sfetsos writes these down but in a way that is neither geometric nor gauge invariant. He applies them to Poisson-Lie duality and derives the generating function. Sfetsos’ formulation does not exploit the fact that there are natural geometric structures associated to these equations. This is what I was trying to do in but failed due to a bad choice of variables $`(x,p)`$ leading to an absence of manifest $`B`$ field gauge invariance. The formulation presented here uses the variables $`(x,\pi )`$ and is manifestly gauge invariant. In Section II-2.2.2 I give a geometric interpretation of $`B`$ field gauge invariance. In this article I work in terms of adapted frame fields. In this way, the formalism has an immediate interpretation in terms of $`H`$-structures on the bundle of frames. In fact the discussion presented in Section II-4.1 is done in a sub-bundle of the bundle of frames.
The framework developed in this work allows one to attack a variety of interesting questions. Are there any interesting restrictions on the manifolds $`M`$ and $`\stackrel{~}{M}`$? We show in Section 6 that the manifolds $`M`$ and $`\stackrel{~}{M}`$ have to admit flat orthogonal connections. We know for any manifold $`M`$ there always exists a natural symplectic manifold $`P=T^{}M`$, the cotangent bundle. We can ask what type of dualities arises from the standard symplectic structure on the cotangent bundle? We show that this can only happen if $`M`$ is a Lie group, see Section II-2.2.1. This formalism allows general question to be asked. For example there are a series of PDEs that have to be solved to determine the duality transformations. These PDEs depend on some functions. If these functions are zero then one gets abelian duality, if some are made nonzero then you get nonabelian duality, *etc*. This is a framework that can be used for a systematic study of duality. It opens up the possibility to study dualities involving parallelizable manifolds that are not Lie groups such as $`S^7`$ or sub-bundles of the frame bundle. This work indicates that duality is a very rich geometrical framework ripe for study and we have only scratched the surface.
## 2 The symplectic structure
We review briefly the notion of a “generating function” in canonical transformations because our methods introduce a secondary symplectic structure into the formulation of target space duality and it is important to understand the difference between the two.
Assume you have symplectic manifolds, $`P`$ and $`\stackrel{~}{P}`$, with respective symplectic forms $`\omega `$ and $`\stackrel{~}{\omega }`$. Consider $`P\times \stackrel{~}{P}`$ with standard projections $`\mathrm{\Pi }:P\times \stackrel{~}{P}P`$ and $`\stackrel{~}{\mathrm{\Pi }}:P\times \stackrel{~}{P}\stackrel{~}{P}`$. You can make $`P\times \stackrel{~}{P}`$ into a symplectic manifold by choosing as symplectic form $`\mathrm{\Omega }=\stackrel{~}{\mathrm{\Pi }}^{}\stackrel{~}{\omega }\mathrm{\Pi }^{}\omega `$. By definition, a canonical or symplectic transformation $`f:P\stackrel{~}{P}`$ satisfies $`f^{}\stackrel{~}{\omega }=\omega `$. We describe $`f`$ by its graph $`\mathrm{\Gamma }_fP\times \stackrel{~}{P}`$. It is clear $`f:P\stackrel{~}{P}`$ will be symplectic if and only if $`\mathrm{\Omega }|_{\mathrm{\Gamma }_f}=0`$. Locally we have $`\omega =d\theta `$ and $`\stackrel{~}{\omega }=d\stackrel{~}{\theta }`$. Thus we see that $`\stackrel{~}{\theta }\theta `$ is a closed $`1`$-form on $`\mathrm{\Gamma }_f`$. Consequently there exists locally a function $`F:\mathrm{\Gamma }_f`$ such that $`\stackrel{~}{\theta }\theta =dF`$. This function $`F`$ is called the “generating function” for the symplectic transformation. The reason is that if in local Darboux coordinates we have that $`\theta =pdq`$ and $`\stackrel{~}{\theta }=\stackrel{~}{p}d\stackrel{~}{q}`$ then we have that $`F`$ is locally a function of only $`q`$ and $`\stackrel{~}{q}`$, $`\stackrel{~}{p}=F/\stackrel{~}{q}`$ and $`p=F/q`$. We can now use the inverse function theorem to construct the map from $`(q,p)`$ to $`(\stackrel{~}{q},\stackrel{~}{p})`$. Note that $`dim\mathrm{\Gamma }_f=2n`$ and therefore $`F`$ is a function of $`2n`$ variables. Had we chosen $`\theta =qdp`$ then we would have that $`F`$ is a function of $`\stackrel{~}{q}`$ and $`p`$. In this case it is worthwhile to observe $`F=\stackrel{~}{q}p`$ generates the identity transformation. We mention this because the identity transformation is not in the class of transformations generated by functions of $`q`$ and $`\stackrel{~}{q}`$.
All this generalizes to field theory. We discuss only the case of $`(1+1)`$ dimensions. Let $`P(M)`$ be the path space of M. By this we mean the set of maps $`\{\gamma :NM\}`$ where $`N`$ can be $``$, $`S^1`$ or $`[0,\pi ]`$ depending on whether we are discussing infinite strings, closed strings or open strings. Most of the discussion in this article is local and so we do not specify $`N`$. In the case of a sigma model with target space $`M`$, the basic configuration space is $`P(M)`$ with associated phase space $`P(T^{}M)`$. If $`(x,\pi )`$ are coordinates on $`T^{}M`$ then the symplectic structure on $`P(T^{}M)`$ is given by
$$\delta \pi (\sigma )\delta x(\sigma )d\sigma .$$
(2.1)
In what follows we are interested in looking for canonical transformations between a sigma model with target space $`M`$ and one with target space $`\stackrel{~}{M}`$ of the same dimensionality. We say that a sigma model with geometrical data $`(M,g,B)`$ is dual to a sigma model $`(\stackrel{~}{M},\stackrel{~}{g},\stackrel{~}{B})`$ if there exists a canonical transformation $`F:P(T^{}M)P(T^{}\stackrel{~}{M})`$ that preserves the hamiltonian densities, $`F^{}\stackrel{~}{}=`$, where the hamiltonian density is given by (3.1).
In the case of “abelian duality” where the target space is a circle you can choose the generating function to be
$$F[x,\stackrel{~}{x}]=\stackrel{~}{x}\frac{dx}{d\sigma }𝑑\sigma .$$
This leads to the standard duality relations $`\pi (\sigma )=d\stackrel{~}{x}/d\sigma `$ and $`\stackrel{~}{\pi }(\sigma )=dx/d\sigma `$.
The nonabelian duality relations follow from the following natural choice for generating function. Assume the target space is a simple connected compact Lie group $`G`$ with Lie algebra $`𝔤`$. The dual manifold is the Lie algebra with an unusual metric. The generating function is very natural:
$$F[g,\stackrel{~}{X}]=\mathrm{Tr}\left(\stackrel{~}{X}g^1\frac{dg}{d\sigma }\right)𝑑\sigma ,$$
where $`\stackrel{~}{X}`$ is a Lie algebra valued field.
We now consider a class of generating functions for target space duality that leads to a linear relationship between $`(dx/d\sigma ,\pi (\sigma ))`$ and the corresponding variables on the dual space. On $`M\times \stackrel{~}{M}`$ choose locally a $`1`$-form $`\alpha =\alpha _i(x,\stackrel{~}{x})dx^i+\stackrel{~}{\alpha }_i(x,\stackrel{~}{x})d\stackrel{~}{x}^i`$. We can define a natural “generating function” on $`P(M\times \stackrel{~}{M})`$ by
$$F[x(\sigma ),\stackrel{~}{x}(\sigma )]=\alpha =\left(\alpha _i(x(\sigma ),\stackrel{~}{x}(\sigma ))\frac{dx^i}{d\sigma }+\stackrel{~}{\alpha }_i(x(\sigma ),\stackrel{~}{x}(\sigma ))\frac{d\stackrel{~}{x}^i}{d\sigma }\right)𝑑\sigma .$$
(2.2)
We only consider target space duality that arises from this type of canonical transformation.
Let $`v`$ be a vector field along the path $`(x(\sigma ),\stackrel{~}{x}(\sigma ))M\times \stackrel{~}{M}`$ with compact support which represents a deformation of the path. Note that $`\delta _vF=_v\alpha =\iota _v𝑑\alpha `$. In the previous formula $`_v\alpha =\iota _vd\alpha +d\iota _v\alpha `$ is the Lie derivative with respect to $`v`$. Since $`v`$ has compact support, the exact term can be neglected. Thus the variation of $`F`$ is determined by the exact $`2`$-form $`\beta =d\alpha `$:
$$\delta _vF=\iota _v\beta .$$
(2.3)
We use $`\beta `$ to construct the duality transformation. If $`x`$ and $`\stackrel{~}{x}`$ are respectively local coordinates on $`M`$ and $`\stackrel{~}{M}`$ then
$$\beta =\frac{1}{2}l_{ij}(x,\stackrel{~}{x})dx^idx^j+m_{ij}(x,\stackrel{~}{x})d\stackrel{~}{x}^idx^j+\frac{1}{2}\stackrel{~}{l}_{ij}(x,\stackrel{~}{x})d\stackrel{~}{x}^id\stackrel{~}{x}^j,$$
(2.4)
where $`\stackrel{~}{l}`$: $`l_{ij}=l_{ji}`$ and $`\stackrel{~}{l}_{ij}=\stackrel{~}{l}_{ji}`$. The three $`n\times n`$ matrix functions $`l,\stackrel{~}{l},m`$ are used to construct the canonical transformation on the infinite dimensional phase space. A brief calculation shows that the canonical transformations are
$`\pi _i(\sigma )`$ $`=`$ $`m_{ji}(x,\stackrel{~}{x}){\displaystyle \frac{d\stackrel{~}{x}^j}{d\sigma }}+l_{ij}(x,\stackrel{~}{x}){\displaystyle \frac{dx^j}{d\sigma }},`$ (2.5)
$`\stackrel{~}{\pi }_i(\sigma )`$ $`=`$ $`m_{ij}(x,\stackrel{~}{x}){\displaystyle \frac{dx^j}{d\sigma }}+\stackrel{~}{l}_{ij}(x,\stackrel{~}{x}){\displaystyle \frac{d\stackrel{~}{x}^j}{d\sigma }}.`$ (2.6)
The invertibility of the canonical transformation between $`P(T^{}M)`$ and $`P(T^{}\stackrel{~}{M})`$ requires $`m`$ to be an invertible matrix. This implies that $`\beta `$ is of maximal rank, *i.e.* a symplectic form<sup>4</sup><sup>4</sup>4 It is possible for $`\beta `$ to be symplectic and have $`m=0`$ but this will not define an invertible canonical transformation between $`P(T^{}M)`$ and $`P(T^{}\stackrel{~}{M})`$. For example, if $`M`$ and $`\stackrel{~}{M}`$ are symplectic manifolds with respective symplectic forms $`\omega `$ and $`\stackrel{~}{\omega }`$ then choose $`\beta =\stackrel{~}{\omega }\omega `$. on $`M\times \stackrel{~}{M}`$.
It is important to recognize that there are two very different symplectic structures in this problem. The first one is the standard symplectic structure on phase space $`P(T^{}M)`$ given by (2.1). The second one on $`M\times \stackrel{~}{M}`$ given by $`\beta `$ arises from the class of generating functions (2.2) we are considering. The generating function arguments are local and suggest that the symplectic structure on $`M\times \stackrel{~}{M}`$ may be generalized to a symplectic manifold $`P`$ which “contains” $`M\times \stackrel{~}{M}`$. In the cartesian product $`M\times \stackrel{~}{M}`$ you have natural cartesian projections $`\mathrm{\Pi }_c:M\times \stackrel{~}{M}M`$ and $`\stackrel{~}{\mathrm{\Pi }}_c:M\times \stackrel{~}{M}\stackrel{~}{M}`$. The product structure can be generalized by the introduction of the concept of a bifibration. A $`2n`$ dimensional manifold $`P`$ is said to be a *bifibration* if there exists $`n`$ dimensional manifolds $`M`$ and $`\stackrel{~}{M}`$ and projections $`\mathrm{\Pi }:PM`$ and $`\stackrel{~}{\mathrm{\Pi }}:P\stackrel{~}{M}`$ such that the respective fibers are diffeomorphic to coverings spaces of $`\stackrel{~}{M}`$ and $`M`$ and they are also transverse. This means that if $`pP`$ then $`\mathrm{ker}\mathrm{\Pi }_{}|_p\mathrm{ker}\stackrel{~}{\mathrm{\Pi }}_{}|_p=T_pP`$ where $`\mathrm{\Pi }_{}`$ and $`\stackrel{~}{\mathrm{\Pi }}_{}`$ are the differential maps of the projections. Note that the cartesian product manifold $`P=M\times \stackrel{~}{M}`$ is an example of a bifibration. If the product projection $`\mathrm{\Pi }\times \stackrel{~}{\mathrm{\Pi }}:PM\times \stackrel{~}{M}`$ is injective<sup>5</sup><sup>5</sup>5The definition of a fiber bundle implies that $`\mathrm{\Pi }\times \stackrel{~}{\mathrm{\Pi }}`$ is surjective. then $`P=M\times \stackrel{~}{M}`$. A covering space example is given by $`P=^2`$ and $`M=\stackrel{~}{M}=S^1`$ with $`\mathrm{\Pi }:(x,\stackrel{~}{x})e^{ix}`$ and $`\stackrel{~}{\mathrm{\Pi }}:(x,\stackrel{~}{x})e^{i\stackrel{~}{x}}`$.
We introduce the following terminology illustrated in Figure 1.
At a point $`pP`$ we have a splitting of the tangent space $`T_pP=H_pV_p`$ where the “horizontal tangent space” $`H_p`$ is tangent to the fiber of $`\stackrel{~}{\mathrm{\Pi }}`$, and the “vertical tangent space” $`V_p`$ is tangent to the fiber of $`\mathrm{\Pi }`$. A symplectic form $`\beta `$ is said to be *bifibration compatible* if for every $`pP`$ one has the following nondegeneracy conditions:
1. Given $`YV_p`$, if for all $`XH_p`$ one has $`\beta (X,Y)=0`$ then $`Y=0`$.
2. Given $`XH_p`$, if for all $`YV_p`$ one has $`\beta (X,Y)=0`$ then $`X=0`$.
This is the coordinate independent way of stating that the matrix $`m_{ij}`$ is invertible. The reader can verify that the symplectic form given in footnote 4 fails the above. Our abstract scenario is a bifibration<sup>6</sup><sup>6</sup>6It may be possible to generalize from a bifibration to a bifoliation. It is not clear to the author what is the most general formulation. with a bifibration compatible symplectic form. We will refer to such a manifold as a *special symplectic bifibration*. We believe that the formulation of canonical transformations in path space in terms of $`\beta `$ is probably more fundamental than the use of a generating function. This is probably analogous to the ascendant role the symplectic $`2`$-form has taken in symplectic geometry because of global issues. The $`2`$-form $`\beta `$ may play a role in the quantum aspects of duality maybe in some geometric quantization type of framework.
A scenario for how a natural generating function of type (2.2) might arise is the following. For any $`M`$, the cotangent bundle $`T^{}M`$ is a symplectic manifold. Firstly, one has to investigate whether the cotangent bundle admits a second fibration transverse to the defining one. Secondly, it may be necessary to deform the original symplectic structure. In the case of a Lie group $`G`$, the cotangent bundle is trivial and is thus a product $`T^{}G=G\times 𝔤`$ where we have used the metric on $`G`$ to identify the Lie algebra $`𝔤`$ with its vector space dual $`𝔤^{}`$.
## 3 Hamiltonian structure
The discussion in the Section 2 is general and makes no reference to the hamiltonian. The hamiltonian only played an indirect role because we chose a class of canonical transformations which are linear with respect to $`dx/d\sigma `$ and $`\pi (\sigma )`$ in anticipation of future application to the nonlinear sigma model. The nonlinear sigma model has target space a riemannian manifold $`M`$ with metric $`g`$ and a $`2`$-form field $`B`$. The hamiltonian density and the momentum density are respectively given by
$``$ $`=`$ $`{\displaystyle \frac{1}{2}}g^{ij}(x)\left(\pi _iB_{ik}{\displaystyle \frac{dx^k}{d\sigma }}\right)\left(\pi _jB_{jl}{\displaystyle \frac{dx^l}{d\sigma }}\right)+{\displaystyle \frac{1}{2}}g_{ij}(x){\displaystyle \frac{dx^i}{d\sigma }}{\displaystyle \frac{dx^j}{d\sigma }},`$ (3.1)
$`𝒫`$ $`=`$ $`\pi _i(\sigma ){\displaystyle \frac{dx^i}{d\sigma }}.`$ (3.2)
We are interested whether we can find a canonical transformation with generating function of type (2.2) which will map the hamiltonian density and momentum density into that of another sigma model (the dual sigma model) characterized by target space $`\stackrel{~}{M}`$, metric tensor $`\stackrel{~}{g}`$ and $`2`$-form $`\stackrel{~}{B}`$.
It winds up that working in coordinates is not the best way of attacking the problem. It is best to use moving frames à la Cartan and Chern. Let $`(\theta ^1,\mathrm{},\theta ^n)`$ be a local orthonormal coframe<sup>7</sup><sup>7</sup>7Because we will be working in orthonormal frames we do not distinguish an upper index from a lower index in a tensor. for $`M`$. The Cartan structural equations are
$`d\theta ^i`$ $`=`$ $`\omega _{ij}\theta ^j,`$
$`d\omega _{ij}`$ $`=`$ $`\omega _{ik}\omega _{kj}+{\displaystyle \frac{1}{2}}R_{ijkl}\theta ^k\theta ^l,`$
where $`\omega _{ij}=\omega _{ji}`$ is the riemannian connection<sup>8</sup><sup>8</sup>8The riemannian connection is the unique torsion free metric compatible connection. A metric compatible connection will also be referred to as an orthogonal connection. In general an orthogonal connection can have torsion.. Next we define $`dx/d\sigma `$ in the orthonormal frame to be $`x_\sigma `$ by requiring that $`\theta ^i=x^i{}_{\sigma }{}^{}d\sigma `$. If $`\pi `$ is now the canonical momentum density in the orthonormal frame then in this frame (3.1) and (3.2) become
$``$ $`=`$ $`{\displaystyle \frac{1}{2}}(\pi _iB_{ik}x^k{}_{\sigma }{}^{})(\pi _iB_{il}x^l{}_{\sigma }{}^{})+{\displaystyle \frac{1}{2}}x^i{}_{\sigma }{}^{}x_{}^{i}{}_{\sigma }{}^{},`$ (3.3)
$`𝒫`$ $`=`$ $`\pi _ix^i{}_{\sigma }{}^{}=(\pi _iB_{ij}x^j{}_{\sigma }{}^{})x^i{}_{\sigma }{}^{}.`$ (3.4)
In this coframe we can write (2.4) as
$$\beta =\frac{1}{2}l_{ij}(x,\stackrel{~}{x})\theta ^i\theta ^j+m_{ij}(x,\stackrel{~}{x})\stackrel{~}{\theta }^i\theta ^j+\frac{1}{2}\stackrel{~}{l}_{ij}(x,\stackrel{~}{x})\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j.$$
(3.5)
We use the same letters $`l,m,\stackrel{~}{l}`$ but the meaning above is different from (2.4). In this notation equations (2.5) and (2.6) become
$`\pi _i(\sigma )`$ $`=`$ $`m_{ji}(x,\stackrel{~}{x})\stackrel{~}{x}^j{}_{\sigma }{}^{}+l_{ij}(x,\stackrel{~}{x})x^j{}_{\sigma }{}^{},`$ (3.6)
$`\stackrel{~}{\pi }_i(\sigma )`$ $`=`$ $`m_{ij}(x,\stackrel{~}{x})x^j{}_{\sigma }{}^{}+\stackrel{~}{l}_{ij}(x,\stackrel{~}{x})\stackrel{~}{x}^j{}_{\sigma }{}^{},`$ (3.7)
In matrix notation the above may be written as
$$\left(\begin{array}{cc}m^t& 0\\ \stackrel{~}{l}& I\end{array}\right)\left(\begin{array}{c}\stackrel{~}{x}_\sigma \\ \stackrel{~}{\pi }\end{array}\right)=\left(\begin{array}{cc}l& I\\ m& 0\end{array}\right)\left(\begin{array}{c}x_\sigma \\ \pi \end{array}\right)$$
Rewrite the above in the form
$$\left(\begin{array}{cc}m^t& 0\\ \stackrel{~}{n}& I\end{array}\right)\left(\begin{array}{c}\stackrel{~}{x}_\sigma \\ \stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}_\sigma \end{array}\right)=\left(\begin{array}{cc}n& I\\ m& 0\end{array}\right)\left(\begin{array}{c}x_\sigma \\ \pi Bx_\sigma \end{array}\right),$$
where
$`n`$ $`=`$ $`lB,`$ (3.8)
$`\stackrel{~}{n}`$ $`=`$ $`\stackrel{~}{l}\stackrel{~}{B}.`$ (3.9)
The rewriting above is closely related to (A.2), see below. This equation is not very interesting in this form but it becomes much more interesting when rewritten as
$`\left(\begin{array}{c}\stackrel{~}{x}_\sigma \\ \stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}_\sigma \end{array}\right)`$ $`=`$ $`\left(\begin{array}{cc}m^t& 0\\ \stackrel{~}{n}& I\end{array}\right)^1\left(\begin{array}{cc}n& I\\ m& 0\end{array}\right)\left(\begin{array}{c}x_\sigma \\ \pi Bx_\sigma \end{array}\right)`$ (3.18)
$`=`$ $`\left(\begin{array}{cc}(m^t)^1n& (m^t)^1\\ \stackrel{~}{n}(m^t)^1n+m& \stackrel{~}{n}(m^t)^1\end{array}\right)\left(\begin{array}{c}x_\sigma \\ \pi Bx_\sigma \end{array}\right).`$ (3.23)
Notice that equation (3.23) gives us a linear transformation between $`(x{}_{\sigma }{}^{},\pi Bx{}_{\sigma }{}^{})`$ and $`(\stackrel{~}{x}{}_{\sigma }{}^{},\stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}{}_{\sigma }{}^{})`$. The preservation of the hamiltonian density means that this linear transformation must be in $`\mathrm{O}(2n)`$. If in addition you want to preserve the momentum density then this transformation must be in $`\mathrm{O}_\mathrm{Q}(n,n)`$, the group of $`2n\times 2n`$ matrices isomorphic to $`\mathrm{O}(n,n)`$ which preserves the quadratic form
$$Q=\left(\begin{array}{cc}0& I_n\\ I_n& 0\end{array}\right).$$
(3.24)
In the formula above, $`I_n`$ is the $`n\times n`$ identity matrix. Properties of $`\mathrm{O}_\mathrm{Q}(n,n)`$ and its relation with $`\mathrm{O}(2n)`$ are reviewed in Appendix A. They key observation<sup>9</sup><sup>9</sup>9I do not understand geometrically why $`\beta `$ automatically induces this pseudo-orthogonal matrix. is that the matrix appearing in (3.23) is automatically in $`\mathrm{O}_\mathrm{Q}(n,n)`$ which means that our canonical transformation automatically preserves the canonical momentum density (3.4). As previously mentioned to preserve the hamiltonian density (3.3) is it necessary that the matrix above also be in $`\mathrm{O}(2n)`$. Thus the matrix
$$\left(\begin{array}{cc}(m^t)^1n& (m^t)^1\\ \stackrel{~}{n}(m^t)^1n+m& \stackrel{~}{n}(m^t)^1\end{array}\right)$$
(3.25)
must be in $`\mathrm{O}(2n)\mathrm{O}_\mathrm{Q}(n,n)`$, a compact group locally isomorphic to $`\mathrm{O}(n)\times \mathrm{O}(n)`$, see Appendix A. Using the equations in the appendix we learn that the condition that (3.25) be in the intersection $`\mathrm{O}(2n)\mathrm{O}_\mathrm{Q}(n,n)`$ is that
$`mm^t`$ $`=`$ $`I\stackrel{~}{n}^2,`$ (3.26)
$`m^tm`$ $`=`$ $`In^2,`$ (3.27)
$`mn`$ $`=`$ $`\stackrel{~}{n}m.`$ (3.28)
We can now simplify (3.25) to
$$\left(\begin{array}{cc}(m^t)^1n& (m^t)^1\\ (m^t)^1& (m^t)^1n\end{array}\right)$$
(3.29)
To better understand the above is is worthwhile using the conjugation operation (A.4) and switch the quadratic from from $`Q`$ to
$$\left(\begin{array}{cc}I& 0\\ 0& +I\end{array}\right).$$
Under this conjugation operation (3.23) becomes
$`\left(\begin{array}{c}\stackrel{~}{x}{}_{\sigma }{}^{}(\stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}{}_{\sigma }{}^{})\\ \stackrel{~}{x}{}_{\sigma }{}^{}+(\stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}{}_{\sigma }{}^{})\end{array}\right)`$ (3.32)
$`=\left(\begin{array}{cc}(m^t)^1(I+n)& 0\\ 0& (m^t)^1(In)\end{array}\right)\left(\begin{array}{c}x{}_{\sigma }{}^{}(\pi Bx{}_{\sigma }{}^{})\\ x{}_{\sigma }{}^{}+(\pi Bx{}_{\sigma }{}^{})\end{array}\right).`$ (3.37)
This leads to the pair of equations
$`(\stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}{}_{\sigma }{}^{})+\stackrel{~}{x}_\sigma `$ $`=`$ $`+T_+[(\pi Bx{}_{\sigma }{}^{})+x{}_{\sigma }{}^{}],`$ (3.38)
$`(\stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}{}_{\sigma }{}^{})\stackrel{~}{x}_\sigma `$ $`=`$ $`T_{}[(\pi Bx{}_{\sigma }{}^{})x{}_{\sigma }{}^{}],`$ (3.39)
where
$$T_\pm =(m^t)^1(In)\mathrm{O}(n).$$
(3.40)
An equivalent way of writing the above is $`m=T_\pm (I\pm n)`$. Also note that $`T_+`$ and $`T_{}`$ are not independent. They are related by $`T_{}^1T_+=(I+n)^1(In)`$ which is the Cayley transform of $`n`$. It is often convenient to think that (3.5) is determined by two orthogonal matrices $`T_\pm \mathrm{O}(n)`$ with
$$n=(T_++T_{})^1(T_+T_{}).$$
(3.41)
## 4 Gauge invariance
It is well known that the sigma model $`(M,g,B)`$ has a gauge invariance given by $`BB+dA`$ where $`A`$ is a $`1`$-form on $`M`$. We can manifest these gauge transformations within the class (3.40) of canonical transformation by considering $`\alpha (\alpha +A)`$ which transforms $`\pi `$ appropriately. An observation and a change of viewpoint will give us a manifestly gauge invariant formulation. Notice that both the left hand side and right hand side of equation (3.37) is manifestly gauge invariant. This suggests that $`m,n,\stackrel{~}{n}`$ may be gauge invariant. Looking at (3.8) and (3.9) and incorporating the remark about how we implement gauge invariance we see that $`n`$ and $`\stackrel{~}{n}`$ are gauge invariant quantities, *i.e.*, the gauge transformations are implemented by shifting $`l,\stackrel{~}{l}`$ respectively by $`dA`$ and $`d\stackrel{~}{A}`$. This suggest that instead of working with $`\beta `$ it may be worthwhile to work with $`\gamma `$ defined by
$$\gamma =\frac{1}{2}n_{ij}(x,\stackrel{~}{x})\theta ^i\theta ^j+m_{ij}(x,\stackrel{~}{x})\stackrel{~}{\theta }^i\theta ^j+\frac{1}{2}\stackrel{~}{n}_{ij}(x,\stackrel{~}{x})\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j$$
(4.1)
where $`\gamma `$ is not closed but satisfies
$$d\gamma =H\stackrel{~}{H}$$
(4.2)
where $`H=dB`$ and $`\stackrel{~}{H}=d\stackrel{~}{B}`$. More correctly one has $`d\gamma =\mathrm{\Pi }^{}H\stackrel{~}{\mathrm{\Pi }}^{}\stackrel{~}{H}`$. We have now achieved a gauge invariant formulation.
## 5 The geometry of $`P`$
To gain further insight into relations between the geometry of $`M`$ and $`\stackrel{~}{M}`$ is it best to work in $`P`$ which you may think of it locally being $`M\times \stackrel{~}{M}`$. We can use the freedom of working in $`P`$ to simplify results and then project back to either $`M`$ or $`\stackrel{~}{M}`$.
There are two closely related ways of simplifying the geometry. One way is to work in the bundle of orthonormal frames. The other is to *adapt* the orthonormal frames to the problem at hand similar to the way one uses Darboux frames to study surfaces in classical differential geometry. The former gives a global formulation but the latter is more familiar to physicists hence we choose the latter. All our computations will be local and can be patched together to define global objects.
The first thing to observe is that the existence of the double fibration allows us to naturally define a riemannian metric on $`P`$ by pulling back the metrics on $`M`$ and $`\stackrel{~}{M}`$ and declaring that the fibers are orthogonal to each other. In a similar fashion we pullback local coframes and get local coframes on $`P`$. These orthonormal coframes satisfy the Cartan structural equations
$`d\theta ^i`$ $`=`$ $`\omega _{ij}\theta ^j,`$ (5.1)
$`d\stackrel{~}{\theta }^i`$ $`=`$ $`\stackrel{~}{\omega }_{ij}\stackrel{~}{\theta }^j,`$ (5.2)
$`d\omega _{ij}`$ $`=`$ $`\omega _{ik}\omega _{kj}+{\displaystyle \frac{1}{2}}R_{ijkl}\theta ^k\theta ^l,`$ (5.3)
$`d\stackrel{~}{\omega }_{ij}`$ $`=`$ $`\stackrel{~}{\omega }_{ik}\stackrel{~}{\omega }_{kj}+{\displaystyle \frac{1}{2}}\stackrel{~}{R}_{ijkl}\stackrel{~}{\theta }^k\stackrel{~}{\theta }^l.`$ (5.4)
Once we begin working on $`P`$ then we have the freedom to independently rotate $`\theta `$ and $`\stackrel{~}{\theta }`$ at each point. Once we do this these coframes will no longer be pullbacks but this doesn’t matter because it does not change the metric on each fiber. We are going to exploit this freedom to relate the geometry of $`M`$ to that of $`\stackrel{~}{M}`$ in a way similar to the way the intrinsic curvature of a submanifold is related to the total curvature of the space and the curvature of the normal bundle. Note that with these choices there is a natural group of $`\mathrm{O}(n)\times \mathrm{O}(n)`$ gauge transformations on the tangent bundle of $`P`$ which is compatible with the metric structure and the bifibration.
## 6 Constraints from the algebraic structure of $`\gamma `$
First we derive various constraints that follow from the algebraic constraints on $`\gamma `$ imposed by the preservation of $``$ and $`𝒫`$. Equations (3.27) and (3.28) tell us that
$$m=T(I+n)\text{and}\stackrel{~}{n}=TnT^t$$
(6.1)
where $`T\mathrm{O}(n)`$. Since $`T`$ “connects” a $`\theta `$ to a $`\stackrel{~}{\theta }`$ we see that its covariant differential is given by
$$dT_{ij}+\stackrel{~}{\omega }_{ik}T_{kj}+\omega _{jk}T_{ik}=+T_{ijk}\theta ^k\stackrel{~}{T}_{ijk}\stackrel{~}{\theta }^k,$$
(6.2)
where the components of the covariant differential in the $`M`$ direction is $`+T_{ijk}`$ and in the $`\stackrel{~}{M}`$ direction is $`\stackrel{~}{T}_{ijk}`$. The negative sign is introduced for future convenience. Notice that $`T_{ijk}`$ and $`\stackrel{~}{T}_{ijk}`$ are tensors defined on $`P`$ whose existence is guaranteed by the existence of the tensor $`T_{ij}`$ on $`P`$.
We now invoke a “symmetry breaking mechanism” to reduce the structure group of gauge transformations from $`\mathrm{O}(n)\times \mathrm{O}(n)`$ to $`\mathrm{O}(n)`$. At each point in $`P`$ we can rotate $`\stackrel{~}{\theta }`$ (or $`\theta `$) and make $`T=I`$ because under these gauge transformations $`T\stackrel{~}{R}TR^1`$ where $`(R,\stackrel{~}{R})\mathrm{O}(n)\times \mathrm{O}(n)`$. The isotropy group of $`T=I`$ is the diagonal $`\mathrm{O}(n)`$. This is no different than giving a scalar field a vacuum expectation value to break the symmetry. This symmetry breaking leads to an identification at each point of $`P`$ of the “vertical” and “horizontal” tangent spaces. This does not tell us that the metrics are the same but allows us to identify an orthonormal frame in one with an orthonormal frame in the other. Let us be a bit more precise and abstract on the reduction of the structure group and the identification of the “vertical” and “horizontal” tangent spaces. We already mentioned that at $`pP`$ one has $`T_pP=H_pV_p`$. The tensor $`m(p)`$ may be viewed as an element of $`V_p^{}H_p^{}`$. Because there is a metric on $`V_p`$ we can reinterpret $`m`$ as giving us an invertible linear transformation $`\stackrel{ˇ}{m}:H_pV_p`$. We also have a metric on $`H_p`$ and thus we can study the orbit of $`m(p)`$ under the action of $`\mathrm{O}(n)\times \mathrm{O}(n)`$. Our previous discussion shows that a “canonical” form for $`m(p)`$ may be taken to be $`m(p)=I+n(p)`$ with isotropy group being the diagonal $`\mathrm{O}(n)`$. If $`(e_1,\mathrm{},e_n)`$ is an orthonormal basis at $`H_p`$ and $`(\stackrel{~}{e}_1,\mathrm{},\stackrel{~}{e}_n)`$ is the corresponding orthonormal basis at $`V_p`$ then they are related by $`\stackrel{ˇ}{m}(p)e_i=\stackrel{~}{e}_j(\delta _{ji}+n_{ji}(p))`$.
From now on we assume we have adapted our coframes such that $`T=I`$ and
$`m_{ij}`$ $`=`$ $`\delta _{ij}+n_{ij},`$ (6.3)
$`n_{ij}`$ $`=`$ $`\stackrel{~}{n}_{ij}.`$ (6.4)
In this frame, $`\gamma `$ simplifies to
$$\gamma =\stackrel{~}{\theta }^i\theta ^i+n_{ij}\stackrel{~}{\theta }^i\theta ^j\frac{1}{2}n_{ij}\theta ^i\theta ^j\frac{1}{2}n_{ij}\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j.$$
(6.5)
The duality equations are particularly simple now and they are given by
$`(\stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}{}_{\sigma }{}^{})+\stackrel{~}{x}_\sigma `$ $`=`$ $`(\pi Bx{}_{\sigma }{}^{})+x{}_{\sigma }{}^{},`$ (6.6)
$`(\stackrel{~}{\pi }\stackrel{~}{B}\stackrel{~}{x}{}_{\sigma }{}^{})\stackrel{~}{x}_\sigma `$ $`=`$ $`T_{}[(\pi Bx{}_{\sigma }{}^{})x{}_{\sigma }{}^{}],`$ (6.7)
Where the orthogonal matrix $`T_{}`$ is the Cayley transform of $`n`$:
$$T_{}=\frac{I+n}{In}.$$
(6.8)
The matrix $`T_{}`$ is not arbitrary because there are constraints on $`n_{ij}`$ as we will see later on. Without constraints on $`T_{}`$ there are interesting solutions to (6.6) and (6.7) which map spaces of constant positive curvature into spaces of negative constant curvature or more generally dual symmetric spaces<sup>10</sup><sup>10</sup>10O. Alvarez, unpublished..
We can now exploit equation (6.2) to relate the connections in the adapted coframing. Inserting $`T=I`$ into the above leads to
$$\stackrel{~}{\omega }_{ij}\omega _{ij}=+T_{ijk}\theta ^k\stackrel{~}{T}_{ijk}\stackrel{~}{\theta }^k.$$
(6.9)
Thus we see that in the reduction of the structure group we have generated torsion and that this torsion satisfies $`T_{ijk}=T_{jik}`$ and $`\stackrel{~}{T}_{ijk}=\stackrel{~}{T}_{jik}`$. We now define an orthogonal connection on our adapted frames by
$$\psi _{ij}=\omega _{ij}+T_{ijk}\theta ^k=\stackrel{~}{\omega }_{ij}+\stackrel{~}{T}_{ijk}\stackrel{~}{\theta }^k.$$
(6.10)
First we define the components of the covariant derivatives of $`T`$ and $`\stackrel{~}{T}`$ by
$`dT_{ijk}+(\omega T)_{ijk}`$ $`=`$ $`T_{ijkl}^{}\theta ^l+T_{ijkl}^{\prime \prime }\stackrel{~}{\theta }^l,`$ (6.11)
$`d\stackrel{~}{T}_{ijk}+(\stackrel{~}{\omega }\stackrel{~}{T})_{ijk}`$ $`=`$ $`\stackrel{~}{T}_{ijkl}^{}\theta ^l+\stackrel{~}{T}_{ijkl}^{\prime \prime }\stackrel{~}{\theta }^l.`$ (6.12)
In the above $`(\omega T)`$ and $`(\stackrel{~}{\omega }\stackrel{~}{T})`$ are abbreviations for standard expressions. We have chosen to use the connections $`\omega `$ and $`\stackrel{~}{\omega }`$ rather than $`\psi `$ in the definition of the covariant derivative for the following reasons: if $`T_{ijk}`$ is the pullback of a tensor on $`M`$ then $`T_{ijkl}^{\prime \prime }=0`$; if $`\stackrel{~}{T}_{ijk}`$ is the pullback of a tensor on $`\stackrel{~}{M}`$ then $`\stackrel{~}{T}_{ijkl}^{}=0`$. A notational remark is that a primed tensor denoted the covariant derivative in the $`M`$ direction and a doubly primed tensor denotes the covariant derivative in the $`\stackrel{~}{M}`$ direction. Doubly primed does not mean second derivative.
The curvature of this connection may be computed by either using the expression involving $`\omega `$ or the one involving $`\stackrel{~}{\omega }`$. A straightforward computation of the curvature matrix $`2`$-form
$$\mathrm{\Psi }_{ij}=d\psi _{ij}+\psi _{ik}\psi _{kj}$$
(6.13)
in these two ways leads to the following expressions
$`\mathrm{\Psi }_{ij}`$ $`=`$ $`T_{ijlm}^{\prime \prime }\theta ^l\stackrel{~}{\theta }^m`$
$`+`$ $`{\displaystyle \frac{1}{2}}\left[R_{ijlm}(T_{ijlm}^{}T_{ijml}^{})+(T_{ikl}T_{kjm}T_{ikm}T_{kjl})\right]\theta ^l\theta ^m,`$
and
$`\mathrm{\Psi }_{ij}`$ $`=`$ $`\stackrel{~}{T}_{ijlm}^{}\stackrel{~}{\theta }^l\theta ^m`$
$`+`$ $`{\displaystyle \frac{1}{2}}\left[\stackrel{~}{R}_{ijlm}(\stackrel{~}{T}_{ijlm}^{\prime \prime }\stackrel{~}{T}_{ijml}^{\prime \prime })+(\stackrel{~}{T}_{ikl}\stackrel{~}{T}_{kjm}\stackrel{~}{T}_{ikm}\stackrel{~}{T}_{kjl})\right]\stackrel{~}{\theta }^l\stackrel{~}{\theta }^m.`$
Comparing these two expression we learn that the curvature two form matrix is given by
$$\mathrm{\Psi }_{ij}=d\psi _{ij}+\psi _{ik}\psi _{kj}=T_{ijlm}^{\prime \prime }\theta ^l\stackrel{~}{\theta }^m.$$
(6.14)
The following constraints must also hold
$`R_{ijlm}(T_{ijlm}^{}T_{ijml}^{})+(T_{ikl}T_{kjm}T_{ikm}T_{kjl})`$ $`=`$ $`0,`$ (6.15)
$`\stackrel{~}{R}_{ijlm}(\stackrel{~}{T}_{ijlm}^{\prime \prime }\stackrel{~}{T}_{ijml}^{\prime \prime })+(\stackrel{~}{T}_{ikl}\stackrel{~}{T}_{kjm}\stackrel{~}{T}_{ikm}\stackrel{~}{T}_{kjl})`$ $`=`$ $`0,`$ (6.16)
$`T_{ijlm}^{\prime \prime }+\stackrel{~}{T}_{ijml}^{}`$ $`=`$ $`0.`$ (6.17)
Form (6.14) is reminiscent of a Kähler manifold where the curvature is of type $`dzd\overline{z}`$ and there are no $`dzdz`$ or $`d\overline{z}d\overline{z}`$ components. The absence of these many curvature components is due to the reduction of the structure group from $`\mathrm{O}(2n)`$ to $`\mathrm{O}(n)`$ at the expense of generating torsion.
There are a variety of equivalent ways of interpreting the above. The most geometric is to observe that $`\psi _{ij}`$ defines a connection on $`P`$ and thus a connection when restricted to any of the fibers. For example, let $`M_{\stackrel{~}{x}}=\stackrel{~}{\mathrm{\Pi }}^1(\stackrel{~}{x})`$ be a horizontal fiber. Notice that along this fiber $`\stackrel{~}{\theta }=0`$ and thus $`\mathrm{\Psi }_{ij}=0`$. Since $`M_{\stackrel{~}{x}}`$ is isometric to $`M`$ we have found a flat orthogonal connection (generally with torsion) on $`M`$. Note that this is true for all horizontal fibers. One can make a similar statement about the vertical fibers. We have our first major result.
> Target space duality requires that the manifolds $`M`$ and $`\stackrel{~}{M}`$ respectively admit flat orthogonal connections. The connection $`\psi _{ij}`$ is flat when restricted to either $`M`$ or $`\stackrel{~}{M}`$.
At a more algebraic level equations (6.15) and (6.16) are the standard equations for “parallelizing” the curvature by torsion. A manifold $`M`$ is said to be parallelizable if the tangent bundle is a product bundle $`TM=M\times ^n`$. This means that you can globally choose a frame on $`M`$. The existence of a flat connection on a manifold does not imply parallelizability. The reason is that in a non-simply connected manifold there is an obstruction to globally choosing a frame if there is holonomy. If the manifold is simply connected and the connection is flat then it is parallelizable. Finally we observe that if a manifold is parallelizable then there are an infinite number of other possible parallelizations<sup>11</sup><sup>11</sup>11I would like to thank I.M. Singer for the ensuing argument.. Assume we have an orthogonal parallelization, *i.e.*, a choice of orthonormal frame at each point. Given any other orthogonal parallelization we can always make a rotation point by point so that both frames agree at the point. Thus the space of all orthogonal parallelizations is given by the set of maps from $`M`$ to $`\mathrm{O}(n)`$.
Note that given two distinct points $`\stackrel{~}{x}_1,\stackrel{~}{x}_2\stackrel{~}{M}`$, the tensor $`T_{ijk}`$ on the respective horizontal fibers $`M_{\stackrel{~}{x}_1}`$ and $`M_{\stackrel{~}{x}_2}`$ do not have to be the same. There are many flat orthogonal connections on $`M`$ as can be seen by a variant parallelizability argument. In fact you could in principle have a multiparameter family parametrized by $`\stackrel{~}{M}`$.
There is a special case of interest when $`T_{ijk}`$ is the pullback of a tensor on $`M`$. In this case a previous remark tells us that $`T_{ijkl}^{\prime \prime }=0`$ and consequently by (6.17) we also have $`\stackrel{~}{T}_{ijkl}^{}=0`$. Therefore $`\stackrel{~}{T}_{ijk}`$ is also the pullback of a tensor on $`\stackrel{~}{M}`$. This means that the same torsion tensors make the connection flat on all the fibers. Note that in this case $`\mathrm{\Psi }_{ij}=0`$ and the orthogonal connection $`\psi _{ij}`$ is a flat connection on $`P`$.
> If $`T_{ijk}`$ is the pullback of a tensor on $`M`$ then $`\stackrel{~}{T}_{ijk}`$ is the pullback of a tensor on $`\stackrel{~}{M}`$ and $`\mathrm{\Psi }_{ij}=0`$. In this case $`\psi _{ij}`$ is a flat connection on $`P`$.
## 7 Simple examples
The equation $`d\gamma =H\stackrel{~}{H}`$ introduces relations among $`H,\stackrel{~}{H},T_{ijk}`$ and $`\stackrel{~}{T}_{ijk}`$. First we point out some facts.
### 7.1 The case of $`n_{ij}=0`$
As a warmup we study the case where $`n_{ij}=0`$. In this case $`\gamma =\stackrel{~}{\theta }^i\theta ^i`$ and we compute $`d\gamma `$ by using the Cartan structural equations (5.1), (5.2) and the condition which follows from the reduction of the symmetry group (6.9). A brief computation yields
$$d\gamma =T_{kij}\theta ^i\theta ^j\stackrel{~}{\theta }^k\stackrel{~}{T}_{ijk}\theta ^i\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k.$$
First we learn that the $`3`$-forms $`H`$ and $`\stackrel{~}{H}`$ vanish. Next we see that $`T_{kij}=T_{kji}`$ and $`\stackrel{~}{T}_{ijk}=\stackrel{~}{T}_{ikj}`$. We remind the reader that a tensor $`S_{ijk}`$ which is skew symmetric under $`ij`$ and symmetric under $`jk`$ is zero. Thus we conclude that $`T_{ijk}=\stackrel{~}{T}_{ijk}=0`$. It follows from equations (6.15) and (6.16) that $`R_{ijkl}=\stackrel{~}{R}_{ijkl}=0`$. Since the Riemannian curvatures vanish we know that $`M`$ and $`\stackrel{~}{M}`$ are manifolds with universal cover $`^n`$. There are no other possibilities if $`n_{ij}=0`$. For example you can have $`M=𝕋^k\times ^{nk}`$. This is the case of abelian duality. Other potential singular cases of interest are orbifolds or cones which are flat but have holonomy due to the presence of singularities.
### 7.2 The case of a Lie group
We verify that the standard nonabelian duality results are reproducible in this formalism. We present a schematic discussion here because the Lie group example is a special case of a more general result presented in Section II-2.2.1. Let $`G`$ a compact simple Lie group with Lie algebra $`𝔤`$. Let $`(e_i,\mathrm{},e_n)`$ is an orthonormal basis for $`𝔤`$ with respect to the Killing form. The structure constants $`f_{ijk}`$ are defined by $`[e_i,e_j]=f_{kij}e_k`$. In this case the structure constants are totally antisymmetric. Let $`\theta ^i`$ be the associated Maurer-Cartan forms satisfying the Maurer-Cartan equations
$$d\theta ^i=\frac{1}{2}f_{ijk}\theta ^j\theta ^k.$$
(7.1)
Because of the Killing form we can identify the Lie algebra $`𝔤`$ with its vector space dual $`𝔤^{}`$. We choose $`P`$ to be the cotangent bundle $`T^{}G`$ which is a product bundle $`T^{}G=G\times 𝔤^{}=G\times 𝔤`$. If $`(p_1,\mathrm{},p_n)`$ are the standard coordinates on the cotangent bundle with respect to the orthonormal frame then the we take $`\alpha `$ in (2.2) to be $`\alpha =p_i\theta ^i`$, the canonical $`1`$-form on $`T^{}G`$. Therefore $`\beta =d\alpha `$ is the standard symplectic form on $`T^{}G`$ given by
$$\beta =dp_i\theta ^i\frac{1}{2}p_if_{ijk}\theta ^j\theta ^k.$$
(7.2)
By looking at reference one can see that the orthonormal coframe $`(\stackrel{~}{\theta }^1,\mathrm{},\stackrel{~}{\theta }^n)`$ on the fiber $`𝔤^{}`$ is given by $`dp_j=\stackrel{~}{\theta }^i(\delta _{ij}+f_{kij}p_k)`$. This suggests that $`m_{ij}=(\delta _{ij}+f_{kij}p_k)`$ and that in this basis the symmetry breaking is manifest and thus $`n_{ij}=f_{kij}p_k`$. Thus we expect that $`\gamma `$ is given by
$$\gamma =\frac{1}{2}f_{kij}p_k\theta ^i\theta ^j+(\delta _{ij}+f_{kij}p_k)\stackrel{~}{\theta }^i\theta ^j\frac{1}{2}f_{kij}p_k\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j.$$
(7.3)
Note that $`d\gamma =\stackrel{~}{H}`$ because the modification of going from the closed form $`\beta `$ to $`\gamma `$ involved a term of the type $`n_{ij}\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j`$. To verify this we observe that $`\stackrel{~}{\theta }^i=dp_jm_{ji}^1`$ and thus $`n_{ij}\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j`$ only depends on $`p`$ and $`dp`$, therefore, its exterior derivative can only be of type $`dpdpdp\stackrel{~}{\theta }\stackrel{~}{\theta }\stackrel{~}{\theta }`$. In fact $`\frac{1}{2}f_{kij}p_k\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j`$ is the standard representation for the $`2`$-form $`\stackrel{~}{B}`$.
If we write $`d\stackrel{~}{\theta }^i=\frac{1}{2}\stackrel{~}{f}_{ijk}\theta ^j\theta ^k`$ then a straightforward exercise shows that
$$\stackrel{~}{f}_{ijk}=(m_{jm}f_{mkl}m_{km}f_{mjl})m_{li}^1.$$
By using (B.1) one can compute $`\stackrel{~}{\omega }_{ij}`$. It is now an algebraic exercise to compute parallelizing torsions $`T_{ijk}`$ and $`\stackrel{~}{T}_{ijk}`$.
## 8 The case of a general connection $`\psi `$
### 8.1 General theory
We already saw that the connection $`\psi _{ij}`$ on $`P`$ gives a flat connection on both $`M`$ and $`\stackrel{~}{M}`$, a necessary condition for $`M`$ and $`\stackrel{~}{M}`$ to be target space duals of each other. We are going to take the following approach. Assume we are given a $`\psi _{ij}`$ on $`P`$, how do we determine $`n_{ij}`$? We will derive PDEs that $`n_{ij}`$ must satisfy. If there exist solutions to these PDEs then we automatically have a duality between the sigma model on $`M`$ and the one on $`\stackrel{~}{M}`$. for It is worthwhile to rewrite the Cartan structural equations in terms of $`\psi _{ij}`$:
$`d\theta ^i`$ $`=`$ $`\psi _{ij}\theta ^j{\displaystyle \frac{1}{2}}f_{ijk}\theta ^j\theta ^k,`$ (8.1)
$`d\stackrel{~}{\theta }^i`$ $`=`$ $`\psi _{ij}\stackrel{~}{\theta }^j{\displaystyle \frac{1}{2}}\stackrel{~}{f}_{ijk}\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k,`$ (8.2)
$`d\psi _{ij}`$ $`=`$ $`\psi _{ik}\psi _{kj}T_{ijlm}^{\prime \prime }\theta ^l\stackrel{~}{\theta }^m.`$ (8.3)
where $`f_{ijk}=f_{ikj}`$, $`\stackrel{~}{f}_{ijk}=\stackrel{~}{f}_{ikj}`$ and $`T_{ijkl}^{\prime \prime }=T_{jikl}^{\prime \prime }`$. The structure functions $`f_{ijk}`$ and $`\stackrel{~}{f}_{ijk}`$ are related to $`T_{ijk}`$ and $`\stackrel{~}{T}_{ijk}`$ by
$`f_{ijk}=T_{ijk}T_{ikj},`$ $`T_{ijk}={\displaystyle \frac{1}{2}}(f_{ijk}f_{jik}f_{kij}),`$ (8.4)
$`\stackrel{~}{f}_{ijk}=\stackrel{~}{T}_{ijk}\stackrel{~}{T}_{ikj},`$ $`\stackrel{~}{T}_{ijk}={\displaystyle \frac{1}{2}}(\stackrel{~}{f}_{ijk}\stackrel{~}{f}_{jik}\stackrel{~}{f}_{kij}).`$ (8.5)
We define the components $`n_{ijk}^{},n_{ijk}^{\prime \prime },f_{ijkl}^{},f_{ijkl}^{\prime \prime },\stackrel{~}{f}_{ijkl}^{},\stackrel{~}{f}_{ijkl}^{\prime \prime }`$ of the covariant derivatives of $`n_{ij},f_{ijk},\stackrel{~}{f}_{ijk}`$ with respect to the connection $`\psi _{ij}`$ by
$`dn_{ij}+\psi _{ik}n_{kj}+\psi _{jk}n_{ik}`$ $`=`$ $`n_{ijk}^{}\theta ^k+n_{ijk}^{\prime \prime }\stackrel{~}{\theta }^k.`$ (8.6)
$`df_{ijk}+\psi _{il}f_{ljk}+\psi _{jl}f_{ilk}+\psi _{kl}f_{ijl}`$ $`=`$ $`f_{ijkl}^{}\theta ^l+f_{ijkl}^{\prime \prime }\stackrel{~}{\theta }^l,`$ (8.7)
$`d\stackrel{~}{f}_{ijk}+\psi _{il}\stackrel{~}{f}_{ljk}+\psi _{jl}\stackrel{~}{f}_{ilk}+\psi _{kl}\stackrel{~}{f}_{ijl}`$ $`=`$ $`\stackrel{~}{f}_{ijkl}^{}\theta ^l+\stackrel{~}{f}_{ijkl}^{\prime \prime }\stackrel{~}{\theta }^l.`$ (8.8)
There are several important constraints which follow from $`d^2\theta =d^2\stackrel{~}{\theta }=0`$:
$`\left(f_{ijkl}^{}+f_{mjk}f_{iml}\right)\theta ^j\theta ^k\theta ^l`$ $`=`$ $`0,`$ (8.9)
$`f_{ijkl}^{\prime \prime }`$ $`=`$ $`T_{ijkl}^{\prime \prime }T_{ikjl}^{\prime \prime },`$ (8.10)
$`\left(\stackrel{~}{f}_{ijkl}^{\prime \prime }+\stackrel{~}{f}_{mjk}\stackrel{~}{f}_{iml}\right)\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k\stackrel{~}{\theta }^l`$ $`=`$ $`0`$ (8.11)
$`\stackrel{~}{f}_{ijkl}^{}`$ $`=`$ $`(T_{ijlk}^{\prime \prime }T_{iklj}^{\prime \prime }).`$ (8.12)
Note that $`T_{ijkl}^{\prime \prime }=0`$ if and only if $`f_{ijkl}^{\prime \prime }=\stackrel{~}{f}_{ijkl}^{}=0`$, *i.e.*, $`f_{ijk}`$ and $`\stackrel{~}{f}_{ijk}`$ are respectively pullbacks in accord with a previous remark. The $`d^2\psi _{ij}=0`$ constraints are not used in this report and will not be given.
To derive the PDE satisfied by $`n_{ij}`$ we compute $`d\gamma `$:
$`d\gamma `$ $`=`$ $`H\stackrel{~}{H}`$ (8.13)
$`=`$ $`{\displaystyle \frac{1}{2}}n_{i}^{}{}_{j}{}^{}{}_{k}{}^{}\theta ^i\theta ^j\theta ^k+{\displaystyle \frac{1}{2}}f_{i}^{}{}_{j}{}^{}{}_{k}{}^{}n_l^i\theta ^j\theta ^k\theta ^l`$
$``$ $`{\displaystyle \frac{1}{2}}n_{i}^{\prime \prime }{}_{j}{}^{}{}_{k}{}^{}\theta ^i\theta ^j\stackrel{~}{\theta }^kn_{i}^{}{}_{j}{}^{}{}_{k}{}^{}\theta ^i\theta ^k\stackrel{~}{\theta }^j`$
$`+`$ $`{\displaystyle \frac{1}{2}}f_{i}^{}{}_{j}{}^{}{}_{k}{}^{}\theta ^j\theta ^k\stackrel{~}{\theta }^i{\displaystyle \frac{1}{2}}f_{i}^{}{}_{j}{}^{}{}_{k}{}^{}n_l^i\theta ^j\theta ^k\stackrel{~}{\theta }^l`$
$`+`$ $`n_{i}^{\prime \prime }{}_{j}{}^{}{}_{k}{}^{}\theta ^i\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k{\displaystyle \frac{1}{2}}n_{i}^{}{}_{j}{}^{}{}_{k}{}^{}\theta ^k\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j`$
$``$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{f}_{i}^{}{}_{j}{}^{}{}_{k}{}^{}\theta ^i\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k{\displaystyle \frac{1}{2}}\stackrel{~}{f}_{i}^{}{}_{j}{}^{}{}_{k}{}^{}n_l^i\theta ^l\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k`$
$``$ $`{\displaystyle \frac{1}{2}}n_{i}^{\prime \prime }{}_{j}{}^{}{}_{k}{}^{}\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k+{\displaystyle \frac{1}{2}}\stackrel{~}{f}_{i}^{}{}_{j}{}^{}{}_{k}{}^{}n_l^i\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k\stackrel{~}{\theta }^l.`$
If we write the closed $`3`$-forms in components as
$$H=\frac{1}{3!}H_{ijk}\theta ^i\theta ^j\theta ^k,\stackrel{~}{H}=\frac{1}{3!}\stackrel{~}{H}_{ijk}\stackrel{~}{\theta }^i\stackrel{~}{\theta }^j\stackrel{~}{\theta }^k,$$
(8.14)
where $`H_{ijk}`$ and $`\stackrel{~}{H}_{ijk}`$ are totally skew symmetric then we immediately see that
$`n_{ijk}^{}+n_{jki}^{}+n_{kij}^{}`$ $`=`$ $`H_{ijk}+(f_{lij}n_{lk}+f_{ljk}n_{li}+f_{lki}n_{lj}),`$ (8.15)
$`n_{ijk}^{\prime \prime }+n_{jki}^{\prime \prime }+n_{kij}^{\prime \prime }`$ $`=`$ $`+\stackrel{~}{H}_{ijk}+(\stackrel{~}{f}_{lij}n_{lk}+\stackrel{~}{f}_{ljk}n_{li}+\stackrel{~}{f}_{lki}n_{lj}),`$ (8.16)
$`(n_{kij}^{}n_{kji}^{})n_{ijk}^{\prime \prime }`$ $`=`$ $`(f_{kij}n_{lk}f_{lij})=m_{kl}f_{lij},`$ (8.17)
$`n_{ijk}^{}+(n_{kij}^{\prime \prime }n_{kji}^{\prime \prime })`$ $`=`$ $`+(\stackrel{~}{f}_{kij}+n_{lk}\stackrel{~}{f}_{lij})=\stackrel{~}{f}_{lij}m_{lk}.`$ (8.18)
The number of linearly independent equations above is $`\frac{1}{3}n(n1)(2n1)`$. The best way to see this is that if we define $`\xi _\pm ^i=(\theta ^i\stackrel{~}{\theta }^i)`$ then the term containing $`n_{ij}`$ in $`\gamma `$ is basically $`n_{ij}\xi _+^i\xi _+^j`$. If the components of the covariant derivatives of $`n_{ij}`$ in this basis are $`n_{ijk}^\pm `$ then $`d(n_{ij}\xi _+^i\xi _+^i)n_{ijk}^+\xi _+^k\xi _+^i\xi _+^j+n_{ijk}^{}\xi _{}^k\xi _+^i\xi _+^j\mathrm{}`$. The stuff in ellipsis does not involves derivatives of $`n_{ij}`$. Since $`n_{ijk}^+`$ is linearly independent of $`n_{ijk}^{}`$ we see that the number of equations we get is $`\frac{1}{3!}n(n1)(n2)+n\times \frac{1}{2}n(n1)`$. The first remark we make is that the PDEs given by (8.6) generally make an overdetermined system if $`n>1`$. The reason is that there are $`\frac{1}{3}n(n1)(2n1)`$ equations for $`\frac{1}{2}n(n1)`$ functions $`n_{ij}`$. This means that for a solution to exist integrability conditions arising from $`d^2n_{ij}=0`$ must be satisfied.
Let $`t_{ijk}=t_{jik}`$ be a tensor in $`(^2V)V`$ for some $`n`$ dimensional vector space $`V`$ with inner product. The vector space $`(^2V)V`$ has an orthogonal decomposition into $`(^3V)((^2V)V)_{\mathrm{mixed}}`$ where the latter are the tensors of mixed symmetry under the permutation group. The orthogonal projectors $`𝔄`$ (antisymmetrization) and $`𝔐`$ (mixed) that respectively project onto $`^3V`$ and $`((^2V)V)_{\mathrm{mixed}}`$ are
$`(𝔄t)_{ijk}`$ $`=`$ $`{\displaystyle \frac{1}{3}}(t_{ijk}+t_{jki}+t_{kij}),`$ (8.19)
$`(𝔐t)_{ijk}`$ $`=`$ $`{\displaystyle \frac{1}{3}}(2t_{ijk}t_{jki}t_{kij}).`$ (8.20)
A detailed analysis (see below) of equations (8.15), (8.16), (8.17) and (8.18) shows that they determine $`𝔄n^{}`$, $`𝔄n^{\prime \prime }`$ and $`𝔐(n^{}+n^{\prime \prime })`$. These equations do not provide information about $`𝔐(n^{}n^{\prime \prime })`$.
To solve the equations above it is best on introduce the following auxiliary tensors:
$`V_{ijk}`$ $`=`$ $`H_{ijk}(f_{lij}n_{lk}+f_{ljk}n_{li}+f_{lki}n_{lj}),`$ (8.21)
$`\stackrel{~}{V}_{ijk}`$ $`=`$ $`\stackrel{~}{H}_{ijk}+(\stackrel{~}{f}_{lij}n_{lk}+\stackrel{~}{f}_{ljk}n_{li}+\stackrel{~}{f}_{lki}n_{lj}),`$ (8.22)
$`W_{ijk}`$ $`=`$ $`(f_{kij}n_{lk}f_{lij})=m_{kl}f_{lij},`$ (8.23)
$`\stackrel{~}{W}_{ijk}`$ $`=`$ $`(\stackrel{~}{f}_{kij}+n_{lk}\stackrel{~}{f}_{lij})=\stackrel{~}{f}_{lij}m_{lk}.`$ (8.24)
They are all skew symmetric under the interchange $`ij`$ and $`V,\stackrel{~}{V}`$ are totally antisymmetric. Given a value for $`n_{ij}`$, these tensor are determined by the geometric data which specifies the sigma models. This data is not independent because these tensors are linearly related due to the right hand sides of (8.15), (8.16), (8.17) and (8.18).
A little algebra shows that
$$n^{}+n^{\prime \prime }=WV=\stackrel{~}{W}+\stackrel{~}{V}.$$
(8.25)
All the content of (8.15), (8.16), (8.17) and (8.18) is contained in (8.15), (8.16) and (8.25). These equations place constraints on $`V,\stackrel{~}{V},W,\stackrel{~}{W}`$. Immediate conclusions are that
$`W+\stackrel{~}{W}`$ $`=`$ $`V+\stackrel{~}{V},`$ (8.26)
$`𝔐(W+\stackrel{~}{W})`$ $`=`$ $`0,`$ (8.27)
$`𝔄W`$ $`=`$ $`{\displaystyle \frac{2}{3}}V+{\displaystyle \frac{1}{3}}\stackrel{~}{V},`$ (8.28)
$`𝔄\stackrel{~}{W}`$ $`=`$ $`{\displaystyle \frac{1}{3}}V+{\displaystyle \frac{2}{3}}\stackrel{~}{V}.`$ (8.29)
In deriving the last two equation we used (8.15), (8.16) and applied the $`𝔄`$ operator to (8.25). The equations above imply linear algebraic relations among the data that defines the sigma models. They tell us that there exists a tensor $`U_{ijk}`$ of mixed symmetry, *i.e.*, $`U_{ijk}=U_{jik}`$ and $`𝔄U=0`$ such that
$`W`$ $`=`$ $`+U+{\displaystyle \frac{2}{3}}V+{\displaystyle \frac{1}{3}}\stackrel{~}{V},`$ (8.30)
$`\stackrel{~}{W}`$ $`=`$ $`U+{\displaystyle \frac{1}{3}}V+{\displaystyle \frac{2}{3}}\stackrel{~}{V}.`$ (8.31)
Collating all our information we can now write down the $`\frac{1}{3}n(n1)(2n1)`$ first order linear PDEs that determine $`n_{ij}`$:
$`𝔄n^{}`$ $`=`$ $`{\displaystyle \frac{1}{3}}V,`$ (8.32)
$`𝔄n^{\prime \prime }`$ $`=`$ $`+{\displaystyle \frac{1}{3}}\stackrel{~}{V},`$ (8.33)
$`𝔐(n^{}+n^{\prime \prime })`$ $`=`$ $`+U.`$ (8.34)
There is no equation for $`𝔐(n^{}n^{\prime \prime })`$. It is worthwhile to note that
$`n^{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}U+{\displaystyle \frac{1}{2}}𝔐(n^{}n^{\prime \prime }){\displaystyle \frac{1}{3}}V,`$ (8.35)
$`n^{\prime \prime }`$ $`=`$ $`{\displaystyle \frac{1}{2}}U{\displaystyle \frac{1}{2}}𝔐(n^{}n^{\prime \prime })+{\displaystyle \frac{1}{3}}\stackrel{~}{V}.`$ (8.36)
You can envision using this formalism in four basic scenarios.
1. Test to see if two sigma models $`(M,g,B)`$ and $`(\stackrel{~}{M},\stackrel{~}{g},\stackrel{~}{B})`$ are dual to each other. This entails the construction of the symplectic manifold $`P`$.
2. Given a sigma model $`(M,g,B)`$ and a symplectic manifold $`P`$, naturally associated with $`M`$, can you construct the dual sigma model $`(\stackrel{~}{M},\stackrel{~}{g},\stackrel{~}{B})`$?
3. Given a symplectic manifold $`P`$ that admits a bifibration, attempt to construct dual sigma models.
4. Find all symplectic manifolds $`P`$ that admit dual sigma models.
### 8.2 Covariantly constant $`n_{ij}`$
Here we show that the assumption of covariantly constant $`n_{ij}`$ leads to a flat connection on $`P`$. Assume that in our adapted coframes the $`n_{ij}`$ are covariantly constant with respect to the $`\psi `$ connection, *i.e.*, $`n_{ijk}^{}=n_{ijk}^{\prime \prime }=0`$. In this case it is immediate from (8.17) and (8.18) that $`f_{ijk}=\stackrel{~}{f}_{ijk}=0`$. Subsequently we see from (8.15) and (8.16) that $`H=\stackrel{~}{H}=0`$. From (8.10) we see that $`T_{ijkl}^{\prime \prime }=0`$ and thus the curvature vanishes, $`\mathrm{\Psi }_{ij}=0`$. We are mostly interested in local properties so we might as well assume $`P`$ is parallelizable. We can use parallel transport with respect to this connection to get a global framing. In this special framing the connection coefficients vanish and thus we can make the substitution $`\psi _{ij}=0`$ in all the equations in Section 8.1. Note that the orthonormal coframes satisfy $`d\theta ^i=d\stackrel{~}{\theta }^i=0`$ and thus $`M`$ and $`\stackrel{~}{M}`$ are manifolds with cover $`^n`$. Following up on remarks made in Section 7.1 we see that this is the case of abelian duality but with constant $`n_{ij}`$ in the adapted frames corresponding to constant $`B_{ij}`$ and $`\stackrel{~}{B}_{ij}`$.
### 8.3 Case of $`\stackrel{~}{f}_{ijk}=0`$.
What is the most general manifold $`M`$ whose dual $`\stackrel{~}{M}`$ has cover $`^n`$? Note that by (8.5) we have that $`\stackrel{~}{T}_{ijk}=0`$ and thus $`T_{ijlm}^{\prime \prime }=\stackrel{~}{T}_{ijml}^{}=0`$. This means that the curvature (6.14) of the connection $`\psi _{ij}`$ vanishes. Again using the remarks just made we can choose a parallel framing such that $`\psi _{ij}=0`$. Since $`\stackrel{~}{f}_{ijk}=0`$ we have that $`d\stackrel{~}{\theta }^i=0`$ and thus locally there exists functions $`\stackrel{~}{x}^i`$ such that $`\stackrel{~}{\theta }^i=d\stackrel{~}{x}^i`$. We also have that $`d\theta ^i=\frac{1}{2}f_{ijk}\theta ^j\theta ^k`$. Previous arguments also tell us that $`f_{ijk}`$ is the pullback of a tensor on $`M`$. From (8.22) we see that $`\stackrel{~}{V}=\stackrel{~}{H}`$ and from (8.24) we have that $`\stackrel{~}{W}=0`$. Equation (8.31) tells us that $`U=0`$ and $`V=2\stackrel{~}{V}=2\stackrel{~}{H}`$. Inserting into (8.30) we find that
$$W_{ijk}=m_{kl}f_{lij}=(\delta _{kl}+n_{kl})f_{lij}=\stackrel{~}{H}_{ijk}.$$
(8.37)
An elementary consequence of this equation is that if $`\stackrel{~}{H}=0`$ then $`f_{ijk}=0`$ and $`M`$ is also a manifold with cover $`^n`$. Inserting the above into (8.21) we find that
$$\stackrel{~}{H}_{ijk}=H_{ijk}(f_{ijk}+f_{jki}+f_{kij}).$$
(8.38)
The left hand side is the pullback of a tensor on $`\stackrel{~}{M}`$ and the right hand side is the pullback of a tensor on $`M`$ thus each side must be constant. We have learned that $`\stackrel{~}{H}_{ijk}`$ are constants. We assume that $`n_{ij}`$ are not constant (see Section 8.2). From (8.37) we expect the $`f_{ijk}`$ not to be constant. Let $`\mu _{ijk}`$ be a tensor of mixed symmetry then from (8.35) and (8.36) we see that
$$dn_{kl}=\left(\mu _{klm}\frac{2}{3}\stackrel{~}{H}_{klm}\right)\theta ^m+\left(\mu _{klm}+\frac{1}{3}\stackrel{~}{H}_{klm}\right)\stackrel{~}{\theta }^m.$$
(8.39)
The answer to the question, “What is the most general manifold $`M`$ whose dual $`\stackrel{~}{M}`$ has cover $`^n`$?” and the construction of the duality transformation is given by the general solution to the following system of exterior differential equations:
$`(\delta _{kl}+n_{kl})d\theta ^l`$ $`=`$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{H}_{kij}\theta ^i\theta ^j,`$ (8.40)
$`d\stackrel{~}{\theta }^i`$ $`=`$ $`0,`$ (8.41)
$`dn_{kl}`$ $`=`$ $`\left(\mu _{klm}{\displaystyle \frac{2}{3}}\stackrel{~}{H}_{klm}\right)\theta ^m+\left(\mu _{klm}+{\displaystyle \frac{1}{3}}\stackrel{~}{H}_{klm}\right)\stackrel{~}{\theta }^m.`$ (8.42)
As an example consider the special case of $`\stackrel{~}{H}=0`$. From (8.37) we have that $`f_{ijk}=0`$. We are now asking, “What is the most general duality transformation between manifolds with cover $`^n`$?” The equations above tell us that there exists functions $`x^i`$ and $`\stackrel{~}{x}^j`$ such that $`\theta ^i=dx^i`$ and $`\stackrel{~}{\theta }^j=d\stackrel{~}{x}^j`$. Equation (8.42) becomes $`dn_{kl}=\mu _{klm}dy^m`$ where $`y^i=x^i\stackrel{~}{x}^i`$. We learn that $`n_{ij}`$ is a function of $`y`$ only. Since the tensor $`\mu `$ has mixed symmetry we see that $`d(n_{ij}(y)dy^idy^j)=0`$ and thus we conclude that locally there exists functions $`r_i`$ of the independent variables $`y^j`$ such that
$$\frac{1}{2}n_{ij}(y)dy^idy^j=d\left(r_i(y)dy^i\right).$$
We now have all the information required to construct the duality transformation. The duality transformations are given by
$`\stackrel{~}{\pi }+\stackrel{~}{x}_\sigma `$ $`=`$ $`\pi +x{}_{\sigma }{}^{},`$
$`\stackrel{~}{\pi }\stackrel{~}{x}_\sigma `$ $`=`$ $`T_{}(x\stackrel{~}{x})[\pi x{}_{\sigma }{}^{}]`$
with $`T_{}(y)=(I+n(y))(In(y))^1`$. By taking the sum and difference of the equations above one gets ODEs that can be solved for $`(\stackrel{~}{x}(\sigma ),\stackrel{~}{\pi }(\sigma ))`$ given $`(x(\sigma ),\pi (\sigma ))`$.
## Acknowledgments
I would like to thank O. Babelon, L. Baulieu, T. Curtright, L.A. Ferreira, D. Freed, S. Kaliman, C-H Liu, R. Nepomechie, N. Reshetikhin, J. Sánchez Guillén, N. Wallach and P. Windey for discussions on a variety of topics. I would also like to thank Jack Lee for his *Mathematica* package Ricci that was used to perform some of the computations. I am particularly thankful to R. Bryant and I.M. Singer for patiently answering my many questions about differential geometry. This work was supported in part by National Science Foundation grant PHY–9870101.
Appendices
## Appendix A Facts about orthogonal groups
We collate some basic properties of orthogonal groups in this section. An orthogonal matrix in $`\mathrm{O}(n,n)`$ may be written in terms of $`n\times n`$ blocks as
$$\left(\begin{array}{cc}A& B\\ C& D\end{array}\right)$$
where $`A^tA+C^tC=I`$, $`B^tB+D^tD=I`$ and $`B^tA+D^tC=0`$. A matrix in the Lie algebra $`𝔰𝔬(n,n)`$ may be written as
$$\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)$$
where $`a=a^t`$, $`d=d^t`$ and $`c=b^t`$.
The group $`\mathrm{O}_\mathrm{Q}(n,n)\mathrm{GL}(2n,)`$ is defined to be the linear transformations which leave
$$Q=\left(\begin{array}{cc}0& I_n\\ I_n& 0\end{array}\right).$$
invariant. A matrix in $`\mathrm{O}_\mathrm{Q}(n,n)`$ may be written in $`n\times n`$ blocks as
$$\left(\begin{array}{cc}W& X\\ Y& Z\end{array}\right)$$
(A.1)
where $`W^tZ+Y^tX=I`$, $`W^tY+Y^tW=0`$ and $`X^tZ+Z^tX=0`$. It is very important to observe that matrices of the form
$$\left(\begin{array}{cc}I& 0\\ Y& I\end{array}\right)$$
(A.2)
where $`Y=Y^t`$ are in $`\mathrm{O}_\mathrm{Q}(n,n)`$. A matrix in the Lie algebra $`𝔰𝔬_\mathrm{Q}(n,n)`$ may be written as
$$\left(\begin{array}{cc}w& x\\ y& z\end{array}\right)$$
(A.3)
where $`z=w^t`$, $`y=y^t`$ and $`x=x^t`$.
Conjugating by the orthogonal matrix
$$T=\frac{1}{\sqrt{2}}\left(\begin{array}{cc}I_n& I_n\\ I_n& I_n\end{array}\right)$$
leads one to the observation that
$$T\left(\begin{array}{cc}0& I_n\\ I_n& 0\end{array}\right)T^1=\left(\begin{array}{cc}I_n& 0\\ 0& I_n\end{array}\right).$$
(A.4)
Therefore the group $`\mathrm{O}_\mathrm{Q}(n,n)`$ is isomorphic to $`\mathrm{O}(n,n)`$.
We will also need to identify the compact group $`\mathrm{O}(2n)\mathrm{O}_\mathrm{Q}(n,n)`$. To do this we observe that at the Lie algebra level $`𝔰𝔬(2n)𝔰𝔬_\mathrm{Q}(n,n)`$ is given by matrices of the form
$$\left(\begin{array}{cc}a& b\\ b& a\end{array}\right)$$
where $`a^t=a`$ and $`b^t=b`$. Conjugating by $`T`$ one sees that
$$T\left(\begin{array}{cc}a& b\\ b& a\end{array}\right)T^1=\left(\begin{array}{cc}ab& 0\\ 0& a+b\end{array}\right).$$
Therefore, the intersection $`𝔰𝔬(2n)𝔰𝔬_\mathrm{Q}(n,n)`$ is conjugate to $`𝔰𝔬(n)𝔰𝔬(n)`$.
## Appendix B On the structure functions
In a local orthonormal coframe one has the Cartan structural equations (5.1). Locally one can always write $`d\theta ^i=\frac{1}{2}f_{ijk}\theta ^j\theta ^k`$ for some “structure functions” $`f_{ijk}`$ which are skew symmetric in $`jk`$. If we write the riemannian connection as $`\omega _{ij}=\omega _{ijk}\theta ^k`$ then
$$\omega _{ijk}=\frac{1}{2}(f_{kij}f_{ijk}+f_{jik}).$$
(B.1)
This allows us to reconstruct all the local Riemannian geometry of the manifold in terms of the structure functions. |
warning/0003/quant-ph0003134.html | ar5iv | text | # Generalized Noiseless Quantum Codes utilizing Quantum Enveloping Algebras
## 1. Introduction
Quantum computation is a new and quickly developing area of science. Its power comes from using quantum parallelism of computations. This new paradigm for computation was envisioned by Feynman . For many years quantum computation looked as an unrealistic dream. The reason is unavoidable decoherence due to the interaction of quantum devices with a classical environment, which destroys quantum coherent states. Then the advantage of parallelism is lost, and this makes quantum computation impossible. This situation changed radically when the quantum error correcting codes were invented . This fact plus remarkable progress in experimental manipulation with individual qubits make the dream coming true.
In this paper we study a special implementation of so-called noiseless quantum codes, also known as error avoiding quantum codes . Such codes were proposed in as an alternative or, more likely, supplement to the error correcting quantum codes. In error avoiding quantum codes were built using group theoretic methods. The idea is that among quantum states of the system there exist distinguished ones which, despite interaction with the environment, do not underlie decoherence. Important assumption was that qubits of the quantum register interact with a coherent environment. This assumption, besides the assumption on dynamical symmetry of the system, turned out to be essential for the introduction of the states protected against decoherence. Namely, it is possible to introduce collective variables describing the qubits composing the register. The singlet state of the qubits turned out to be protected against corruption.
An attempt to describe a more general situation was made in . There the semigroup technique was used to describe general evolution of the system interacting with environment. In comparison to , the generalization consisted in consideration of various degrees of coherence of the environment on the distances of the order of the length of the register. Besides full coherence, lack of any coherence and partial coherence were considered, too. Basic results on error protected states are similar to the ones obtained in . Additional noises were considered, and it was shown that their influence on the error protected states is negligible up to the first order of a small parameter characterizing the noise. Performing quantum computations with the error protected states was shown to be realistic.
In the paper general criteria for error avoiding quantum codes were formulated. First, the existing codes were divided into three groups: error correcting codes, error avoiding codes and error preventing codes. Error correcting codes detect and correct errors. Error preventing codes only detect errors, without correcting them. After the classification was made the general theory of error avoiding codes was formulated, in the manner following the paper , where the general theory of error correcting quantum codes was presented. It was found that the error avoiding codes are derived from the subspaces of the Hilbert space that are common eigenspaces of the operators $`A_a`$ describing the evolution of the system. If $`\rho _i`$ and $`\rho _f`$ are the initial and the final density matrices of the system respectively, then, under assumption that initially the system is not entagled with the environment, $`\rho _f=_aA_a\rho _iA_a^+`$, where $`_aA_a^+A_a=I`$.
¿From the paper the conditions for the error correcting codes are known. Namely, if the vectors $`|i>`$ form an orthonormal basis of the code, the condition
(1.1)
$$<i|A_a^+A_b|j>=\gamma _{ab}\delta _{ij}$$
should be satisfied, where $`\gamma _{ab}`$ is a Hermitian matrix. All known error correcting codes have $`\gamma _{ab}`$ nondegenerate. The matrix is then expressible, after some unitary redefinition of $`A_a`$, as a diagonal matrix with positive entries.
As shown in error avoiding codes also satisfy (1.1) but are maximally degenerate (in the diagonal form only one diagonal element is different from zero). Moreover the matrix $`\gamma _{ab}`$ is expressible as $`\gamma _{ab}=\gamma _a^{}\gamma _b`$, where $`\gamma _a`$ are eigenvalues of the respective operators $`A_a`$ for the states from the code. This general approach does not use the group theoretic language, so that we do not know if there exist error avoiding codes of different origin than the group theoretic (or quantum group theoretic) one. It shows however usefulness of the error avoiding codes for quantum computing (especially when used simultaneously with the error correcting codes).
The aim of this paper is to study error avoiding codes in a more general framework than the group-theoretic one. More precisely, we shall discuss the problematics of error avoiding codes in the framework of quantum groups.
Our motivation is that the quantum group framework enables introduction of a more general dynamical symmetry of the system, comparing to the standard group theoretic one. By construction, it covers also the dynamical symmetry connected with classical groups, since the classical groups are all special cases of quantum groups. We hope to take this way into account certain fluctuations from the exact group theoretic dynamical symmetries required by the standard noiseless codes.
In there was expressed a hope that deviations from the proposed ideal situation should not destroy too much of the quantum coherence and consequently the error protected states should become ‘almost error protected’ under these conditions. Our considerations show there exist particular perturbations for which the error protected states ‘remain’ exactly protected despite these perturbations.
The plan of the paper is as follows. In the second section we define and briefly discuss the notion of dynamical symmetry connected with quantum groups. In the third section we formulate and prove main theorems concerning error protected states. In the fourth section we present simple examples to illustrate the general results. We conclude the paper with the discussion of possible extensions of this work. In the Appendix we give a very brief review of the basic material on quantum groups and their representations.
## 2. Dynamical Symmetry Coming From Quantum Groups
Symmetry proved to be one of the basic notions in physics. Dynamical symmetry of a physical system is defined in terms of its Hamiltonian, which should be expressible as a linear combination of operators generating a representation of the appropriate Lie algebra. There is a large class of systems possessing such a property. Dynamical symmetry of a system should not necessary be visible at a first sight. Nevertheless, searching for such a symmetry is highly rewarding, since one can apply to the systems with a dynamical symmetry powerful methods developed on the ground of the theory of Lie algebras and their representations, like the method of coherent states . Dynamical symmetry proved to be also important in searching for physical systems possessing very specific quantum states—which can not be corrupted despite their interaction with the environment . Such systems provide noiseless quantum codes that are of potential great interest for constructing quantum computers. Noiseless quantum codes can be either alternative to error correcting codes, which are elaborate methods of coding information, recognizing errors and correcting them , or a valuable supplement to such codes.
It turns out that analogous codes exist for systems with dynamical symmetry based on quantum groups instead of Lie groups. The goal of this section is to define the notion of dynamical symmetry associated to quantum groups. In the next section we shall apply our new concept of dynamical symmetry to prove our main theorems concerning error protected states. Then, we shall study some systems providing noiseless quantum codes.
Basic mathematical concepts and tools that will be used in the paper are briefly presented in the Appendix.
A generalization of the concept of dynamical symmetry can be defined only when there are well established notions of a Lie algebra, and the corresponding universal enveloping algebra, associated to a given quantum group $`G`$. In the theory of quantum groups, all these notions essentially depend on an appropriately chosen differential calculus over $`G`$.
The quantum group $`G`$ will be represented by a non-commutative $`C^{}`$-algebra $`A`$, playing the role of the algebra of ‘continuous functions’ defined on the quantum space $`G`$, together with a coproduct map $`\varphi :AAA`$ (corresponding to the standard product in the case of classical groups). Effectively, all caclulations will be performed within an everywhere dense \*-subalgebra $`𝒜A`$, playing the role of polynomial functions over $`G`$. Actually, $`𝒜`$ is a Hopf \*-algebra in a natural way.
Suppose that on $`G`$ is defined a \*-covariant, left-covariant first-order differential calculus $`\mathrm{\Gamma }`$. Let $`L`$ be the associated quantum Lie algebra. If the calculus $`\mathrm{\Gamma }`$ is in addition right-covariant, we can introduce the universal enveloping algebra $`U(L)`$. Every representation $`v:VV𝒜`$ of $`G`$ in a finite-dimensional vector space $`V`$ naturally induces (as in the classical theory) a representation $`\delta =\delta _v`$ of $`L`$ and $`U(L)`$ in $`V`$.
Definitions of all these objects are sketched in the Appendix.
We consider an open quantum system, represented by a Hilbert state space $`V=H_S`$. The system interacts with its environment (bath) which is described by a Hilbert space $`H_B`$. Here it is assumed for simplicity that all Hilbert state spaces are finite dimensional—however, everything could be incorporated into the infinite-dimensional case.
###### Definition 2.1.
We say that a system has quantum dynamical symmetry described by the quantum group $`G`$ and its quantum Lie algebra $`L`$ if the following conditions are satisfied:
$`(`$i$`)`$ The evolution of the system is governed by the Hamiltonian
$$\text{End}(H_SH_B)\text{End}(H_S)\text{End}(H_B).$$
$`(`$ii$`)`$ The Hamiltonian is a hermitian operator $`^{}=`$.
$`(`$iii$`)`$ The Hamiltonian is of the form:
(2.1)
$$=P_1(l_1,\mathrm{},l_n)T_1+\mathrm{}+P_N(l_1,\mathrm{},l_n)T_N$$
where $`P_1,\mathrm{},P_N`$ are polynomial expressions of infinitesimal generators $`l_i=\delta (e_i)`$ and $`\{e_i\}`$ is a basis in $`L`$. Finally $`T_1,\mathrm{},T_N`$ are hermitian operators
$$T_\alpha :H_BH_B.$$
Such systems with quantum dynamical symmetry can be explored by generalized methods known from the theory of systems with classical dynamical symmetry, for example by the method of quantum coherent states . Let us observe that the terms in (2.1) can be reorganized in such a way, that the Hamiltonian takes a more familiar form:
(2.2)
$$=_S\mathrm{id}+\mathrm{id}_B+_I$$
where $`_S`$ is the system’s Hamiltonian, $`_B`$ is the Hamiltonian of the environment and $`_I`$ is the ‘interaction hamiltonian’ uniquely defined as the part of $``$ traceless in both tensor factors.
## 3. Error Protected States $`\&`$ Noiseless Quantum Codes
In this section we present our main theorems on error protected states, and on noiseless quantum codes. We assume that we deal with a (open) quantum system with dynamical symmetry of a quantum group, and all other features as described in the previous section. The vectors that are $`v`$-invariant, where $`v`$ is a representation of the quantum group $`G`$, are of vital importance for our further discussion. Let us give their definition now. Let $`v:VV𝒜`$ be an arbitrary representation of $`G`$ in a finite-dimensional vector space $`V`$, and let $`\delta =\delta _v:L\text{End}(V)`$ be the associated representation of $`L`$. To further simplify the considerations, we shall consider the case when the quantum group is ‘connected’ in the sense that only scalar elements of $`𝒜`$ are annihilated by the differential $`d:𝒜\mathrm{\Gamma }`$.
Then the following equivalence holds for every vector $`\psi V`$
$$v(\psi )=\psi 1\delta (x)\psi =0,xL$$
Let us assume that the calculus $`\mathrm{\Gamma }`$ is in addition bicovariant. This enables us to introduce the quantum universal enveloping algebra $`U(L)`$, and to discuss the representations of $`U(L)`$ associated with the representations of the quantum group $`G`$. Let us also introduce the map $`\chi :U(L)`$, with the properties $`\chi (L)=0`$, $`\chi (1)=1`$, extended then to the whole $`U(L)`$ by multiplicativity. The representation $`\delta `$ uniquely (as in the standard theory) extends from $`L`$ to $`U(L)`$. The above two conditions are further equivalent to
$$\delta (q)\psi =\chi (q)\psi ,qU(L).$$
The proof of these equivalences is quite straightforward, but it needs some additional definitions and constructions, which we would rather skip in this paper as they are going too far in the formalism. Vectors satisfying any of the above conditions are called v-invariant. The $`v`$-invariant vectors are very important for the study of quantum registers (which are open systems with a quantum dynamical symmetry). Such vectors give us examples of the error protected states.
Our main theorem reads:
###### Theorem 3.1.
Unitary evolution described by the Hamiltonian $``$ possessing a quantum dynamical symmetry given by $`(G,L)`$ preserves the $`v`$-invariant vectors and associated states of the system, even when the other states of the system are corrupted due to decoherence.
###### Proof.
Let us take as an initial vector $`\psi \zeta H_SH_B`$, where $`\psi `$ is $`v`$-preserved in the sense defined above. Then the unitary evolution defined by
$$U(t)=\mathrm{exp}(\frac{i}{\mathrm{}}t)$$
with $``$ of the form (2.2) gives
$$\mathrm{exp}(\frac{i}{\mathrm{}}t)(\psi \zeta )=\psi \mathrm{exp}(\frac{i}{\mathrm{}}_{\mathrm{eff}}t)\zeta $$
where
$$_{\mathrm{eff}}=\chi (P_1)T_1+\mathrm{}+\chi (P_N)T_N.$$
This proves the statement. ∎
Interesting property of $`_{\mathrm{eff}}`$ is that the coefficients $`\chi (P_i)`$ should somehow reflect the structure given by $`G`$ and its Lie algebra $`L`$.
Now we can easily prove generalization of the theorems 1 and 2 given in the paper . We follow the notation from . Let $`\rho _S\text{End}(H_S)`$ and $`\rho _B\text{End}(H_B)`$ be the (mixed quantum) states of the system and the environment respectively. If the overall system is initially in the state $`\rho (0)=\rho _S\rho _B`$, then $`\rho (t)=U(t)\rho (0)U(t)^+`$, so that the evolution is unitary. The induced evolution on $`H_S`$ is given exactly like in by $`L_t:\rho _S\text{tr}_B\rho (t)`$, where $`\text{tr}_B`$ is the trace over $`H_B`$. Then the following theorem is fulfilled:
###### Theorem 3.2.
Let $`_N`$ be the manifold of states built over the space of vectors invariant under the representation $`v`$, and $`\rho _S_N`$. Then for any initial bath state $`\rho _B`$ the induced evolution on $`H_S`$ is trivial,
$$L_t[\rho _S(t)]=\rho _S,t>0.$$
###### Proof.
Theorem 3.1 allows us to reduce the proof of 3.2 to the proof of the first theorem of . ∎
The invariant vectors are generalizations of the singlet states pointed out in as the states of the quantum register which are not corrupted by interaction with the environment.
We should stress that the Hamiltonian of the system + environment should not necessary contain terms with trivial representation in the space of the system and in the space of the environment, so that it can be even of the pure interaction form.
Before we present simple examples illustrating the general theory and explicitly showing the ‘error protected’ states, let us discuss interesting question of the structure of the Hilbert space of the quantum computer registers, and discuss physical implications. The register usually consists of a number of copies of the same quantum system, often having two possible states for example spin ‘up’ and spin ‘down’ (a qubit).
Dynamical symmetry is defined in the Hilbert space that originates from the Hilbert spaces for individual qubits being described as carrier spaces of unitary representations
$$v_i:V_iV_i𝒜i=1,\mathrm{},n$$
of our quantum group $`G`$.
The register Hilbert space is the tensor product of the representation spaces,
$$V=V_1V_2\mathrm{}V_n$$
in which $`G`$ acts by the direct product
$$v=v_1\times v_2\times \mathrm{}\times v_n$$
of representations $`v_i`$. Each of the representations $`v_i`$ induces a representation $`\delta _i`$ of the corresponding quantum universal enveloping algebra. The representation $`v`$ induces the representation $`\delta `$ of the quantum universal enveloping algebra, and one can easily prove the following relation:
(3.1)
$$\delta (x)(\varphi _1\mathrm{}\varphi _n)=\underset{k=1}{\overset{n}{}}\underset{\alpha I[k]}{}\varphi _1\mathrm{}\delta _k(x^\alpha )\varphi _k\eta _{k+1}^\alpha \mathrm{}\eta _n^\alpha $$
where
$$\tau _{nk}\left(\left\{\varphi _{k+1}\mathrm{}\varphi _n\right\}x\right)=\underset{\alpha I[k]}{}x^\alpha \left\{\eta _{k+1}^\alpha \mathrm{}\eta _n^\alpha \right\}$$
and $`\tau _{nk}:V_{k+1}\mathrm{}V_nLLV_{k+1}\mathrm{}V_n`$ are the appropriate ‘flip-over’ operators naturally associated to the differential calculus.
The formula (3.1) differs from the corresponding formula for the classical case of addition angular momenta in quantum mechanics ($`\tau `$ is in the classical case just the standard transposition). It is easy to see that qubits in the register are not treated on the same footing. It could be associated to some effects due to, not taken into account in , linear extension of the register, or to fluctuations of the fields due to nonideal structure of boundaries of the register and their influence. Anyway, it is possible to realize a system with weaker symmetry than the one presented in but still possessing error protected states. It is known that similar deviations from exact dynamical symmetry of Lie groups lead to better mass (or energy) formulas in both nuclear/particle physics, and molecular physics . Therefore, one can look also among such systems for possible candidates for registers of quantum computers.
In a physically plausible conjecture was expressed, that small deviations from ideal properties assumed of the system should lead to small errors in the error-protected states. Actually, we have shown that there exist systems with special kind of deviation from the assumed symmetry, which nevertheless still have error protected states.
## 4. Examples
Let us switch now to some simple examples that would highlight our general ideas. The first example of a quantum group presented systematically in the literature was a quantum version of the standard $`SU(2)`$ group , where the theory of representations was developed together with various geometrical aspects and a construction of a natural three-dimensional left-covariant differential calculus. This calculus is not bicovariant, and the minimal dimension for a bicovariant calculus over the quantum $`SU(2)`$ is 4 (this four-dimensional calculus is analyzed in detail in ). Generalization of the results concerning this particular quantum group leads to the general theory of compact matrix quantum groups , the definition of quantum spheres and their geometry , deep generalization of the Tannaka-Krein duality , and also the theory of quantum principal bundles together with the corresponding gauge theory on quantum spaces, first formulated in and then developed systematically in (see also ). Also in the $`C^{}`$-algebraic framework the quantum homogeneous bundles were defined and the example of such a bundle with quantum spheres as fibers was given . Different approaches to quantum groups were developed in , where quantum groups are treated from the point of view of deformations of universal enveloping algebras, Yang-Baxter equations and completely solvable systems.
We shall use the quantum version of $`SU(2)`$ in our examples. This is relatively simple from computational viewpoint, but highly non-trivial and very suggestive for the aims of this paper. First, we remind some basic facts about this group, which is denoted by $`S_\mu U(2)`$. Here the deformation parameter $`\mu `$ takes the values $`\mu [1,1]\{0\}`$, and $`\mu =1`$ corresponds to the classical $`SU(2)`$ group.
In our further considerations important role will play the fact that irreducible unitary representations of $`S_\mu U(2)`$ are classified by the half-integers, like the representations of $`SU(2)`$. The fundamental representation corresponds, as in the classical case, to spin $`j=\frac{1}{2}`$ (see Appendix for more details). The Clebsch-Gordan decompositions of tensor products of the representations of the $`S_\mu U(2)`$ into irreducible representations look similar (concerning the multiplicities of the appearance of irreducible components in the products of representations) as in the classical case:
$$\stackrel{k}{\stackrel{}{u\times \mathrm{}\times u}}=\underset{jJ}{}n_{j,k}u_j$$
with the numbers $`n_{j,k}`$ the same as in the classical case. In particular, the decomposition of the second tensor power of the fundamental representation is $`u_{1/2}^2=u_0u_1`$, where $`u_0`$ and $`u_1`$ are the $`1`$-dimensional and the $`3`$-dimensional irreducible representations, respectively. One can describe these representations more explicitly after introducing the orthonormal basis in the representation space $`V=^2`$ of $`u_{1/2}`$, which will be denoted $`|+`$, $`|`$ for the purpose of being easily recognizable by physicists. The tensor square $`u_{1/2}^2`$ is realized in $`VV^4`$, and the orthonormal basis in this space is $`|+|+`$, $`|+|`$, $`||+`$, $`||`$. It is an easy exercise to find that the invariant subspaces of $`u_{1/2}^2`$ are spanned by:
(4.1)
$$\frac{1}{\sqrt{1+\mu ^2}}(|+|\mu ||+)$$
and
(4.2)
$$\begin{array}{c}|+|+\\ \frac{\mu }{\sqrt{1+\mu ^2}}(|+|+\frac{1}{\mu }||+)\\ ||\end{array}$$
The formula (4.1) generalizes the standard singlet, and the formula (4.2) generalizes the standard triplet. In analogy to the classical case the even tensor powers of the fundamental representation decompose into irreducible representations in such a way that the one-dimensional representation appears a number of times, and the number is identical as in the classical case. These singlets will be preserved by the dynamics.
### 4.1. First Example
In the first example we treat a system which is very similar to the one considered in . Namely, as a model of the environment (bath) we consider a system of harmonic oscillators, described by the Hamiltonian
$$_B=\underset{k}{}\omega _kb_k^{}b_k,$$
acting in the Hilbert space $`H_B`$. The register consists in this simplest case of two qubits. In contrast to the case considered by Zanardi and Rasetti , the system consisting of the register and the bath has the dynamical symmetry not of the classical but of the quantum $`SU(2)`$ group. As already mentioned, in the quantum group context it is necessary to chose the differential calculus, prior to establishing the notion of the dynamical symmetry. The closest to the classical case seems to be introduction of the $`3D`$ left-covariant calculus . In other words, the quantum Lie algebra $`L`$ is $`3`$-dimensional. Let us denote by $`K_i`$ the operators representing the basis vectors $`l_i`$, in an arbitrary representation of $`L`$ (here $`i\{1,2,3\}`$). The following recurrent formulas enable us to compute explicitly the operators $`K_i`$, in the arbitrary tensor product of elementary $`2`$-dimensional representations (qubits):
(4.3) $`K_3(\psi |+)`$ $`={\displaystyle \frac{1}{2}}\psi |++{\displaystyle \frac{1}{\mu ^2}}K_3(\psi )|+`$
(4.4) $`K_3(\psi |)`$ $`=\mu ^2K_3(\psi )|{\displaystyle \frac{1}{2}}\psi |`$
(4.5) $`K_j(\psi |+)`$ $`={\displaystyle \frac{1}{2}}\psi |++{\displaystyle \frac{1}{\mu }}K_j(\psi )|+`$
(4.6) $`K_j(\psi |)`$ $`=\mu K_j(\psi )|{\displaystyle \frac{1}{2}}\psi |,`$
where $`j\{1,2\}`$.
In such a case the bath-register interaction Hamiltonian which is the quantum group analog of the Hamiltonian used in reads:
$$_I=K_+T+K_{}T^{}+K_3T^{},$$
where $`K_\pm =K_1\pm iK_2`$, and $`T,T^{}`$ are operators acting in the bath Hilbert state-space. Relating to the corresponding formulas in , the operators $`T`$ and $`T^{}`$ are obtained as the appropriate linear combinations of the creation and annihilation operators describing relevant elementary excitations of the bath. The operators $`K_j`$ are acting in the $`4`$-dimensional 2-qubit space. In other words, it is formally of the same shape as in , however the ‘spin’ operators are different as explained above. The singlet state of the register is error-protected in the sense discussed above.
### 4.2. Second Example
In the second example the only difference with the first example is the register consists now of any even number of qubits, instead of just two. The spin operators $`K_j`$ are referring to the total register system, and are calculated by applying the above listed inductive rules (4.6).
It is important to stress that the number of singlet states is just the same as in the classical $`SU(2)`$ case. This is a consequence of the mentioned similarity between the representation theories for quantum and classical $`SU(2)`$ groups. The dimension of the singlet state space depends on the number of qubits in the way described in . All these states are clearly protected from corruption due to decoherence.
### 4.3. Remark
The group $`SU(2)`$ appears as a dynamical symmetry group mainly in the context of the dynamics of spin systems. It appears less frequently in the context of dynamics of different systems, like bosonic particles. Therefore, the examples usually begin with the fundamental representation $`u_{1/2}`$ of $`SU(2)`$, as the elementary building blocks of the Hilbert space of the system. However, such a fundamental block could be any of $`u_j`$ with $`j`$ half-integer. Since for example $`u_j\times u_j`$ contains the singlet $`u_0`$ in its splitting into irreducible representations. this could be a starting point for building error protected states with the help of bosonic particles. Similar considerations are true also for the quantum group $`S_\mu U(2)`$. It seems, though, that a physical realization of such systems is more complicated and creates more technical problems.
Let us stress that in the examples we considered following the paper , all qubits are coupled to the same, coherent, environment. As was stressed in coupling to the same environment of the qubits gives more possibility to get error protected states than coupling to independent environments. However, our methods seem to be general enough to deal with the cases of coupling with independent environments as well, till the system has a dynamical symmetry of the type introduced in this paper.
## 5. Conclusions
In this paper we introduced the general notion of dynamical symmetry associated with quantum groups and Lie algebras. Then we applied this notion to construct error protected states for open systems with such symmetry. The states can be useful for quantum computation. They are close analogs of their standard group theoretical counterparts. As a result, the error protected states obtained in strictly group theoretical dynamical symmetry context have counterparts preserved when the symmetry is slightly deviated towards the quantum group theoretic one. Recently various authors introduced a technique of quantum computation which dynamically eliminates errors, by a quantum counterpart of the classical so called ‘bang-bang’ technique. Zanardi has shown that the technique called ‘symmetrizing’ can be interpreted group theoretically as control of the systems forcing the systems with dynamical symmetry of a Lie group to be in states which are error protected. This very interesting observation should increase interest in error avoiding quantum codes. Since mathematically the technique seems to rely on invariant measures on the groups, it is applicable not only to the systems with the dynamical symmetry of finite groups discussed in the paper, but also to the systems with dynamical symmetry of the compact (even locally compact) Lie groups, which all possess Haar measure necessary for such construction. One should observe that the same is also true for compact (or locally compact) quantum groups, since these objects possess the Haar measure, too. It seems the generalization of the results by Zanardi to the quantum group case is straightforward, but its physical interpretations are less clear. Work on this issue is in progress.
## Acknowledgements
M. D urd evich likes to thank Theoretical Division of Los Alamos National Laboratory for the hospitality, H.E. Makaruk and R. Owczarek like to acknowledge the hospitality of the Math Institute of National Autonomous University of Mexico. Realization of this research was partially supported by Investigation Project in106879 of DGAPA/UNAM
## Appendix A Quantum Spaces and Quantum Groups
Classical theorem by Gelfand and Naimark states that compact topological spaces are in a natural correspondence with commutative unital $`C^{}`$-algebras. These $`C^{}`$-algebras consist of continuous complex-valued functions on the corresponding spaces.
Let $`X`$ be a compact topological space and $`C(X)`$ be the associated algebra of continuous complex-valued functions on $`X`$. The linear structure on $`C(X)`$ is given by the obvious conditions: $`(f+g)(x)=f(x)+g(x)`$, and $`(\alpha f)(x)=\alpha f(x)`$. The product in the algebra is $`(fg)(x)=f(x)g(x)`$ and the \*-involution is given by $`f^{}(x)=\overline{f(x)}`$.
There is a natural norm in $`C(X)`$ given by
$$||f||=sup\{|f(x)|:xX\}.$$
In such a way is introduced a structure of (commutative) $`C^{}`$-algebra in $`C(X)`$. Conversely, every commutative unital $`C^{}`$-algebra is of this form—according to classical Gelfand-Naimark theory. Actually, the Gelfand-Naimark theory can be generalized to the level of locally-compact spaces, giving us a correspondence between arbitrary commutative (not necessarily unital) $`C^{}`$-algebras and locally compact topological spaces. This correspondence is functorial, in the sense of category theory.
These facts lead to a generalized concept of space, which is understood as ‘the underlying space’ of a general $`C^{}`$-algebra, about which we no longer assume it should be commutative. Generalized spaces of this type are called quantum spaces. The reason for the adjective ‘quantum’ follows from the observation that as in the classical quantization scheme a commutative algebra of functions is substituted by a noncommutative algebra of operators acting in a Hilbert space. The latter is indeed the case since all $`C^{}`$-algebras can be realized as algebras of operators acting in some Hilbert spaces.
Interesting algebra and geometry appears when the classical topological spaces are equipped with an additional structure: differential-geometric, metric, Lie group, and so on. A very important class of quantum spaces constitute the quantum groups, which are understandable as quantum spaces equipped with a group structure.
Let us explain now, very briefly, what is exactly a compact quantum group. Let us start from a classical compact topological group $`G`$. This means that $`G`$ is a compact topological space equiped with a group structure, such that the product map $`:G\times GG`$ is continuous (it can be shown that in the compact case continuity of the product implies continuity of the inverse map). At the dual level, the product map is represented by a \*-homomorphism $`\varphi :AAA`$, where $`A=C(G)`$.
More precisely, we first naturally identify
$$\stackrel{k}{\stackrel{}{A\mathrm{}A}}=C(\stackrel{k}{\stackrel{}{G\times \mathrm{}\times G}})k2$$
and define
$$\varphi (f)(g_1,g_2)=f(g_1g_2),fAg_1,g_2G.$$
The associativity property of the product is equivalent to the coassociativity property
$$(\varphi \mathrm{id})\varphi =(\mathrm{id}\varphi )\varphi .$$
It can be shown that the remaining two group axioms (the existence of the neutral element and the existence of the inverse elements) are equivalent to a single assumption that the elements of the form $`a\varphi (b)`$ as well as of the form $`\varphi (b)a`$, where $`a,bA`$, span two everywhere dense linear subspaces of $`AA`$.
Generalizing this to the quantum level, we define a group structure on a quantum space $`G`$ as a \*-homomorhism $`\varphi :AAA`$ such that the diagram
$$\begin{array}{ccc}A& \stackrel{\varphi }{}& AA\\ \varphi & & \mathrm{id}\varphi & & \\ AA& \underset{\varphi \mathrm{id}}{}& AAA\end{array}$$
is commutative, and such that
$`AA`$ $`=\overline{\left\{{\displaystyle a\varphi (b)}\right|a,bA\}}`$
$`AA`$ $`=\overline{\left\{{\displaystyle \varphi (b)a}\right|a,bA\}}.`$
where the bar means appropriate closure.
As a very important special case of this structure, let us mention matrix groups. These structures are given by triplets $`(A,\varphi ,u)`$ consisting of a $`C^{}`$-algebra $`A`$, a \*-homomorphism $`\varphi :AAA`$ and a matrix $`uM_n(A)`$ (all $`n\times n`$-matrices with coefficients from $`A`$) which is (together with the conjugate matrix $`\overline{u}`$) invertible in $`M_n(A)`$ and such that
$`(`$i$`)`$ The \*-algebra $`𝒜`$ generated by the entries $`u_{ij}`$ is everywhere dense in $`A`$;
$`(`$ii$`)`$ The following identity holds:
$$\varphi (u_{ij})=\underset{k}{}u_{ik}u_{kj}.$$
In this case we have the inclusion
$$\varphi (𝒜)𝒜_{\mathrm{alg}}𝒜.$$
Let us stress that the above mentioned coassociativity and density properties are satisfied automatically.
Matrix groups generalize compact Lie groups (if $`A`$ is commutative the theory reduces to standard compact matrix groups).
The algebra $`𝒜`$ plays the role of the algebra of polynomial functions over $`G`$. The matrix $`uM_n(A)`$ correspond to the fundamental representation of the group $`G`$.
## Appendix B Differential calculus on quantum groups, Quantum Lie Algebras, Quantum universal envelopes
### B.1. Quantum Lie Algebras
There is a very important notion of a differential structure defined for quantum groups. The definitions of a quantum Lie algebra and of a quantum universal enveloping algebra depend on the introduced differential calculus. Therefore, we begin from giving the definition of the differential calculus.
First-order differential calculi are defined as certain bimodules $`\mathrm{\Gamma }`$ over $`𝒜`$, equipped with a differential $`d:𝒜\mathrm{\Gamma }`$. The space $`\mathrm{\Gamma }`$ is a noncommutative counterpart of the usual module of $`1`$-forms over a classical group, and $`d`$ generalizes the standard differential of functions.
It is important to mention that there is not a unique prescription to construct a differential calculus over a quantum group, and generally a given quantum group will possess a variety of non-equivalent calculi, each of them having a potential significance. It is surprising that the same situation appears in classical theory, where one can also use the methods of quantum groups to construct new differential calculi over the standard Lie groups. This opens interesting new possibilities in the study of classical Lie groups. In particular, it opens a possibility to extend the notion of dynamical symmetry, in the framework of classical groups.
In the quantum group theory a special role is played by so-called left-covariant and bicovariant differential calculi. In these cases we can introduce the analogs of left and left/right actions of the group $`G`$ on $`\mathrm{\Gamma }`$. If the module $`\mathrm{\Gamma }`$ is left-covariant, then we can define its subspace $`\mathrm{\Gamma }_{\text{inv}}`$, consisting of left-invariant ‘$`1`$-forms’. Quantum Lie algebra is then defined as the corresponding dual space, in other words $`L=\mathrm{\Gamma }_{\text{inv}}^{}`$.
If the calculus is bicovariant, then we can introduce a natural braid operator $`\sigma :LLLL`$, generalizing the classical transposition. Furthermore, in analogy with classical theory, we can define a quantum Lie bracket in the space $`L`$ generalizing the classical Lie bracket . The Lie bracket is defined as a linear operator $`C:LLL`$, and we can equivalently write $`[x,y]=C(xy)`$. This bracket satisfies the appropriate generalized Jacobi identity and braided-antisymmetricity conditions.
Following the classical theory, the quantum universal enveloping algebra for $`(L,[,])`$ is defined as a unital associative algebra $`U(L)`$ generated by relations
(B.1)
$$xy\underset{i}{}y_ix_i=[x,y],$$
where $`x,yL`$ and $`_iy_ix_i=\sigma (xy)`$.
### B.2. Representations of Quantum Groups and Quantum Lie Algebras
Having the Lie bracket and using (B.1) one can define representations of quantum Lie algebras and of the corresponding quantum universal enveloping algebras. It can be shown that every representation $`v`$ of $`G`$ in a finite-dimensional vector space $`V`$ naturally gives rise to a representation $`S:U(L)\text{End}(V)`$ of the quantum universal enveloping algebra. Namely, let $`v:VV𝒜`$ be a (left) representation of the quantum group $`G`$ in a finite dimensional complex vector space $`V`$, in other words $`v`$ is linear, satisfies the condition
$$(\mathrm{id}\varphi )v=(v\mathrm{id})v$$
and $`v`$ is invertible, understood as an element of $`\mathrm{End}(V)𝒜`$. This corresponds to the classical requirements on representations of groups saying that the product of group elements is represented by composition of operators representing these elements, and the neutral element of a group is represented by the identity operator.
Every representation $`v`$ of $`G`$ in $`V`$ naturally generates a representation
$$\delta =\delta _v:U(L)\text{End}(V)$$
of $`U(L)`$ in $`V`$ (if the differential calculus is bicovariant) or only of the Lie algebra $`L`$, $`\delta :L\text{End}(V)`$ (if the differential calculus is left-covariant).
Moreover, if the differential calculus is \*-covariant, which means that in the module $`\mathrm{\Gamma }`$ of 1-forms is defined the \*-operation $`{}_{}{}^{}:\mathrm{\Gamma }\mathrm{\Gamma }`$ induced by $``$ in $`𝒜`$, it makes sense to speak about hermiticity of the representation $`\delta `$. Namely, the -operation on $`\mathrm{\Gamma }`$ naturally induces the -structure on the quantum Lie algebra $`L`$, via the formula $`<f^{},\psi >=<f,\psi ^{}>`$ where $`fL=\mathrm{\Gamma }_{\text{inv}}^{}`$ and $`\psi \mathrm{\Gamma }_{\text{inv}}`$.
### B.3. Quantum $`SU(2)`$ group
This quantum group is based on a $`C^{}`$-algebra $`A`$ generated by elements $`\{\alpha ,\alpha ^{},\gamma ,\gamma ^{}\}`$ satisfying the following relations:
$$\alpha \alpha ^{}+\mu ^2\gamma ^{}\gamma =1\alpha ^{}\alpha +\gamma ^{}\gamma =1$$
$$\gamma ^{}\gamma =\gamma \gamma ^{}$$
$$\alpha \gamma =\mu \gamma \alpha \alpha \gamma ^{}=\mu \gamma ^{}\alpha ,$$
where $`\mu [1,1]\{0\}`$. The comultiplication $`\varphi :AAA`$ is given by
$$\varphi (\alpha )=\alpha \alpha \mu \gamma ^{}\gamma \varphi (\alpha ^{})=\alpha ^{}\alpha ^{}\mu \gamma \gamma ^{}$$
$$\varphi (\gamma )=\gamma \alpha +\alpha ^{}\gamma \varphi (\gamma ^{})=\gamma ^{}\alpha ^{}+\alpha \gamma ^{}$$
The theory of representations of $`S_\mu U(2)`$ is very interesting from the point of view of our examples. This theory has many similarities to its classical counterpart–the theory of representations of $`SU(2)`$. Classical $`SU(2)`$ is obtained as a special case $`\mu =1`$.
The fundamental representation of $`S_\mu U(2)`$ is defined by the matrix
$$u=\left(\begin{array}{cc}\alpha & \mu \gamma ^{}\\ \gamma & \alpha ^{}\end{array}\right).$$
It is easy to see that the defining relations for $`S_\mu U(2)`$ are equivalent to the unitarity property
$$u^{}u=uu^{}=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right).$$
The fundamental representation enables us to build all other representations by using direct sums, tensor products and reduction procedures. Irreducible representations $`u_j`$ are numbered by half-integers $`j`$, and are $`2j+1`$-dimensional. Every representation of an arbitrary compact quantum group can be decomposed into irreducible ones. |
warning/0003/astro-ph0003259.html | ar5iv | text | # Measuring the galaxy power spectrum and scale-scale correlations with multiresolution-decomposed covariance – I. method
## 1 Introduction
Measuring the galaxy power spectrum has been and is being a central subject of the large scale structure study. Although the power spectrum is only a second order statistical measure of the deviations of a random field, $`\delta (𝐱)`$, of mass density from homogeneity, it directly reflects the physical scales of the processes that affect structure formation. Mathematically, the positive definiteness of the power spectrum is useful for constraining the parameter space in comparing predictions with data. Since the ongoing and upcoming redshift surveys of galaxies will provide data of galaxy distribution with highly improved quality and a larger quantity, it also requests to develop the methods of measuring the power spectrum more precise and computationally efficient.
Different methods of the power spectrum measurements adopt different representations, or decomposition of the covariance $`Cov=\delta (𝐱)\delta (𝐱^{})`$, where $`\mathrm{}`$ stands for an ensemble average. For a representation given by a set of basis functions $`\psi _i(𝐱)`$ (sometimes referred as weight function), the random field is described by the variables
$$X_i=\delta (𝐱)\psi _i(𝐱)𝑑𝐱,$$
(1)
and the covariance is given by $`Cov_{ij}=X_iX_j`$. If the covariance in this representation is exactly or approximately diagonalized, the diagonal elements $`|X_i|^2`$ would be a fair estimate of the power spectrum, or band-power spectrum. Thus, measuring power spectrum mathematically is almost a synonym of diagonalizing the covariance of the density field $`\delta (𝐱)`$, or calculating the eigenvalues of the covariance matrix.
Traditionally, the Fourier decomposition, and then, the Fourier power spectrum are the popular tool to analyze a cosmic density field, because the Fourier transform retains the translation invariance of a homogeneous and isotropic universe. However, the observed sample given by redshift surveys are not translation invariant due to the selection effect and irregular geometry of the surveys. To effectively compare the predicted power spectrum with the observed galaxy distributions, the basis functions of the decomposition should be chosen to incorporate with the selection effect, sampling, and complex geometry of the data. As a result, various decompositions for measuring the galaxy power spectrum have been proposed (Tegmark, et al. 1998 and reference therein). An ideal estimator of the power spectrum should match the following conditions
* $`X_i`$s are independent from each other, i.e. the data is decomposed into mutually exclusive chunks;
* $`X_i`$s retains all the information of the original data, i.e. the decomposed chunks are collectively exhaustive;
* It is computationally feasible;
* It allows us to take account of the systematic effects, such as redshift distortion, evolution, morphology-dependence, galactic extinction etc.
These ideal estimators are believed to be information lossless, i.e. retaining all information of the power spectrum in the original data.
We will study, in this paper, the estimator based on the multiscale decomposition, i.e. the discrete wavelet transform (DWT) representation. The DWT power spectrum estimator has been applied to measure the power spectrum from samples of the Ly-$`\alpha `$ forests of QSO’s absorption spectra (Pando & Fang 1998a.) The result has demonstrated that the DWT power spectrum estimator can match the conditions listed above, especially it is very helpful to overcome the difficulties of complex geometry and sampling. Within the framework of DWT, this paper will present a general working scheme for extracting the statistical characters from the observational data, in which the selection effect, sampling and binning are addressed.
It has been recognized recently that the non-Gaussian behavior of $`X_i`$ is substantial for a precise measurement of the power spectrum. The accuracy of a power spectrum estimation is significantly affected by the so-called power spectrum correlations induced by non-linear clustering (Meiksin & White, 1998, Scoccimarro, Zaldarriaga & Hui 1999). The power spectrum correlation is also found to be essential for recovering the initial power spectrum by a Gaussianization of observed distribution (Weinberg 1992, Narayanan & Weinberg 1998, Feng & Fang 1999). Thus, beyond the conditions mentioned above for an ideal power spectrum estimator, one should add one more requirement that the power spectrum correlation caused by the non-linear clustering and Poisson sampling are calculable. We will show that the power spectrum correlations, or the scale-scale correlations, can be calculated in the DWT analysis.
Moreover, for popular models of the cold dark matter cosmogony, including the standard cold dark matter models (SCDM), open CDM model (OCDM), and flat CDM (LCDM), the scale-scale correlations have been found to be negligible on large scales, and the non-local scale-scale correlations are also negligible even on small scales (Fang, Deng & Fang 2000). That is, the effect of the power spectrum correlations is largely suppressed in the DWT representation. We will show how to take the advantage of this suppression for a scale-by-scale approach of measuring the power spectrum.
The paper will be organized as follows. §2 gives a brief description of the DWT decomposition of the covariance of density random field. The physical meaning and mathematical properties of the $`j`$ diagonal and $`j`$ off-diagonal components of the covariance will also be discussed. In §3, an optimized band power spectrum estimator based on the DWT $`j`$ diagonal covariance is proposed. In addition, the scale-scale correlation extracting from the $`j`$ off-diagonal components of the covariance is investigated. This correlation gives the scale range in which the power spectrum obtained by the $`j`$ diagonalization are information lossless. We then present the algorithm for estimating the DWT band power spectrum from observed galaxy catalog. It includes the DWT binning (§4), and the DWT technique of dealing with Poisson sampling and selection (§5). The discussions and conclusions are given in §6. A brief introduction of the DWT analysis is given in Appendix.
## 2 Covariance of density fluctuations in the DWT representation
### 2.1 DWT decomposition of density fields
For the sake of simplicity, we analyze a 1-D density distribution $`\rho (x)`$ in the range $`0<x<L`$, which is assumed to be a stationary random field. The density contrast is defined by $`\delta (x)=(\rho (x)\overline{\rho })/\overline{\rho }`$, where $`\overline{\rho }=\rho (x)`$, and $`\mathrm{}`$ stands for ensemble average. It would be straightforward to extend the most results to 2-D and 3-D. Some specific problems related with higher dimension extension will be discussed in §6. In addition, the redshift distortion will not be taken into account in this paper.
To ensure a multiscale decomposition of $`\delta (x)`$ to be information-lossless, the natural working scheme is to adopt discrete wavelet transformation (DWT) within the framework of multiresolution analysis (MRA). The mathematical construction of MRA theory is briefly sketched in Appendix A.
Let $`\delta ^P(x)`$ be the periodic extension of $`\delta (x)`$, i.e., $`\delta ^P(x)=\delta (x[x/L]L)`$, where $`[\eta ]`$ denotes integer part of $`\eta `$. From eq.(A36), the density contrast $`\delta ^P(x)`$ can be decomposed in term of orthonormal wavelet basis
$$\delta ^P(x)=\underset{j=0}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{+\mathrm{}}{}}\stackrel{~}{ϵ}_{j,l}\psi _{j,l}(x),$$
(2)
The wavelet function coefficient (WFC), $`\stackrel{~}{ϵ}_{j,l}`$, is given by the inner product of
$$\stackrel{~}{ϵ}_{j,l}=\psi _{j,l}|\delta _{\mathrm{}}^{\mathrm{}}\delta ^P(x)\psi _{j,l}(x)𝑑x.$$
(3)
which describes the density fluctuation on scale $`L/2^j`$ at position $`lL/2^j`$. The WFCs are the variables of the random field in the DWT representation. The original distributions can be exactly and unredundantly reconstructed from these decomposed variables.
By using the periodized wavelet function defined by
$$\psi _{j,l}^P(x)=\left(\frac{2^j}{L}\right)^{1/2}\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\psi [2^j(\frac{x}{L}+n)l].$$
(4)
where $`\psi `$ is the basic wavelet function \[eq.(A21)\], eq.(1) becomes
$$\delta ^P(x)=\underset{j=0}{\overset{\mathrm{}}{}}\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}\psi _{j,l}^P(x),$$
(5)
The WFC can then be computed by
$$\stackrel{~}{ϵ}_{j,l}^P=_0^L\delta ^P(x)\psi _{j,l}^P(x)𝑑x$$
(6)
We will always use the periodized functions below, and drop the superscript $`P`$.
Furthermore, $`\psi _{j,l}(x)`$ is admissible \[eq.(A27)\], which implies that $`\psi _{j,l}(x)`$ has zero mean if it is integrable,
$$\psi _{j,l}(x)𝑑x=0.$$
(7)
It then follows from eq.(2) that
$$\stackrel{~}{ϵ}_{j,l}=0$$
(8)
The Fourier decomposition of the field $`\delta (x)`$ is given by
$$\delta (x)=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\delta _ne^{i2\pi nx/L},$$
(9)
where $`n`$ is an integer, and the Fourier coefficients, $`\delta _n`$, is
$$\delta _n=n|\delta \frac{1}{L}_0^L\delta (x)e^{i2\pi nx/L}𝑑x,$$
(10)
Since both the bases of the Fourier transform and the DWT are orthogonal and complete in the space of 1-D functions with period length $`L`$, we have
$$\underset{j=0}{\overset{\mathrm{}}{}}\underset{l=0}{\overset{2^j1}{}}n|\psi _{j,l}\psi _{j,l}|n^{}=\delta _{n,n^{}}^K$$
(11)
where $`\delta _{n,n^{}}^K`$ is the Kronecker Delta function, and $`n|\psi _{j,l}`$ the Fourier transform of the wavelet $`\psi _{j,l}`$ given by
$$\widehat{\psi }_{j,l}(n)n|\psi _{j,l}=_0^L\psi _{j,l}(x)e^{i2\pi nx/L}𝑑x.$$
(12)
Considering the wavelet $`\psi _{j,l}(x)`$ is related to the basic wavelet $`\psi (\eta )`$ by eq.(A11), eq.(12) can be rewritten as
$$\widehat{\psi }_{j,l}(n)=\left(\frac{2^j}{L}\right)^{1/2}\widehat{\psi }(n/2^j)e^{i2\pi nl/2^j},$$
(13)
where $`\widehat{\psi }(n)`$ is the Fourier transform of the basic wavelet
$$\widehat{\psi }(n)=_0^L\psi (\eta )e^{i2\pi n\eta }𝑑\eta .$$
(14)
Substituting expansion (9) into eq.(6) yields
$$\stackrel{~}{ϵ}_{j,l}=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\delta _n_0^Le^{i2\pi nx/L}\psi _{j,l}(x)𝑑x=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\delta _n\widehat{\psi }_{j,l}(n).$$
(15)
Similarly, inserting expansion (5) into eq.(10) we have
$`\delta _n`$ $`=`$ $`{\displaystyle \frac{1}{L}}{\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{l=0}{\overset{2^j1}{}}}\stackrel{~}{ϵ}_{j,l}\widehat{\psi }_{j,l}(n)`$
$`=`$ $`{\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{l=0}{\overset{2^j1}{}}}\left({\displaystyle \frac{1}{2^jL}}\right)^{1/2}\stackrel{~}{ϵ}_{j,l}e^{i2\pi nl/2^j}\widehat{\psi }(n/2^j),n0.`$
Equations (15) and (16) show that both the Fourier variables $`\delta _n`$ and the DWT variables $`\stackrel{~}{ϵ}_{j,l}`$ are complete.
However, the statistical properties of the Fourier mode $`n`$ and the DWT mode $`(j,l)`$ are quite different. For a non-Gaussian field consisting of randomly homogeneously distributed clumps with a non-Gaussian probability distribution function(PDF), the one-point distributions of the real and imaginary components of the Fourier modes could be still Gaussian. That is because the Fourier modes are subject to the central limit theorem of random fields (Adler 1981). Even though the non-Gaussian clumps are correlated, the central limit theorem still holds if the two-point correlation function of the clumps approaches zero fast sufficiently (Fan & Bardeen, 1995.) Thus, the non-Gaussian information could be lost in the Fourier representation if the phases of the Fourier coefficients are missing.
On the other hand, the DWT basis doesn’t suffer from the central limit theorem. A key condition necessary for the central limit theorem to hold is that the modulus of the decomposition basis are less than $`C/\sqrt{L}`$, where $`L`$ is the size of the sample and $`C`$ is a constant (Ivanov & Leonine 1989). The Fourier basis obviously satisfy this condition because of $`(1/\sqrt{L})|\mathrm{sin}2\pi nx/L|<C/\sqrt{L}`$, where $`C`$ is independent of $`x`$ and $`n`$. While the DWT basis is compactly supported (Appendix A), and its modulus does not satisfy the condition $`<C/\sqrt{L}`$. Consequently, for the non-Gaussian fields, the one-point distributions of the Fourier variables $`|\delta _n|`$ could be Gaussian, while for the DWT variable $`\stackrel{~}{ϵ}_{j,l}`$, the one-point distributions show non-Gaussian (Pando & Fang 1998b.)
### 2.2 The WFC covariance and DWT power spectrum
In the DWT representation, the covariance $`\delta (x)\delta (x^{})`$ is expressed by a matrix $`\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}`$ with subscripts $`(j,l);(j^{},l^{})`$. The elements of $`j=j^{}`$, $`l=l^{}`$ will be called diagonals, while $`j=j^{}`$ called $`j`$ diagonals, and $`jj^{}`$ the $`j`$ off-diagonals.
The Parseval’s theorem for the DWT decomposition is (Fang & Thews 1998)
$$\frac{1}{L}_0^L|\delta (x)|^2𝑑x=\underset{j=0}{\overset{\mathrm{}}{}}\frac{1}{L}\underset{l=0}{\overset{2^j1}{}}|\stackrel{~}{ϵ}_{j,l}|^2,$$
(17)
which implies that the power of perturbations can be divided into modes, $`(j,l)`$. $`|\stackrel{~}{ϵ}_{j,l}|^2`$ describes the power of the mode $`(j,l)`$. One can then define the DWT power spectrum by the diagonals of the covariance matrix, i.e.<sup>3</sup><sup>3</sup>3The DWT power spectrum, or called scalogram, has been extensively applied in signal analysis (e.g. Mallat 1999.)
$$P_{j,l}=\stackrel{~}{ϵ}_{j,l}^2.$$
(18)
Since the random variables $`\stackrel{~}{ϵ}_{j,l}`$ are complete, one can define a Gaussian field $`\delta (x)`$ by requiring that all the variables $`\stackrel{~}{ϵ}_{j,l}`$ are distributed as a Gaussian process with the covariance
$$\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}=P_{j,l}\delta _{j,j^{}}\delta _{l,l^{}},$$
(19)
and the zero ensemble average of all higher order cumulants of $`\stackrel{~}{ϵ}_{j,l}`$. Thus, a Gaussian field is completely described by its DWT power spectrum $`P_{j,l}`$. For a homogeneous Gaussian field, the DWT power spectrum $`P_{j,l}`$ is $`l`$-independent, i.e. $`P_{j,l}=P_j`$.
Using eqs.(15) and (16), the covariance in the Fourier and DWT representations can be converted from one form to another by
$$\widehat{\delta }_n\widehat{\delta }_n^{}^{}=\underset{j,j^{}=0}{\overset{+\mathrm{}}{}}\underset{l=0}{\overset{2^j1}{}}\underset{l^{}=0}{\overset{2^j^{}1}{}}\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}\widehat{\psi }_{j,l}(n)\widehat{\psi }_{j^{},l^{}}^{}(n^{})$$
(20)
and conversely
$$\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}=\underset{n,n^{}=\mathrm{}}{\overset{+\mathrm{}}{}}\widehat{\delta }_n\widehat{\delta }_n^{}^{}\widehat{\psi }_{j^{},l^{}}(n^{})\widehat{\psi }_{j,l}^{}(n).$$
(21)
Therefore, for a homogeneous Gaussian field given by the DWT power spectrum $`P_j`$, eq. (20) implies
$$\delta _n\delta _n^{}^{}=P(n)\delta _{n,n^{}},$$
(22)
where
$$P(n)=\underset{j=0}{\overset{\mathrm{}}{}}P_j\left|\widehat{\psi }\left(\frac{n}{2^j}\right)\right|^2.$$
(23)
In the derivation of eqs.(22), we used
$$\underset{l=0}{\overset{2^j1}{}}e^{i2\pi (nn^{})l/2^j}=\delta _{n,n^{}}.$$
(24)
Eq.(22) shows that for a homogeneous Gaussian $`P_j`$, the Fourier power spectrum $`P(n)`$ is uniquely determined by the DWT power spectrum $`P_j`$.
However, the reversed relation doesn’t exist, i.e. one cannot show that the DWT covariance is given by eq.(19) with $`P_{j,l}=P_j`$ if the Fourier covariance is given by eq.(22). This indicates that the Fourier and WFC covariance are not equivalent. For instance, fields consisting of homogeneously distributed non-Gaussian clumps generally do not satisfy eq.(19) with a $`l`$-independent $`P_{j,l}`$, but do so for eq.(22). That is, eq.(19) with a $`l`$-independent $`P_{j,l}`$ places stronger constrains on the random field than eq.(22), and therefore, eq.(22) will hold when eq.(19) with a $`l`$-independent $`P_{j,l}`$ holds, but not generally true for the converse.
### 2.3 $`j`$ off-diagonals of the WFC covariance
We now identify the physical meaning of the $`j`$ off-diagonal components of the WFC covariance.
When the “fair sample hypothesis” (Peebles 1980) holds, or equivalently, the random field is ergodic, the $`2^j`$ WFCs $`\stackrel{~}{ϵ}_{j,l}`$, $`l=0\mathrm{}2^j1`$, for a given $`j`$ can be taken as $`2^j`$ independent measurements, because they are measured by projecting onto the mutually orthogonal basis $`\psi _{j,l}(x)`$. Accordingly, the $`2^j`$ WFCs form a statistical ensemble on the scale $`j`$. This ensemble represents actually the one-point distribution of the fluctuations of the DWT modes at a given scale $`j`$. The average over $`l`$ is thus a fair estimation of the ensemble average.
For a Gaussian field, these one-point distributions are Gaussian. However, even if the one-point distributions for all $`j`$ are Gaussian, the density field $`\delta (x)`$ could still be non-Gaussian. That is simply due to the statistical properties of the WFCs $`\stackrel{~}{ϵ}_{j,l}`$ for indices $`j`$ and $`l`$ are independent. It is easy to construct a density field $`\delta (x)`$ for which the WFCs $`\stackrel{~}{ϵ}_{j,l}`$ are Poisson or Gaussian in its one-point distribution with respect to $`l`$, while highly non-Gaussian in terms of $`j`$ (Greiner, Lipa & Carruthers 1995). A simple example is demonstrated as follows. Suppose the one-point distribution of the 2<sup>j</sup> WFCs, $`\stackrel{~}{ϵ}_{j,l}`$, on a scale $`j`$, is Gaussian. If the WFCs on the scale $`j+1`$ is incorporated with those on the scale $`j`$, e.g.,
$`\stackrel{~}{ϵ}_{j+1,2l}`$ $`=`$ $`a\stackrel{~}{ϵ}_{j,l},`$ (25)
$`\stackrel{~}{ϵ}_{j+1,2l+1}`$ $`=`$ $`b\stackrel{~}{ϵ}_{j,l},`$
where $`a`$ and $`b`$ are arbitrary constants, the one-point distribution of the 2<sup>j+1</sup> WFCs $`\stackrel{~}{ϵ}_{j+1,l}`$ is also Gaussian. However, the coherent structure given by eq.(25) leads to a strong correlation between $`\stackrel{~}{ϵ}_{j+1,l}`$ and $`\stackrel{~}{ϵ}_{j,l}`$, i.e. the scale $`j+1`$ fluctuations are always proportional to those on the scale $`j`$ at the same position. This is a local scale-scale correlation. One can also design non-local scale-scale correlation by
$`\stackrel{~}{ϵ}_{j+1,2l}`$ $`=`$ $`a\stackrel{~}{ϵ}_{j,l+\mathrm{\Delta }l},`$ (26)
$`\stackrel{~}{ϵ}_{j+1,2l+1}`$ $`=`$ $`b\stackrel{~}{ϵ}_{j,l+\mathrm{\Delta }l},`$
where $`\mathrm{\Delta }l=1,2..`$. Eq.(26) leads to a strong correlation between the fluctuations on scales $`j+1`$ and $`j`$, but at two places with distance $`\mathrm{\Delta }l`$.
Hence, in terms of the DWT representation, a homogeneous Gaussian field requires that (1) the one-point distributions of the WFCs with respect to $`l`$ are Gaussian, and (2) the distributions of WFCs with different $`j`$’s are uncorrelated, such as
$$\stackrel{~}{ϵ}_{j+1,l}\stackrel{~}{ϵ}_{j,l^{}}=0.$$
(27)
Correspondingly, in the Fourier representation, a Gaussian field also has two requirements (1) the one-point distributions of the amplitudes of the Fourier mode $`|\delta _n|`$ are Gaussian; (2) the phases of $`\delta _n`$ are random. Therefore, eq.(27) is the DWT counterpart of the Fourier random phase. However, it is difficult, or practically impossible, to capture the phase information of each Fourier modes. The local scale-scale correlation is overlooked with the Fourier covariance.
In summary, the $`j`$ off-diagonals of the WFC covariance provide the information of the scale-scale correlation. This non-Gaussian feature arises from mode-mode coupling of gravitational clustering, and cannot be measured by the higher order cumulants of the one-point distribution for a given scale $`j`$, rather, the cross correlation between the different scales. The covariance of a system without scale-scale correlation will be $`j`$-diagonal, i.e.
$$\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}=\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}=0,jj^{},$$
(28)
where eq.(8) has been used at the last step.
## 3 Statistical information extracting from the WFC covariance
### 3.1 $`j`$-diagonalization of the WFC covariance
It has been known that the DWT is powerful for data compression. For very wide types of stochastic clustering processes, the off-diagonal components of the covariance are strongly suppressed in the DWT representation. This suppression is especially efficient for selfsimilar clustering. For instance, one can show analytically that the covariance in the DWT representation is exactly diagonal for some popular hierarchical models of structure formations, such as the block model and its variants (Meneveau & Sreenivasan 1987, Cole & Kaiser 1988). In this respect, the DWT basis represents the adequate normal coordinates. In other words, the DWT analysis can be understood as a Proper Orthonormal Decomposition (POD), or a Karhunen-Loève transformation (e.g. Aubry et al. 1988), in regard to the second order correlations of these stochastic clustering processes.
For more realistic models and observed samples, the WFC covariance is not fully diagonal, but mostly $`j`$-diagonal. In fact, this character has been evident from the measurement of the fourth order scale-scale correlation in the observational samples such as the Ly$`\alpha `$ forest lines (Pando et al. 1998), the transmitted flux of QSO absorption spectrum (Feng & Fang 1999) and the APM bright galaxy catalog (Feng, Deng & Fang 2000). A common conclusion is that the scale-scale correlations are very weak, and negligible on large scales, i.e. $`\stackrel{~}{ϵ}_{j,l}^2\stackrel{~}{ϵ}_{j^{},l^{}}^2=\stackrel{~}{ϵ}_{j,l}^2\stackrel{~}{ϵ}_{j^{},l^{}}^2`$ for $`jj^{}`$ and $`j,j^{}J_{ss}`$, where $`J_{ss}`$ denotes the scale above which the scale-scale correlation is not significant. It is also true for the mass distributions and 2-D and 3-D mock catalog of galaxies in the CDM family of models (Feng, Deng & Fang 2000). This result indicates $`\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}=\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}=0`$ for $`jj^{}`$ and $`j,j^{}J_{ss}`$. Of course, the typical scale $`J_{ss}`$ relies on the models or observational samples.
Therefore, on large spatial scales, $`jJ_{ss}`$, the WFC covariance is already $`j`$-diagonal. Within this range, the covariance matrix is decomposed into $`j`$ sub-matrices $`\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j,l^{}}`$. This guide us to design the first statistics – the DWT band-power spectrum.
### 3.2 The DWT band-power spectrum
Because the model-predicted power spectrum is currently expressed in the Fourier representation, any statistical estimator designed for measuring the power spectrum from real data should have simple relation with the Fourier power spectrum.
Since we have only one realization of the cosmic mass field, no ensemble is available for each mode $`n`$. One cannot measure the Fourier power spectrum $`P(n)`$, as it is from the variance of the amplitude $`|\delta _n|`$ of mode $`n`$. Generally, a power spectrum estimator is to measure banded power spectrum as
$$P_j=\underset{n}{}W_j(n)P(n),$$
(29)
where $`W_j(n)`$ is a window function, which is localized in the $`n`$ (or Fourier) space. The problem that arises here is, what is the criterion for a reasonable banding? and how to optimize the banded power spectrum? The DWT representation provides a natural and reasonable way for the banding.
As discussed in §2.2, for an ergodic field, the $`2^j`$ WFCs $`\stackrel{~}{ϵ}_{j,l}`$ at a given $`j`$ formed an one point distribution of the fluctuations at the scale $`j`$. Therefore, the DWT power spectrum at the scale $`j`$ can be defined as the variance of the one-point distribution, i.e.,
$$P_j=\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}(\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j,l})^2.$$
(30)
Because of the zero mean of WFC $`\stackrel{~}{ϵ}_{j,l}`$, \[eq.(8)\]. $`P_j`$ can be written as, statistically,
$$P_j=\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}|\stackrel{~}{ϵ}_{j,l}|^2=\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}P_{j,l},$$
(31)
which is an ergodicity-allowed spatial average of $`P_{j,l}`$, and is usually referred as DWT power spectrum. As we will show below, Eq.(31) gives an estimator of band-average Fourier power spectrum.
The DWT power spectrum eq.(31) is certainly less detailed than the power spectrum $`P(n)`$ or $`P_{j,l}`$. However, the numbers $`P_j`$ are probably the maximum of statistically valuable band-power spectrum which can be extracted from one realization of an ergodic field. The optimum of this banding can be seen via the phase space $`\{x,k\}`$, where the wavenumber $`k=2\pi n/L`$. Generally a set of orthogonal and complete basis of multiresolution analysis decomposes the entire phase space into elements with different shape, but their volume always satisfies the uncertainty relation, $`\mathrm{\Delta }x\mathrm{\Delta }k2\pi `$. The ordinary Fourier transform is not a multiresolution decomposition, but always takes highest resolution of $`k`$, i.e. $`\mathrm{\Delta }k0`$, and lowest resolution of $`x`$, $`\mathrm{\Delta }x\mathrm{}`$.
To apply the ergodicity, we chopped the survey volume $`L`$ into pieces $`\mathrm{\Delta }x`$. If $`\mathrm{\Delta }x`$ is too large, or $`L/\mathrm{\Delta }x`$ too small, the ensemble contains few members, and thus there will be larger vertical errors placed on the estimated power spectrum. In order to minimize this error, we may make the size of chopped pieces $`\mathrm{\Delta }x`$ to be small. Correspondingly, the width of window function $`\mathrm{\Delta }k=2\pi /\mathrm{\Delta }x`$ will broaden, and the scale resolution will be poor, i.e., there will be a large horizonal error bar placed on the estimated power spectrum. Thus, the optimal chopping can be achieved by a compromise between these two trade-off factors $`L/\mathrm{\Delta }x`$ and $`\mathrm{\Delta }k`$. Generally, $`1/\mathrm{\Delta }x`$ is proportional to the resolvable wavenumber, i.e.
$$1/\mathrm{\Delta }xk.$$
(32)
therefore, the optimized banding $`\mathrm{\Delta }k\mathrm{\Delta }x=2\pi `$ requires
$$\frac{\mathrm{\Delta }k}{k}=\mathrm{\Delta }\mathrm{ln}k1.$$
(33)
That is, the optimized banding is in logarithmic spacing. To detect small scale fluctuations (larger wavenumber $`k`$), the size of the pieces $`\mathrm{\Delta }x`$ is chosen to be smaller. To detect large scale fluctuations (smaller wavenumber), the size of the pieces $`\mathrm{\Delta }x`$ is chosen to be larger. The wavelets $`\psi _{j,l}(x)`$ is constructed by dilating (i.e. changing scale) of the generating function by a factor $`2^j`$ (Appendix A). Therefore, we have $`\mathrm{\Delta }\mathrm{ln}k1`$. In this sense, the DWT is an optimized multiscale decomposition (Farge 1992). Because the set of wavelet basis is complete, one cannot have more independent bands than $`P_j`$.
Under the assumption of a homogeneous Gaussian field, the DWT power spectrum eq.(31) can be rewritten as
$$P_j=\frac{1}{2^j}\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}|\widehat{\psi }(n/2^j)|^2P(n).$$
(34)
where eqs.(11), (21) and (22) have been used. Comparing with eq.(29), clearly, $`P_j`$ is a band-averaged Fourier power spectrum with the window function
$$W_j(n)=\frac{1}{2^j}|\widehat{\psi }(n/2^j)|^2.$$
(35)
Generally, the function $`\widehat{\psi }(n)`$ is non-zero in two narrow wavenumber ranges centered at $`n=\pm n_p`$ with width $`\mathrm{\Delta }n_p`$. Therefore $`P_j`$ is the band spectrum centered at
$$\mathrm{ln}n_j=j\mathrm{log}2+\mathrm{log}n_p,$$
(36)
with the band width as
$$\mathrm{\Delta }\mathrm{log}n=\mathrm{\Delta }n_p/n_p$$
(37)
which stays constant logarithmically. Eqs.(36) and (37) show that the countable data set $`\{P_j,j=1,2\mathrm{}\}`$ represents scale-by-scale band-averaged Fourier power spectrum with the logarithmic spacing of wavenumber. $`P_j`$ is completely determined by the Fourier power spectrum, and therefore, it should be effective for constraining the parameters contained in the Fourier power spectrum.
The band-power spectrum (31) can also be written as, alternatively,
$$P_j=\frac{1}{2^j}\text{tr}Cov_{l,l^{}}^j$$
(38)
where the matrix $`Cov_{l,l^{}}^j`$ is the $`j`$ submatrix of the covariance, i.e.
$$Cov_{l,l^{}}^j=\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j,l^{}}.$$
(39)
Therefore, $`P_j`$’s exhaust all information of the $`j`$ diagonals of the WFC covariance. Eq.(38) shows that we actually need not to diagonalize each $`j`$ submatrix, as $`P_j`$ is given by the trace of the $`j`$ submatrix.
### 3.3 Scale-scale correlations in second and higher orders
In the range of $`j>J_{ss}`$, the scale-scale correlations become significant, the DWT covariance will no longer be diagonal or $`j`$-diagonal.
In this scale range, we should do somewhat diagonalization of the DWT covariance. However, the scale-scale correlation may lead to large errors of the diagonalization, even the diagonalization becomes impossible. Let us consider the example of the scale-scale correlation given by eq.(25). In this case, the variable $`\stackrel{~}{ϵ}_{j+1;l}`$ actually is linearly dependent on $`\stackrel{~}{ϵ}_{j,l+\mathrm{\Delta }l}`$, and therefore the matrix $`\stackrel{~}{ϵ}_{j+1,l}\stackrel{~}{ϵ}_{j,l^{}}`$ is singular. It cannot be diagonalized. For instance, for scales $`j=1,2`$, the covariance matrix now is
$$\left(\begin{array}{ccc}\stackrel{~}{ϵ}_{1,0}\stackrel{~}{ϵ}_{1,0}& \stackrel{~}{ϵ}_{1,0}\stackrel{~}{ϵ}_{2,0}& \stackrel{~}{ϵ}_{1,0}\stackrel{~}{ϵ}_{2,1}\\ \stackrel{~}{ϵ}_{2,0}\stackrel{~}{ϵ}_{1,0}& \stackrel{~}{ϵ}_{2,0}\stackrel{~}{ϵ}_{2,0}& \stackrel{~}{ϵ}_{2,0}\stackrel{~}{ϵ}_{2,1}\\ \stackrel{~}{ϵ}_{2,1}\stackrel{~}{ϵ}_{1,0}& \stackrel{~}{ϵ}_{2,1}\stackrel{~}{ϵ}_{2,0}& \stackrel{~}{ϵ}_{2,1}\stackrel{~}{ϵ}_{2,1}\end{array}\right)=\stackrel{~}{ϵ}_{1,0}^2\left(\begin{array}{ccc}1& a& b\\ a& a^2& ab\\ b& ab& b^2\end{array}\right)$$
(40)
Obviously, this matrix cannot be diagonalized.
More seriously, if the matrix elements have some uncorrelated errors due to measurements, i.e. $`\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}\pm \mathrm{\Delta }\stackrel{~}{ϵ}_{j,l,j^{}l^{}}`$, the matrix (40) looks diagonalizable. However in this case the minors of the matrix are given by the errors $`\mathrm{\Delta }\stackrel{~}{ϵ}_{j,l,j^{}l^{}}`$, and therefore, the diagonalization will be largely contaminated by the errors.
This example indicates that when the scale-scale correlations appear, the number of the independent variables, and then the signal-to-noise ratio. will decrease. we should not extract the statistical properties of the covariance by a diagonalization.
Fortunately, our ultimate goal is not the mathematical diagonalization, but discrimination among physical models of the structure formation. An alternative to the full diagonalization is to take the following two measures: (1) Using the $`j`$-diagonals of each $`j`$ to calculate the band-power spectrum $`P_j`$ \[eq.(31)\]; (2) using the $`j`$ off-diagonals to calculate the second order scale-scale correlations. The second order scale-scale correlations is defined as
$`C_{j,j^{}}(\mathrm{\Delta }l)`$ $`=`$ $`{\displaystyle \frac{1}{2^j}}{\displaystyle \underset{l=0}{\overset{2^j1}{}}}\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}},j>j^{},`$ (41)
$`l^{}`$ $`=`$ $`\mathrm{mod}[l/2^{jj^{}}]+\mathrm{\Delta }l.`$
Like the band-power spectrum \[eqs.(30) and (31)\], $`C_{j,j^{}}(\mathrm{\Delta }l)`$ is defined by an ergodicity-allowed average. $`C_{j,j^{}}(\mathrm{\Delta }l)`$ measures the second order correlation between fluctuations on scale $`j`$ and $`j^{}`$ at positions $`l`$ and $`l^{}`$. Since cosmic density field is homogeneous, the correlation depends only on the difference between $`l`$ and $`l^{}`$, i.e. $`\mathrm{\Delta }lL/2^j^{}`$. For an initially Gaussian field, the scale-scale correlations are developed during the non-linear evolution of the gravitational clustering.
Now, we can use the two statistics $`P_j`$ and $`C_{j,j^{}}`$ to discriminate among models. Actually, the two statistics discrimination would be more worth than the full diagonalization. For instance, the model-predicted galaxy power spectra on smaller scales are generally degenerate with respect to cosmological parameters, i.e. models with different cosmological parameters can yield the same galaxy power spectrum. This is because one always can choose the bias model parameters to fit the prediction with the observations. Therefore, to remove the degeneracy, an independent measure for constraining the bias models is necessary. The scale-scale correlation is found to be sensitive to the bias model (Feng, Deng & Fang 2000). Thus, for model discrimination, the $`j`$-diagonal power spectrum plus scale-scale correlation would be more useful than a full-diagonalization.
In a word, in the scale range of $`j>J_{ss}`$, we will extract the valid statistical information from the covariance by $`P_j`$ and $`C_{j,j^{}}(\mathrm{\Delta }l)`$.
It should be pointed out that even when all $`C_{j,j^{}}(\mathrm{\Delta }l)`$ vanish, one cannot conclude that the system is scale-scale uncorrelated. In other words, that a decomposition $`X_i`$ yields a diagonal covariance doesn’t mean that the modes $`X_i`$ are really statistical uncorrelated. There are many clustering models which have diagonal covariance, but mode-mode statistics are correlated on higher orders (Greiner, Lipa & Carruthers 1995.) A diagonal decomposition means only that mode-mode is uncorrelated on second order.
The higher order generalization of $`C_{j,j^{}}(\mathrm{\Delta }l)`$ is straightforward. For instance one can measure the fourth order scale-scale correlations by
$`C_{j,j^{}}^2(\mathrm{\Delta }l)`$ $`=`$ $`{\displaystyle \frac{1}{2^j}}{\displaystyle \underset{l=0}{\overset{2^j1}{}}}\stackrel{~}{ϵ}_{j,l}^2\stackrel{~}{ϵ}_{j^{},l^{}}^2,j>j^{},`$ (42)
$`l^{}`$ $`=`$ $`\mathrm{mod}[l/2^{jj^{}}]+\mathrm{\Delta }l.`$
This correlation $`C_{j,j^{}}^2(\mathrm{\Delta }l=0)`$ is essentially the same as the so called band-band correlation defined by
$$T=\frac{P_jP_{j+1}}{P_jP_{j+1}}.$$
(43)
It has been shown that the precision of the Fourier band-power spectrum estimator depends on the band-band correlation $`T`$ (Meiksin & White 1998.) In the DWT representation, we arrive at the similar conclusion that when $`C_{j,j^{}}(\mathrm{\Delta }l)`$ or $`C_{j,j^{}}^2(\mathrm{\Delta }l)`$ are non-zero, i.e. when the DWT covariance is not $`j`$ diagonal, we should test models by both the band-power spectrum and scale-scale correlations. For samples of large scale structure, the scale-scale correlations $`C_{j,j^{}}^2(\mathrm{\Delta }=0)`$ has been found to be significant on scales less about 10 $`h^1`$ Mpc (Pando et al 1998, Feng, Deng & Fang 2000.)
## 4 The DWT algorithm of data binning
In the following two sections, we will discuss the algorithm for estimating the band power spectrum $`P_j`$ and scale-scale correlations $`C_{j,j^{}}(\mathrm{\Delta }l)`$ from galaxy redshift surveys, and other samples of large scale structures.
If the position measurement is perfectly precise, the observed galaxy distribution can be written as
$$\rho ^g(x)=\underset{i=1}{\overset{N_g}{}}w_i\delta ^D(xx_i),$$
(44)
where $`N_g`$ is the total number of galaxies, $`\{x_i\}`$ the position of the $`i`$-th galaxy, $`0x_iL`$, $`w_i`$ its weight, and $`\delta ^D`$ is the Dirac-$`\delta `$ function. However, the position measurement has error due to finite spatial resolution, and therefore, the distribution usually is somewhat given by a binned histogram.
The binning is performed by a convolution of the data with a binning function $`W(x)`$ as
$$\stackrel{~}{\rho ^g}(x)=\mathrm{\Pi }(x)W(xx^{})\rho ^g(x^{}),dx^{}$$
(45)
in which $`\mathrm{\Pi }(x)`$ is the sampling function defined as $`\mathrm{\Pi }(x)=_l\delta ^D(xlL/2^j)`$, where $`l`$ labels the $`l`$-th bin. Obviously, the mesh-defined density distribution is given by $`\stackrel{~}{\rho ^g}(x)=_l\rho _l^g\delta ^D(xlL/2^j)`$, where $`\rho _l^g=W(lL/2^jx^{})\rho ^g(x^{})𝑑x^{}`$ is a mass assignment at the $`l`$-th bin.
It is well known that the binning eq.(45) will result in spurious features of the Fourier power spectrum on scale around the Nyquist frequency of the FFT grid (e.g. Jing 1992, Percival & Walden 1993, Baugh & Efstathiou 1994). Mathematically, eq.(45) implies a decomposition by the weight function $`W(x)`$. In other word, $`W(lL/2^jx^{})`$ are playing the role of a scaling functions (or sampling function.) If the scaling functions are orthogonal and complete, the one cannot recovered the original field without distortion. This may cause some spurious features, such as the aliasing effect in the FFT. In the DWT analysis, the binning or sampling are always done by an orthogonal and complete decomposition, one can expected that the spurious features and false correlations can be completely avoided.
### 4.1 Binning with wavelets
The WFCs $`\stackrel{~}{ϵ}_{j,l}`$ are assigned at regular grids $`l=0\mathrm{}2^{j1}`$. It is actually a binning of data. In this case, the binning is automatically realized by the orthogonal projection onto wavelet space, and no extra weight function is required. In result, the contamination due to the sampling error is naturally eliminated.
With eq.(6), one can directly calculate the WFCs of the galaxy distribution (44) by
$$\stackrel{~}{ϵ}_{j,l}^g=\underset{i=1}{\overset{N_g}{}}w_i\psi _{j,l}(x_i).$$
(46)
The errors of $`\stackrel{~}{ϵ}_{j,l}^g`$ can also be calculated from the errors of $`x_i`$.
Since we used the periodized distribution $`\delta (x)`$ in eq.(6), the discontinuity between the data at two boundaries may introduce false coefficients. Yet, this possible false signal is only related to boundaries. One can expected that this false coefficients will not be important for detecting power spectrum on scales much less than $`L`$. This boundary effect has been tested numerically by using simulated samples over a finite length divided in 512 bins with two different boundary conditions (A) periodic boundary conditions; (B) zero padding. The results show that the spectrum can be correctly reconstructed by the DWT regardless of the boundary conditions on scales equal to and less than 64 bins (Pando & Fang 1998).
Note has to be taken of the difference between usual mass assignment and the DWT projection (46). In the former, the mass assignment is given by partitioning the mass on the grids according to the binning function $`W(x)`$, and the binning data are the mesh-defined densities. Whereas for the DWT projection, the binning data, i.e. the WFCs $`\stackrel{~}{ϵ}_{j,l}^g`$ are not the mesh-defined densities, but the fluctuations on scale $`j`$ at position $`l`$, which is obviously not positive-definite.
### 4.2 Binning with scaling functions
In the DWT analysis, the mass assignment is realized by the scaling function $`\varphi _{j,l}(x)`$ \[eq.(A30)\]. Besides the orthogonality eqs.(A33) and (A34), the basic scaling function $`\varphi (\eta )`$ (which is not yet periodized!) satisfies the so-called “partition of unity” as (Daubechies 1992)
$$\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\varphi (\eta l)=1.$$
(47)
One can also define the periodized scaling function as
$$\varphi _{j,l}^P(x)=\left(\frac{2^j}{L}\right)^{1/2}\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\varphi [2^j(\frac{x}{L}+n)l].$$
(48)
Thus, eq.(47) can be rewritten as
$$\underset{l=0}{\overset{2^j1}{}}\frac{L}{2^j}\varphi _{j,l}^P(x)=1$$
(49)
We will only use the periodized scaling function below, and drop the superscript $`P`$.
With the periodized scaling function, the eqs.(A39) - (A41) give
$$\rho (x)=\rho ^J(x)+\underset{j=J}{\overset{\mathrm{}}{}}\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}\psi _{j,l}(x),$$
(50)
where
$$\rho ^J(x)=\underset{l=0}{\overset{2^J1}{}}ϵ_{J,l}\varphi _{J,l}(x).$$
(51)
The scaling function coefficients (SFCs) $`ϵ_{J,l}`$ is given by
$$ϵ_{J,l}=_0^L\rho (x)\varphi _{J,l}(x)𝑑x$$
(52)
Subjecting the distribution (44) to the transform eq.(50), we have
$$\rho ^g(x)=\underset{l=0}{\overset{2^J1}{}}ϵ_{J,l}^g\varphi _{J,l}(x)+\underset{j=J}{\overset{\mathrm{}}{}}\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}^g\psi _{j,l}(x),$$
(53)
where
$$ϵ_{J,l}^g=\underset{i=1}{\overset{N_g}{}}w_i\varphi _{J,l}(x_i).$$
(54)
Using eqs.(44) and (54), eq.(49) yields
$$\underset{l=0}{\overset{2^j1}{}}\frac{L}{2^j}ϵ_{j,l}^g=\underset{i=1}{\overset{N_g}{}}w_i.$$
(55)
This shows that the $`i`$-th galaxy is assigned onto grid $`l`$ by number $`(L/2^j)w_i\varphi _{J,l}(x_i)`$. Therefore, the SFC $`(L/2^j)ϵ_{j,l}^g`$ is the mass assignment of $`\rho ^g(x)`$.
### 4.3 The DWT binning and FFT
Given a galaxy distribution eq.(44), its Fourier transform is evaluated by the trigonometric summation
$$\widehat{\rho }^g(n)=\underset{i=1}{\overset{N_g}{}}w_ie^{i2\pi nx_i/L},$$
(56)
and the power spectrum is $`|\widehat{\rho }^g(n)|^2`$. However, the power spectrum given by the FFT of $`\stackrel{~}{\rho }^g(x)`$ \[eq.(45)\] is
$$|\widehat{\stackrel{~}{\rho }}_l^g(n)|^2=\underset{n^{}=\mathrm{}}{\overset{\mathrm{}}{}}|\widehat{W}(n+2^jn^{})|^2|\widehat{\rho }^g(n+2^jn^{})|^2$$
(57)
where $`\widehat{W}(n)`$ is the FT of the binning function $`W(x)`$. The power spectrum (57) is obviously not equal to the power spectrum $`|\widehat{\rho }^g(n)|^2`$. The power spectrum (57) is given by a superpositions of the power spectrum $`|\widehat{\rho }^g(n+2^jn^{})|^2`$ on all scales $`n+2^jn^{}`$. This is the “aliasing” effect (Hockney & Eastwood 1989, Hoyle, et al. 1999).
In the DWT representation, the FT of eq.(53) yields
$$\widehat{\rho }^g(n)=\underset{l=0}{\overset{2^J1}{}}ϵ_{J,l}^g\widehat{\varphi }_{J,l}(n)+\underset{j=J}{\overset{\mathrm{}}{}}\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}^g\widehat{\psi }_{j,l}(n)$$
(58)
where the function $`\widehat{\varphi }_{j,l}(n)`$ is the Fourier transform of $`\varphi _{j,l}(x)`$, i.e.
$$\widehat{\varphi }_{j,l}(n)=_{\mathrm{}}^{\mathrm{}}\varphi _{j,l}(x)e^{i2\pi nx/L}𝑑x.$$
(59)
Using the definition of $`\varphi _{j,l}(x)`$ \[eq.(A30)\], eq.(59) becomes
$$\widehat{\varphi }_{j,l}(n)=\left(\frac{2^j}{L}\right)^{1/2}\widehat{\varphi }(n/2^j)e^{i2\pi nl/2^j}$$
(60)
where $`\widehat{\varphi }(n)`$ is the Fourier transform of the basic scaling function $`\varphi (\eta )`$
$$\widehat{\varphi }(n)=_{\mathrm{}}^{\mathrm{}}\varphi (\eta )e^{i2\pi n\eta }𝑑\eta .$$
(61)
Eq.(58) gives then
$$\widehat{\rho }^g(n)=\left(\frac{2^J}{L}\right)^{1/2}\widehat{\varphi }(n/2^J)\underset{l=0}{\overset{2^J1}{}}ϵ_{J,l}^ge^{i2\pi nl/2^J}+\underset{j=J}{\overset{\mathrm{}}{}}\left(\frac{2^j}{L}\right)^{1/2}\widehat{\varphi }(n/2^J)\widehat{\psi }(n/2^j)\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}^ge^{i2\pi nl/2^j},$$
(62)
Since $`\widehat{\psi }(n/2^j)`$ is localized in $`n/2^jn_p`$, the second terms in the r.h.s. of eq.(62) are important only for $`n2^Jn_p`$. Thus, the Fourier transform $`\widehat{\rho }^g(n)`$ can be evaluated by
$$\widehat{\rho }^g(n)=\widehat{\varphi }(n/2^J)\widehat{F}(n/2^J),n2^Jn_p$$
(63)
where
$$\widehat{F}(n/2^J)=\left(\frac{2^J}{L}\right)^{1/2}\underset{l=0}{\overset{2^J1}{}}ϵ_{J,l}e^{i2\pi nl/2^J}.$$
(64)
$`\widehat{F}`$ can be calculated by the standard FFT technique. Therefore, the FT of the galaxy distribution $`\rho ^g(x)`$ can be evaluated directly by FFT of its SFC mass assignment $`ϵ_{J,l}^g`$. Eqs.(63) and (64) is actually a scale-adaptive FFT for estimating the power spectrum of an irregular data set. This algorithm computes $`\widehat{\rho }^g(n)`$ up to the scales $`n2^Jn_p`$, where the adapted scale $`J`$ can be chosen as high as the scales to be studied.
## 5 The DWT algorithm on the Poisson sampling
The observed or the mock galaxy distributions $`\rho ^g(x)`$ are considered to be a Poisson sampling with an intensity $`\rho ^M(x)=\overline{\rho }(x)[1+\delta (x)]`$, where $`\overline{\rho }(x)`$ is the galaxy distribution if galaxy clustering is absent, and given by the selection function (Peebles 1980). A proper power spectrum estimator should be effective to obtain the power spectrum debiased from the Poisson sampling. It has been realized that, to handle the Poisson sampling with a non-uniform selection function, the decomposition basis $`\psi _i(𝐱)`$ \[eq.(1)\] is required to have zero average (e.g. Tegmark et al. 1998), i.e.
$$\psi _i(x)𝑑x=0.$$
(65)
This is what we can take the advantage of the DWT analysis, as for the wavelets $`\psi _{j,l}(x)`$, eq.(65) always holds due to the admissibility \[eq.(7)\].
### 5.1 Algorithm for the DWT covariance affected by Poisson sampling
Considering the Poisson sampling, the characteristic function of the galaxy distribution $`\rho ^g(x)`$ is
$$Z[e^{i{\scriptscriptstyle \rho ^g(x)u(x)𝑑x}}]=\mathrm{exp}\left\{𝑑x\rho ^M(x)[e^{iu(x)}1]\right\},$$
(66)
and the correlation functions of $`\rho ^g(x)`$ are given by
$$\rho ^g(x_1)\mathrm{}\rho ^g(x_n)_P=\frac{1}{i^n}\left[\frac{\delta ^nZ}{\delta u(x_1)\mathrm{}\delta u(x_n)}\right]_{u=0},$$
(67)
where $`\mathrm{}_P`$ is the average for the Poisson sampling. We have then
$$\rho ^g(x)_P=\rho ^M(x),$$
(68)
and
$$\rho ^g(x)\rho ^g(x^{})_P=\rho ^M(x)\rho ^M(x^{})+\delta ^D(xx^{})\rho ^M(x).$$
(69)
This equation yields
$$\delta (𝐱)\delta (𝐱^{})=1+\frac{\rho ^g(x)\rho ^g(x^{})_P}{\overline{\rho }(x)\overline{\rho }(x^{})}\delta ^D(xx^{})\frac{1}{\overline{\rho }(x)}.$$
(70)
Since $`\overline{\rho }(x)`$ is not subject to a Poisson process, the second term of the r.h.s. of eq.(70) can be rewritten as $`[\rho ^g(x)/\overline{\rho }(x)][\rho ^g(x^{})/\overline{\rho }(x^{})]_P`$. Using eq. (44), we have
$$\frac{\rho ^g(x)}{\overline{\rho }(x)}=\underset{i=1}{\overset{N_g}{}}\frac{1}{\overline{\rho }(x_i)}w_i\delta ^D(xx_i).$$
(71)
in which the factor $`\overline{\rho }(x_i)`$ can be absorbed into the weight factors $`w_i`$. The WFC covariance is given by
$$\stackrel{~}{ϵ}_{j,l}\stackrel{~}{ϵ}_{j^{},l^{}}=\stackrel{~}{ϵ}_{j,l}^g\stackrel{~}{ϵ}_{j^{},l^{}}^g_P\frac{\psi _{j,l}(x)\psi _{j^{},l^{}}(x)}{\overline{\rho }(x)}𝑑x.$$
(72)
The first term in r.h.s of eq.(70) disappears as all the basis functions $`\psi _{j,l}(x)`$ are admissible \[eq.(7)\].
### 5.2 The estimators for the DWT band power spectrums
If the selection function varies slowly on a scale $`j`$, i.e.
$$\frac{d\mathrm{ln}\overline{\rho }(x)}{dx}2^j/L,$$
(73)
we have approximately,
$$\frac{\psi _{j,l}(x)\psi _{j^{},l^{}}(x)}{\overline{\rho }(x)}𝑑x=\frac{1}{\overline{\rho }(x_l)}\delta _{j,j^{}}\delta _{l,l^{}},$$
(74)
where $`\overline{\rho }(x_l)`$ is the number density of galaxies averaged over a volume of $`L/2^j`$ at $`l`$. In this case, the band-power spectrum is simplified as
$$P_j=\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}^g\stackrel{~}{ϵ}_{j,l}^g_P\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}\frac{1}{\overline{\rho }(x_l)}.$$
(75)
The second term in the r.h.s. is the variance from the Poisson process. Since the Poisson process does not change the ergodicity, the average over $`l`$ in eq.(75) is already a fair estimation for the ensemble average. Therefore, one can drop $`\mathrm{}_P`$ in eq.(75), and the estimation of the DWT band power spectrum is given by
$$P_j=\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}^g\stackrel{~}{ϵ}_{j,l}^g\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}\frac{1}{\overline{\rho }(x_l)}.$$
(76)
The second term is for subtracting the contribution of the discreteness effect (or shot noise) in the Poisson sampling from the power spectrum. $`P_j`$ is debiased from the Poisson process.
### 5.3 The estimators for the scale-scale corrections
Similarly, one can calculate the debiased scale-scale correlations from a galaxy sample $`\rho ^g(x)`$. From eq.(70), the term of the Poisson process is free from scale-scale correlation, the second order scale-scale correlation can be calculated from the WFCs of the galaxy distribution without the correction for the shot noise
$$C_{j,j^{}}(\mathrm{\Delta }l)=\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}^g\stackrel{~}{ϵ}_{j^{},\mathrm{mod}[l/2^{jj^{}}]+\mathrm{\Delta }l}^g.j>j^{}.$$
(77)
However, the Poisson process is not free from higher order scale-scale correlations. For instance, to estimate the band-band correlations eq.(42), we use eq.(67) with $`n=4`$. It gives
$`C_{j,j^{}}^2`$ $`=`$ $`{\displaystyle \frac{1}{2^j}}[{\displaystyle \underset{l=0}{\overset{2^j1}{}}}(\stackrel{~}{ϵ}_{j,l}^g)^2(\stackrel{~}{ϵ}_{j^{},l^{}}^g)^2`$
$`2{\displaystyle \underset{l=0}{\overset{2^j1}{}}}{\displaystyle \frac{\psi _{j,l}(x)\psi _{j^{},l^{}}(x)}{\overline{\rho }(x)}𝑑x\frac{\psi _{j,l}(x^{})\psi _{j^{},l^{}}(x^{})}{\overline{\rho }(x^{})}𝑑x^{}}`$
$`{\displaystyle \underset{l=0}{\overset{2^j1}{}}}{\displaystyle }{\displaystyle \frac{\psi _{j,l}^2(x)}{\overline{\rho }(x)}}dx{\displaystyle }{\displaystyle \frac{\psi _{j^{},l^{}}^2(x^{})}{\overline{\rho }(x^{})}}dx{\displaystyle \underset{l=0}{\overset{2^j1}{}}}{\displaystyle }{\displaystyle \frac{\psi _{j,l}^2(x)\psi _{j^{},l^{}}^2(x)}{\overline{\rho }^3(x)}}dx].`$
where $`j>j^{}`$ and $`l^{}=\mathrm{mod}[l/2^{jj^{}}]+\mathrm{\Delta }l`$. The last three terms are the scale-scale correlations $`C_{j,j^{}}^2`$ from the Poisson sampling. Exactly, the factor $`\overline{\rho }(x)`$ in the Poisson terms should be $`\rho ^M(x)=\overline{\rho }(x)[1+\delta (x)]`$, but we ignored the contributions of $`\delta (x)`$ at the moment.
If the selection function is slowly varying on scales $`j`$ and $`j^{}`$ \[eq.(73)\], we have
$`C_{j,j^{}}^2`$ $`=`$ $`{\displaystyle \frac{1}{2^j}}[{\displaystyle \underset{l=0}{\overset{2^j1}{}}}(\stackrel{~}{ϵ}_{j;l}^g)^2(\stackrel{~}{ϵ}_{j^{};l^{}}^g)^2`$
$`{\displaystyle \underset{l=0}{\overset{2^j1}{}}}{\displaystyle \frac{1}{\overline{\rho }(x_l)\overline{\rho }(x_l^{})}}{\displaystyle \underset{l=0}{\overset{2^j1}{}}}{\displaystyle }{\displaystyle \frac{\psi _{j,l}^2(x)\psi _{j^{},l^{}}^2(x)}{\overline{\rho }^3(x)}}dx].`$
The second and third terms correct for the shot noise on the 4-th order. Numerical results showed that for typical samples of galaxy survey the local ($`l^{}=l`$) scale-scale correlation of the Poisson sampling is significant on small scales (Feng, Deng & Fang 2000.)
## 6 Discussions and conclusions
We presented the method of extracting the band-power spectrum from observed data and simulation sample via a DWT multiresolution decomposition. The DWT scale-by-scale approach provides a physical insight into the covariance matrix of the cosmic mass field.
A key indicator of the DWT power spectrum estimator is the scale-scale and/or the band-band correlations, which can be calculated directly from the DWT covariance and the WFCs. In the scale range that the scale-scale correlations are negligible, the DWT covariance is $`j`$(scale)-diagonal, and it is already a lossless estimation of a banded power spectrum $`P_j`$. This DWT band power spectrum is optimized in the sense that the spatial resolution is adaptive automatically to the scales of the density perturbations.
In the scale range that the scale-scale (or band-band) correlations are significant, the diagonalization of the covariance may not yield an accurate power spectrum, but seriously contaminated by errors. In this case, an effective confrontation between the observed sample and model-prediction may not be given by a full diagonalized covariance, but both of the DWT power spectrum and scale-scale correlations. With the DWT representation, one can calculate the scale-scale correlation as well as the DWT power spectrum. Therefore, the DWT covariance is also useful when scale-scale correlation is strong.
In summary, the basic DWT algorithm is proceeded in the following steps,
1. Calculation of the WFCs $`\stackrel{~}{ϵ}_{J,l}^g`$ and/or the SFCs $`ϵ_{J,l}^g`$ from the data $`\rho ^g(x)`$, where $`J`$ corresponds to the highest resolution of the samples.
2. Calculation of the WFCs $`\stackrel{~}{ϵ}_{j,l}^g`$ for various scale $`j`$.
3. Calculate the band-power spectrum $`P_j`$, and scale-scale correlations $`C_{j,j^{}}`$.
4. In the $`j`$ range of $`C_{j,j^{}}0`$, testing models or constraining parameters by comparing the model-predicted DWT band-power spectrum $`P_j`$ with observed results.
5. In the $`j`$ range of $`C_{j,j^{}}0`$, testing model or constraining parameters by comparing the model-predicted DWT band-power spectrum and scale-scale correlations with observed results.
Since the DWT is computationally powerful, the above-mentioned algorithm is found to be numerically efficient and flexiable (Yang et al. 2000.) Moreover, the developed method is open in the sense that based on the WFCs and SFCs one can add subsequent items to realize the further goals related to the power spectrum measurement and model discrimination. Some of these problems are discussed below.
### 6.1 Higher dimensions and complex geometry
The DWT analysis in a 2 and/or 3-D space $`𝐱`$ can be performed by the bases of the 1-D bases direct product, i.e.
$$\psi _{(j_1,j_2,j_3),(l_1,l_2,l_3)}(x_1,x_2,x_3)=\psi _{j_1,l_1}(x_1)\psi _{j_2,l_2}(x_2)\psi _{j_3,l_3}(x_3).$$
(80)
In this case, the three scales $`(j_1,j_2,j_3)`$ of the WFCs can be different for different directions. One can define radial scales by
$$k=2\pi \left[\left(\frac{2^{j_1}}{L_1}\right)^2+\left(\frac{2^{j_2}}{L_2}\right)^2+\left(\frac{2^{j_3}}{L_3}\right)^2\right]^{1/2},$$
(81)
where $`L_1\times L_2\times \times L_3`$ is the 3-D box.
For 2 and 3-D samples, one can also decompose by the mixed direct product of 1-D wavelets and scaling functions. For instance, a 3-D sample can be decomposed by bases
$$\psi _{(j_1,j_2,j_3),(l_1,l_2,l_3)}^{(1,2)}(x_1,x_2,x_3)=\varphi _{j_1,l_1}(x_1)\psi _{j_2,l_2}(x_2)\psi _{j_3,l_3}(x_3).$$
(82)
where the scaling functions $`\varphi _{j,l}`$ actually play the role of chopping a 3-D sample into $`2^{j_1}`$ 2-D slices in the $`x_1`$ direction, $`l_1=0,\mathrm{}2^{j_1}1.`$ Like the binning by the scaling function (§4.2), the chopping eq.(82) will not cause spurious features.
The problem of complex geometry of samples can be treated by using the locality of the $`\psi _{j,l}`$ (Pando & Fang 1998a). The locality property allows the WFCs to be independent of the data outside an “influence” cone. The WFCs $`\stackrel{~}{ϵ}_{j,l}`$ is only determined by data in the interval $`[(lL/2^{j+1}(\mathrm{\Delta }x)/2^{j+1},(lL/2^{j+1}+(\mathrm{\Delta }x)/2^{j+1}]`$, where $`\mathrm{\Delta }x`$ is the width of the basic wavelet $`\psi `$. With this property, any complex geometry of samples can be regularized into a 2 or 3-D box by zero padding in the field between the sample geometry and the box. Since all WFCs at the zero padding zone are zero, one can use the DWT to analyze the regular box, but not treat the WFCs related to the zero padding as the variables of valid degrees of freedom.<sup>4</sup><sup>4</sup>4About DWT on manifold, see also W. Sweldens http://cm.bell-labs.com/who/wim or http://www.wavelet.org
### 6.2 Non-Gaussianity and power spectrum detection
We have emphasized that the information of the non-Gaussian features are important for a precise detection of the power spectrum, or band power spectrum. That is because, from the covariance, one can only find statistically uncorrelated (or statistical orthogonal) bases or modes on second order. For non-Gaussian fields, the modes statistically uncorrelated on second order might be statistically correlated at the 3rd and 4th orders. On the other hand, the power spectrum is of second order, and therefore, the power spectrum estimates at different scales might not be statistically uncorrelated if there are 3rd and 4th order correlations. The accuracy of a power spectrum estimation is affected by the higher order statistical correlations.
For instance, a popular bias model for galaxy formation employ the selection probability functions as (Cole et al. 1998)
$$P(\delta (𝐫))\mathrm{exp}\left[\alpha \frac{\delta _s(𝐫)}{\sigma _s}\right],$$
(83)
where $`\alpha `$ is const, and $`\delta _s(𝐫)`$ and $`\sigma _s`$ are smoothed density field and variance. Therefore, if the density field is Gaussian, the galaxy distribution given by the Poisson sampling with the intensity eq.(83) will be lognormal. The baryonic distribution is sometimes also modeled by a lognormal relation with the underlying Gaussian mass field (Bi, Ge & Fang 1995, Bi & Davidsen 1997). As having been well known, for lognormal distribution, the most likely value can be significantly different from their mean value. In this case, to estimate the accuracy of a power spectrum detection, the higher order cumulant statistics is needed.
In the DWT analysis, the $`2^j`$ WFCs give the one point distribution of the fluctuations on scale $`j`$. Therefore, the third and forth cumulants can be calculated by
$$S_j=\frac{1}{P_j^{3/2}}\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}(\stackrel{~}{ϵ}_{j,l}\overline{\stackrel{~}{ϵ}_{j,l}})^3.,$$
(84)
$$K_j\frac{1}{P_j^2}\frac{1}{2^j}\underset{l=0}{\overset{2^j1}{}}(\stackrel{~}{ϵ}_{j,l}\overline{\stackrel{~}{ϵ}_{j,l}})^43$$
(85)
These are, respectively, the skewness and kurtosis spectra. It is not difficult to generalize eqs.(84) and (85) to more higher orders.
### 6.3 Selection of the basis of the multiresolution analysis
In computing the samples of redshift surveys, there are two coordinate systems having been widely used: 1. parallel plane system; 2. spherical shell system. For system 1, the volume of the survey can be approximated as a box, and therefore, the wavelets of eqs.(80) and (82) are suitable for the decomposition. For the system 2, we should use the wavelets on 2-D spherical surface. With the development of the DWT analysis, the bank of the DWT analysis has stored more and more sets of the orthogonal and complete basis for the multiresolution decomposition of different geometries. The multiscale analysis on geometry beyond above-mention two simple cases is being feasible.
### 6.4 Systematic effects
The influence of various systematic effects on the power spectrum detection has only been studied very preliminarily. The linear effect of redshift distortion on the power spectrum detection has been well studied (e.g. Hamilton 1995). It is not difficult to incorporate the linear theory of the redshift distortion with the DWT analysis. A key operator of the mapping a real space distribution into redshift space is $`(1a(^2/z^2)^2)`$, where coefficient $`a`$ is const. To diagonalize this differential-integral operator, the Fourier representation is certainly the best. However, it has been shown that this operator is quasidiagonal in the DWT representation (Farge 1996).
Moreover, it would be straightforward to include a scale-dependent bias in the DWT representation. The redshift distortion is usually calculated under the assumption that the galaxy distribution $`\rho ^g(𝐱)`$ is linearly related to the underlying mass field $`\rho (𝐱)`$, i.e. $`\rho ^g(𝐫)=b\rho (𝐫)`$, where $`b`$ is the bias parameter. However, observations have indicated that the bias parameters probably are scale-dependent (Fang, Deng & Xia 1998.) It is easy to introduce scale-dependent bias in the DWT representation. For instance one can define a bias parameter on scale by $`\stackrel{~}{ϵ}_{j,l}^g=b_j\stackrel{~}{ϵ}_{j,l}`$.
LLF acknowledges support from the National Science Foundation of China (NSFC) and World Laboratory Scholarship. This project was done during LLF’s visiting to the Department of Physics, University of Arizona. This work was supported in part by the LWL foundation. We thank anonymous referee for helpful comments.
## Appendix A The discrete wavelet transform (DWT) of density fields
Let us briefly introduce the DWT analysis of the cosmic mass density fields, for the details of mathematical stuffs refers to the classical papers by Mallat (1989a,b,c); Meyer (1992); Daubechies, (1992) and references therein, and for physical applications, refers to Fang & Thews (1998) and references therein. Some other cosmological applications of wavelets can also be found at, e.g., Pando, Vills-Gabaud & Fang (1998), Hobson, Jones & Lasenby (1999), Sanz et al. (1999), Tenorio et al. (1999), Xu, Fang, & Wu (2000), Cayon, et al (2000).
### A.1 Expansion by scaling functions
We consider here a 1-D mass density distribution $`\rho (x)`$ or contrast $`\delta (x)=[\rho (x)\overline{\rho }]/\overline{\rho }`$, which are mathematically random fields over a spatial range $`0xL`$. It is not difficult to extend all results developed in this section into 2-D and 3-D because the DWT bases for higher dimension can be constructed by a direct product of 1-D bases.
First, we introduce the scaling functions for the Haar wavelets. There are top-hat window functions defined by
$$\varphi _{j,l}^H(x)=\{\begin{array}{cc}1\hfill & \text{for }Ll2^jxL(l+1)2^j\hfill \\ 0\hfill & \text{otherwise.}\hfill \end{array},$$
(A1)
where the superscript $`H`$ is stand for Haar. The scaling function, $`\varphi _{j,l}^H(x)`$ actually gives a window at resolution scale $`L/2^j`$ and position $`Ll2^jxL(l+1)2^j`$. With the scaling function, the mean of density contrast distribution in the spatial range $`Ll2^jxL(l+1)2^j`$ can be expressed as
$$ϵ_{j,l}=\frac{2^j}{L}_0^L\delta (x)\varphi _{j,l}^H(x)𝑑x.$$
(A2)
The number $`ϵ_{j,l}`$ is called the scaling function coefficient(SFC). Using SFCs, one can construct a density contrast field as
$$\delta ^j(x)=\underset{l=0}{\overset{2^j1}{}}ϵ_{j,l}\varphi _{j,l}^H(x).$$
(A3)
This is the density contrast $`\delta (x)`$ smoothed on scale $`L/2^j`$, or for simple, $`j`$-scale.
The scaling function $`\varphi _{j,l}^H(x)`$ can be rewritten
$$\varphi _{j,l}^H(x)=\varphi ^H(2^jx/Ll),$$
(A4)
where
$$\varphi ^H(\eta )=\{\begin{array}{cc}1\hfill & \text{for 0 }\eta \text{ 1}\hfill \\ 0\hfill & \text{otherwise.}\hfill \end{array}$$
(A5)
$`j`$, $`l`$ are integers, with $`j0`$, and $`0l2^j1`$. $`\varphi ^H(\eta )`$ is called the basic scaling function. The scaling function $`\varphi _{j,l}^H(x)`$ is thus a translation and dilation of the basic scaling function.
The functions $`\varphi _{j,l}^H(x)`$ are orthogonal with respect to $`l`$, i.e.
$$_0^L\varphi _{j,l}^H(x)\varphi _{j,l^{}}^H(x)𝑑x=\frac{L}{2^j}\delta _{l,l^{}}$$
(A6)
where $`\delta _{l,l^{}}`$ is Kronecker delta function. Thus, eq.(A3) gives functions in the function space $`V_j`$ spanned by bases $`\varphi _{j,l}^H(x)`$. $`V_j`$ is a closed subspaces of $`L_2(R)`$, i.e. $`V_jL_2(R)`$. It is easy to show that
$$\varphi _{j,l}^H(x)=\varphi _{j+1,2l}^H(x)+\varphi _{j+1,2l+1}^H(x)$$
(A7)
$$ϵ_{j,l}=\frac{1}{2}(ϵ_{j+1,2l}+ϵ_{j+1,2l+1}).$$
(A8)
Therefore, $`V_jV_{j+1}`$ for all $`j`$. Thus, the orthogonal projectors $`P_j`$ onto $`V_j`$, i.e. $`P_jfV_j`$, satisfy
$$\underset{j\mathrm{}}{lim}P_jf=f,$$
(A9)
for all $`fL_2(R)`$. A multiresolution analysis is then defined by the sequence of subspaces $`V_j`$.
### A.2 Expansion by wavelets
Eqs. (A7) and (A8) show that $`\delta ^j(x)`$ contains less information than $`\delta ^{j+1}(x)`$, because information on scale $`j+1`$ have been smoothed out by eq. (A8). It would be nice not to lose any information during the smoothing from $`j+1`$ to $`j`$ \[eq.(A8)\]. This can be accomplished if the differences, $`\delta ^{j+1}(x)\delta ^j(x)`$, between the smoothed distributions on succeeding scales are somehow retained. This is, if we are able to retain these differences, this scheme will then make it possible to smooth the distribution and yet not lose any information as a result of the smoothing.
To calculate the differences, we define the difference function, or wavelet, as
$$\psi ^H(\eta )=\{\begin{array}{cc}1\hfill & \text{for }0\eta 1/2\hfill \\ 1\hfill & \text{for }1/2\eta 1\hfill \\ 0\hfill & \text{otherwise.}\hfill \end{array}$$
(A10)
This is the basic Haar wavelet. As with the scaling functions, one can construct a set of wavelets $`\psi _{j,l}^H(x)`$ by dilating and translating eq.(A10) as
$$\psi _{j,l}^H(x)=\psi ^H(2^jx/Ll).$$
(A11)
The Haar wavelets are orthogonal with respect to both indexes $`j`$ and $`l`$, i.e.
$$_0^L\psi _{j^{},l^{}}^H(x)\psi _{j,l}^H(x)𝑑x=\left(\frac{L}{2^j}\right)\delta _{j^{},j}\delta _{l^{},l}.$$
(A12)
For a given $`j`$, $`\psi _{j,l}^H(x)`$ is also orthogonal to the scaling functions $`\varphi _{j^{},l}^H(x)`$ with $`j^{}j`$, i.e.
$$_0^L\varphi _{j^{},l^{}}^H(x)\psi _{j,l}^H(x)𝑑x=0,\mathrm{if}j^{}j.$$
(A13)
From eqs.(A4) and (A11), we have
$$\begin{array}{cc}\varphi _{j,2l}^H(x)\hfill & =\frac{1}{2}(\varphi _{j1,l}^H(x)+\psi _{j1,l}^H(x)),\hfill \\ & \\ \varphi _{j,2l+1}^H(x)\hfill & =\frac{1}{2}(\varphi _{j1,l}^H(x)\psi _{j1,l}^H(x)).\hfill \end{array}$$
(A14)
Thus, the difference $`\delta ^{j+1}(x)\delta ^j(x)`$ is given by
$$\delta ^{j+1}(x)\delta ^j(x)=\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}\psi _{j1,l}^H(x),$$
(A15)
where $`\stackrel{~}{ϵ}_{J1,l}`$ are called the wavelet function coefficients(WFC), which is given by
$$\stackrel{~}{ϵ}_{j,l}=\frac{2^j}{L}\delta (x)\psi _{j,l}^H(x)𝑑x.$$
(A16)
Using the relation (A15) repeatedly, we have
$$\delta ^j(x)=\delta ^0(x)+\underset{j^{}=0}{\overset{j1}{}}\underset{l=0}{\overset{2^j^{}1}{}}\stackrel{~}{ϵ}_{j^{},l}\psi _{j^{},l}^H(x).$$
(A17)
This is an expansion of the function $`\delta ^j(x)`$ with respect to the basis $`\psi _{j,l}^H(x)`$, and $`\delta ^0(x)`$ is the mean of $`\delta (x)`$ in the range $`L`$. We have $`\delta ^0(x)=0`$ if $`\delta (x)`$ is density contrast. Considering (A9), for any $`f(x)L^2(R)`$ in $`L`$ with mean $`\overline{f}=0`$ we have
$$f(x)=\underset{j=0}{\overset{\mathrm{}}{}}\underset{l=0}{\overset{2^j1}{}}\stackrel{~}{ϵ}_{j,l}\psi _{j,l}^H(x),$$
(A18)
and
$$\stackrel{~}{ϵ}_{j,l}=\frac{2^j}{L}_0^Lf(x)\psi _{j,l}^H𝑑x.$$
(A19)
For a given $`j`$, the wavelets $`\psi _{j,l}^H(x)`$ form a space $`W_j`$ which is the orthogonal complements of $`V_j`$ in $`V_{j+1}`$, i.e. $`V_{j+1}=V_jW_j`$. Thus, every $`f^jV_j`$ has a unique decomposition $`f^j=f^{j1}+d^{j1}`$ with $`f^{j1}V_{j1}`$ and $`d^{j1}W_{j1}`$. Since $`W_jV_{j+1}`$ and $`W_j`$ is orthogonal to $`V_j`$, $`W_j`$ is also orthogonal to $`W_{j1}`$ and $`W_{j+1}`$. Thus, all the spaces $`W_j`$ are mutually orthogonal. Since $`V_j`$ contains only $`W_j^{}`$ with $`j^{}<j`$, $`V_j`$ is orthogonal to all $`W_j^{}`$ with $`j^{}j`$.
### A.3 Compactly supported orthogonal basis
In terms of the subspace $`V_j`$, the basic scaling function $`\varphi (\eta )`$ and basic $`\psi (\eta )`$ belong to $`V_0`$ and $`W_0`$ respectively, and they can be expressed by the basis of $`V_1`$, $`\varphi (2\eta l)`$, i.e.
$$\varphi (\eta )=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}a_l\varphi (2\eta l),$$
$$\psi (\eta )=\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}b_l\varphi (2\eta l),$$
(A20)
where $`a_l`$ and $`b_l`$ are called the filter coefficients.
If we require that the scaling function $`\varphi (\eta )`$ is normalized, eq.(A21) yields
$$\underset{l}{}a_l=2.$$
(A21)
Requiring orthogonality for $`\varphi (x)`$ with respect to discrete integer translations, i.e.
$$_{\mathrm{}}^{\mathrm{}}\varphi (\eta m)\varphi (\eta )𝑑\eta =\delta _{m,0},$$
(A22)
we have
$$\underset{l}{}a_la_{l+2m}=2\delta _{0,m}.$$
(A23)
The orthogonality between $`\varphi `$ and $`\psi `$ means
$$_{\mathrm{}}^{\mathrm{}}\psi (\eta )\varphi (\eta l)𝑑\eta =0.$$
(A24)
Therefore, one has
$$b_l=(1)^la_{1l}.$$
(A25)
Furthermore, the wavelet $`\psi (\eta )`$ has to be admissible
$$_{\mathrm{}}^+\mathrm{}\psi (\eta )𝑑\eta =0,$$
(A26)
so we need
$$\underset{l}{}b_l=0.$$
(A27)
The conditions (A22), (A24), (A26) and (A28) for the filter coefficients were employed to construct families of scaling functions and wavelets. The simplest solution of the filter coefficients is $`a_0=a_1=b_0=b_1=1`$ and all others 0. This solution gives the Haar wavelet. After the Haar wavelet, the simplest solution for the filter coefficients is
$`a_0=(1+\sqrt{3})/4,`$ $`a_1=(3+\sqrt{3})/4,`$ (A28)
$`a_2=(3\sqrt{3})/4,`$ $`a_3=(1\sqrt{3})/4.`$
This is the Daubechies 4 wavelet (D4). It is compactly supported and continuous.
With these wavelets, the multiresolution analysis can be performed in the similar way as developed in last two sections for the Haar wavelets. The scaling functions and wavelets for spanning the subspace $`V_j`$ and $`W_j`$ are given, respectively, by a translation and dilation of the basic scaling function and basic wavelet
$$\varphi _{j,l}(x)=\left(\frac{2^j}{L}\right)^{1/2}\varphi (2^jx/Ll)$$
(A29)
and
$$\psi _{j,l}(x)=\left(\frac{2^j}{L}\right)^{1/2}\psi (2^jx/Ll).$$
(A30)
The wavelets are orthonormal, i.e.
$$\psi _{j,l}(x)\psi _{j^{},l^{}}(x)𝑑x=\delta _{j,j^{}}\delta _{l,l^{}}.$$
(A31)
Eqs.(A23) and (A25) yield also
$$\varphi _{j,l}(x)\varphi _{j,l^{}}(x)𝑑x=\delta _{l,l^{}},$$
(A32)
and
$$\varphi _{j,l}(x)\psi _{j^{},l^{}}(x)𝑑x=0j^{}j.$$
(A33)
The set of $`\psi _{j,l}`$ and $`\varphi _{0,m}(x)`$ with $`0j<\mathrm{}`$ and $`\mathrm{}<l,m<\mathrm{}`$ form a complete, orthonormal basis in the space of functions with period length $`L`$.
Thus, a density field $`\rho (x)`$ with period length $`L`$ can be expanded as (Fang & Thews 1998)
$$\rho (x)=\overline{\rho }+\overline{\rho }\underset{j=0}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\stackrel{~}{ϵ}_{j,l}\psi _{j,l}(x),$$
(A34)
or the density contrast $`\delta (x)=(\rho (x)\overline{\rho })/\overline{\rho }`$ is
$$\delta (x)=\underset{j=0}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\stackrel{~}{ϵ}_{j,l}\psi _{j,l}(x),$$
(A35)
where
$$\overline{\rho }=L^1_0^L\rho (x)𝑑x$$
(A36)
and
$$\stackrel{~}{ϵ}_{j,l}=_{\mathrm{}}^{\mathrm{}}\delta (x)\psi _{j,l}(x)𝑑x.$$
(A37)
More generally, we have
$$\rho (x)=\rho ^J(x)+\overline{\rho }\underset{j=J}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{+\mathrm{}}{}}\stackrel{~}{ϵ}_{j,l}\psi _{j,l}(x),$$
(A38)
where $`\rho ^J(x)`$ is the density field smoothed on scale $`J`$
$$\rho ^J(x)=\underset{l=\mathrm{}}{\overset{+\mathrm{}}{}}ϵ_{J,l}\varphi _{J,l}(x).$$
(A39)
and the scaling function coefficient(SFC) $`ϵ_{J,l}`$ is given by
$$ϵ_{J,l}=_{\mathrm{}}^+\mathrm{}\rho (x)\varphi _{J,l}(x)𝑑x.$$
(A40) |
warning/0003/math0003099.html | ar5iv | text | # Bochner-Kähler metrics
## 1. Introduction
In Riemannian geometry, the decomposition of the curvature tensor into its irreducible summands under the orthogonal group is regarded as fundamental. There are three such summands, the scalar curvature, the traceless Ricci curvature, and the Weyl curvature.<sup>1</sup><sup>1</sup>1The Weyl curvature exists as a nontrivial summand only when the dimension $`n`$ of the underlying manifold is 4 or more. When $`n=4`$, the Weyl curvature is further reducible under the special orthogonal group, but not the full orthogonal group. The metrics for which one or more of these irreducible tensors vanishes have been the subject of much research and a great deal is now known about restrictions on the topology of the complete or compact examples. For example, consult , where the bulk of the work is devoted to studying the metrics for which the traceless Ricci curvature vanishes, i.e., the Einstein metrics. The metrics in dimensions 4 or higher for which the Weyl curvature vanishes are the conformally flat metrics. While such metrics are trivial to describe locally, their global geometry is rather delicate, so that classifying the complete or compact examples remains a challenge.
In Kähler geometry, the corresponding decomposition of the curvature tensor into its irreducible summands under the unitary group is not quite as familiar, although it has been known since the 1949 work of Bochner . (For a more recent treatment, see \[3, 2.63\].) The Kähler decomposition bears some resemblance to the Riemannian one, there being three irreducible summands, the scalar curvature, the traceless Ricci curvature, and what has become known as the *Bochner* curvature.<sup>2</sup><sup>2</sup>2N.B.: The Bochner curvature is one component of the Weyl curvature, but not the only component. For example, in complex dimension $`2`$ the Bochner curvature is the anti-self-dual part of the Weyl curvature. See §2.1.3. Bochner’s interest in this latter tensor was due to its appearance in certain Weitzenbock-type formulae. In , he proved some cohomological vanishing theorems for compact Kähler manifolds with vanishing Bochner tensor or, more generally, for manifolds for which the pointwise norm of the Bochner tensor was sufficiently small relative to the smallest eigenvalue of the Ricci tensor.
While Kähler metrics with vanishing scalar curvature or vanishing traceless Ricci curvature (i.e., Kähler-Einstein metrics) have been much studied, those with vanishing Bochner tensor, now known as *Bochner-Kähler* metrics, have received considerably less attention. For surveys of what has been known up to now about these metrics, the reader might consult , , , , or in addition to §2 of the present article. One will be struck by the paucity of examples. For example, up until now, every known complete Bochner-Kähler metric was also locally symmetric. (The symmetric examples are the products of the form $`M_c^p\times M_c^{np}`$ where $`M_c^p`$ denotes the $`p`$-dimensional complex space form of constant holomorphic sectional curvature $`c`$.)
At first glance, one might expect the theory of Bochner-Kähler manifolds to parallel the theory of conformally flat manifolds. However, this expectation is quickly abandoned. Unlike the local description of conformally flat metrics, a local description of Bochner-Kähler metrics is far from trivial. In fact, no such description was known until now.
Theorem 1 and Corollary 1 show that the space of isometry classes of germs of $`C^5`$ Bochner-Kähler metrics in complex dimension $`n`$ can be naturally regarded as a closed semi-algebraic subset $`F_n^{2n+1}`$ (with a nonempty interior). More precisely, if $`M`$ is a complex $`n`$-manifold endowed with a $`C^5`$ Bochner-Kähler metric $`g`$, there is a mapping $`f:MF_n^{2n+1}`$ (which is a polynomial function of the curvature tensor of $`g`$ and its first two covariant derivatives) with the property that $`f(x)=f(y)`$ for $`x,yM`$ if and only if the germ of $`g`$ at $`x`$ is holomorphically isometric to the germ of $`g`$ at $`y`$. Moreover, I show that for every $`vF_n`$, there is a Bochner-Kähler metric $`g`$ on a neighborhood $`U`$ of $`0^n`$ so that the associated classifying map $`f:UF_n`$ satisfies $`f(0)=v`$. (This existence theorem relies on some old results of Élie Cartan that are not readily available in the current literature, so I have included an appendix that exposes these results in a form convenient for the applications in this article.) A by-product of this analysis is that any $`C^5`$ Bochner-Kähler metric is necessarily real-analytic.<sup>3</sup><sup>3</sup>3Presumably, any $`C^2`$ Bochner-Kähler metric is real-analytic, but I have not shown this. Accordingly, for the rest of the article, I assume that the Bochner-Kähler metrics under consideration are real-analytic.
Theorem 1 suggests that a notion of ‘analytic continuation’ of Bochner-Kähler metrics might be useful. Elements $`v_1,v_2F_n`$ are said to be *analytically connected* if there is a connected Bochner-Kähler manifold $`(M^n,g)`$ for which $`f(M)`$ contains both $`v_1`$ and $`v_2`$. This is an equivalence relation, so denote the analytically connected equivalence class of $`vF_n`$ by $`[v]F_n`$. In Theorem 3, I construct a polynomial submersion $`C:^{2n+1}^{n+1}`$ and show that it is constant on each $`[v]`$. Eventually, Theorem 7 will show that each fiber $`C^1(c)F_n`$ consists of a finite number of analytically connected equivalence classes and explicitly identify each one as a (not necessarily closed) semi-algebraic set of (real) dimension at most $`n`$. Thus, the components of $`C`$ furnish a set of ‘coarse moduli’ for Bochner-Kähler metrics. The image $`C(F_n)^{n+1}`$ (which will be explicitly identified below) has nonempty interior, so it makes sense to say that, roughly speaking, the moduli space of Bochner-Kähler metrics in complex dimension $`n`$ has real dimension $`n+1`$.
Since each equivalence class $`[v]F_n`$ has real dimension $`mn`$ at its smooth points, this suggests that a connected Bochner-Kähler manifold of complex dimension $`n`$ must always have a non-trivial local isometry ‘group’, acting with some cohomogeneity $`mn`$. In Theorem 2 and Proposition 4 , I show that when $`M^n`$ is simply-connected, the Lie algebra $`𝔤`$ of Killing fields for a Bochner-Kähler structure on $`M`$ does indeed have dimension at least $`n`$ and I compute its precise dimension for each analytically connected equivalence class $`[v]F_n`$. Moreover, for each $`vF_n`$, I compute the dimension of the orbit of the local isometry pseudogroup through an $`xM`$ with $`f(x)=v`$. In particular, I show in §3.3.3 how to compute the cohomogeneity $`m`$ for each $`vF_n`$. (Interestingly enough, it turns out that $`m`$ cannot, in general, be computed from the coarse moduli $`C(v)`$ alone. This is a reflection of the fact that not all of the equivalence classes $`[v]`$ are closed sets in $`F_n`$.) The ultimate conclusion is that a Bochner-Kähler metric always possesses a rather high degree of infinitesimal symmetry.
Perhaps the greatest surprise and what, ultimately, turns out to be the key to understanding the geometry of Bochner-Kähler metrics is that the Lie algebra $`𝔤`$ contains a canonical central subalgebra $`𝔷`$ whose dimension $`m`$ is the same as that of $`[f(x)]F_n`$ for some (and hence any) $`xM`$. This infinitesimal torus action can be described explicitly as follows: Let $`\mathrm{\Omega }`$ be the Kähler form and let $`\rho =\mathrm{Ric}(\mathrm{\Omega })`$ be its associated Ricci form \[3, 2.44\]. Define a ‘renormalized’ Ricci 2-form $`\eta `$ by
$$\eta =\frac{1}{2(n+1)(n+2)}(\mathrm{tr}_\mathrm{\Omega }\rho )\mathrm{\Omega }\frac{1}{2(n+2)}\rho $$
and define $`p_h(t)`$ by the formula $`(t\mathrm{\Omega }\eta )^n=p_h(t)\mathrm{\Omega }^n`$. Thus,
$$p_h(t)=t^nh_1t^{n1}+\mathrm{}+(1)^nh_n$$
where $`h_j:M`$ is a certain symmetric polynomial of degree $`j`$ in the eigenvalues of the Ricci tensor. Then Theorem 4 asserts that the $`\mathrm{\Omega }`$-Hamiltonian vector fields $`X_j`$ defined by $`X_j\text{ }\text{ }\mathrm{\Omega }=dh_j`$ for $`1jn`$ are Killing fields for the metric $`g`$ and that they Lie commute, i.e., span a torus $`𝔷𝔤`$. Of course, this infinitesimal action is Poisson since $`h=(h_1,\mathrm{},h_n):M^n`$ is a momentum mapping by definition.
As is shown in §3.4.2, the map $`h:M^n`$ can be written as $`\mathrm{\Psi }f`$ where $`\mathrm{\Psi }`$ is a weighted homogeneous polynomial mapping from $`^{2n+1}`$ to $`^n`$. When $`M`$ is connected, the maps $`h`$ and $`f`$ have the same fibers. The image of $`h`$ is $`m`$-dimensional and lies in an affine subspace $`𝔞^n`$ of dimension $`m`$ (the same $`mn`$ as defined above). This number $`m`$ is defined to be the *cohomogeneity* of the Bochner-Kähler structure. Theorem 5 shows that, in fact, $`p_h(t)`$ has a polynomial factor $`p_{h^{\prime \prime }}(t)`$ with constant coefficients and of degree $`nm`$. Thus, $`p_h(t)=p_{h^{\prime \prime }}(t)p_h^{}(t)`$ where
$$p_h^{}(t)=t^mh_1^{}t^{m1}+\mathrm{}+(1)^mh_m^{}$$
and the functions $`h_j^{}:M^m`$ for $`1jm`$ are smooth. Theorem 5 also shows that, outside a (possibly singular) complex submanifold $`NM`$ (called the *exceptional locus*), the *reduced momentum mapping* $`h^{}=(h_1^{},\mathrm{},h_m^{}):M^m`$ is a submersion. This singular locus is the union of a number of totally geodesic complex submanifolds of $`M`$. Let $`M^{}=MN`$ be its complement, the *regular locus*.
Theorem 7 yields a polynomial embedding $`\iota _v:[v]^m`$ of each $`m`$-dimensional analytically connected equivalence class $`[v]`$ into $`^m`$ as a convex polytope, i.e., an intersection of half-spaces (which can be open or closed). The embedding $`\iota _v`$ satisfies $`h^{}=\iota _vf`$ when $`f(M)`$ lies in $`[v]`$. Moreover, $`h^{}`$ maps $`M^{}`$ into the interior of the polytope. Theorem 8 shows that the interior of $`\iota _v\left([v]\right)`$ carries a canonical Riemannian metric so that $`h^{}:M^{}\iota _v\left([v]\right)^{}`$ is a Riemannian submersion. In fact, this metric on $`\iota _v\left([v]\right)^{}`$ has rational polynomial coefficients when expressed in terms of linear coordinates on $`^m`$. These metrics are related to certain metrics considered by Guillemin in his study of Kähler structures on toric varieties , as will be explained.
Since the metric on the polytope is very explicitly computed, this allows conclusions to be drawn about the existence of complete Bochner-Kähler metrics based on the geometry of the polytopes. In Proposition 8, I show that if there is a complete Bochner-Kähler metric whose moduli image lies in $`[v]`$, then $`[v]`$ must be bounded (which turns out to be the same as saying that its corresponding polytope is bounded). Essentially, it turns out that when $`[v]`$ is unbounded, any attempt to ‘analytically continue’ the metric to a maximal domain will run into curvature blow-up at finite distance. Since there are very few $`[v]`$ that are bounded, this considerably narrows the search for complete examples.
On the other hand, Proposition 9 shows that if $`[v]`$ is compact but is not a single point, then there is no complete Bochner-Kähler manifold whose moduli image lies in $`[v]`$. In this case, the problem is not curvature blow-up but is, instead, the presence of essential orbifold singularities in any attempted completion.
A corollary of Proposition 9 is that the only compact Bochner-Kähler manifolds are the compact quotients of the known symmetric ones. This result renders vacuous or trivial many of the results in the literature about Bochner-Kähler metrics. For example, the only Kähler $`n`$-manifold satisfying the conditions of \[5, Theorems 8.25 and 8.26\] is $`^n`$ endowed with a constant multiple of the Fubini-Study metric. The conclusions of these theorems (which concern the vanishing of various cohomology groups) are trivial for these manifolds.
Theorem 9 provides explicit models for the Bochner-Kähler metrics in dimension $`n`$ that are of cohomogeneity $`n`$ (i.e., the *least* symmetric ones) on the regular locus. It constructs, for each $`n`$-dimensional class $`[v]F_n`$, a Bochner-Kähler metric on $`\iota _v\left([v]\right)^{}\times ^n`$ with the following universal embedding property: If $`(M,g)`$ is a Bochner-Kähler $`n`$-manifold with $`f(M)[v]`$, then the universal cover $`\stackrel{~}{M^{}}`$ can be isometrically immersed into $`\iota _v\left([v]\right)^{}\times ^n`$, lifting the momentum submersion $`h:M^{}\iota _v\left([v]\right)^{}`$. Completeness issues can then be addressed by studying the model metric on $`\iota _v\left([v]\right)^{}\times ^n`$. These metrics are closely related to the metrics studied in and . In particular, Abreu’s results in can be generalized to show that the above metrics are actually extremal in the sense of Calabi.
Theorem 10 provides a contractible $`n`$-parameter family of complete Bochner-Kähler metrics on $`^n`$ and proves that every simply-connected, complete Bochner-Kähler manifold that is not homogeneous is isometric to a unique member of this family.
Thus, the set of complete Bochner-Kähler manifolds is very restricted. However, if one is willing to consider orbifolds, it turns out that there are many nontrivial complete Bochner-Kähler metrics on orbifolds. I include some discussion of these at the end of the article. In fact, by Theorem 11, every weighted projective space carries a Bochner-Kähler metric,<sup>4</sup><sup>4</sup>4A natural guess would be that this metric is the one that comes by symplectic reduction from the standard metric on $`^{n+1}`$ via the weighted $`S^1`$-action that defines the weighted projective space. However, this ‘reduced’ metric is never Bochner-Kähler except in the case of equal weights. presumably unique up to constant multiples, though I have not shown this. For example, the Fubini-Study metric is, up to isometry and constant multiples, the unique Bochner-Kähler metric on $`^n`$. For more detail on this, see §4.3.2 and §4.4.6.
Finally, in §5, I collect some miscellaneous and incidental remarks about generalizations and related problems. In particular, I comment on how this work in the dimension 2 case is related to the recent work of Apostolov and Gauduchon that classifies the self-dual Hermitian Einstein metrics in (real) dimension 4 and use the normal forms constructed in this article to produce the first known complete examples of such metrics that are of cohomogeneity 2 (the maximum possible, as it turns out).
## 2. The Structure Equations of Bochner-Kähler Metrics
First, some standard notation. Let $`^n`$ (thought of as columns of height $`n`$ whose entries are complex numbers) be endowed with its usual Hermitian inner product, in which $`z,w={}_{}{}^{t}\overline{z}w`$ for all $`w,z^n`$. Let $`\mathrm{U}(n)M_n()`$ denote the group of unitary matrices and let $`𝔲(n)M_n()`$ denote its Lie algebra, i.e., the space of skew-Hermitian $`n`$-by-$`n`$ matrices. As is customary, the conjugate transpose operation will be denoted by a superscript asterisk. Thus, $`z,w=z^{}w`$, and $`aM_n()`$ lies in $`𝔲(n)`$ if and only if $`a^{}=a`$.
### 2.1. The unitary coframe bundle
Let $`(M,g,\mathrm{\Omega })`$ be a Kähler manifold, i.e., $`M`$ is an $`n`$-dimensional complex manifold and $`g`$ is an Hermitian metric on $`M`$ whose associated Kähler 2-form $`\mathrm{\Omega }`$ is closed. As is customary, let $`J:TMTM`$ be the associated almost complex structure endomorphism.
For $`xM`$, let $`P_x`$ be the set of unitary isomorphisms $`u:T_xM^n`$. Then $`P=_{xM}P_x`$ is a principal right $`\mathrm{U}(n)`$-bundle over $`M`$, with the basepoint projection $`\pi :PM`$ given by $`\pi \left(P_x\right)=x`$ and $`\mathrm{U}(n)`$-action given by $`ua=a^1u`$ for $`a\mathrm{U}(n)`$.
#### 2.1.1. The first and second structure equations
Let $`\omega `$ be the $`^n`$-valued $`1`$-form on $`P`$ defined by the rule $`\omega (v)=u\left(\pi ^{}(v)\right)`$ for all $`vT_uP`$. Then $`\pi ^{}\mathrm{\Omega }=\frac{i}{2}\omega ^{}\omega `$.
Because the structure $`(M,g,\mathrm{\Omega })`$ is Kählerian, there exists a unique $`𝔲(n)`$-valued 1-form $`\varphi `$ on $`P`$ satisfying the *first structure equation* of É. Cartan,
(2.1)
$$d\omega =\varphi \omega .$$
The *second structure equation* of É. Cartan takes the form
(2.2)
$$d\varphi =\varphi \varphi +\frac{1}{2}R(\omega \omega ^{}),$$
where $`R:P\mathrm{Hom}(𝔲(n),𝔲(n))`$ is the *Kähler curvature function*. The adjoint representation of $`\mathrm{U}(n)`$ on $`𝔲(n)`$ induces a representation $`\rho `$ of $`\mathrm{U}(n)`$ on $`\mathrm{Hom}(𝔲(n),𝔲(n))`$. The curvature function $`R`$ is equivariant with respect to this action, i.e., $`R(ua)=\rho (a^1)\left(R(u)\right)`$ for $`a\mathrm{U}(n)`$.
The first Bianchi identity is $`0=d(d\omega )=R(\omega \omega ^{})\omega `$. Thus, $`R`$ takes values in the subspace $`𝒦\left(𝔲(n)\right)`$ consisting of those elements $`r\mathrm{Hom}(𝔲(n),𝔲(n))`$ that satisfy
$$r(xy^{}yx^{})z+r(yz^{}zy^{})x+r(zx^{}xz^{})y=0,x,y,z^n.$$
#### 2.1.2. Tensors, vector fields, and symmetries
The reader will recall that any (real or complex) representation $`\chi :\mathrm{U}(n)\mathrm{Aut}(V)`$ defines a (tensor) vector bundle $`P_\chi =P\times _\chi V`$ over $`M`$. A section $`\sigma `$ of $`P_\chi `$ is then uniquely defined by a function $`s:PV`$ that satisfies the equivariance condition $`s(ua)=\chi (a^1)\left(s(u)\right)`$ for all $`uP`$ and $`a\mathrm{U}(n)`$ and $`\sigma (x)=[u,s(u)]_\chi `$ for some (and hence any) $`uP_x`$. The function $`s`$ is said to *represent* $`\sigma `$. For example, $`R`$ represents the Kähler curvature tensor.
For notational simplicity, I will use $`\chi `$ also to denote the induced map on Lie algebras; thus, $`\chi :𝔲(n)\mathrm{End}(V)`$. The $`\mathrm{U}(n)`$-equivariance of a representative function $`s:PV`$ implies that the $`1`$-form $`ds+\chi (\varphi )s`$ is $`\pi `$-semibasic. Thus, there exists a linear mapping $`Ds:P\mathrm{Hom}_{}(^n,V)`$ satisfying
$$ds+\chi (\varphi )s=Ds(\omega ).$$
Naturally, $`Ds`$ represents the covariant derivative of the section $`\sigma `$ represented by $`s`$.
For example, the standard inclusion $`\iota :\mathrm{U}(n)\mathrm{Aut}(^n)`$ yields $`P_\iota TM`$. A vector field $`Z`$ on $`M`$ is represented by the function $`z:P^n`$ defined by $`z(u)=u\left(Z_{\pi (u)}\right)`$. Now, $`\mathrm{Hom}_{}(^n,^n)=\mathrm{Hom}_{}(^n,^n)\mathrm{Hom}_{}(^n,^n)C`$ where $`C:^n^n`$ is conjugation. Thus, since $`\mathrm{Hom}_{}(^n,^n)=M_n()`$, there are functions $`z^{}`$ and $`z^{\prime \prime }`$ on $`P`$ with values in $`M_n()`$ so that
$$dz+\varphi z=z^{}\omega +z^{\prime \prime }\overline{\omega }.$$
These functions have the $`\mathrm{U}(n)`$-equivariance
$$z^{}(ua)=a^1z^{}(u)a,z^{\prime \prime }(ua)=a^1z^{\prime \prime }(u)\overline{a}$$
and thus represent tensors on $`M`$. In fact, $`z^{}`$ represents $`^{1,0}(ZiJZ)`$ while $`z^{\prime \prime }`$ represents $`^{0,1}(ZiJZ)=\overline{}(ZiJZ)`$.
In particular, $`Z`$ is the real part of a holomorphic vector field, namely $`ZiJZ`$, if and only if $`z^{\prime \prime }=0`$. Moreover, computation shows that
$$\pi ^{}\left(Z\text{ }\text{ }\mathrm{\Omega }\right)=\frac{i}{2}(z^{}\omega \omega ^{}z),$$
implying, in particular, that
$$\pi ^{}\left(𝔏_Z\mathrm{\Omega }\right)=\frac{i}{2}\omega ^{}\left(z^{}+(z^{})^{}\right)\omega \frac{i}{2}\omega ^{}z^{\prime \prime }\overline{\omega }\frac{i}{2}\overline{\omega }^{}(z^{\prime \prime })^{}\omega .$$
Thus, the flow of $`Z`$ is both holomorphic and symplectic (and hence an infinitesimal symmetry of the Kähler structure) if and only if $`z^{\prime \prime }=0`$ and $`z^{}+(z^{})^{}=0`$. In such a case, $`Z=\pi ^{}(Z^{})`$ where $`Z^{}`$ is the vector field on $`P`$ that satisfies
$$\omega (Z^{})=z,\varphi (Z^{})=z^{}.$$
The flow of $`Z^{}`$ preserves both $`\omega `$ and $`\varphi `$. In fact,
$$𝔏_Z^{}\omega =d\left(\omega (Z^{})\right)+Z^{}\text{ }\text{ }(\varphi \omega )=dz+\varphi zz^{}\omega =0,$$
so the flow of $`Z^{}`$ does indeed preserve $`\omega `$. Moreover, since $`\varphi `$ is the unique $`𝔲(n)`$-valued $`1`$-form that satisfies $`d\omega =\varphi \omega `$, the flow of $`Z^{}`$ must preserve $`\varphi `$ as well.
Conversely, any vector field on $`P`$ whose flow preserves both $`\omega `$ and $`\varphi `$ is of the form $`Z^{}`$ where $`Z`$ is a symmetry vector field of the Kähler structure.
If $`Z`$ is a symmetry vector field of the Kähler structure and $`Z`$ vanishes at $`xM`$, then $`Z(x)T_xMT_x^{}M`$ is both skew-symmetric and commutes with the complex structure $`J_x`$. Moreover, the flow $`\mathrm{\Phi }_Z`$ of $`Z`$ is complete on the open geodesic ball $`B_\delta (x)`$ for all sufficiently small $`\delta >0`$ and is isometric there. Let $`z:P^n`$ represent $`Z`$. Then $`z(u)=0`$ for all $`uP_x`$ and, by the above discussion, $`z^{}(u)`$ belongs to $`𝔲(n)`$. In particular, the linear transformation $`a=u^1z^{}(u)u:T_xMT_xM`$ is a well-defined skew-Hermitian transformation of $`T_xM`$.
Then, for all $`vT_xM`$ with $`|v|<\delta `$,
$$\mathrm{\Phi }_Z(t,\mathrm{exp}_x(v))=\mathrm{exp}_x\left(e^{at}v\right),$$
i.e., the map $`\mathrm{exp}_x:B_\delta (0_x)B_\delta (x)`$ intertwines the linear 1-parameter subgroup action on $`T_xM`$ generated by exponentiating $`a`$ with the flow of $`Z`$.
This has two consequences that will be needed in this article (see §4.3.3). First, exponentiating the kernel of $`a`$ gives the component of the fixed locus of the flow of $`Z`$ that passes through $`x`$, which is therefore a totally geodesic complex submanifold of $`M`$. Second, when $`M`$ is connected, the flow of $`Z`$ will be periodic of period $`T`$ if and only if the eigenvalues of $`z^{}(u)`$ generate the discrete subgroup of $`i`$ that consists of the integral multiples of $`2\pi i/T`$.
A ‘micro-local’ version of symmetry will be useful. Two coframes $`u,vP`$ are said to be *equivalent* if there is a connected $`u`$-neighborhood $`U`$, a connected $`v`$-neighborhood $`V`$ and a diffeomorphism $`\psi :UV`$ that satisfies $`p(u)=v`$ and $`p^{}(\omega _V)=\omega _U`$. (It follows, as a consequence, that $`p^{}(\varphi _V)=\varphi _U`$.) Such a $`p`$, when it exists, is unique once $`U`$ is specified and is locally of the form $`p(w)=w(\overline{p}^{})^1`$ for some local isomorphism $`\overline{p}:\pi (U)\pi (V)`$ of the Kähler structure on $`M`$.
If $`u`$ and $`v`$ are equivalent, then $`R(u)=R(v)`$; in fact, $`D^kR(u)=D^kR(v)`$ for all $`k0`$.<sup>5</sup><sup>5</sup>5The converse is not generally true, though it is when the Kähler structure is real-analytic. Let $`\mathrm{\Gamma }P\times P`$ consist of the equivalent pairs. Then the set $`\overline{\mathrm{\Gamma }}=\mathrm{\Gamma }/\mathrm{U}(n)`$ (where the $`\mathrm{U}(n)`$-action is the diagonal one on $`P\times P`$) can be identified with the set of pointed local isomorphisms of the Kähler structure on $`M`$. For want of a better name, I will refer to $`\overline{\mathrm{\Gamma }}`$ as the *symmetry pseudo-groupoid* of the Kähler structure.
For any $`x`$, the set $`\overline{\mathrm{\Gamma }}x`$ is defined to consist of the points $`\pi (v)`$ where $`\pi (u)=x`$ and $`(u,v)`$ lies in $`\mathrm{\Gamma }`$. Thus, $`\overline{\mathrm{\Gamma }}xM`$ consists of the points $`yM`$ about which the Kähler structure is locally isomorphic to the Kähler structure about $`x`$. Even though $`\overline{\mathrm{\Gamma }}`$ is not a group, I will, by an extension of the usual language, refer to $`\overline{\mathrm{\Gamma }}x`$ as the *$`x`$-orbit* of the symmetry pseudo-groupoid of the Kähler structure. For any $`xM`$, the $`x`$-orbit is a smooth (but not necessarily closed) submanifold of $`M`$.
The subset $`\overline{\mathrm{\Gamma }}_x=\left(\mathrm{\Gamma }(P_x\times P_x)\right)/\mathrm{U}(n)`$ actually is a group in a natural way, canonically represented as a closed subgroup of $`\mathrm{U}(T_xM)`$ as the (local) rotations about $`x`$ that preserve the metric and complex structure. This group will be known as the *stabilizer* of $`x`$.
#### 2.1.3. Curvature decomposition
Now, the curvature representation $`𝒦\left(𝔲(n)\right)`$ is a $`\mathrm{U}(n)`$-invariant subspace of $`\mathrm{Hom}(𝔲(n),𝔲(n))`$. It is known that $`𝒦\left(𝔲(n)\right)`$ is isomorphic as a $`\mathrm{U}(n)`$-module to $`S_{}^{2,2}(^n)=\left(S^{2,0}(^n)_{}S^{0,2}(^n)\right)_{}`$, the real-valued quartic functions on $`^n`$ that are complex quadratic and complex conjugate quadratic.
Now, for each $`p>0`$, the $`\mathrm{U}(n)`$-invariant Hermitian inner product on $`^n`$ induces a surjective $`\mathrm{U}(n)`$-equivariant ‘trace’ (also called a ‘contraction’ or ‘Laplacian’)
$$\mathrm{tr}:S_{}^{p,p}(^n)S_{}^{p1,p1}(^n).$$
Its kernel $`S_{,0}^{p,p}(^n)S_{}^{p,p}(^n)`$ is an irreducible $`\mathrm{U}(n)`$-module .
It follows that there is an isomorphism of $`\mathrm{U}(n)`$-modules
(2.3)
$$𝒦\left(𝔲(n)\right)S_{}^{2,2}(^n)S_{,0}^{1,1}(^n)S_{,0}^{2,2}(^n),$$
where the $`\mathrm{U}(n)`$-irreducible modules on the right hand side have (real) dimensions $`1`$, $`n^21`$, and $`\frac{1}{4}n^2(n1)(n+3)`$, respectively. Thus, there are unique $`\mathrm{U}(n)`$-invariant subspaces $`𝒦_i𝒦\left(𝔲(n)\right)`$ satisfying $`𝒦_iS_{}^{i,i}(^n)`$ for $`i=0`$, $`1`$, and $`2`$.
The Kähler curvature function $`R`$ can therefore be written as a sum
$$R=R_0+R_1+R_2$$
where $`R_i`$ takes values in $`𝒦_i`$ and represents a section of the bundle $`S_{}^{i,i}(TM)`$, i.e., a tensor associated to the Kähler structure $`\mathrm{\Omega }`$.
The function $`R_0`$ represents the scalar curvature, $`R_1`$ represents the traceless Ricci tensor, and $`R_2`$ represents the *Bochner tensor*, identified in 1949 by S. Bochner . When $`n=1`$, both $`R_1`$ and $`R_2`$ are zero by definition, but when $`n2`$, all three tensors are nonzero for the generic Kähler metric.
The Kähler structures for which $`R_0`$ vanishes are the scalar-flat Kähler structures. When $`n2`$, those for which $`R_1`$ vanishes are the Kähler-Einstein structures and those for which $`R_2`$ vanishes are known as *Bochner-Kähler* structures.
###### Remark 1 (The Riemannian analogy).
Bochner’s decomposition of the Kähler curvature bears a resemblance to the more familiar decomposition of the Riemann curvature tensor of a Riemannian metric into the scalar curvature, the traceless Ricci tensor, and the Weyl curvature tensor. However, this resemblance is somewhat misleading.
While the scalar curvature and the Ricci curvature in the two cases do correspond, the Weyl curvature tensor of a Kähler metric is not simply the Bochner curvature tensor. For example, when $`n=2`$, so that the underlying manifold has dimension $`4`$ and is canonically oriented, the Bochner tensor turns out to be $`W^{}`$, the anti-self-dual part of the Weyl curvature. Thus, in complex dimension $`2`$, the Bochner-Kähler metrics are the same as the self-dual Kähler metrics.<sup>6</sup><sup>6</sup>6These metrics have been studied from this point of view. For example, see and the forthcoming . For further comments on this relationship, see §5.3.
Bochner observed that the Weyl curvature of a Kähler metric breaks up into two or three irreducible components under the action of $`\mathrm{U}(n)\mathrm{O}(2n)`$, one of which is the Bochner curvature tensor. One of the other components is equivalent to the scalar curvature while, when $`n>2`$, another is equivalent to the traceless Ricci curvature. Thus, when $`n>2`$, the vanishing of the Weyl curvature of a Kähler metric implies that the metric is flat. In particular, when $`n>2`$, a conformally flat Kähler metric is flat. When $`n=2`$, the conformal flatness of a Kähler metric implies only that the structure is Bochner-Kähler, with vanishing scalar curvature.<sup>7</sup><sup>7</sup>7However, as will be seen in Example 1 below, when $`n=2`$ there are essentially only two conformally flat Kähler structures up to local isomorphism and homothety.
### 2.2. Explicit Bochner-Kähler structures
Few explicit examples of Bochner-Kähler structures have been found up to now. The main strategy for constructing examples so far has been to look for examples that satisfy conditions sufficiently stringent to reduce the construction to an ODE problem.
###### Example 1 (Locally symmetric).
The simplest Bochner-Kähler metric is the complex $`n`$-dimensional space $`M_c^n`$ of constant holomorphic sectional curvature $`c`$. (In fact, $`R_1=R_2=0`$ characterizes these metrics.)
Tachibana and Liu \[27, §2\] showed that the products $`M_c^p\times M_c^{np}`$ are Bochner-Kähler for any $`n`$, $`p`$, and $`c`$. Moreover, they showed that any Bochner-Kähler structure that is a product in a nontrivial way is locally isomorphic to $`M_c^p\times M_c^{np}`$.
Matsumoto \[18, Theorem 2\] proved that a Bochner-Kähler structure with constant scalar curvature is locally symmetric. Matsumoto and Tanno then proved that any locally symmetric Bochner-Kähler structure is locally isomorphic to one of the above examples. (For a simple proof, see Proposition 1 below.)
Note that their results, combined with the preceding remark, imply the well-known result that the only conformally flat Kähler structures in dimension $`n=2`$ are those that are locally isometric to $`M_c^1\times M_c^1`$ for some $`c0`$.
###### Example 2 (Rotationally symmetric).
The first examples with nonconstant scalar curvature appear to be due to Tachibana and Liu , who considered Kähler structures of the form
(2.4)
$$\mathrm{\Omega }=\frac{i}{2}\overline{}f\left(|z|^2\right)=\frac{i}{2}dz^{}\left[f^{}\left(|z|^2\right)\mathrm{I}_n+f^{\prime \prime }\left(|z|^2\right)zz^{}\right]dz$$
where $`f`$ is smooth and real-valued on some interval $`I`$. The $`(1,1)`$-form $`\mathrm{\Omega }`$ is positive on $`D=\{z^n|z|^2I\}`$ if and only if $`f^{}(t)+tf^{\prime \prime }(t)>0`$ and $`f^{}(t)>0`$ (when $`n>1`$) for all $`tI[0,\mathrm{})`$.
For $`n2`$, they showed that $`\mathrm{\Omega }`$ is Bochner-Kähler on $`D`$ if and only if $`f^{}`$ satisfies
(2.5)
$$f^{\prime \prime }(t)=\left(atf^{}(t)+k\right)f^{}(t)^2$$
for some constants $`a`$ and $`k`$.<sup>8</sup><sup>8</sup>8 While equation (2.5) makes sense even when $`n=1`$, ‘Bochner-Kähler’ has not yet been defined for $`n=1`$. This will be remedied in §2.3.6 in such a way that the present discussion extends without change to the case $`n=1`$.
For such an $`\mathrm{\Omega }`$, the eigenvalues of $`\mathrm{Ric}(\mathrm{\Omega })`$ with respect to $`\mathrm{\Omega }`$ are
$$\begin{array}{cc}\hfill \rho _1& =2(n+1)k2(n+2)a|z|^2f^{}\left(|z|^2\right),\hfill \\ \hfill \rho _2& =2(n+1)k4(n+2)a|z|^2f^{}\left(|z|^2\right),\hfill \end{array}$$
with $`\rho _1`$ having multiplicity $`n1`$, representing the $`(n1)`$-plane orthogonal to the radial direction, and $`\rho _2`$ having multiplicity $`1`$, representing the radial direction. Thus, the solutions of (2.5) for which $`a0`$ yield Bochner-Kähler structures that are not homogeneous.
Tachibana and Liu integrated the above equation when $`k=0`$, thereby giving explicit examples of Bochner-Kähler structures that are not homogeneous. They do not discuss completeness issues, but it is evident from their formulae that none of their explicit examples are complete.
#### 2.2.1. Further analysis
Now, (2.5) can be integrated even when $`k0`$. Set $`x(t)=tf^{}(t)`$, so that (2.5) becomes
(2.6)
$$tx^{}(t)=x(t)\left(1+kx(t)+ax(t)^2\right).$$
Admissible solutions must satisfy $`x>0`$ when $`t>0`$ and $`x^{}=f^{}+tf^{\prime \prime }>0`$. Now, (2.6) can be integrated by separation of variables
(2.7)
$$\frac{dx}{x(1+kx+ax^2)}=\frac{dt}{t}.$$
*Scaling equivalences.* Relation (2.7) is invariant under scaling $`t`$, which corresponds geometrically to homothety in $`^n`$. Thus, solutions of (2.7) that differ by constant scaling in $`t`$ represent isomorphic Kähler structures and can be regarded as equivalent. Similarly, multiplying $`x`$ by a positive constant corresponds to multiplying the Kähler form $`\mathrm{\Omega }`$ by that constant, so solutions for a given pair of constants $`(k,a)`$ can be regarded as equivalent to the solutions for any other pair $`(\lambda k,\lambda ^2a)`$ with $`\lambda ^+`$.
*The two types of solutions.* For any fixed $`(k,a)^2`$, let $`J_{k,a}`$ be the maximal $`x`$-interval containing $`0`$ on which $`(1+kx+ax^2)`$ is positive. Define a positive function $`F`$ on $`J_{k,a}`$ by the formula
$$\mathrm{log}F(x)=_0^x\frac{k+a\xi }{(1+k\xi +a\xi ^2)}𝑑\xi .$$
One solution to (2.7) can then be written implicitly in the form
$$xF(x)=t.$$
The expression on the left hand side of this equation defines a function on $`J_{k,a}`$ that has positive derivative and that vanishes at $`x=0`$. Let $`I_{k,a}`$ denote the range of this function. (More will be said about this range below.) The above relation can then be solved for $`x`$, yielding a real-analytic solution $`x:I_{k,a}J_{k,a}`$ to (2.6).
This solution satisfies $`x^{}(0)=1`$. Any other solution to (2.6) whose range lies in $`J_{k,a}`$ differs from this one by scaling in $`t`$. These solutions will be said to be of *type one*.
When $`a>0`$ and $`k2\sqrt{a}`$, there is a second, geometrically distinct, admissible solution to (2.6). Under these assumptions, let $`J_{k,a}^{}`$ be the interval $`(p,\mathrm{})`$ where
$$p=\frac{k+\sqrt{k^24a}}{2a}>0$$
is the larger root of $`(1+kp+ap^2)=0`$. (When the two roots are equal, $`p`$ is simply the root.) Define a function $`F^{}`$ on $`J_{k,a}^{}`$ by the formula
$$\mathrm{log}F^{}(x)=_x^{\mathrm{}}\frac{1}{\xi (1+k\xi +a\xi ^2)}𝑑\xi .$$
Then (2.7) can be integrated in the form $`F^{}(x)=t`$. Since the integral diverges to infinity as $`x`$ approaches $`p`$ from above, the function $`F^{}`$ maps $`(p,\mathrm{})`$ diffeomorphically onto $`(0,1)`$. Thus, the equation $`F^{}(x)=t`$ can be solved for $`x`$, yielding a real analytic solution $`x^{}:(0,1)(p,\mathrm{})`$ to (2.6). Any other solution to (2.6) whose range lies in $`J_{k,a}^{}`$ differs from this one by scaling in $`t`$. These solutions will be said to be of *type two*.
*Completeness.* For any admissible solution $`x:IJ`$ to (2.6), consider the Bochner-Kähler structure $`\mathrm{\Omega }`$ got by setting $`f^{}(t)=x(t)/t`$. The differential of arc length $`\sigma `$ along a radial curve $`\{svs^2I\}`$ for any fixed $`v^n`$ with $`|v|=1`$ can be calculated to be
(2.8)
$$d\sigma =\frac{d\left(x(s^2)\right)}{2\sqrt{x(s^2)\left(1+kx(s^2)+ax(s^2)^2\right)}}.$$
This formula permits an analysis of the completeness properties of $`\mathrm{\Omega }`$ without having to write down an explicit formula for $`x`$.
When $`a=0`$, the solution of type one is $`x(t)=t/(1kt)`$ and $`\mathrm{\Omega }`$ has constant holomorphic sectional curvature. This metric is complete on $`^n`$ when $`k=0`$. When $`k>0`$, it is complete on the ball $`|z|^2<k^1`$. When $`k<0`$ it is not complete, since the radial arc length
$$_0^{k^1}\frac{d\xi }{2\sqrt{\xi (1+k\xi )}}$$
is finite. However, in this case, the metric on $`^n`$ extends smoothly to (a multiple of) the Fubini-Study metric on $`^n`$.
When $`a>0`$ and $`1+kp+ap^2=0`$ has no positive root $`p`$, the interval $`J_{k,a}`$ contains some interval of the form $`(\alpha ,\mathrm{})`$ for $`\alpha <0`$. Because the integral
$$_1^{\mathrm{}}\frac{d\xi }{\xi (1+k\xi +a\xi ^2)}$$
converges, the type one solution to (2.6) is defined on an interval $`I_{k,a}=(\delta ,R^2)`$ for $`\delta `$ and $`R^2`$ positive, with $`x(t)`$ tending to infinity as $`t`$ approaches $`R^2`$. Thus, $`\mathrm{\Omega }`$ is defined and nondegenerate on a ball $`|z|<R`$. Since
$$_0^{\mathrm{}}\frac{d\xi }{2\sqrt{\xi (1+k\xi +a\xi ^2)}}<\mathrm{}$$
the metric is not complete. Yet, $`\mathrm{\Omega }`$ cannot be extended beyond $`|z|<R`$ because the two curvatures $`\rho _1`$ and $`\rho _2`$ tend to $`\mathrm{}`$ as $`|z|`$ approaches $`R`$.
Suppose now that $`1+kp+ap^2=0`$ does have at least one positive root. By the $`x`$-scaling argument, it can be assumed that $`p=1`$ is a root and that there is no root in the interval $`(0,1)`$. Thus, $`k=(1+a)`$, so that $`(1+kp+ap^2)=(1p)(1ap)`$, and $`a1`$.
Suppose first that $`a=1`$ (the extreme value), so that $`(1+kp+ap^2)=(1p)^2`$. Since the integral
$$_0^1\frac{d\xi }{\xi (1\xi )^2}$$
diverges at both endpoints, the type one solution $`h`$ to (2.6) is defined on all of $``$ and maps $`[0,\mathrm{})`$ to $`[0,1)`$. Thus, $`\mathrm{\Omega }`$ is defined and nondegenerate on all of $`^n`$. Since
$$_0^1\frac{d\xi }{2\sqrt{\xi (1\xi )^2}}=\mathrm{},$$
this metric is complete on $`^n`$. As $`|z|^2`$ goes to infinity, the curvatures $`\rho _1`$ and $`\rho _2`$ approach $`2n`$ and $`4`$, respectively.
Still assuming $`(1+kp+ap^2)=(1p)^2`$, consider the type two solution to (2.6). The form $`\mathrm{\Omega }`$ is defined and nondegenerate on the punctured ball $`0<|z|<1`$. The arc length integral shows that this metric is complete on a neighborhood of the puncture but not complete near the boundary $`|z|=1`$. Since $`x`$ goes to infinity as $`|z|`$ tends to $`1`$, the curvatures $`\rho _1`$ and $`\rho _2`$ tend to $`\mathrm{}`$ near this boundary. Thus, $`\mathrm{\Omega }`$ cannot be extended beyond the punctured ball $`0<|z|<1`$.
Now suppose $`a<1`$. The integral
$$_0^1\frac{d\xi }{\xi (1\xi )(1a\xi )}$$
still diverges at both endpoints, so the type one solution to (2.6) is defined on an open interval in $``$ that contains $`[0,\mathrm{})`$. Moreover, $`x`$ maps $`[0,\mathrm{})`$ to $`[0,1)`$. Again, $`\mathrm{\Omega }`$ is defined and positive definite on all of $`^n`$. However, now, the elliptic integral
$$_0^1\frac{d\xi }{2\sqrt{\xi (1\xi )(1a\xi )}}$$
is finite, so the metric is not complete. The curvatures $`\rho _1`$ and $`\rho _2`$ approach the limits $`2(n+1)2a`$ and $`2(n+1)(1a)`$, respectively, as $`|z|^2`$ goes to infinity. It can be shown that this Bochner-Kähler structure extends to an ‘orbifold’ Bochner-Kähler structure on $`^n`$ even when $`a0`$. I will not discuss this extension here since its nature will be more clear after the considerations to be taken up in the next section. Unless $`a=0`$ (the Fubini-Study case), this is not a homogeneous metric.
Finally, when $`0<a<1`$, consider the type two solution to (2.6), whose range is $`(a^1,\mathrm{})`$. Since the integral
$$_{a^1+1}^{\mathrm{}}\frac{d\xi }{\xi (1\xi )(1a\xi )}$$
converges, the domain of this solution is $`(0,1)`$. Then $`\mathrm{\Omega }`$ is defined and nondegenerate on the punctured unit ball $`0<|z|<1`$. When $`0<a<1`$, the elliptic integral
$$_{a^1}^{\mathrm{}}\frac{d\xi }{2\sqrt{\xi (1\xi )(1a\xi )}}$$
is finite, so the metric is not complete at either the puncture or the boundary. The curvatures $`\rho _1`$ and $`\rho _2`$ approach $`\mathrm{}`$ as $`|z|`$ approaches $`1`$. However, these curvatures remain bounded and approach a limit when $`z`$ approaches $`0`$. The nature of the singularity at $`|z|=0`$ and whether or not it can be removed will be discussed in §4.
*Conclusion.* Up to constant multiples and scaling, the Ansatz of Tachibana and Liu provides exactly one example of a complete Bochner-Kähler metric (on $`^n`$) that is not locally symmetric.
###### Example 3 (Ejiri metrics).
Ejiri considered a somewhat more general Ansatz, seeking Bochner-Kähler metrics for which the Ricci tensor has at most two distinct eigenvalues, an evident property of the Tachibana-Liu examples and the locally symmetric examples. He showed that when $`n3`$, such examples that are not locally symmetric have cohomogeneity one and that the isometry stabilizer of the general point is $`\mathrm{U}(n1)\mathrm{U}(n)`$. Thus, the problem of describing these examples reduces to an ODE problem, which Ejiri integrated up to a Weierstraß-type equation, thereby producing the desired examples.
In \[Ej,§4\], Ejiri remarked that none of his examples (aside from the locally symmetric ones) were known to be complete. However, since the Tachibana-Liu examples are special cases of his examples, at least one of his examples is complete. In fact, Ejiri’s example in \[10, §4\] of a complete, $`C^2`$ Bochner-Kähler metric on $`^{2n}`$ turns out to be the complete example of Tachibana and Liu on $`^n`$, but presented in unusual coordinates in which it is not fully regular at the origin. This will become apparent in §4.4, when all of the complete Bochner-Kähler metrics in dimension $`n`$ will be classified.
### 2.3. The differential analysis
Now suppose that $`M`$ is a complex manifold of complex dimension $`n2`$ endowed with a Bochner-Kähler structure $`\mathrm{\Omega }`$. As before, let $`\pi :PM`$ be the unitary coframe bundle of $`\mathrm{\Omega }`$ and denote its canonical forms by $`\omega `$, with values in $`^n`$ and $`\varphi `$, with values in $`𝔲(n)`$. Let $`R:P𝒦(𝔲(n))`$ be the Kähler curvature function.
#### 2.3.1. Simplification of the curvature
By definition, $`\mathrm{\Omega }`$ is Bochner-Kähler if and only if $`R_2`$ vanishes identically. The curvature decomposition of §2.1.3 shows that the remaining part of $`R`$ takes values in a representation isomorphic to $`S_{}^{1,1}(^n)`$, the Hermitian symmetric quadratic forms. Now, for any function $`S=S^{}:Pi𝔲(n)M_n()`$, the $`2`$-form
$$\mathrm{\Phi }=S\omega ^{}\omega S\omega \omega ^{}\omega \omega ^{}S+\omega ^{}S\omega \mathrm{I}_n$$
takes values in $`𝔲(n)`$ and satisfies $`\mathrm{\Phi }\omega =0`$ (which is the first Bianchi identity). Moreover, $`\mathrm{\Phi }`$ vanishes if and only if $`S`$ vanishes.
It follows that the assumption that $`\mathrm{\Omega }`$ be Bochner-Kähler is equivalent to the existence of a function $`S:Pi𝔲(n)M_n()`$ for which
(2.9)
$$d\varphi +\varphi \varphi =S\omega ^{}\omega S\omega \omega ^{}\omega \omega ^{}S+\omega ^{}S\omega \mathrm{I}_n.$$
Now, $`S`$ does not represent the Ricci tensor *per se*. However, the identity
$$\pi ^{}\left(\mathrm{Ric}(\mathrm{\Omega })\right)=i\mathrm{tr}(d\varphi +\varphi \varphi )=i\left(\mathrm{tr}(S)\omega ^{}\omega +(n+2)\omega ^{}S\omega \right)$$
shows how $`S`$ is related to the Ricci form. In particular, the scalar curvature of the underlying metric is $`2\mathrm{tr}_\mathrm{\Omega }\left(\mathrm{Ric}(\mathrm{\Omega })\right)=8(n+1)\mathrm{tr}S`$.
#### 2.3.2. Higher Bianchi identities
Now, consider the consequences of differentiating (2.9). Setting $`\sigma =dS+\varphi SS\varphi `$ and taking the exterior derivative of (2.9) leads to the identity
$$\sigma \omega ^{}\omega \sigma \omega \omega ^{}\omega \omega ^{}\sigma \omega ^{}\sigma \omega \mathrm{I}_n=0.$$
This, coupled with the evident identity $`\sigma =\sigma ^{}`$ implies, by a straightforward variant of Cartan’s Lemma, that there must exist a function $`T:P^n`$ so that
(2.10)
$$dS+\varphi SS\varphi =\sigma =T\omega ^{}+\omega T^{}+\frac{1}{2}(T^{}\omega +\omega ^{}T)\mathrm{I}_n.$$
(Equation (2.10) is the second Bianchi identity for Bochner-Kähler structures.)
Setting $`\tau =dT+\varphi TS^2\omega `$ and computing the exterior derivative of (2.10) yields
$$\tau \omega ^{}\omega \tau ^{}+\frac{1}{2}(\tau ^{}\omega \omega ^{}\tau )\mathrm{I}_n=0.$$
By another variant of Cartan’s Lemma, there is a function $`U:P`$ so that
(2.11)
$$dT+\varphi TS^2\omega =\tau =U\omega .$$
(This might be thought of as a sort of *third* Bianchi identity.)
Finally, setting $`\upsilon =dU(T^{}S\omega +\omega ^{}ST)`$ and differentiating (2.11) yields $`\upsilon \omega =0`$, implying that $`\upsilon =0`$, i.e., that
(2.12)
$$dU=T^{}S\omega +\omega ^{}ST.$$
(This is a *fourth* Bianchi identity.) The exterior derivative of (2.12) is an identity.
The collection of formulae
(2.13)
$$\begin{array}{cc}\hfill d\omega & =\varphi \omega ,\hfill \\ \hfill d\varphi & =\varphi \varphi +S\omega ^{}\omega S\omega \omega ^{}\omega \omega ^{}S+\omega ^{}S\omega \mathrm{I}_n,\hfill \\ \\ \hfill dS& =\varphi S+S\varphi +T\omega ^{}+\omega T^{}+\frac{1}{2}(T^{}\omega +\omega ^{}T)\mathrm{I}_n,\hfill \\ \hfill dT& =\varphi T+(U\mathrm{I}_n+S^2)\omega ,\hfill \\ \hfill dU& =T^{}S\omega +\omega ^{}ST.\hfill \end{array}$$
will be referred to as the *structure equations* of a Bochner-Kähler structure.
#### 2.3.3. First consequences
The equations (2.13) allow simple proofs of some known results about Bochner-Kähler structures.
The first part of the following result is due to Matsumoto and the second part is due to Matsumoto and Tanno .
###### Proposition 1.
If a Bochner-Kähler structure has constant scalar curvature, then it is a locally symmetric space. Any locally symmetric Bochner-Kähler structure is locally isometric to $`M_c^p\times M_c^{np}`$ for some $`n`$, $`p`$, and $`c`$.
###### Proof.
Since the pullback of the scalar curvature to $`P`$ is $`8(n+1)\mathrm{tr}S`$, the hypothesis of constant scalar curvature is equivalent to $`d(\mathrm{tr}S)=0`$. Now, by the structure equations
$$d(\mathrm{tr}S)=\frac{1}{2}(n+2)\left(T^{}\omega +\omega ^{}T\right),$$
so $`d(\mathrm{tr}S)=0`$ implies that $`T`$ vanishes identically. However, if $`T`$ vanishes identically, then $`dS=\varphi S+S\varphi `$, so that the curvature tensor is parallel. Thus, the structure is locally symmetric. In particular, the eigenvalues of $`S`$ are all constant.
Also, $`T=0`$ implies that $`S^2=U\mathrm{I}_n`$. This, combined with the constancy of the eigenvalues of $`S`$ implies that $`U`$ is constant and equal to $`s^2`$ for some real number $`s0`$. This, in turn, implies that $`(Ss\mathrm{I}_n)(S+s\mathrm{I}_n)=0`$. Consequently, $`S`$ has at most two distinct eigenvalues. It follows that either $`S=\pm s\mathrm{I}_n`$, in which case the structure has constant holomorphic sectional curvature $`4s`$, or else that there is a symmetric frame reduction of $`P`$ to a $`\left(\mathrm{U}(p)\times \mathrm{U}(np)\right)`$-subbundle $`P^{}P`$ on which
$$S=\left(\begin{array}{cc}s\mathrm{I}_p& 0\\ 0& s\mathrm{I}_{np}\end{array}\right).$$
Thus, the structure is a locally isomorphic to $`M_c^p\times M_c^{np}`$ where $`c=4s`$. ∎
The structure equations also yield a simple proof of the following result of Tachibana and Liu.
###### Proposition 2.
If a Bochner-Kähler structure is locally a nontrivial product, then it is locally isometric to $`M_c^p\times M_c^{np}`$ for some $`n`$, $`p`$, and $`c`$.
###### Proof.
Assume that the Bochner-Kähler structure is locally a nontrivial product. Then for some $`1pn/2`$, there is a $`\left(\mathrm{U}(p)\times \mathrm{U}(np)\right)`$-subbundle $`P^{}P`$ on which $`\varphi `$ is blocked in the form
$$\varphi =\left(\begin{array}{cc}\varphi _1& 0\\ 0& \varphi _2\end{array}\right),$$
where $`\varphi _1`$ takes values in $`𝔲(p)`$ and $`\varphi _2`$ takes values in $`𝔲(np)`$. This forces $`S`$ to be blocked in the corresponding form
$$S=\left(\begin{array}{cc}S_1& 0\\ 0& S_2\end{array}\right).$$
The vanishing of the off-diagonal blocks of the structure equation for $`dS`$ then shows that $`T`$ must be zero, thus implying that the structure is locally symmetric. Now apply Proposition 1. ∎
#### 2.3.4. The structure function
It turns out<sup>9</sup><sup>9</sup>9This was only noticed in hindsight, after the momentum mapping construction of §3.5. to be more convenient to work with $`H=S\frac{1}{n+2}(\mathrm{tr}S)\mathrm{I}_n`$ than to work with $`S`$ directly. Thus, $`S=H+\frac{1}{2}(\mathrm{tr}H)\mathrm{I}_n`$, and the structure equations (2.13) assume the form
(2.14)
$$\begin{array}{cc}\hfill d\omega & =\varphi \omega ,\hfill \\ \hfill d\varphi & =\varphi \varphi +H\omega ^{}\omega H\omega \omega ^{}\omega \omega ^{}H+\omega ^{}H\omega \mathrm{I}_n\hfill \\ & +(\mathrm{tr}H)\left(\omega ^{}\omega \mathrm{I}_n\omega \omega ^{}\right),\hfill \\ \\ \hfill dH& =\varphi H+H\varphi +T\omega ^{}+\omega T^{},\hfill \\ \hfill dT& =\varphi T+\left(H^2+(\mathrm{tr}H)H+V\mathrm{I}_n\right)\omega ,\hfill \\ \hfill dV& =(\mathrm{tr}H)\left(T^{}\omega +\omega ^{}T\right)+\left(T^{}H\omega +\omega ^{}HT\right).\hfill \end{array}$$
where I have also set $`V=U+\frac{1}{4}(\mathrm{tr}H)^2`$. The map $`(H,T,V):Pi𝔲(n)^n`$ will be known as the *structure function*.
While several of these equations seem more complicated than their counterparts in (2.13), the decisive simplification is the formula for $`dH`$ versus the formula for $`dS`$, as will be seen. For later use, I record the identity
(2.15)
$$\pi ^{}\left(\mathrm{Ric}(\mathrm{\Omega })\right)=(n+2)i\omega ^{}\left(H+(\mathrm{tr}H)\mathrm{I}_n\right)\omega $$
which follows from the earlier formula for the Ricci form in terms of $`S`$.
#### 2.3.5. Scaling weights
If $`\mathrm{\Omega }`$ is a Bochner-Kähler structure on a complex manifold $`M`$, then so is $`c\mathrm{\Omega }`$ for any constant $`c>0`$. The unitary coframe bundle of this scaled structure is
$$\sqrt{c}P=\{\sqrt{c}uuP\}.$$
The structure functions on the two bundles $`P`$ and $`\sqrt{c}P`$ then satisfy
(2.16)
$$H\left(\sqrt{c}u\right)=c^1H(u),T\left(\sqrt{c}u\right)=c^{3/2}T(u),V\left(\sqrt{c}u\right)=c^2V(u).$$
This motivates assigning ‘scaling weights’ to the components of the structure function as follows: $`H`$ has scaling weight 1, $`T`$ has scaling weight $`\frac{3}{2}`$, and $`V`$ has scaling weight $`2`$. (Taking positive, rather than negative, scaling weights is a simplifying convention.)
#### 2.3.6. Dimension $`1`$
Equations (2.14) still make sense when $`n=1`$, i.e., when $`M`$ is a complex curve endowed with a positive 2-form $`\mathrm{\Omega }`$ and $`\pi :PM`$ is its $`\mathrm{U}(1)`$-coframe bundle. In this case, $`H`$ is an $``$-valued function while $`T`$ is $``$-valued. The equations (2.14) then simplify to the scalar equations
(2.17)
$$\begin{array}{cc}\hfill d\omega & =\varphi \omega ,\hfill \\ \hfill d\varphi & =6H\omega \overline{\omega },\hfill \\ \\ \hfill dH& =\overline{T}\omega +T\overline{\omega },\hfill \\ \hfill dT& =\varphi T+\left(2H^2+V\right)\omega ,\hfill \\ \hfill dV& =2H\left(\overline{T}\omega +T\overline{\omega }\right)=2HdH\hfill \end{array}$$
Accordingly, when $`n=1`$, the satisfaction of these structure equations can be taken to be the *definition* of the Bochner-Kähler property. Throughout this article, this will be done. It is not difficult to check that the rotationally symmetric analysis of Example 2, extends to the case $`n=1`$ when one takes this as the definition of Bochner-Kähler.
The Gaussian curvature of the associated metric $`g`$ is $`K=12H`$. In fact, the geometric interpretation of the equations (2.17) is just that the $`\mathrm{\Omega }`$-Hamiltonian flow associated to $`K`$ should be $`g`$-isometric. (Compare §2.1.2.) Thus, any constant curvature metric in (complex) dimension $`1`$ is Bochner-Kähler. Moreover, any nonconstant curvature metric in dimension 1 that is Bochner-Kähler has a canonically defined nontrivial Killing field.
Assume that $`M`$ is connected, which implies that $`P`$ is also connected. The last structure equation of (2.17) implies that $`VH^2`$ is a constant $`C_2`$ (the index denotes the scaling weight), and the next-to-last equation of (2.17) then implies that there is a constant $`C_3`$ so that $`|T|^2=H^3+C_2H+C_3`$, or equivalently, that $`|T|^2VH=C_3`$.
These two ‘constants of the structure’ will be generalized considerably in higher dimensions, as will the existence of nontrivial symmetry vector fields.
## 3. Existence and Moduli
### 3.1. Existence
In , Élie Cartan proved a powerful existence and uniqueness theorem that generalizes Lie’s Third Fundamental Theorem from the case of a transitive group action to the case of an intransitive group action.
For the convenience of the reader and because Cartan’s rather sketchy treatment needs amplification on some minor points, a discussion of his theorem is included in the Appendix.
Cartan’s conditions for the existence of a (local) coframing and system of functional invariants satisfying a given set of structure equations are satisfied by the system (2.14). The following result is then an immediate consequence of his general theorem.
###### Theorem 1.
For any $`(H_0,T_0,V_0)i𝔲(n)^n`$, there exists a Bochner-Kähler structure $`\mathrm{\Omega }`$ on a neighborhood $`V`$ of $`0^n`$ whose unitary coframe bundle $`\pi :PV`$ contains a $`u_0P_0=\pi ^1(0)`$ for which $`H(u_0)=H_0`$, $`T(u_0)=T_0`$, and $`V(u_0)=V_0`$. Any two real-analytic Bochner-Kähler structures with this property are isomorphic on a neighborhood of $`0^n`$. Finally, any Bochner-Kähler structure that is $`C^5`$ is real-analytic.
###### Proof.
Since the exterior derivatives of the equations (2.14) are identities, Cartan’s conditions (i.e., his generalization of the Jacobi conditions) are satisfied for these equations as structure equations of a coframing.
Thus, by Theorem A.1 (see the Appendix), for any $`(H_0,T_0,V_0)i𝔲(n)^n`$, there exists a real-analytic manifold $`N`$ of dimension $`n^2+2n`$ on which there are two real-analytic 1-forms $`\omega `$ and $`\varphi `$, taking values in $`^n`$ and $`𝔲(n)`$, respectively, and a real-analytic function $`(H,T,V):Pi𝔲(n)^n`$ with the properties that $`(\omega ,\varphi )`$ is a $`^n𝔲(n)`$-valued coframing on $`N`$, that the equations eq: structure equations ii are satisfied on $`N`$, and that there exists a $`u_0N`$ for which $`(H(u_0),T(u_0),V(u_0))=(H_0,T_0,V_0)`$.
Since $`d\omega =\varphi \omega `$, the equation $`\omega =0`$ defines an integrable plane field of codimension $`2n`$ on $`N`$. After shrinking $`N`$ to an open neighborhood of $`u_0`$ if necessary, an application of the complex Frobenius theorem shows that there is a submersion $`z:N^n`$ with $`z(u_0)=0`$ so that the leaves of this integrable plane field are the fibers of $`\pi `$ and, moreover, that $`dz=p\omega `$ for some function $`p:N\mathrm{GL}(n,)`$ that satisfies $`p(u_0)=\mathrm{I}_n`$.
Since $`\varphi =\varphi ^{}`$, the 2-form
$$\mathrm{\Omega }=\frac{i}{2}\omega ^{}\omega =\frac{i}{2}dz^{}(pp^{})^1dz$$
is closed. Since $`\mathrm{\Omega }`$ is $`z`$-semibasic and since, by definition, the fibers of $`z`$ are connected, it follows that $`\mathrm{\Omega }`$ is actually the pullback to $`N`$ of a closed, positive (1,1)-form on the open set $`V=z(N)^n`$, i.e., a Kähler structure on $`V`$.
Let $`\pi :PV`$ be the unitary coframe bundle of this Kähler structure. Define a mapping $`\tau :NP`$ as follows: If $`z(u)=xV`$, then $`dz_u:T_uNT_x^n^n`$ is surjective and, by construction, has the same kernel as $`\omega _u:T_uN^n`$. Thus, there is a unique linear isomorphism $`\tau (u):T_x^n^n`$ so that $`\omega _u=\tau (u)dz_u`$. In fact, $`\tau (u)`$ is complex linear; using the standard identification $`T_x^n^n`$, one sees that $`\tau (u)`$ becomes $`p(u)^1M_n()`$.
The equation $`\mathrm{\Omega }=\frac{i}{2}\omega ^{}\omega `$ implies that $`\tau (u)`$ is a unitary coframe for all $`uN`$. Since $`(\omega ,\varphi )`$ is a coframing, it follows that $`\tau :NP`$ is an open immersion of $`N`$ into $`P`$. Shrinking $`N`$ again if necessary, it can be assumed that $`\tau `$ embeds $`N`$ as an open subset of $`P`$. Thus, nothing is lost by identifying $`N`$ with this open subset of $`P`$.
The structure equations (2.14) now become identified with the structure equations of the unitary coframe bundle $`P`$, implying that the underlying Kähler structure on $`V`$ is, in fact Bochner-Kähler, and that the structure function $`(H,T,V)`$ takes on the value $`(H_0,T_0,V_0)`$ at $`u_0P`$, as desired. Further details are left to the reader. This completes the existence proof.
Uniqueness in the real-analytic category now follows directly from Theorem A.1. Now, while Cartan states the uniqueness part of Theorem A.1 only in the real-analytic category, uniqueness can actually be proved using only ordinary differential equations (i.e., the Frobenius theorem); the Cauchy-Kowalewski or Cartan-Kähler Theorems are not needed. Thus, his uniqueness result is valid as long as the form $`\mathrm{\Omega }`$ is sufficiently differentiable for $`P`$ to exist as a differentiable bundle and for $`H`$, $`T`$, and $`V`$ to be defined and differentiable. For this to be true, it certainly suffices for $`\mathrm{\Omega }`$ to be $`C^5`$.
Since Cartan’s existence proof produces a real-analytic example, uniqueness then implies that any $`C^5`$ Bochner-Kähler structure is real-analytic. ∎
###### Remark 2 (Minimal Regularity).
With some work, one can show that if $`H`$ and $`T`$ are differentiable, then $`V`$ (which, by (2.11), must exist) must be differentiable as well, thus reducing the regularity needed to apply Cartan’s Theorem to $`C^4`$. However, this is almost certainly not optimal since, presumably, when $`n2`$, any $`C^2`$ Kähler structure that is Bochner-Kähler is real-analytic. However, the above proof does not show this.
*From now on, I will assume that the Bochner-Kähler structures under consideration are real-analytic.*
### 3.2. Local moduli
The group $`\mathrm{U}(n)`$ acts on the space $`i𝔲(n)^n`$ in the usual way:
(3.1)
$$a(h,t,v)=(aha^{},at,v)$$
for $`a\mathrm{U}(n)`$. This action makes the structure function of a Bochner-Kähler structure $`(H,T,V):Pi𝔲(n)^n`$ equivariant with respect to the right bundle action, i.e.,
(3.2)
$$(H(ua),T(ua),V(ua))=a^1(H(u),T(u),V(u)).$$
Consequently, it will be useful to have an understanding of the orbits of $`\mathrm{U}(n)`$ acting on this space.
#### 3.2.1. Orbits
Let $`Wi𝔲(n)^n`$ be the linear subspace consisting of the triples $`(h,t,v)`$ for which $`h`$ is diagonal and $`t`$ is real. Then $`W`$ is a linear subspace of (real) dimension $`2n+1`$. Let $`CW`$ be the ‘chamber’ defined by the inequalities $`h_{1\overline{1}}h_{2\overline{2}}\mathrm{}h_{n\overline{n}}`$ augmented by the conditions that $`t_j0`$, with equality if $`h_{j\overline{ȷ}}=h_{i\overline{ı}}`$ for any $`i<j`$. N.B.: The set $`C`$ has nonempty interior in $`W`$. Note, however, that $`C`$ is not closed when $`n2`$.
###### Proposition 3.
Each $`\mathrm{U}(n)`$-orbit in $`i𝔲(n)^n`$ meets $`C`$ in exactly one point.
###### Proof.
Consider any $`(h,t,v)i𝔲(n)^n`$. Act by an element $`a\mathrm{U}(n)`$ so as to reduce to the case where $`h`$ is diagonal and its (real) eigenvalues are arranged in decreasing order down the diagonal. If there are integers $`ij`$ so that $`h_{j\overline{ȷ}}=h_{i\overline{ı}}`$, suppose that $`i,i+1,\mathrm{},j`$ is a maximal unbroken string with this property. Then the stabilizer of $`h`$ in $`\mathrm{U}(n)`$ will contain a subgroup isomorphic to $`\mathrm{U}(ji+1)`$ that will act as unitary rotations on the subvector $`(t_i,\mathrm{},t_j)`$. Acting by an element of the stabilizer of $`h`$, one can then reduce to the case where $`t_i`$ is real and nonnegative while $`t_{i+1}=\mathrm{}=t_j=0`$. By definition, the resulting new $`(h,t,v)`$ is an element of $`C`$. It is clear from the construction that this element is unique. ∎
###### Corollary 1.
The set of isomorphism classes of germs of Bochner-Kähler structures in dimension $`n`$ is in one-to-one correspondence with the elements of $`C`$.
#### 3.2.2. Invariant polynomials
It is not difficult to exhibit enough $`\mathrm{U}(n)`$-invariant polynomials on $`i𝔲(n)^n`$ to separate the $`\mathrm{U}(n)`$-orbits. For $`k0`$, define the $`\mathrm{U}(n)`$-invariant polynomials
(3.3)
$$a_k(h,t,v)=\mathrm{tr}(h^k),b_{k+3}(h,t,v)=t^{}h^kt,$$
and set $`b_2(h,t,v)=v`$. (The indexing is chosen so as to indicate the scaling weight as defined in §2.3.5. The anomalous definition of $`b_2`$ will be explained below.) Then an easy argument using Proposition 3 shows that the collection of $`2n+1`$ functions
(3.4)
$$\phi =(a_1,\mathrm{},a_n,b_2,b_3,\mathrm{},b_{n+2})$$
separates the $`\mathrm{U}(n)`$-orbits in $`i𝔲(n)^n`$.<sup>10</sup><sup>10</sup>10In fact, by \[22, Theorem 12.1\], the components of $`\phi `$ generate the ring of $`\mathrm{U}(n)`$-invariant polynomials on $`i𝔲(n)^n`$.
When $`n=1`$, the function $`a_1^2+b_{2}^{}{}_{}{}^{2}+b_30`$ is evidently a proper function on $`i𝔲(1)`$ while, for $`n2`$, the function $`a_2+b_{2}^{}{}_{}{}^{2}+b_30`$ is a proper function on $`i𝔲(n)^n`$.
It follows that $`\phi `$ is a proper mapping, implying that $`F_n=\phi \left(i𝔲(n)^n\right)`$ is closed in $`^{2n+1}`$. The set $`F_n^{2n+1}`$ is thus the proper moduli space of orbits.
If $`(h_0,t_0,v_0)i𝔲(n)^n`$ is such that $`h_0`$ has $`n`$ distinct eigenvalues and $`t_0`$ is not orthogonal to any of the eigenvectors of $`h_0`$, then an elementary computation shows that $`\phi ^{}(h_0,t_0,v_0):i𝔲(n)^n^{2n+1}`$ is surjective. It follows from this that $`F_n`$ is the closure of its interior. Of course, $`\phi :CF_n`$ is a bijection.
#### 3.2.3. The moduli mapping
This description of $`F_n`$ can be interpreted as saying that the germs of Bochner-Kähler structures in dimension $`n`$ form a singular space of real dimension $`2n+1`$. It is $`F_n`$ that is the natural moduli space for germs of Bochner-Kähler structures in the following sense: For any Bochner-Kähler structure $`(M,g,\mathrm{\Omega })`$, there is a commutative diagram
(3.5)
$$\begin{array}{ccc}P& \stackrel{(H,T,V)}{}& i𝔲(n)^n\\ \pi & & \phi & & \\ M& \stackrel{f}{}& F_n^{2n+1}\end{array}$$
where $`f:MF_n`$ is a real-analytic map each of whose fibers is an orbit of the symmetry pseudo-groupoid of the Bochner-Kähler structure on $`M`$. This function will be known as the *moduli mapping* of the Bochner-Kähler structure.
#### 3.2.4. Analytic connectedness
However, this description does not really say ‘how many’ Bochner-Kähler structures there are locally since, for a given Bochner-Kähler structure, the map $`f:MF_n`$ might have rather large image in $`F_n`$. A priori, the image could even have dimension as large as $`2n`$, in which case one would be tempted to say that the ‘generic’ Bochner-Kähler structures depend on only one parameter, the parameter that distinguishes the ‘hypersurfaces’ in $`F_n`$ that are the images of generic Bochner-Kähler structure maps. However, as will be shown in the next subsection, this is not the case. Instead, the dimension of the image $`f(M)`$ turns out to be no more than $`n`$ for any Bochner-Kähler structure.
One of the difficulties that arises in discussing this ‘how many’ question is that it turns out that not every connected Bochner-Kähler structure can be regarded as an open subset of a unique ‘maximal’ connected Bochner-Kähler structure (cf. the discussion of the dimension $`n=1`$ at the end of §3.2.5). Even when one restricts attention to the simply-connected, connected Bochner-Kähler structures, this difficulty persists. Compare this situation with that of locally symmetric spaces: Every simply-connected, connected locally symmetric space has an isometric open immersion (sometimes called a developing map) into a unique (complete) simply-connected symmetric space and this immersion is unique up to ambient isometry. The discussion carried out in Example 2 and in §3.2.5 below shows that no such result could hold for Bochner-Kähler structures.
Two elements $`v_1,v_2F_n`$ will be said to be *analytically connected* if there exists a connected Bochner-Kähler manifold $`(M,\mathrm{\Omega })`$ so that both $`v_1`$ and $`v_2`$ lie in $`f(M)`$. An elementary argument shows that this is an equivalence relation. One of the tasks of this article is to describe these equivalence classes explicitly.
#### 3.2.5. Dimension $`1`$
Now, $`i𝔲(1)=`$ and the map $`\phi :^3`$ takes the form
$$\phi (h,t,v)=(h,v,|t|^2).$$
Thus $`F_1^3`$ is the closed upper half-space. The fiber $`\phi ^1(x,y,0)`$ is a single point for each $`(x,y,0)`$ on the boundary of $`F_1`$ while the fiber $`\phi ^1(x,y,z)`$ is a circle when $`(x,y,z)`$ lies in the interior $`F_1^{}`$, i.e., when $`z>0`$.
By Theorem 1, every point of $`F_1`$ lies in the $`f`$-image of some Bochner-Kähler structure in dimension 1.
Let $`(M,g,\mathrm{\Omega })`$ be a connected Bochner-Kähler manifold of dimension $`1`$, so that $`\phi (H,T,V)=(H,V,|T|^2)`$. As was pointed out in §2.3.6, there are constants $`C_2`$ and $`C_3`$ so that
$$VH^2=C_2\text{and}|T|^2HV=C_3.$$
In other words, $`\phi (H,T,V)=(H,H^2+C_2,H^3+C_2H+C_3)`$, implying that the map $`f:M^3`$ has its image either a point (if $`H`$ is constant) or a curve.
For any $`C=(C_2,C_3)`$, let $`p_C(t)=t^3+C_2t+C_3`$ and set
$$\mathrm{\Gamma }_C=\left\{(t,t^2+C_2,t^3+C_2t+C_3)\text{ }p_C(t)0\right\}.$$
Since $`dH=\overline{T}\omega +T\overline{\omega }`$, it follows that $`df`$ vanishes only at those $`xM`$ where $`|T|^2=0`$. In other words, if $`M^{}=f^1(F_1^{})`$ is the locus where $`|T|^2`$ is nonzero, then $`f:M^{}F_1^{}`$ is a submersion onto an open subset of $`\mathrm{\Gamma }_C^{}=\mathrm{\Gamma }_CF_1^{}`$.
Since $`4|T|^2`$ is the squared norm of the gradient of $`dH`$ and the $`\mathrm{\Omega }`$-Hamiltonian of $`H`$ is a Killing field on the surface, it follows that either $`|T|^2`$ vanishes identically or else it vanishes only at isolated points in $`M`$ and then only to second order.
In the former case, $`H`$ is constant on $`M`$. By the structure equations (2.17), since $`T`$ vanishes identically it follows that $`V2H^2`$. Thus, each of the points $`v=(r,2r^2,0)F_1`$ constitutes a single analytically connected equivalence class that is the $`f`$-image of any surface endowed with a metric of constant curvature $`K=12r`$. Note that, in this case, the constants $`C_2`$ and $`C_3`$ assume the values $`C_2=3r^2`$ and $`C_3=2r^3`$, so that $`p_C(t)=(tr)^2(t+2r)`$ has either a double or triple root (if $`r=0`$). Let $`\mathrm{\Pi }=\left\{(r,2r^2,0)\text{ }r\right\}`$ be the parabola of ‘isolated’ classes. These are the only points that can be the value of a constant $`f`$.
Now suppose that $`|T|^2`$ is not identically zero, so that $`M^{}`$ is simply $`M`$ minus a set of isolated points.
When $`p_C(t)`$ has only one real simple root, say $`r_0`$, then $`\mathrm{\Gamma }_C`$ is connected and homeomorphic to a closed half-line. Call this Case 1. In this case $`\mathrm{\Gamma }_C\mathrm{\Pi }=\mathrm{}`$, so that it is not possible for $`M`$ to satisfy $`f(M)\mathrm{\Gamma }_C`$ and have $`f(M)`$ be a point. Since $`f:M^{}\mathrm{\Gamma }_C^{}`$ is a submersion, it follows that if $`f(M)`$ lies in $`\mathrm{\Gamma }_C`$, then $`f(M)`$ is a open subset of $`\mathrm{\Gamma }_C`$. Since $`\mathrm{\Gamma }_C`$ is connected, it follows that $`\mathrm{\Gamma }_C`$ must constitute a single analytically connected equivalence class.
When $`p_C(t)`$ has only one real root, but this root is multiple, the only possibility is that this root is $`t=0`$ and, in fact, $`p_C(t)=t^3`$. Call this Case 2. In this case, $`\mathrm{\Gamma }_C=\{(0,0,0)\}\mathrm{\Gamma }_C^{}`$ where $`\mathrm{\Gamma }_C^{}=\left\{(t,t^2,t^3)\text{ }t>0\right\}`$. In this case, $`f(M)`$ can lie in $`\mathrm{\Gamma }_C`$ only if either $`f(M)=\{(0,0,0)\}`$ or $`f(M)`$ is an open subset of $`\mathrm{\Gamma }_C^{}`$. Since $`\mathrm{\Gamma }_C^{}`$ is connected, it follows that $`\mathrm{\Gamma }_C^{}`$ constitutes a single analytically connected equivalence class.
When $`p_C(t)`$ has two real distinct roots, say $`r_1>r_2`$, one must be double, so there are two possibilities. Case 3-$`i`$ will be that in which $`r_i`$ is the double root.
In Case 3-1, $`p_C(t)=(tr)^2(t+2r)`$ where $`r>0`$. Since $`\mathrm{\Gamma }_C\mathrm{\Pi }=\{(r,2r^2,0)\}`$, define
$$\begin{array}{cc}\hfill \mathrm{\Gamma }_C^a& =\left\{(t,t^23r^2,(tr)^2(t+2r))t>r\right\},\hfill \\ \hfill \mathrm{\Gamma }_C^b& =\left\{(t,t^23r^2,(tr)^2(t+2r))2rt<r\right\}.\hfill \end{array}$$
Then $`\mathrm{\Gamma }_C=\mathrm{\Gamma }_C^b\{(r,2r^2,0)\}\mathrm{\Gamma }_C^a`$, and each of $`\mathrm{\Gamma }_C^a`$, $`\{(r,2r^2,0)\}`$, and $`\mathrm{\Gamma }_C^b`$ is evidently a single analytically connected equivalence class.
In Case 3-2, $`p_C(t)=(tr)^2(t+2r)`$ where $`r<0`$. Still, $`\mathrm{\Gamma }_C\mathrm{\Pi }=\{(r,2r^2,0)\}`$, but now $`\mathrm{\Gamma }_C=\{(r,2r^2,0)\}\mathrm{\Gamma }_C^a`$ where
$$\mathrm{\Gamma }_C^a=\left\{(t,t^23r^2,(tr)^2(t+2r))\text{ }2rt\right\},$$
and each of $`\{(r,2r^2,0)\}`$ and $`\mathrm{\Gamma }_C^a`$ is evidently a single analytically connected equivalence class.
When $`p_C(t)`$ has three distinct real roots, say $`r_0>r_1>r_2`$, then $`r_0+r_1+r_2=0`$ and again $`\mathrm{\Gamma }_C\mathrm{\Pi }=\mathrm{}`$. Call this Case 4. In this case, $`\mathrm{\Gamma }_C`$ has two components
$$\begin{array}{cc}\hfill \mathrm{\Gamma }_C^0& =\left\{(t,t^2+(r_0r_1+r_0r_2+r_1r_2),(tr_0)(tr_1)(tr_2))r_0t\right\},\hfill \\ \hfill \mathrm{\Gamma }_C^1& =\left\{(t,t^2+(r_0r_1+r_0r_2+r_1r_2),(tr_0)(tr_1)(tr_2))r_2tr_1\right\},\hfill \end{array}$$
each of which is a single analytically connected equivalence class.
Now, in all these cases, the metric $`g=\omega \overline{\omega }`$ can be expressed directly in terms of the invariants. Restrict attention to $`M^{}M`$ and note that, by the structure equations, the complex-valued $`1`$-form $`\omega /T`$ is closed and therefore a nowhere vanishing holomorphic $`1`$-form on $`M^{}`$. Since $`|T|^2`$ vanishes only to second order at each of its zeroes, $`\omega /T`$ extends to all of $`M`$ as a meromorphic $`1`$-form with simple poles at the places where $`\omega /T`$ vanishes.
Also, since $`|T|^2=H^3+C_2H+C_3`$, it follows that
$$\frac{dH}{H^3+C_2H+C_3}=\frac{dH}{|T|^2}=\frac{\omega }{T}+\frac{\overline{\omega }}{\overline{T}}.$$
Thus,
$$\frac{\omega }{T}=\frac{dH}{2(H^3+C_2H+C_3)}+2id\theta $$
where $`\theta `$ is locally well-defined on $`M^{}`$ up to a (real) additive constant (the factor of $`2`$ in front of the $`d\theta `$ term provides for consistency with later notation). Consequently, one has the formula
$$g=\omega \overline{\omega }=\frac{dH^2}{4(H^3+C_2H+C_3)}+4(H^3+C_2H+C_3)d\theta ^2.$$
More precisely, the simply-connected cover $`\stackrel{~}{M^{}}`$ admits a developing map $`(H,\theta ):\stackrel{~}{M^{}}^2`$ that isometrically embeds $`\stackrel{~}{M^{}}`$ into the region $`R_C`$ in the $`H\theta `$-plane defined by the inequality $`H^3+C_2H+C_3>0`$, endowed with the above metric.
Using this representation, one can determine which of the Cases above can allow complete Bochner-Kähler metrics in dimension $`1`$.
For example, let $`R`$ be the largest real root of $`p_C(t)`$. Then because the integral
$$_{R+1}^{\mathrm{}}\frac{dH}{\mathrm{\hspace{0.17em}2}\sqrt{H^3+C_2H+C_3}}$$
converges, the metric $`g`$ defined above is not complete at the ‘edge’ $`H=\mathrm{}`$ of the half-plane $`H>R`$. This implies that if $`(M,g,\mathrm{\Omega })`$ is a Bochner-Kähler metric with characteristic polynomial $`p_C`$, satisfying $`HR`$, and having $`H`$ non-constant, then the length of the gradient lines of $`H`$ would be finite in the increasing direction and so could not be complete.
Consequently, a complete Bochner-Kähler metric must have its image lie in a bounded region of $`F_1`$. In particular, the analytically connected component that contains $`f(M)`$ must be bounded. The only bounded analytically connected equivalence classes are
1. Case 3 with $`f(M)=\{(r,2r^2,0)\}`$;
2. Case 3-1 with $`f(M)=\mathrm{\Gamma }_C^b`$; and
3. Case $`4`$ with $`f(M)=\mathrm{\Gamma }_C^1`$.
The case of a single point has already been discussed: There is a unique connected and simply-connected complete example for each $`r`$.
In Case 3-1, with $`f(M)=\mathrm{\Gamma }_C^b`$ with $`r>0`$, the metric $`g`$ on the region $`2r<H<r`$ in the $`H\theta `$-plane is of the form
$$g=\frac{dH^2}{4(Hr)^2(H+2r)}+4(Hr)^2(H+2r)d\theta ^2.$$
Because
$$_0^r\frac{dH}{\mathrm{\hspace{0.17em}2}\sqrt{(Hr)^2(H+2r)}}=\mathrm{},$$
this metric is complete near the ‘edge’ $`H=r`$. However, since
$$_{2r}^0\frac{dH}{\mathrm{\hspace{0.17em}2}\sqrt{(Hr)^2(H+2r)}}<\mathrm{},$$
the metric is not complete near the ‘edge’ $`H=2r`$. In fact, making the substitution $`H+2r=3r\rho ^2`$, the metric takes the form
$$g=\frac{d\rho ^2+\rho ^2(1\rho ^2)^4(18r^2d\theta )^2}{(1\rho ^2)^2},$$
and one recognizes that $`g`$ will extend to a smooth metric at $`\rho =0`$ in polar coordinates $`(\rho ,\theta )`$ on the disk $`\rho <1`$ if and only if $`\theta `$ is taken to be periodic with period $`\pi /(9r^2)`$. This disk endowed with this complete metric is conformally equivalent to $``$. The Gaussian curvature decreases monotonically from $`24r`$ at $`\rho =0`$ to a limiting value of $`12r`$ as $`\rho `$ approaches $`1`$.
Finally, consider Case 4 with image in $`\mathrm{\Gamma }_C^1`$. Let $`r_0>r_1>r_2`$ be the three roots satisfying $`r_0=(r_1+r_2)`$, so that $`H^3+C_2H+C_3=(Hr_0)(Hr_1)(Hr_2)`$. Consider the metric on the strip $`r_2<H<r_1`$ in the $`H\theta `$-plane given by
$$g=\frac{dH^2}{4(Hr_0)(Hr_1)(Hr_2)}+4(Hr_0)(Hr_1)(Hr_2)d\theta ^2.$$
Since
$$_{r_2}^{r_1}\frac{dH}{2\sqrt{(Hr_0)(Hr_1)(Hr_2)}}<\mathrm{},$$
this metric is not complete at either edge $`H=r_i`$ for $`i=1,2`$.
Letting $`H=r_2+v^2`$ and computing as above, one finds that the metric will extend to a smooth metric on a disk about $`v=0`$ in $`(v,\theta )`$ polar coordinates if and only if $`\theta `$ is taken to be periodic with period
$$\tau _2=\frac{\pi }{3r_{2}^{}{}_{}{}^{2}+C_2}>0.$$
Similarly, setting $`H=r_1w^2`$, and computing as above, one finds that the metric will extend to a smooth metric on a disk about $`w=0`$ in $`(w,\theta )`$ polar coordinates if and only if $`\theta `$ is taken to be periodic with period
$$\tau _1=\frac{\pi }{3r_{1}^{}{}_{}{}^{2}+C_2}>0.$$
Now, computation shows that $`\tau _1=\tau _2`$ has no solutions with $`r_1>r_2`$.
*Consequently, there is no complete Bochner-Kähler metric on a surface whose moduli image is $`\mathrm{\Gamma }_C^1`$.*
However, complete Bochner-Kähler metrics on orbifolds do exist: Taking $`r_1=r(q2p)`$ and $`r_2=r(p2q)`$ where $`0<p<q`$ are relatively prime integers and $`r`$ is a positive real number, one can choose a period for $`\theta `$ so that the resulting quotient completes to an orbifold metric on $`S^2`$ with one conical point of order $`1/q`$ and the other of order $`1/p`$.
This orbifold is the weighted projective line $`^{(p,q)}`$, i.e., $`^{\mathrm{\hspace{0.17em}2}}`$ minus the origin modulo the $`^{}`$-action $`\lambda (z,w)=(\lambda ^pz,\lambda ^qw)`$. This compact Riemannian orbifold could reasonably be regarded as the natural complete model for this case. Note that the Gaussian curvature of this metric will be strictly positive if and only if $`q<2p`$.
### 3.3. Infinitesimal symmetries
It turns out that any Bochner-Kähler structure has a nontrivial symmetry pseudo-groupoid $`\overline{\mathrm{\Gamma }}`$. In this subsection, some useful information about the ‘dimension’ and orbits of $`\overline{\mathrm{\Gamma }}`$ will be collected.
For $`(h,t,v)i𝔲(n)^n`$, let $`G_{(h,t,v)}^0\mathrm{U}(n)`$ be the stabilizer of $`(h,t,v)`$ under the action defined in §3.2. Since $`a\mathrm{U}(n)`$ lies in $`G_{(h,t,v)}^0`$ if and only if $`aha^{}=h`$ and $`at=t`$, it follows that $`G_{(h,t,v)}^0`$ is a closed, connected subgroup of $`\mathrm{U}(n)`$.
In fact, $`G_{(h,t,v)}^0`$ is a product of unitary groups and can be described as follows: Let $`h_1>h_2>\mathrm{}>h_\delta `$ be the distinct eigenvalues of $`h`$ and, for $`1\alpha n`$, let $`L_\alpha ^n`$ be the $`h_\alpha `$-eigenspace of $`h`$. Since $`h`$ is Hermitian symmetric, there is an orthogonal direct sum decomposition
$$^n=L_1L_2\mathrm{}L_\delta $$
with $`dimL_\alpha =n_\alpha 1`$. Write $`t=t_1+\mathrm{}+t_\delta `$ where $`t_\alpha `$ lies in $`L_\alpha `$ and let $`t_\alpha ^{}L_\alpha `$ be the subspace of $`L_\alpha `$ that is perpendicular to $`t_\alpha `$. Then, using obvious notation,
$$G_{(h,t,v)}^0=\mathrm{U}(t_1^{})\times \mathrm{U}(t_2^{})\times \mathrm{}\times \mathrm{U}(t_\delta ^{}).$$
The uniqueness part of Cartan’s Theorem A.1 then has the following useful corollary.
###### Corollary 2.
Let $`PM`$ be a Bochner-Kähler structure. Then for any $`uP_x`$, the unitary isomorphism $`u:T_xM^n`$ induces an isomorphism
$$\overline{\mathrm{\Gamma }}_xG_{(H(u),T(u),V(u))}^0.$$
Thus, $`\overline{\mathrm{\Gamma }}_x`$ is isomorphic to a product of unitary groups and, in particular, is connected.
#### 3.3.1. Existence and lower bounds
Roughly speaking, a Bochner-Kähler structure has at least an $`n`$-dimensional ‘infinitesimal symmetry group’. As will be seen below, this lower bound is reached for the ‘generic’ Bochner-Kähler structure.
###### Theorem 2.
Let $`M`$ be a simply-connected complex $`n`$-manifold endowed with a Bochner-Kähler structure $`\mathrm{\Omega }`$. Let $`𝔤𝔛(M)`$ denote the Lie algebra of vector fields on $`M`$ whose flows preserve the complex structure and $`\mathrm{\Omega }`$. Then $`dim_{}𝔤n`$.
###### Proof.
Let $`(M,\mathrm{\Omega })`$ satisfy the assumptions of the theorem, let $`\pi :PM`$ be the unitary coframe bundle, with canonical forms $`\omega `$ and $`\varphi `$, and let $`(H,T,V):Pi𝔲(n)^n`$ be the structure function.
Because $`M`$ is simply-connected and the Bochner-Kähler structure $`\mathrm{\Omega }`$ is real-analytic, any symmetry vector field of the structure defined on a connected open subset of $`M`$ can be uniquely analytically continued to a symmetry vector field on all of $`M`$. Moreover if $`Z𝔛(M)`$ is such a symmetry vector field, then, as discussed in §2.1, there is a unique vector field $`Z^{}`$ on $`P`$ satisfying $`\pi ^{}(Z^{})=Z`$ and $`𝔏_Z^{}\omega =𝔏_Z^{}\varphi =0`$. Conversely, if $`Y`$ is a vector field on $`P`$ satisfying $`𝔏_Y\omega =𝔏_Y\varphi =0`$, then $`Y=Z^{}`$ where $`Z=\pi ^{}(Y)`$ is a symmetry vector field on $`M`$.
In other words, the mapping $`ZZ^{}`$ defines an embedding $`𝔤𝔛(P)`$ that realizes $`𝔤`$ as the Lie algebra of vector fields on $`P`$ whose flows preserve the coframing $`\eta =(\omega ,\varphi )`$. By the structure equations (2.14) the flow of such a vector field must necessarily preserve the structure function $`(H,T,V):Pi𝔲(n)^n`$, which is a submersion onto its (connected) image.
Applying Cartan’s Theorem A.2 (see the Appendix), for any $`uP`$ the evaluation map $`e_u:𝔤T_uP`$ defined by $`e_u(Z)=Z^{}(u)T_uP`$ is a vector space isomorphism between $`𝔤`$ and the kernel of $`(H,T,V)^{}(u):T_uPi𝔲(n)^n`$. Let $`K_uT_uP`$ denote this kernel. Then, by (2.14), the image $`(\omega ,\varphi )(K_u)^ni𝔲(n)`$ consists of the pairs $`(w,f)^ni𝔲(n)`$ that satisfy
$$\begin{array}{cc}\hfill 0& =H(u)ffH(u)+T(u)w^{}+wT(u)^{},\hfill \\ \hfill 0& =fT(u)+\left(H(u)^2+(\mathrm{tr}H(u))H(u)+V(u)\mathrm{I}_n\right)w,\hfill \\ \hfill 0& =(\mathrm{tr}H(u))\left(T(u)^{}w+w^{}T(u)\right)+T(u)^{}H(u)w+w^{}H(u)T(u).\hfill \end{array}$$
By the first of these equations,
$$T(u)^{}w+w^{}T(u)=\mathrm{tr}\left([f,H(u)]\right)=0$$
and
$$2(T(u)^{}H(u)w+w^{}H(u)T(u))=2\mathrm{tr}\left([f,H(u)]H(u)\right)=\mathrm{tr}\left([f,H(u)^2]\right)=0.$$
Thus, the third equation is a consequence of the first and so can be ignored for the rest of this discussion.
Let $`(H,T,V)(u_0)=(H_0,T_0,V_0)i𝔲(n)^n`$. By $`\mathrm{U}(n)`$-equivariance, it is enough to show that the dimension of $`K_{u_0}`$ is at least $`n`$ at any point $`u_0`$ where $`H_0`$ and $`T_0`$ are both real, so assume this for the rest of the argument.
By the structure equations (2.14), the dimension of $`K_{u_0}`$ is equal to the dimension of the space of solutions of the linear equations
(3.6)
$$\begin{array}{cc}\hfill 0& =H_0ffH_0+T_0w^{}+wT_0^{}\hfill \\ \hfill 0& =fT_0+\left(H_{0}^{}{}_{}{}^{2}+(\mathrm{tr}H_0)H_0+V_0\mathrm{I}_n\right)w\hfill \end{array}$$
for $`w^n`$ and $`f𝔲(n)`$.
Consider the solutions of (3.6) for which $`f`$ and $`w`$ are purely imaginary, i.e., where $`f=is`$ and $`w=iy`$ for some symmetric (real) matrix $`s`$ and some $`y^n`$. Then the equations in (3.6) reduce to
(3.7)
$$\begin{array}{cc}\hfill 0& =H_0ssH_0T_0{}_{}{}^{t}y+y{}_{}{}^{t}T_{0}^{},\hfill \\ \hfill 0& =sT_0+\left(H_{0}^{}{}_{}{}^{2}+(\mathrm{tr}H_0)H_0+V_0\mathrm{I}_n\right)y.\hfill \end{array}$$
The right hand side of the first equation of (3.7) takes values in $`𝔰𝔬(n)`$ and the right hand side of the second equation of (3.7) takes values in $`^n`$. Thus, this is $`\frac{1}{2}n(n1)+n`$ equations for the $`\frac{1}{2}n(n+1)+n`$ components of $`s={}_{}{}^{t}s`$ and $`y`$. Consequently, the space of solutions is at least of dimension $`n`$. ∎
#### 3.3.2. The symmetry algebra
For $`(h,t,v)i𝔲(n)^n`$, let $`𝔤_{(h,t,v)}^n𝔲(n)`$ be the space of solutions $`(w,f)^n𝔲(n)`$ of the linear equations
(3.8)
$$\begin{array}{cc}\hfill 0& =hffh+tw^{}+wt^{}\hfill \\ \hfill 0& =ft+\left(h^2+(\mathrm{tr}h)h+v\mathrm{I}_n\right)w\hfill \end{array}$$
As was established in the course of the above proof, $`𝔤_{(h,t,v)}`$ is isomorphic as a vector space to the symmetry algebra $`𝔤`$ of any simply-connected Bochner-Kähler manifold whose structure function assumes the value $`(h,t,v)`$. In fact, the structure equations (2.14) show that, if $`(H(u),T(u),V(u))=(h,t,v)`$, then the vector space isomorphism $`𝔤𝔤_{(h,t,v)}`$ defined by $`X(\omega _u(X^{}),\varphi _u(X^{}))`$ induces a Lie algebra structure on $`𝔤_{(h,t,v)}`$ that is given by the formula
$$[(x,x^{}),(y,y^{})]=(x^{}y+y^{}x,[x^{},y^{}]+\{x,y\}_h)$$
where, for $`x,y^n`$, the element $`\{x,y\}_h`$ in $`𝔲(n)`$ is defined by
$$\begin{array}{cc}\hfill \{x,y\}_h& =h(xy^{}yx^{})(xy^{}yx^{})h+(x^{}yy^{}x)h\hfill \\ & +(x^{}hyy^{}hx)\mathrm{I}_n+(\mathrm{tr}h)\left((x^{}yy^{}x)\mathrm{I}_nxy^{}+yx^{}\right).\hfill \end{array}$$
For $`xM`$, let $`𝔤_x𝔤`$ denote the subalgebra that consists of the vector fields in $`𝔤`$ that vanish at $`x`$. Under the vector space isomorphism $`𝔤𝔤_{(h,t,v)}`$ defined above, $`𝔤_x`$ maps into the subalgebra $`𝔤_{(h,t,v)}^0𝔤_{(h,t,v)}`$ defined by $`w=0`$.
Since information about $`𝔤_{(h,t,v)}^0`$ and $`𝔤_{(h,t,v)}`$ will be needed later, these spaces will now be described more fully.
Fix $`(h,t,v)i𝔲(n)^n`$. Suppose that $`h_1>h_2>\mathrm{}>h_\delta `$ are the distinct eigenvalues of $`h`$, that $`h`$ has $`L_\alpha ^n`$ as its $`h_\alpha `$-eigenspace, and that $`n_\alpha 1`$ is the (complex) dimension of $`L_\alpha `$. Write
$$t=t_1+\mathrm{}+t_\delta $$
where $`t_\alpha `$ lies in $`L_\alpha `$. Define the quantities
$$v_\alpha =h_{\alpha }^{}{}_{}{}^{2}+(\mathrm{tr}h)h_\alpha +v+\underset{\beta \alpha }{}\frac{|t_\beta |^2}{(h_\alpha h_\beta )}.$$
Now define
$$\tau _\alpha =\{\begin{array}{cc}1\hfill & \\ 0\hfill & \\ 2n_\alpha \hfill & \end{array}\text{and}\rho _\alpha =\{\begin{array}{cc}(n_\alpha 1)^2\hfill & \text{if }t_\alpha 0\text{;}\hfill \\ (n_\alpha )^2\hfill & \text{if }t_\alpha =0\text{ but }v_\alpha 0\text{;}\hfill \\ (n_\alpha )^2\hfill & \text{if }t_\alpha =0\text{ and }v_\alpha =0\text{.}\hfill \end{array}$$
###### Proposition 4.
For any $`(h,t,v)i𝔲(n)^n`$,
$$dim𝔤_{(h,t,v)}^0=\rho _1+\mathrm{}+\rho _\delta $$
and
$$dim𝔤_{(h,t,v)}=dim𝔤_{(h,t,v)}^0+\tau _1+\mathrm{}+\tau _\delta .$$
###### Proof.
Because all the integers involved are invariant under the action of $`\mathrm{U}(n)`$, it suffices to prove this formula in the case that $`(h,t,v)`$ lies in $`C`$. Maintaining the notation introduced above, this means that
$$h=\left(\begin{array}{cccc}h_1\mathrm{I}_{n_1}& 0& \mathrm{}& 0\\ 0& h_2\mathrm{I}_{n_2}& \mathrm{}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& \mathrm{}& h_\delta \mathrm{I}_{n_\delta }\end{array}\right)\text{and}t=\left(\begin{array}{c}t_1\\ t_2\\ \mathrm{}\\ t_\delta \end{array}\right),$$
where $`t_\alpha `$ takes values in $`^{n_\alpha }`$ for $`1\alpha \delta `$ and has all of its entries equal to zero except possibly the top one, which is nonnegative.
For $`f𝔲(n)`$, write $`f`$ in ‘block’ form as $`f=(f_{\alpha \overline{\beta }})`$ where $`f_{\alpha \overline{\beta }}=f_{\beta \overline{\alpha }}^{}{}_{}{}^{}`$ takes values in $`n_\alpha `$-by-$`n_\beta `$ complex matrices for $`1\alpha ,\beta \delta `$. Correspondingly, write $`w^n`$ in ‘block’ form as $`w=(w_\alpha )`$ where $`w_\alpha `$ takes values in $`^{n_\alpha }`$. Then the first equation of (3.8) breaks into blocks as
$$0=(h_\alpha h_\beta )f_{\alpha \overline{\beta }}+t_\alpha w_{\beta }^{}{}_{}{}^{}+w_\alpha t_{\beta }^{}{}_{}{}^{}.$$
When $`\alpha =\beta `$, this forces $`t_\alpha w_{\alpha }^{}{}_{}{}^{}+w_\alpha t_{\alpha }^{}{}_{}{}^{}=0`$, so that either $`t_\alpha =0`$, in which case this places no restriction on $`w_\alpha `$ or else $`t_\alpha 0`$, in which case $`w_\alpha `$ must be a purely imaginary multiple of $`t_\alpha `$, say $`w_\alpha =ir_\alpha t_\alpha `$ for some $`r_\alpha `$. In either case, $`w_{\alpha }^{}{}_{}{}^{}t_\alpha `$ is purely imaginary.
When $`\alpha \beta `$, the above equation can be written as
$$f_{\alpha \overline{\beta }}=\frac{t_\alpha w_{\beta }^{}{}_{}{}^{}+w_\alpha t_{\beta }^{}{}_{}{}^{}}{(h_\beta h_\alpha )},\alpha \beta .$$
Substituting this equation into the second equation of (3.8) yields
$$0=f_{\alpha \overline{\alpha }}t_\alpha \underset{\beta \alpha }{}\frac{t_\alpha w_{\beta }^{}{}_{}{}^{}+w_\alpha t_{\beta }^{}{}_{}{}^{}}{(h_\beta h_\alpha )}t_\beta +(h_{\alpha }^{}{}_{}{}^{2}+(\mathrm{tr}h)h_\alpha +v)w_\alpha ,$$
which, by the definition of $`v_\alpha `$ and the purely imaginary nature of $`w_{\beta }^{}{}_{}{}^{}t_\beta `$, can be written in the form
$$0=f_{\alpha \overline{\alpha }}t_\alpha +\left(\underset{\beta \alpha }{}\frac{t_{\beta }^{}{}_{}{}^{}w_\beta }{(h_\beta h_\alpha )}\right)t_\alpha +v_\alpha w_\alpha ,$$
Now, if $`t_\alpha 0`$, then this equation can be written in the form
$$0=\left(f_{\alpha \overline{\alpha }}+\left(iv_\alpha r_\alpha +\underset{\beta \alpha }{}\frac{t_{\beta }^{}{}_{}{}^{}w_\beta }{(h_\beta h_\alpha )}\right)\frac{t_\alpha t_{\alpha }^{}{}_{}{}^{}}{|t_\alpha |^2}\right)t_\alpha ,$$
which is $`2n_\alpha 1`$ real equations for the $`n_{\alpha }^{}{}_{}{}^{2}`$ entries of $`f_{\alpha \overline{\alpha }}`$. In fact, the solutions of this equation can be written in the form
$$f_{\alpha \overline{\alpha }}=f_{\alpha \overline{\alpha }}^{}+\left(iv_\alpha r_\alpha +\underset{\beta \alpha }{}\frac{t_{\beta }^{}{}_{}{}^{}w_\beta }{(h_\beta h_\alpha )}\right)\frac{t_\alpha t_{\alpha }^{}{}_{}{}^{}}{|t_\alpha |^2}$$
where $`f_{\alpha \overline{\alpha }}^{}𝔲(n_\alpha )`$ is any solution to $`f_{\alpha \overline{\alpha }}^{}t_\alpha =0`$, an equation that defines the stabilizer subalgebra of $`t_\alpha `$ in $`𝔲(n_\alpha )`$ and so has a solution space of dimension $`(n_\alpha 1)^2`$.
If $`t_\alpha =0`$, then the equation above simplifies to $`v_\alpha w_\alpha =0`$. If $`v_\alpha 0`$, then this implies that $`w_\alpha =0`$ while if $`v_\alpha =0`$, the equation degenerates to an identity.
In particular, it follows that the equations (3.8) impose no interrelations among the $`w_\alpha `$, just the condition $`w_\alpha =ir_\alpha t_\alpha `$ if $`t_\alpha 0`$, the condition $`w_\alpha =0`$ if $`t_\alpha =0`$ but $`v_\alpha 0`$, and no condition on $`w_\alpha `$ if $`t_\alpha =v_\alpha =0`$.
Moreover, once the $`w_\alpha `$ have been chosen subject to these conditions, the $`f_{\alpha \overline{\beta }}`$ for $`\alpha \beta `$ are completely determined while the $`f_{\alpha \overline{\alpha }}𝔲(n_\alpha )`$ are determined up to a choice of $`f_{\alpha \overline{\alpha }}^{}`$ if $`t_\alpha 0`$ or are freely specifiable if $`t_\alpha =0`$.
The desired dimension formulae follow immediately. ∎
#### 3.3.3. Orbit dimension and slices
The proof of Proposition 4 shows how to compute the dimension of the $`x`$-orbit for any $`xM`$ for which there is a coframe $`uP_x`$ with $`(H(u),T(u),V(u))=(h,t,v)`$. Maintain the notation introduced above for the invariants of $`(h,t,v)`$.
Let $`O_xT_xM`$ be the tangent to the orbit through $`x`$. Then $`u(O_x)^n`$ is the direct sum of the lines $`t_\alpha L_\alpha `$ for those $`\alpha `$ with $`t_\alpha 0`$ and the subspaces $`L_\alpha `$ for those $`\alpha `$ with $`t_\alpha =0`$ and $`v_\alpha =0`$. Thus, the dimension this of the $`x`$-orbit is equal to $`\tau _1+\mathrm{}+\tau _\delta `$.
A more interesting result is the calculation of a near-slice to the orbits near $`x`$. For each $`\alpha `$, let $`t_\alpha ^{}L_\alpha `$ be the (complex) subspace $`\{wL_\alpha t_\alpha ^{}w=0\}`$.
If $`O_x^{}T_xM`$ is the perpendicular to $`O_x`$, then $`u(O_x^{})^n`$ is the direct sum of the subspaces $`it_\alpha t_\alpha ^{}`$ for those $`\alpha `$ with $`t_\alpha 0`$ together with the subspaces $`L_\alpha `$ for those $`\alpha `$ with $`t_\alpha =0`$ and $`v_\alpha 0`$.
Now, from the description of $`𝔤_{(h,t,v)}^0`$, it follows that the flows of the vector fields in $`𝔤_x`$ generate a group of rotations about $`x`$ that, via the unitary identification $`u:T_xM^n`$, is carried isomorphically into the product of the unitary groups $`\mathrm{U}(t_\alpha ^{})`$ for all $`\alpha `$. This is a closed subgroup of $`\mathrm{U}(n)`$ that evidently preserves the subspace $`u(O_x^{})`$.
A near-slice to this action can be constructed as follows: For each $`\alpha `$ with $`t_\alpha 0`$ and $`t_\alpha ^{}=0`$ (i.e., $`n_\alpha =1`$), let $`S_\alpha =it_\alpha `$. If $`t_\alpha 0`$ and $`t_\alpha ^{}0`$ (i.e., $`n_\alpha >1`$), choose a unit vector $`s_\alpha t_\alpha ^{}`$ and let $`S_\alpha =it_\alpha s_\alpha `$. If $`t_\alpha =0`$ and $`v_\alpha 0`$, choose a unit vector $`s_\alpha t_\alpha ^{}=L_\alpha `$ and let $`S_\alpha =s_\alpha `$. Finally, if $`t_\alpha =0`$ and $`v_\alpha =0`$, then set $`S_\alpha =0`$.
Then the direct sum $`S_1\mathrm{}S_\delta u(O_x^{})`$ is of the form $`u(S_x)`$ for a subspace $`S_xO_x^{}`$ that is a near-slice to the action of the isometric rotations about $`x`$ Consequently, the submanifold $`\mathrm{exp}_x(S_x)`$ near $`x`$ meets each orbit in a finite number of points and meets the generic orbit transversely. Let $`m_\alpha =dimS_\alpha `$, so that
$$m_\alpha =\{\begin{array}{cc}2\hfill & \text{if }t_\alpha 0\text{ and }n_\alpha >1\text{;}\hfill \\ 1\hfill & \text{if }t_\alpha 0\text{ and }n_\alpha =1\text{;}\hfill \\ 1\hfill & \text{if }t_\alpha =0\text{ but }v_\alpha 0\text{;}\hfill \\ 0\hfill & \text{if }t_\alpha =0\text{ and }v_\alpha =0\text{.}\hfill \end{array}$$
Then the ‘generic’ orbit in $`M`$ has codimension $`m=m_1+\mathrm{}+m_\delta `$. Since $`m_\alpha n_\alpha `$ for all $`\alpha `$, it follows that $`mn`$.
#### 3.3.4. Minimal symmetry
By Proposition 4, if $`(h,t,v)C`$ satisfies $`h_{i\overline{ı}}>h_{j\overline{ȷ}}`$ for $`i<j`$ and $`t_i>0`$ for all $`i`$ then $`dim𝔤_{(h,t,v)}=n`$. Thus, any simply-connected Bochner-Kähler manifold whose structure function assumes such a $`(h,t,v)`$ must have its symmetry algebra $`𝔤`$ be of dimension $`n`$ exactly. Moreover, from the above discussion, it follows that the generic orbits of such a Bochner-Kähler structure have codimension $`n`$, the maximum possible.
Thus, a ‘generic’ Bochner-Kähler structure has its infinitesimal symmetry algebra of dimension $`n`$ as well as cohomogeneity equal to $`n`$.
### 3.4. Constants of the structure
In the previous subsection, it was shown that the structure function $`(H,T,V):Pi𝔲(n)^n`$ has rank at most $`n^2+n`$. Since the image of $`P`$ under this map is $`\mathrm{U}(n)`$-invariant, it is natural to look for a set of $`\mathrm{U}(n)`$-invariant polynomials whose simultaneous level sets will contain the images of structure functions. In this section, I will exhibit $`n+1`$ such polynomials and show that they are independent.
#### 3.4.1. Conserved polynomials
Let $`\mathrm{\Omega }`$ be any Bochner-Kähler structure on a connected complex manifold $`M`$ and let $`\pi :PM`$ be the unitary coframe bundle, with canonical forms $`\omega `$ and $`\varphi `$ and structure functions $`S`$, $`T`$, and $`V`$ as above.
By (2.14), there are identities
$$\begin{array}{cc}\hfill d(\mathrm{tr}H)& =(T^{}\omega +\omega ^{}T),\hfill \\ \hfill d(\mathrm{tr}H^2)& =2(T^{}H\omega +\omega ^{}HT),\hfill \end{array}$$
Thus, by the last equation of (2.14)
$$dV=(\mathrm{tr}H)d(\mathrm{tr}H)+\frac{1}{2}d(\mathrm{tr}H^2).$$
Since $`P`$ is connected, there is a constant $`C_2`$ for which
$$V\frac{1}{2}\mathrm{tr}(H^2)\frac{1}{2}(\mathrm{tr}H)^2=C_2.$$
I am now going to show that this example can be generalized by constructing $`n`$ additional polynomials on $`i𝔲(n)^n`$ that have this constancy property.
Define $`A_k`$ and $`B_k`$ for $`k0`$ by the formulae
(3.9)
$$\begin{array}{cc}\hfill A_k& =\mathrm{tr}(H^k),\hfill \\ \hfill B_0& =1,B_1=\mathrm{tr}H,B_2=V,\hfill \\ \hfill B_k& =T^{}H^{k3}T,k3.\hfill \end{array}$$
Because these functions are constant on the fibers of $`\pi :PM`$, they can be regarded as the pullbacks to $`P`$ of well-defined smooth functions on $`M`$. In what follows, I will usually treat them as functions on $`M`$. For convenience, define $`A_k=B_k=0`$ when $`k<0`$.
Also, for $`0kn`$, let $`h_k`$ denote the $`k`$-th elementary symmetric function of the eigenvalues of $`H`$. These functions can be expressed as polynomials in the $`A_k`$ and hence are smooth functions on $`M`$. For example, $`h_0=1`$, $`h_1=A_1`$, $`h_2=\frac{1}{2}(A_{1}^{}{}_{}{}^{2}A_2)`$, etc. For convenience, set $`h_k=0`$ for $`k<0`$ or $`k>n`$.
###### Theorem 3.
For any connected Bochner-Kähler $`n`$-manifold $`(M,\mathrm{\Omega })`$, the functions
(3.10)
$$C_k=B_kh_1B_{k1}+h_2B_{k2}\mathrm{}+(1)^{k1}h_{k1}B_1+(1)^kh_kB_0.$$
are locally constant for $`2kn+2`$.
###### Example 4 (Lowest constants).
For example, in addition to the evident constancy of the function
$$C_2=B_2h_1B_1+h_2B_0=B_2\frac{1}{2}A_2\frac{1}{2}A_{1}^{}{}_{}{}^{2},$$
one has the constancy of
$$C_3=B_3h_1B_2+h_2B_1h_3B_0=B_3A_1B_2\frac{1}{3}\left(A_3A_{1}^{}{}_{}{}^{3}\right).$$
The reader may notice that the above formula for $`C_k`$ makes sense for $`k=1`$ and for $`k>n+2`$. Now, the expression $`C_1`$ is just $`B_1h_1=B_1A_1`$, which vanishes by definition. When $`kn+3`$, applying the Cayley-Hamilton theorem to the definition of $`C_k`$ yields
$$C_k=T^{}H^{kn3}(H^nh_1H^{n1}+h_2H^{n2}\mathrm{}+(1)^nh_n\mathrm{I}_n)T=0.$$
However, when $`2kn+2`$, the expression $`C_k`$ is a nontrivial polynomial of weighted degree $`k`$ in the variables $`A_j`$ and $`B_j`$. In fact, the above expressions for $`(C_2,\mathrm{},C_{n+2})`$ can obviously be solved for $`(B_2,\mathrm{},B_{n+2})`$.
###### Proof.
Define 1-forms $`\alpha _0=0`$ and
(3.11)
$$\alpha _{k+1}=T^{}H^k\omega +\omega ^{}H^kT,\text{for }k0.$$
(The indexing is determined by ‘scaling weight’ considerations.) The $`\alpha _k`$ are visibly $`\pi `$-semibasic, but they are also invariant under the $`\mathrm{U}(n)`$-action on $`P`$. Thus, they are the $`\pi `$-pullbacks of well-defined 1-forms on $`M`$. Consequently, they will, by abuse of language, be treated as 1-forms on $`M`$.
The first step will be to prove the following identities for all $`k0`$:
(3.12)
$$\begin{array}{cc}\hfill dA_k& =k\alpha _k,d\alpha _k=0,\hfill \\ \hfill dB_k& =B_0\alpha _k+B_1\alpha _{k1}+B_2\alpha _{k2}+\mathrm{}+B_{k1}\alpha _1.\hfill \end{array}$$
Now, the first set of identities is just a calculation. The case $`k=0`$ is obvious, so assume $`k>0`$. Using $`\mathrm{tr}(PQ)=\mathrm{tr}(QP)`$ and the structure equation, one has
$$\begin{array}{cc}\hfill dA_k& =d\left(\mathrm{tr}(H^k)\right)=k\mathrm{tr}(H^{k1}dH)\hfill \\ & =k\mathrm{tr}\left(H^{k1}(T\omega ^{}+\omega T^{})\right)=k\alpha _k.\hfill \end{array}$$
(The terms in $`dH`$ involving $`\varphi `$ cancel since $`A_k`$ is constant on the fibers of $`\pi `$.) Taking the exterior derivative of this relation and dividing by $`k`$ yields
$$0=d\alpha _k.$$
The second set is a little more complicated, but still just a calculation. The case $`k=0`$ is trivial, and the case $`k=1`$ follows from the fact that $`B_1=A_1`$, so $`dB_1=dA_1=\alpha _1=B_0\alpha _1`$.
Because $`B_2=V`$, the second identity for $`k=2`$ is just the structure equation for $`dV`$. Also,
$$\begin{array}{cc}\hfill dB_3& =dT^{}T+T^{}dT=\omega ^{}\left(H^2+(\mathrm{tr}H)H+V\mathrm{I}_n\right)T+T^{}\left(H^2+(\mathrm{tr}H)H+V\mathrm{I}_n\right)\omega \hfill \\ & =B_0\alpha _3+B_1\alpha _2+B_2\alpha _1,\hfill \end{array}$$
verifying the formula when $`k=3`$. Thus, suppose from now on that $`k>3`$ and compute (again ignoring terms involving $`\varphi `$, which must cancel)
$$\begin{array}{cc}\hfill dB_k& =dT^{}H^{k3}T+T^{}H^{k3}dT+\underset{l=0}{\overset{k4}{}}T^{}H^ldHH^{kl4}T\hfill \\ & =\omega ^{}\left(H^2+(\mathrm{tr}H)H+V\mathrm{I}_n\right)H^{k3}T+T^{}H^{k3}\left(H^2+(\mathrm{tr}H)H+V\mathrm{I}_n\right)\omega \hfill \\ & +\underset{l=0}{\overset{k4}{}}T^{}H^l(T\omega ^{}+\omega T^{})H^{kl4}T\hfill \\ & =\alpha _k+(\mathrm{tr}H)\alpha _{k1}+V\alpha _{k2}+\underset{l=0}{\overset{k4}{}}B_{l+3}\omega ^{}H^{kl4}T+T^{}H^l\omega B_{kl1}\hfill \\ & =B_0\alpha _k+B_1\alpha _{k1}+B_2\alpha _{k2}+\underset{l=0}{\overset{k4}{}}B_{l+3}\alpha _{kl3}\hfill \\ & =B_0\alpha _{k+2}+B_1\alpha _{k1}+B_2\alpha _k+B_3\alpha _{k3}+\mathrm{}+B_{k1}\alpha _1.\hfill \end{array}$$
Thus, the formulae (3.12) are established.
Now, I claim that the functions $`h_k`$ satisfy the differential equations
(3.13)
$$dh_k=h_{k1}\alpha _1h_{k2}\alpha _2+\mathrm{}+(1)^{k+1}h_0\alpha _k.$$
Granting (3.13) for the moment, computation gives
(3.14)
$$\begin{array}{cc}\hfill dC_k& =d\left(\underset{j=0}{\overset{k}{}}(1)^jh_jB_{kj}\right)=\underset{j=0}{\overset{k}{}}(1)^j\left(dh_jB_{kj}+h_jdB_{kj}\right)\hfill \\ & =\underset{j=0}{\overset{k}{}}(1)^j\left(\underset{l=0}{\overset{j}{}}(1)^{jl+1}h_l\alpha _{jl}B_{kj}+\underset{l=0}{\overset{kj}{}}h_jB_l\alpha _{kjl}\right)\hfill \\ & =\underset{j=0}{\overset{k}{}}\underset{l=0}{\overset{j}{}}(1)^{l+1}h_lB_{kj}\alpha _{jl}+\underset{j=0}{\overset{k}{}}\underset{l=0}{\overset{kj}{}}(1)^jh_jB_l\alpha _{kjl}\hfill \\ & =\underset{l=0}{\overset{k}{}}\underset{j=l}{\overset{k}{}}(1)^{l+1}h_lB_{kj}\alpha _{jl}+\underset{l=0}{\overset{k}{}}\underset{j=0}{\overset{kl}{}}(1)^lh_lB_j\alpha _{kjl}\hfill \\ & =\underset{l=0}{\overset{k}{}}\underset{j=l}{\overset{k}{}}(1)^{l+1}h_lB_{kj}\alpha _{jl}+\underset{l=0}{\overset{k}{}}\underset{j=l}{\overset{k}{}}(1)^lh_lB_{kj}\alpha _{jl}\hfill \\ & =0.\hfill \end{array}$$
It remains to verify (3.13). This is a classical identity: Let $`\lambda _1,\mathrm{},\lambda _n`$ be free variables, let $`s_k`$ be the $`k`$-th elementary function of the $`\lambda _i`$, and let $`p_k`$ be the $`k`$-th power function of the $`\lambda _i`$, i.e., $`p_k=\lambda _{1}^{}{}_{}{}^{k}+\mathrm{}+\lambda _{n}^{}{}_{}{}^{k}`$. For any constant $`t`$, one has
$$(1+s_1t+s_2t^2+\mathrm{}+s_nt^n)=(1+\lambda _1t)\mathrm{}(1+\lambda _nt).$$
Taking the logarithm and then computing the differential of both sides yields
$$\frac{(ds_1t+ds_2t^2+\mathrm{}+ds_nt^n)}{(1+s_1t+s_2t^2+\mathrm{}+s_nt^n)}=\underset{i=0}{\overset{n}{}}\frac{td\lambda _i}{1+\lambda _it}.$$
Expanding the right hand side out as a formal geometric power series in $`t`$ and collecting like powers of $`t`$ yields
$$\frac{(ds_1t+ds_2t^2+\mathrm{}+ds_nt^n)}{(1+s_1t+s_2t^2+\mathrm{}+s_nt^n)}=\underset{k=1}{\overset{\mathrm{}}{}}\left(\frac{(1)^{k+1}}{k}dp_k\right)t^k.$$
It follows that
$$\frac{(dh_1t+dh_2t^2+\mathrm{}+dh_nt^n)}{(1+h_1t+h_2t^2+\mathrm{}+h_nt^n)}=\underset{k=1}{\overset{\mathrm{}}{}}\left(\frac{(1)^{k+1}}{k}dA_k\right)t^k=\underset{k=1}{\overset{\mathrm{}}{}}(1)^{k+1}\alpha _kt^k.$$
Thus,
(3.15)
$$dh_1t+\mathrm{}+dh_nt^n=(1+h_1t+\mathrm{}+h_nt^n)(\alpha _1t\alpha _2t^2+\alpha _3t^3\mathrm{}),$$
which, after equating coefficients of like powers of $`t`$ on each side, is (3.13). ∎
#### 3.4.2. The moduli map
The map $`f:MF_n^{2n+1}`$ defined in §3.2 satisfies
$$f=(A_1,\mathrm{},A_n,B_2,B_3,\mathrm{},B_{n+2}).$$
As is well known,<sup>11</sup><sup>11</sup>11 In fact, this mapping can be computed by comparing like powers of $`t`$ in the formal series expansions of the identity
$$1+h_1t+\mathrm{}+h_nt^n=\mathrm{exp}(A_1t\frac{1}{2}A_2t^2+\frac{1}{3}A_3t^3\mathrm{}).$$
The mapping in the other direction can be computed by taking the logarithm of both sides, expanding the left side as a series in $`t`$, and then comparing like powers. there is a unique, invertible weighted-homogeneous polynomial mapping $`\mathrm{\Sigma }:^n^n`$ that satisfies
$$\mathrm{\Sigma }(A_1,\mathrm{},A_n)=(h_1,\mathrm{},h_n).$$
From the definition of the $`C_k`$ as polynomials in $`B_j`$ and $`h_j`$, it follows easily that $`\mathrm{\Sigma }`$ can be extended to an invertible, weighted-homogeneous polynomial mapping $`\mathrm{\Delta }:^{2n+1}^{2n+1}`$ so that
$$\mathrm{\Delta }f=(h_1,\mathrm{},h_n,C_2,C_3,\mathrm{},C_{n+2}).$$
Thus, the fibers and the rank of the map $`\mathrm{\Delta }f:M^{2n+1}`$ are the same as for the map $`f:M^{2n+1}`$. In particular, the fibers of $`\mathrm{\Delta }f`$ are the orbits of the symmetry pseudo-groupoid of the Bochner-Kähler structure.
By Theorem 3, when $`M`$ is connected, the functions $`C_k`$ are constants, so the fibers and rank of $`f`$ are the same as the fibers and rank of the map $`h:M^n`$ defined by
$$h=(h_1,\mathrm{},h_n).$$
### 3.5. Central symmetries
The symmetry algebra $`𝔤`$ of a Bochner-Kähler structure $`\mathrm{\Omega }`$ on a connected $`M`$ turns out to contain a canonical central subalgebra $`𝔷`$, whose dimension is equal to the infinitesimal cohomogeneity of the structure.
#### 3.5.1. An isometry vector field
Consider the vector field $`Z_2^{}`$ on $`P`$ that satisfies
(3.16)
$$\omega (Z_2^{})=2iT,\varphi (Z_2^{})=2i(H^2+\mathrm{tr}(H)H+V\mathrm{I}_n).$$
(The indexing is dictated by scaling weight considerations.) Since $`\omega \varphi :T_uP^n𝔲(n)`$ is an isomorphism for all $`uP`$ and since $`H`$ is Hermitian symmetric while $`V`$ is real, this does indeed define a unique vector field $`Z_2^{}`$. Because of the $`\mathrm{U}(n)`$-equivariance of the structure functions, the vector field $`Z_2^{}`$ is invariant under the right $`\mathrm{U}(n)`$-action. Consequently, there is a unique vector field $`Z_2`$ on $`M`$ that is $`\pi `$-related to $`Z_2^{}`$, i.e., that satisfies $`\pi ^{}\left(Z_2^{}(u)\right)=Z_2\left(\pi (u)\right)`$. The structure equation $`dT+\varphi T=(H^2+\mathrm{tr}(H)H+V\mathrm{I}_n)\omega `$ coupled with the discussion in §2.1.2 shows that $`Z_2`$ is the real part of a holomorphic vector field on $`M`$.
Since
(3.17)
$$\pi ^{}(Z_2\text{ }\text{ }\mathrm{\Omega })=Z_2^{}\text{ }\text{ }\left(\frac{i}{2}\omega ^{}\omega \right)=\left(T^{}\omega +\omega ^{}T\right)=d\left(\mathrm{tr}(H)\right),$$
the flow of $`Z_2`$ is the $`\mathrm{\Omega }`$-Hamiltonian flow associated to $`h_1`$.
Alternatively, one can see directly that the (local) flow of $`Z_2`$ preserves the Bochner-Kähler structure. It has already been observed that $`Z_2`$ is $`\pi `$-related to the $`\mathrm{U}(n)`$-invariant vector field $`Z_2^{}`$ on $`P`$. The defining formulae for $`Z_2^{}`$ yields
$$\begin{array}{cc}\hfill 𝔏_{Z_2^{}}\omega & =d\left(\omega (Z_2^{})\right)+Z_2^{}\text{ }d\omega =d\left(\omega (Z_2^{})\right)+Z_2^{}\text{ }(\varphi \omega )\hfill \\ & =d(2iT)\varphi (Z_2^{})\omega +\varphi \omega (Z_2^{})\hfill \\ & =2i\left(dT(H^2+\mathrm{tr}(H)H+V\mathrm{I}_n)\omega +\varphi T\right)=0.\hfill \end{array}$$
Thus, the flow of $`Z_2^{}`$ preserves $`\omega `$. In turn, this implies that the flow of $`Z_2^{}`$ preserves $`\varphi `$ (since $`\varphi `$ is the unique $`𝔲(n)`$-valued 1-form that satisfies $`d\omega =\varphi \omega `$). Thus, the vector field $`Z_2^{}`$ is $`\pi `$-related to the symmetry vector field $`Z_2`$ of the $`\mathrm{U}(n)`$-structure that $`P`$ defines, i.e., the original Bochner-Kähler structure.
###### Remark 3 (Matsumoto’s observation).
To my knowledge it was Matsumoto \[18, Theorem 2\] who first observed that one could construct a holomorphic vector field on $`M`$ by $`\mathrm{\Omega }`$-dualizing the exterior derivative of the scalar curvature, at least when $`M`$ is compact. The vector field that he constructs is, up to a constant complex multiple, the same as the one whose real part is $`Z_2`$. The above argument shows that compactness actually plays no role; the holomorphicity of $`Z_2iJZ_2`$ is a purely local fact. Apparently, Matsumoto did not realize that some complex multiple of his vector field had a real part whose flow of $`Z_2`$ was not only holomorphic but isometric as well.
#### 3.5.2. The central algebra
I am now going to show that $`Z_2`$ is the first of a sequence of real parts of holomorphic vector fields on $`M`$ whose representative functions can be written down explicitly in terms of $`H`$ and $`T`$. For example, the next term in the sequence will be seen to be the vector field $`Z_3`$ whose representative function is $`z_3=2i\left(H\mathrm{tr}(H)\mathrm{I}_n\right)T`$.
###### Theorem 4.
For every $`k`$ in the range $`0kn1`$, the function
$$z_{k+2}=2i(1)^k\left(H^kh_1H^{k1}+h_2H^{k2}+\mathrm{}+(1)^kh_k\mathrm{I}_n\right)T$$
is the representative function of a vector field $`Z_{k+2}𝔤`$. Moreover, the span $`𝔷`$ of $`Z_2,\mathrm{},Z_{n+1}`$ lies in the center of $`𝔤`$.
For $`xM`$, the subspace $`𝔷_x=\mathrm{span}\{Z_2(x),\mathrm{},Z_{n+1}(x)\}`$ is the $`\mathrm{\Omega }`$-complement to $`\mathrm{ker}df_x`$, where $`f:M^{2n+1}`$ is the moduli mapping of §3.2.3.
If $`M`$ is connected, then $`dim𝔷n`$ is the maximum over $`xM`$ of $`dim𝔷_x`$. If $`dim𝔷=n`$, then $`𝔤=𝔷`$.
###### Remark 4.
While the formula for $`z_{k+2}`$ makes sense for all $`k0`$, this expression vanishes identically when $`kn`$, due to the Cayley-Hamilton Theorem.
###### Proof.
A computation like that done in the proof of (3.12) shows that, for $`k0`$,
(3.18)
$$d(H^kT)+\varphi H^kTH^kT\alpha _0+H^{k1}T\alpha _1+\mathrm{}+H^0T\alpha _kmod\omega .$$
Using this identity, a calculation analogous to (3.14) yields
$$d(z_{k+2})+\varphi z_{k+2}0mod\omega .$$
Thus, according to §2.1.2, each $`z_{k+2}`$ is the representative function of a vector field $`Z_{k+2}`$ on $`M`$ whose local flow is holomorphic.
Letting $`\stackrel{~}{Z}_{k+2}`$ be any vector field on $`P`$ so that $`\omega (\stackrel{~}{Z}_{k+2})=z_{k+2}`$,
(3.19)
$$\begin{array}{cc}\hfill \pi ^{}\left(Z_{k+2}\text{ }\mathrm{\Omega }\right)& =\stackrel{~}{Z}_{k+2}\text{ }\left(\frac{i}{2}\omega ^{}\omega \right)=\frac{i}{2}\left(z_{k+2}^{}\omega \omega ^{}z_{k+2}\right)\hfill \\ & =(1)^k(T^{}(H^kh_1H^{k1}+\mathrm{}+(1)^kh_k\mathrm{I}_n)\omega \hfill \\ & +\omega ^{}(H^kh_1H^{k1}+\mathrm{}+(1)^kh_k\mathrm{I}_n)T)\hfill \\ & =(1)^{k+1}\left(\alpha _{k+1}h_1\alpha _k+h_2\alpha _{k1}+\mathrm{}+(1)^kh_k\alpha _1\right)\hfill \\ & =dh_{k+1}=\pi ^{}\left(dh_{k+1}\right),\hfill \end{array}$$
by virtue of (3.13). Thus, $`Z_{k+2}\text{ }\text{ }\mathrm{\Omega }=dh_{k+1}`$, so that $`Z_{k+2}`$ is the $`\mathrm{\Omega }`$-Hamiltonian vector field associated to $`h_{k+1}`$.
Since the flow of $`Z_{k+2}`$ is both holomorphic and symplectic, $`Z_{k+2}`$ belongs to $`𝔤`$, as claimed.
Since the representative function $`z_{k+2}`$ is constructed as a polynomial in $`H`$ and $`T`$, which are invariant under the $`Y^{}`$-flow on $`P`$ for any vector field $`Y𝔤`$, it follows that $`Z_{k+2}`$ is invariant under the flow of any $`Y𝔤`$, i.e., $`[Y,Z_{k+2}]=0`$ for any $`Y𝔤`$. Thus, the $`Z_{k+2}`$ for $`0kn1`$ span a central subalgebra $`𝔷𝔤`$.
For any $`xM`$, the nondegeneracy of $`\mathrm{\Omega }`$ implies that $`𝔷_xT_xM`$ is the $`\mathrm{\Omega }`$-dual of
$$\mathrm{span}\{dh_1(x),\mathrm{},dh_n(x)\}=\mathrm{span}\{dA_1(x),\mathrm{},dA_n(x)\}$$
(see §3.4.2). The map $`f:M^{2n+1}`$ has components given by $`A_i`$ and $`B_i`$. By Theorem 3 and §3.4.2, each $`B_i`$ can be written on each connected component of $`M`$ as a weighted polynomial in $`A_1,\mathrm{},A_i`$ with constant coefficients. Thus, the kernel of $`df_x`$ is the same the kernel of $`(dA)_x`$ where $`A=(A_1,\mathrm{},A_n)`$, which establishes the stated $`\mathrm{\Omega }`$-complementarity.
Now suppose that $`M`$ is connected. For each $`k0`$, each $`xM`$, and each $`uP_x`$,
$$u\left(𝔷_x\right)=\mathrm{span}\left\{iH(u)^kT(u)0kn1\right\}.$$
Since $`H(u)`$ is Hermitian symmetric, it follows that the dimension of $`𝔷_x`$ over $``$ is the largest integer $`m_xn`$ so that the vectors $`T(u),H(u)T(u),\mathrm{},H(u)^{m_x1}T(u)`$ are linearly independent (over either $``$ or $``$) and, moreover, that $`𝔷_xJ𝔷_x=(0)_x`$.
Let $`M^{}M`$ be the (nonempty) open set consisting of those $`xM`$ for which $`m_x`$ achieves the maximum value $`mn`$. If $`m=n`$, then $`dim𝔷_x=n`$ for all $`xM^{}`$ and, since $`dim𝔷_xdim𝔷n`$ for all $`xM`$, it follows that $`dim𝔷=n`$. If $`m=0`$, then $`T`$ vanishes identically, implying that $`𝔷=(0)`$. If $`0<m<n`$, let $`k`$ satisfy $`mk<n`$. Then, on $`M^{}`$, the vector field $`Z_{k+2}`$ is a linear combination of the independent vector fields $`Z_2,\mathrm{},Z_{m+1}`$. Thus, there exist smooth real-valued functions $`w_2,\mathrm{}w_{m+1}`$ on $`M^{}`$ so that
$$Z_{k+2}=w_2Z_2+\mathrm{}+w_{m+1}Z_{m+1}.$$
Consequently,
$$Z_{k+2}iJZ_{k+2}=w_2(Z_2iJZ_2)+\mathrm{}+w_{m+1}(Z_{m+1}iJZ_{m+1}).$$
However, the left hand side of this equation is a holomorphic vector field while the holomorphic vector fields $`(Z_2iJZ_2),\mathrm{},(Z_{m+1}iJZ_{m+1})`$ are linearly independent (over $`)`$ at each point of $`M^{}`$. It follows that the functions $`w_2,\mathrm{},w_{m+1}`$ are real-valued holomorphic functions on $`M^{}`$ and hence must be constants. Since $`M`$ is connected, the identity
$$Z_{k+2}=w_2Z_2+\mathrm{}+w_{m+1}Z_{m+1}.$$
must hold on all of $`M`$. In other words, the vector fields $`Z_2,\mathrm{},Z_{m+1}`$ are a basis of $`𝔷`$, as desired.
If $`dim𝔷=n`$, then at any point $`xM^{}`$, the vectors $`Z_2(x),\mathrm{},Z_{n+1}(x)`$ are linearly independent. If $`\pi (u)=x`$, then $`\{H^k(u)T(u)0kn1\}`$ are linearly independent, implying (since $`H(u)`$ is diagonalizable) that $`T(u)`$ does not lie any sum of fewer that $`n`$ distinct eigenspaces of $`H(u)`$. By §3.3.4, this implies that the differential of the mapping $`(H,T,V):Pi𝔲(n)^n`$ has kernel of dimension equal to $`n`$, so $`dim𝔤=n`$, as desired. ∎
#### 3.5.3. The momentum mapping
The proof of Theorem 4 shows that the map $`h:M^n`$ defined by
(3.20)
$$h=(h_1,h_2,h_3,\mathrm{},h_n)$$
is a momentum mapping for the infinitesimal torus action generated by $`𝔷`$.<sup>12</sup><sup>12</sup>12More precisely, the infinitesimal $`n`$-torus action on $`M`$ is given by the Lie algebra homomorphism $`^n𝔷𝔛(M)`$ defined by the explicit generators $`Z_2,\mathrm{},Z_{n+1}`$.
As already remarked, there is an invertible weighted-homogeneous polynomial mapping $`\mathrm{\Delta }:^{2n+1}^{2n+1}`$ that satisfies
$$\mathrm{\Delta }(A_1,\mathrm{},A_n,B_2,\mathrm{},B_{n+2})=(h_1,\mathrm{},h_n,C_2,\mathrm{},C_{n+2}).$$
Thus, the fibers of $`\mathrm{\Delta }f:M^{2n+1}`$ are the orbits of the symmetry pseudo-groupoid of the underlying Bochner-Kähler structure. By Theorem 3, if $`M`$ is connected, then the functions $`C_k`$ are constant. Thus, for connected $`M`$, the fibers of $`h`$ are the orbits of this symmetry pseudo-groupoid.
### 3.6. Cohomogeneity and the momentum polynomial
Assume that $`M`$ is connected and that $`dim𝔷=m`$.
#### 3.6.1. Cohomogeneity
The proof of Theorem 4 shows that $`dh_k`$ for $`k>m`$ is a constant linear combination of the differentials $`dh_1,\mathrm{},dh_m`$ and that these latter $`1`$-forms are linearly independent on an open subset of $`M`$. Thus, $`h(M)`$ lies in an $`m`$-dimensional affine subspace $`𝔞^n`$ and, moreover, contains an open subset of $`𝔞`$. Since the fibers of $`h`$ are the orbits of the symmetry pseudo-groupoid, it is reasonable to call the number $`m`$ the *cohomogeneity* of the Bochner-Kähler structure.
#### 3.6.2. Cohomogeneity
Let $`t`$ be a parameter and define the *momentum polynomial* $`p_h(t)`$ of $`M`$ by the formula
$$p_h(t)=t^nh_1t^{n1}+\mathrm{}+(1)^nh_n.$$
Of course, $`p_h(t)=det(t\mathrm{I}_nH)`$ is the characteristic polynomial of the Hermitian symmetric matrix $`H`$, so all of its roots are real.
###### Theorem 5.
If $`(M,\mathrm{\Omega })`$ is a connected Bochner-Kähler manifold of cohomogeneity $`m`$, then $`nm`$ of the roots of $`p_h(t)`$ are constant and, outside a closed, proper, complex analytic subvariety $`NM`$, the remaining $`m`$ roots are distinct, real-analytic, and functionally independent.
###### Proof.
If $`m=0`$, then, in particular, $`h_1`$ is constant, so $`(M,\mathrm{\Omega })`$ has constant scalar curvature. By Proposition 1, $`(M,\mathrm{\Omega })`$ is locally homogeneous, so all of the eigenvalues of $`H`$ are constant. (By Proposition 1, there are at most two distinct eigenvalues.) In this case, $`N`$ can be taken to be empty.
Suppose from now on that $`m>0`$. Technically, I should treat the cases $`m=n`$ and $`m<n`$ separately, but the argument for $`m=n`$ differs from that for $`m<n`$ by trivial notational changes, so I will not explicitly assume $`m<n`$ but, rather, let the reader make the necessary modifications for the case $`m=n`$.
By Theorem 4, the differentials $`dh_1,\mathrm{},dh_m`$ are linearly independent exactly where the vector fields $`Z_2,\mathrm{},Z_{m+1}`$ are linearly independent. In particular, the locus $`NM`$ where $`dh_1\mathrm{}dh_m`$ vanishes is also where the holomorphic $`m`$-vector
$$(Z_2ıJZ_2)(Z_3ıJZ_3)\mathrm{}(Z_{m+1}ıJZ_{m+1})$$
vanishes. Thus, $`N`$ is a closed, proper, complex analytic subvariety of $`M`$ and so has real codimension at least $`2`$. Consequently, its complement $`M^{}M`$ is a connected, open, dense subset of $`M`$.
Since $`dim𝔷_x=m`$ for all $`xM^{}`$, a subbundle $`P_0P`$ can be defined over $`M^{}`$ by saying that $`u\pi ^1(M^{})`$ lies in $`P_0`$ if and only if $`u(𝔷_x)=i^m^m^n`$. Then $`\pi :P_0M^{}`$ is a smooth principal $`\mathrm{O}(m)\times \mathrm{U}(nm)`$-bundle over $`M^{}`$.
Pull the forms $`\omega `$ and $`\varphi `$ and the functions $`H`$, $`T`$, and $`V`$ back to $`P_0`$. By definition, for every $`uP_0`$, the vectors $`T(u),H(u)T(u),\mathrm{},H(u)^{m1}T(u)`$ span $`^m^n`$ and, moreover, $`H(u)^m^m`$. Thus, there exists a function $`T^{}:P_0^m`$, a function $`H^{}`$ on $`P_0`$ with values in the open set of symmetric $`m`$-by-$`m`$ (real) matrices with $`m`$ distinct eigenvalues, and a function $`H^{\prime \prime }`$ on $`P_0`$ with values in Hermitian symmetric $`(nm)`$-by-$`(nm)`$ matrices so that
(3.21)
$$T=\left(\begin{array}{c}T^{}\\ 0\end{array}\right),H=\left(\begin{array}{cc}H^{}& 0\\ 0& H^{\prime \prime }\end{array}\right).$$
Write $`\varphi =\varphi ^{}`$ in $`(m,nm)`$-block form as
(3.22)
$$\varphi =\left(\begin{array}{cc}\varphi ^{}& \tau ^{}\\ \tau & \varphi ^{\prime \prime }\end{array}\right)$$
where, of course, $`\varphi ^{}`$ and $`\varphi ^{\prime \prime }`$ take values in skew-Hermitian matrices of dimensions $`m`$ and $`nm`$, respectively. The lower right-hand $`(nm)`$-by-$`(nm)`$ block of the $`dH`$ equation in (2.14) then becomes
(3.23)
$$dH^{\prime \prime }=\varphi ^{\prime \prime }H^{\prime \prime }+H^{\prime \prime }\varphi ^{\prime \prime }.$$
Consequently, the eigenvalues of $`H^{\prime \prime }`$ are constant on $`P_0`$. Let
$$det(t\mathrm{I}_{nm}H^{\prime \prime })=p_{h^{\prime \prime }}(t)=t^{nm}h_1^{\prime \prime }t^{nm}+\mathrm{}+(1)^{nm}h_{nm}^{\prime \prime }$$
be the characteristic polynomial of $`H^{\prime \prime }`$, where the $`h_i^{\prime \prime }`$ are constants. Then, on $`M^{}`$ at least, $`p_{h^{\prime \prime }}(t)`$ divides $`p_h(t)`$. Using the Euclidean algorithm, write
$$p_h(t)=p_{h^{\prime \prime }}(t)q(t)+r(t)$$
where $`q`$ and $`t`$ are polynomials in $`t`$ and where the degree of $`r`$ is at most $`nm1`$. The coefficients of $`q`$ and $`r`$ are constant linear combinations of the coefficients in $`p_h`$ and so are continuous. Since the coefficients of $`r`$ vanish on $`M^{}`$, which is dense in $`M`$, it follows that $`r`$ vanishes identically on $`M`$. Thus, $`p_{h^{\prime \prime }}(t)`$ divides $`p_h(t)`$ on all of $`M`$.
Defining real-analytic functions $`h_1^{},\mathrm{},h_m^{}`$ on $`M`$ by
$$q(t)=t^mh_1^{}t^{m1}+\mathrm{}+(1)^mh_m^{},$$
one sees that $`q(t)=p_h^{}(t)=det(t\mathrm{I}_mH^{})`$ on $`M^{}`$, i.e., that $`p_h(t)=p_h^{}(t)p_{h^{\prime \prime }}(t)`$.
Of course the roots of $`p_h^{}(t)`$ on $`M^{}`$ equal the eigenvalues of $`H^{}`$ on $`P_0`$ and so are distinct and therefore real-analytic on $`M^{}`$. Since the $`h_i`$ are constant coefficient linear combinations of the $`h_j^{}`$, it follows that there is a constant $`a`$ so that
$$dh_1\mathrm{}dh_m=adh_1^{}\mathrm{}dh_m^{}.$$
Obviously, $`a`$ is nonzero and $`dh_1^{}\mathrm{}dh_m^{}`$ is nonvanishing on $`M^{}`$. Since the roots of $`p_h^{}(t)`$ are distinct on $`M^{}`$, and the $`h_i^{}`$ are the elementary symmetric functions of these roots, it follows that these roots must be functionally independent on $`M^{}`$, as claimed. ∎
###### Remark 5.
Theorem 5 accounts for the $`nm`$ constant coefficient linear relations among the momenta $`h_1,\mathrm{}h_n`$ implicit in the initial discussion. They are just the $`nm`$ coefficients of the remainder polynomial $`r(t)`$.
#### 3.6.3. Reduced momentum
The mapping $`h^{}=(h_1^{},\mathrm{},h_m^{}):M^m`$ will be known as the *reduced momentum mapping* of $`M`$. The proof of Theorem 5 shows that $`h^{}:M^{}^m`$ is a submersion onto its image.
The polynomial $`p_h^{}(t)`$ will be referred to as the *reduced momentum polynomial* of $`M`$. The roots of $`p_h^{}(t)`$ are real on $`M^{}`$, which is dense in $`M`$, so the roots of $`p_h^{}(t)`$ are real at every point of $`M`$. For each $`xM`$, let
$$\lambda _1(x)\lambda _2(x)\mathrm{}\lambda _m(x)$$
be the roots of $`p_h^{}(t)`$, counted with multiplicity. By a standard argument based on the Stone-Weierstraß theorem, the functions $`\lambda _i:M`$ are continuous.<sup>13</sup><sup>13</sup>13Note that $`\lambda _i`$ will be real-analytic even at $`xN`$ as long as it is a simple root of $`p_h^{}(t)`$ at $`x`$. Thus, the reduced momentum polynomial factors continuously as
$$p_h^{}(t)=(t\lambda _1)(t\lambda _2)\mathrm{}(t\lambda _m).$$
###### Example 5 (Low cohomogeneity).
Suppose $`M`$ is locally isometric to $`M_c^p\times M_c^{np}`$. Then, looking back at the proof of Proposition 1 and the definition of $`H`$, one computes that
$$p_h(t)=\left(t+\frac{c(np+1)}{2(n+2)}\right)^p\left(t\frac{c(p+1)}{2(n+2)}\right)^{np}.$$
Since this example is locally homogeneous, i.e., $`m=0`$, it follows that $`p_h(t)=p_{h^{\prime \prime }}(t)`$.
On the other hand, for Example 2 (i.e., rotationally symmetric),
$$p_h(t)=\left(t\frac{k}{(n+2)}\right)^{n1}\left(t\frac{k}{(n+2)}a|z|^2f^{}\left(|z|^2\right)\right).$$
As long as $`a0`$, these examples have cohomogeneity $`m=1`$, with
$$p_{h^{\prime \prime }}(t)=\left(t\frac{k}{(n+2)}\right)^{n1}\text{and}p_h^{}(t)=\left(t\frac{k}{(n+2)}a|z|^2f^{}\left(|z|^2\right)\right).$$
## 4. Global Geometry and Symmetries
Throughout this section it will be assumed that $`M`$ is a connected complex $`n`$-manifold endowed with a Bochner-Kähler structure $`\mathrm{\Omega }`$. All the notation introduced earlier will be retained.
### 4.1. The characteristic polynomials
In this section, two constant coefficient polynomials will be introduced that are invariants of the analytically connected equivalence class of the Bochner-Kähler structure. Also a formula (Theorem 6) will be developed to compute them from the value of the structure function at a single point.
#### 4.1.1. The characteristic polynomial
Let $`C_k`$ for $`k=2,\mathrm{},n+2`$ be the constants introduced in Theorem 3. For the sake of convenience, set $`C_0=1`$ and $`C_k=0`$ for $`k=1`$ and $`k>n+2`$. Let $`t`$ be a real parameter. Then by Theorem 3 (plus the remark following it),
$$\begin{array}{cc}\hfill \underset{k=0}{\overset{\mathrm{}}{}}C_kt^k& =\underset{k=0}{\overset{\mathrm{}}{}}\underset{l=0}{\overset{\mathrm{}}{}}(1)^lh_lB_{kl}t^k\hfill \\ & =\left(\underset{k=0}{\overset{\mathrm{}}{}}(1)^lh_lt^l\right)\left(\underset{j=0}{\overset{\mathrm{}}{}}B_jt^j\right)\hfill \\ & =det(\mathrm{I}_ntH)\left(1+h_1t+Vt^2+t^3\underset{k=3}{\overset{\mathrm{}}{}}T^{}(tH)^{k3}T\right)\hfill \\ & =det(\mathrm{I}_ntH)\left(1+h_1t+Vt^2+t^3T^{}(\mathrm{I}_ntH)^1T\right)\hfill \\ & =det(\mathrm{I}_ntH)(1+h_1t+Vt^2)+t^3T^{}\mathrm{Cof}(\mathrm{I}_ntH)T\hfill \end{array}$$
where $`\mathrm{Cof}(\mathrm{I}_ntH)`$ is the signed cofactor matrix<sup>14</sup><sup>14</sup>14The signed cofactor matrix of any $`n`$-by-$`n`$ matrix $`R`$ is the (unique) homogeneous polynomial matrix of degree $`(n1)`$ that satisfies the identity $`R\mathrm{Cof}(R)=det(R)\mathrm{I}_n`$. of $`\mathrm{I}_ntH`$.
The cautious reader may object that the second factors on the second and third lines need not converge for all $`t`$. However, every $`uP`$ has an open neighborhood on which $`T^{}T`$ and $`\mathrm{tr}(H^{}H)`$ are bounded, so that the series is bounded by a geometric series and hence converges for $`|t|`$ sufficiently small. The upshot of this is that the two series converge absolutely and uniformly on compact subsets of a certain open neighborhood of $`P\times 0`$ in $`P\times `$, so equality of the first and last terms holds on that open subset. The left hand side is evidently a polynomial in $`t`$ of degree at most $`n+2`$ and the final form of the right hand side is also a polynomial in $`t`$, so it follows that these first and last expressions are equal for all $`t`$.
Replacing $`t`$ by $`t^1`$ and multiplying through by $`t^{n+2}`$ gives the form of the identity that will be most useful, namely:
(4.1)
$$\underset{k=0}{\overset{n+2}{}}C_kt^{n+2k}=det(t\mathrm{I}_nH)(t^2+h_1t+V)+T^{}\mathrm{Cof}(t\mathrm{I}_nH)T.$$
The polynomial $`p_C(t)=t^{n+2}+C_2t^n+C_3t^{n2}+\mathrm{}+C_{n+2}`$ will be said to be the *characteristic polynomial* of the Bochner-Kähler structure.
###### Example 6 (Low cohomogeneity).
Suppose $`M`$ is locally isometric to $`M_c^p\times M_c^{np}`$. Then, looking back at the proof of Proposition 1 and the definition of $`H`$, one computes that
$$p_C(t)=\left(t+(np+1)r\right)^{p+1}\left(t(p+1)r\right)^{np+1},\text{where}r=\frac{c}{2(n+2)}.$$
For Example 2 (i.e., rotationally symmetric), the formula is
$$p_C(t)=\left(t2r\right)^n\left[\left(t+nr\right)^2\frac{1}{4}k^2+a\right],\text{where}r=\frac{k}{2(n+2)}.$$
#### 4.1.2. The reduced characteristic polynomial
Let $`P_1`$ be the set of those $`uP_0`$ that satisfy the condition that $`H^{}(u)`$ be diagonal, with eigenvalues arranged in descending order, and that each of the entries $`T_i(u)`$ of $`T^{}(u)^m`$ be positive. (See §3.6 for definitions.) Then $`P_1`$ is a $`\{\mathrm{I}_m\}\times \mathrm{U}(nm)`$-bundle over $`M^{}`$. Using the identities derived in §3.6, equation (4.1) can be written as
(4.2)
$$\frac{p_C(t)}{p_{h^{\prime \prime }}(t)}=\left(t^2+h_1t+V\right)\underset{j=1}{\overset{m}{}}(t\lambda _j)+\underset{i=1}{\overset{m}{}}T_i^2\underset{ji}{}(t\lambda _j).$$
In particular, $`p_{h^{\prime \prime }}(t)`$ divides $`p_C(t)`$. Denote the quotient by $`p_D(t)`$. It is a monic polynomial with constant coefficients of degree $`m+2`$ and will be called the *reduced characteristic polynomial*.
###### Example 7 (Low cohomogeneity).
Suppose $`M`$ is locally isometric to $`M_c^p\times M_c^{np}`$. Then, $`m=0`$ and
$$p_D(t)=\left(t+(np+1)r\right)\left(t(p+1)r\right),\text{where}r=\frac{c}{2(n+2)}.$$
For Example 2, where $`m=1`$, the formula is
$$p_D(t)=\left(t2r\right)\left[\left(t+nr\right)^2\frac{1}{4}k^2+a\right],\text{where}r=\frac{k}{2(n+2)}.$$
###### Proposition 5.
Every root of $`p_{h^{\prime \prime }}(t)`$ is also a root of $`p_D(t)`$.
###### Proof.
Let $`p_{h^{\prime \prime }}(t)`$ have roots $`\lambda _{m+1}\lambda _{m+2}\mathrm{}\lambda _n`$, counting multiplicity, and set
(4.3)
$$\mathrm{\Lambda }=\left(\begin{array}{cccc}\lambda _{m+1}& 0& \mathrm{}& 0\\ 0& \lambda _{m+2}& \mathrm{}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& \mathrm{}& \lambda _n\end{array}\right).$$
Let $`P_2P_1`$ consist of the coframes $`uP_1`$ for which $`H^{\prime \prime }(u)=\mathrm{\Lambda }`$. This $`P_2`$ is a bundle over $`M^{}`$ with structure group $`\{\text{I}_m\}\times G_\mathrm{\Lambda }`$, where $`G_\mathrm{\Lambda }\mathrm{U}(nm)`$ is the group of unitary matrices commuting with $`\mathrm{\Lambda }`$. All calculations will now take place on $`P_2`$.
Adopt the index range convention $`1i,j,km<a,b,c<n`$. Thus, for example $`T_i>0`$ but $`T_a=0`$. Also, $`H_{a\overline{ı}}=0`$. The $`(a,i)`$-entry of the structure equation (2.14) for $`dH`$ becomes (no sum over $`i`$)
(4.4)
$$(\lambda _a\lambda _i)\varphi _{a\overline{ı}}+T_i\omega _a=0.$$
Meanwhile, the structure equation for $`dT_a`$ becomes
(4.5)
$$(\lambda _{a}^{}{}_{}{}^{2}+h_1\lambda _a+V)\omega _a\underset{i=1}{\overset{m}{}}T_i\varphi _{a\overline{ı}}=0.$$
Combining these equations yields
$$\begin{array}{cc}\hfill \left((\lambda _{a}^{}{}_{}{}^{2}+h_1\lambda _a+V)\underset{i=1}{\overset{m}{}}(\lambda _a\lambda _i)\right)\omega _a& =\left(\underset{i=1}{\overset{m}{}}(\lambda _a\lambda _i)\right)\underset{j=1}{\overset{m}{}}T_j\varphi _{a\overline{ȷ}}\hfill \\ & =\underset{j=1}{\overset{m}{}}\left(\underset{ij}{}(\lambda _a\lambda _i)\right)T_j(\lambda _a\lambda _j)\varphi _{a\overline{ȷ}}\hfill \\ & =\left(\underset{j=1}{\overset{m}{}}T_{j}^{}{}_{}{}^{2}\underset{ij}{}(\lambda _a\lambda _i)\right)\omega _a.\hfill \end{array}$$
Since $`\omega _a`$ is nonzero on $`P_2`$, it follows that
$$p_D(\lambda _a)=\left(\lambda _{a}^{}{}_{}{}^{2}+h_1\lambda _a+V\right)\underset{j=1}{\overset{m}{}}(\lambda _a\lambda _j)+\underset{i=1}{\overset{m}{}}T_i^2\underset{ji}{}(\lambda _a\lambda _j)=0,$$
as desired. ∎
#### 4.1.3. Point data
In this subsubsection, a formula will be developed for the characteristic polynomials of a connected Bochner-Kähler structure in terms of a single value of its structure function. The formula for $`p_C`$ is, of course, already given by (4.1). However, the formula for $`p_D`$ is somewhat more subtle.
First, recall the concepts introduced in §3.3, suitably modified for the present section. For any $`(H_0,T_0,V_0)i𝔲(n)^n`$, let $`H_1,H_2,\mathrm{},H_\delta `$ be the distinct eigenvalues of $`H_0`$. Let $`L_\alpha ^n`$ be the eigenspace of $`H_0`$ belonging to the eigenvalue $`H_\alpha `$ and let $`n_\alpha 1`$ be the (complex) dimension of $`L_\alpha `$. Write
(4.6)
$$T_0=T_1+\mathrm{}+T_\delta $$
where $`T_\alpha `$ lies in $`L_\alpha `$ for $`1\alpha \delta `$. Define the quantities
(4.7)
$$V_\alpha =H_{\alpha }^{}{}_{}{}^{2}+(\mathrm{tr}H_0)H_\alpha +V_0+\underset{\beta \alpha }{}\frac{|T_\beta |^2}{(H_\alpha H_\beta )}$$
and
(4.8)
$$m_\alpha =\{\begin{array}{cc}2\hfill & \text{if }T_\alpha 0\text{ and }n_\alpha >1\text{;}\hfill \\ 1\hfill & \text{if }T_\alpha 0\text{ and }n_\alpha =1\text{;}\hfill \\ 1\hfill & \text{if }T_\alpha =0\text{ and }V_\alpha 0\text{;}\hfill \\ 0\hfill & \text{if }T_\alpha =0\text{ and }V_\alpha =0\text{.}\hfill \end{array}$$
###### Theorem 6.
If $`(M^n,g,\mathrm{\Omega })`$ is a connected Bochner-Kähler manifold whose structure function $`(H,T,V)`$ assumes the value $`(H_0,T_0,V_0)`$, then
(4.9)
$$p_C(t)=\underset{\alpha =1}{\overset{\delta }{}}(tH_\alpha )^{n_\alpha }\left[t^2+(\mathrm{tr}H_0)t+V_0+\underset{\alpha =1}{\overset{\delta }{}}\frac{|T_\alpha |^2}{(tH_\alpha )}\right]$$
and
(4.10)
$$p_D(t)=\underset{\alpha =1}{\overset{\delta }{}}(tH_\alpha )^{m_\alpha }\left[t^2+(\mathrm{tr}H_0)t+V_0+\underset{\alpha =1}{\overset{\delta }{}}\frac{|T_\alpha |^2}{(tH_\alpha )}\right]\text{.}$$
###### Proof.
The formula for $`p_C`$ follows directly from (4.1), so the formula for $`p_D`$ will follow from the equivalent statement
(4.11)
$$p_{h^{\prime \prime }}(t)=\underset{\alpha =1}{\overset{\delta }{}}(tH_\alpha )^{n_\alpha m_\alpha },$$
and this is what will be proved.
By §3.3.3, the generic orbit of the symmetry pseudo-groupoid has codimension equal to $`m_1+\mathrm{}+m_\delta `$. By §3.5.3 and Theorem 5, the orbits of the symmetry pseudo-groupoid in $`M^{}`$ (which is open and dense in $`M`$) have codimension $`m`$. Consequently, $`m=m_1+\mathrm{}+m_\delta `$, so that $`nm=(n_1m_1)+\mathrm{}+(n_\delta m_\delta )`$. Moreover, the inequality $`n_\alpha m_\alpha `$ follows immediately from the definitions.
Thus, by the very definition of $`p_{h^{\prime \prime }}(t)`$, it will suffice to show that, for each $`\alpha `$, the polynomial $`p_h(t)`$ has a constant root $`H_\alpha `$ of multiplicity at least $`n_\alpha m_\alpha `$. By Theorem 5, it suffices to to show this constancy in an open neighborhood of the point $`xM`$ for which there exists a $`uP_x`$ satisfying $`(H(u),T(u),V(u))=(H_0,T_0,V_0)`$.
Thus, let $`w^n`$ be a nonzero vector and let $`c:(\epsilon ,\epsilon )M`$ be the constant speed geodesic satisfying $`u\left(\dot{c}(0)\right)=w`$. Then $`c`$ can be lifted uniquely to a curve $`\gamma :(\epsilon ,\epsilon )P`$ that satisfies $`\gamma (0)=u`$ and $`\gamma ^{}\varphi =0`$ (i.e., the coframe field $`\gamma `$ is parallel along $`c`$). Because $`c`$ is a constant speed geodesic, $`\gamma `$ also satisfies $`\gamma ^{}(\omega )=wds`$, where $`s`$ is the parameter on $`(\epsilon ,\epsilon )`$.
Because the polynomial $`p_h(t)`$ is invariant under the action of the symmetry pseudo-groupoid, it suffices to consider only geodesics with initial velocities orthogonal to the subspace $`O_xT_xM`$ that is the tangent to the orbit through $`x`$. Thus, I will assume that if $`T_\alpha 0`$, then $`T_\alpha ^{}w`$ is real and that if $`T_\alpha =V_\alpha =0`$, then $`w`$ is orthogonal to $`L_\alpha `$.
For simplicity, set $`H(s)=H\left(\gamma (s)\right)`$, $`T(s)=T\left(\gamma (s)\right)`$, and $`V(s)=V\left(\gamma (s)\right)`$. Then these functions on $`(\epsilon ,\epsilon )`$ satisfy the initial conditions $`(H(0),T(0),V(0))=(H_0,T_0,V_0)`$ and the system of ordinary differential equations
(4.12)
$$\begin{array}{cc}\hfill \dot{H}& =Tw^{}+wT^{},\hfill \\ \hfill \dot{T}& =\left(H^2+(\mathrm{tr}H)H+V\mathrm{I}_n\right)w,\hfill \\ \hfill \dot{V}& =(\mathrm{tr}H)\left(T^{}w+w^{}T\right)+\left(T^{}Hw+w^{}HT\right).\hfill \end{array}$$
Let $`Li𝔲(n)`$ be any fixed element that satisfies $`Lw=LT_0=0`$ and $`[L,H_0]=0`$. Because of the latter equation, $`L`$ preserves each of the eigenspaces of $`H_0`$, i.e., the subspaces $`L_\alpha `$. Consequently, $`LT_\alpha =0`$ and $`Lw_\alpha =0`$ for all $`\alpha `$. Conversely, if $`Li𝔲(n)`$ preserves the eigenspaces of $`H_0`$ and annihilates $`T_\alpha `$ and $`w_\alpha `$ for all $`\alpha `$, then it satisfies satisfies $`Lw=LT_0=0`$ and $`[L,H_0]=0`$.
Now, the above differential equations imply the differential equations
(4.13)
$$\begin{array}{cc}\hfill [L,\dot{H}]& =LTw^{}w(LT)^{},\hfill \\ \hfill L\dot{T}& =\left([L,H]H+H[L,H]+(\mathrm{tr}H)[L,H]\right)w,\hfill \end{array}$$
so that the quantities $`([L,H(s)],LT(s))`$ satisfy a linear system of ordinary differential equations with vanishing initial condition at $`s=0`$. Consequently $`[L,H(s)]`$ and $`LT(s)`$ vanish identically for all $`s`$, as does $`Lw`$ (for trivial reasons). By the above characterization of those $`Li𝔲(n)`$ that satisfy $`[L,H_0]=LT_0=Lw=0`$, this implies that the subspace $`K_\alpha L_\alpha `$ that is perpendicular to $`T_\alpha `$ and $`w_\alpha `$ is necessarily an eigenspace of $`H(s)`$ that is perpendicular to $`T_\alpha (s)`$ and $`w`$ for all $`s`$.
If $`K_\alpha 0`$, then there is a well-defined eigenvalue $`H_\alpha (s)`$ of $`H(s)`$ associated to $`K_\alpha `$. In particular,
(4.14)
$$\dot{H}_\alpha (s)y=\dot{H}(s)y=\left(T(s)w^{}+wT(s)^{}\right)y=0$$
for all $`yK_\alpha `$. Of course, this implies that $`\dot{H}_\alpha (s)=0`$, i.e., that $`H_\alpha (s)=H_\alpha (0)=H_\alpha `$ for all $`s`$.
There are now four cases to consider:
If $`\alpha `$ is such that $`T_\alpha 0`$ and $`n_\alpha >1`$, then $`dimK_\alpha `$ is either $`n_\alpha 1`$ or $`n_\alpha 2`$, depending on whether $`w_\alpha `$ is zero or not. In either case, $`dimK_\alpha n_\alpha 2=n_\alpha m_\alpha `$, so $`H_\alpha `$ is a root of $`p_h(t)`$ of multiplicity at least $`n_\alpha m_\alpha `$, as desired.
If $`\alpha `$ is such that $`T_\alpha 0`$ and $`n_\alpha =1`$, then $`m_\alpha =1`$ and $`K_\alpha =0`$. In this case, of course, $`n_\alpha m_\alpha =0`$, so $`H_\alpha `$ is trivially a root of of $`p_h(t)`$ of multiplicity at least $`n_\alpha m_\alpha `$, as desired.
If $`\alpha `$ is such that $`T_\alpha =0`$ but $`V_\alpha 0`$, then $`dimK_\alpha `$ is either $`n_\alpha `$ or $`n_\alpha 1`$, depending on whether $`w_\alpha `$ is zero or not. In either case, $`dimK_\alpha n_\alpha 1=n_\alpha m_\alpha `$, so $`H_\alpha `$ is a root of $`p_h(t)`$ of multiplicity at least $`n_\alpha m_\alpha `$, as desired.
Finally, if $`T_\alpha =0`$ and $`V_\alpha =0`$, then $`w_\alpha =0`$ by the above condition on $`c`$. Thus $`K_\alpha =L_\alpha `$, so that $`dimK_\alpha =n_\alpha 0=n_\alpha m_\alpha `$ and, again, $`H_\alpha `$ is a root of $`p_h(t)`$ of multiplicity at least $`n_\alpha m_\alpha `$, as desired. ∎
The following result will be needed in the next subsection. Its proof follows by inspection of the formula for $`p_D(t)`$ and the definition of the $`m_\alpha `$ and so will be omitted.
###### Corollary 3.
No root of $`p_h^{}(t)`$ is a multiple root of $`p_D(t)`$.
### 4.2. Momentum cells
In general, the two characteristic polynomials do not completely determine the analytically connected equivalence class of a Bochner-Kähler structure. However, as will be seen in this subsection, they do determine it up to a finite number (at most $`m+1`$) of possibilities (Theorem 7).
#### 4.2.1. The roots of $`p_D`$
It turns out that the reality and multiplicity properties of the roots of $`p_D`$ are severely constrained.
###### Proposition 6.
One of the following cases holds:
1. $`p_D`$ has $`m`$ real, distinct roots, all of order $`1`$;
2. $`p_D`$ has $`m`$ real, distinct roots, one of order $`3`$ and the rest of order $`1`$;
3. $`p_D`$ has $`m+1`$ real, distinct roots, one of order $`2`$ and the rest of order $`1`$;
4. $`p_D`$ has $`m+2`$ real, distinct roots, all of order $`1`$.
###### Proof.
Substituting $`t=\lambda _i`$ into (4.2) yields
(4.15)
$$p_D(\lambda _i)=T_i^2\underset{ji}{}(\lambda _i\lambda _j).$$
Since $`\lambda _1>\lambda _2>\mathrm{}>\lambda _m`$ and $`T_i>0`$ on $`M^{}`$, it follows that $`(1)^{i1}p_D(\lambda _i)>0`$ for $`1im`$ holds on $`M^{}`$.
Equivalently, for every $`xM^{}`$, the polynomial $`p_D(t)`$ has an even number of real roots (counted with multiplicity) greater than $`\lambda _1(x)`$ and an odd number of real roots (counted with multiplicity) in each open interval $`(\lambda _i(x),\lambda _{i+1}(x))`$ for $`1i<m`$. Moreover, since $`(1)^mp_D(t)`$ is positive for all $`t`$ sufficiently negative, $`p_D(t)`$ has an odd number of real roots (counted with multiplicity) less than $`\lambda _m(x)`$. These considerations imply that $`p_D(t)`$ has at least $`m`$ distinct real roots.
If $`p_D`$ has exactly $`m`$ real roots, then the above parity conditions show that they must all have odd order. Since $`p_D(t)`$ has degree $`m+2`$, either Case 1 or Case 2 must hold.
If $`p_D`$ has exactly $`m+1`$ real, distinct roots, then Case 3 must hold.
If $`p_D`$ has exactly $`m+2`$ real, distinct roots, then Case 4 must hold. ∎
###### Remark 6 (Global inequalities).
By continuity, the inequality $`(1)^{i1}p_D(\lambda _i)0`$ holds on $`M`$.
###### Remark 7 (Root labeling).
I will use the following convention to label the real roots of $`p_D`$: When $`p_D`$ has $`m`$ distinct real roots, denote them by $`r_1>r_2>\mathrm{}>r_m`$; when $`p_D`$ has $`m+1`$ distinct real roots, denote them by $`r_1>\mathrm{}>r_{m+1}`$; and when $`p_D`$ has $`m+2`$ distinct real roots, denote them by $`r_0>r_1>\mathrm{}>r_{m+1}`$. In all cases, the list of real roots of $`p_D`$, in descending order, will be denoted by $`r`$.
If $`r_i`$ is any real root of $`p_D(t)`$, the number of roots $`\{\lambda _1(x),\mathrm{},\lambda _m(x)\}`$ that are strictly greater than $`r_i`$ is independent of $`xM^{}`$, so I will denote this common value by $`\mu _i`$. The function $`\mu `$ is constrained as follows:
Cases 1 and 2: Necessarily, $`\mu _i=i`$.
Case 3: Let $`r_i`$ be the double root. Then $`\mu _i`$ is either $`i`$ (SubCase (3-$`i`$,$`a`$); impossible when $`i=m+1`$) or $`i1`$ (Subcase (3-$`i`$,$`b`$)). Moreover, $`\mu _j=j`$ for $`j<i`$, while $`\mu _j=j1`$ for $`j>i`$.
Case 4: There is an integer $`im`$ so that $`\mu _i=i`$ (Subcase 4-$`i`$). Then $`\mu _j=j+1`$ for $`j<i`$ while $`\mu _j=j1`$ for $`j>i`$.
#### 4.2.2. Momentum cells
Since
$$p_h^{}(r_i)=\underset{j=1}{\overset{m}{}}(r_i\lambda _j),$$
it follows that $`(1)^{\mu _i}p_h^{}(r_i)>0`$ on $`M^{}`$. By continuity, the inequality
$$(1)^{\mu _i}\left(r_{i}^{}{}_{}{}^{m}h_1^{}r_{i}^{}{}_{}{}^{m1}+h_2^{}r_{i}^{}{}_{}{}^{m2}\mathrm{}+(1)^mh_m^{}\right)0$$
holds on $`M`$, with strict inequality on $`M^{}`$. Moreover, by Corollary 3, if $`r_i`$ is a multiple root of $`p_D`$, then it is not a root of $`p_h^{}(t)`$ at any point of $`M`$, so that the above inequality is strict on all of $`M`$.
The image $`h^{}(M)^m`$ therefore lies in the intersection of the closed half-spaces $`\overline{H(r_i,\mu _i)}`$ defined by the inequalities
(4.16)
$$(1)^{\mu _i}\left(r_{i}^{}{}_{}{}^{m}r_{i}^{}{}_{}{}^{m1}x_1+r_{i}^{}{}_{}{}^{m2}x_2\mathrm{}+(1)^mx_m\right)0$$
as $`r_i`$ ranges over the simple real roots of $`p_D`$ and the open half-space $`H(r_i,\mu _i)`$ defined by
(4.17)
$$(1)^{\mu _i}\left(r_{i}^{}{}_{}{}^{m}r_{i}^{}{}_{}{}^{m1}x_1+r_{i}^{}{}_{}{}^{m2}x_2\mathrm{}+(1)^mx_m\right)>0$$
if $`r_i`$ is a multiple root of $`p_D`$. This intersection will be referred to as the *momentum cell* $`C(p_D,\mu )^m`$. Note that $`h^{}(M^{})`$ lies in $`C(p_D,\mu )^{}`$, the interior of $`C(p_D,\mu )`$, and that $`h^{}:M^{}C(p_D,\mu )^{}`$ is a submersion onto its image.
#### 4.2.3. Possible momentum cells
More generally, if $`p_D(t)`$ is any monic polynomial of degree $`m+2`$ with real coefficients that falls into one of the Cases 1 through 4 of Proposition 6, define the *possible momentum cells of $`p_D`$* as follows:
If $`p_D`$ falls into Case 1 or Case 2, define $`\mu _i=i`$ for $`1im`$ and let $`C(p_D,\mu )`$ be defined by the inequalities (16) (strict or not depending on the multiplicity of the roots). This cell $`C(p_D,\mu )`$ is a closed, unbounded, convex polytope in Case 1, but is not closed in Case 2, since one of the faces is missing.
If $`p_D`$ falls into Case 3, with $`r_i`$ being the double root, define $`\mu _j=j`$ for $`j<i`$ and $`\mu _j=j1`$ for $`j>i`$, while $`\mu _i`$ is allowed to be one of $`i`$ (type $`a`$) or $`i1`$ (type $`b`$). Let $`C(p_D,\mu )`$ be defined by the inequalities (16) (strict or not depending on the multiplicity of the roots). Thus, there are two possible momentum cells except in the case that $`r_{m+1}`$ is the double root, in which case, there is only one possible cell. When there are two cells, neither is closed and their closures share the missing face. The only subcase with a bounded cell is Subcase (3-1,$`b`$).
When $`p_D`$ falls into Case 4, choose an integer $`i`$ in the range $`0im`$ and define $`\mu `$ so that $`\mu _j=j+1`$ for $`j<i`$, while $`\mu _i=i`$ and $`\mu _j=j1`$ for $`j>i`$. Let $`C(p_D,\mu )`$ be defined by the inequalities (4.16). Thus, there are $`m+1`$ possible momentum cells, one for each possible choice of $`i`$. Each of these cells is a closed polytope and they are mutually disjoint. When $`\mu _m=m`$ (i.e., $`\lambda _m>r_{m1}`$ on $`M^{}`$), this ‘highest’ cell has $`m`$ faces. Each of the other cells has $`m+1`$ faces. The only bounded cell falls in Subcase 4-0, i.e., $`\mu _0=0`$ (implying that $`\mu _i=i1`$ for all $`1im+1`$).
Figures 1, 2, and 3 show the possible momentum cells when $`m=2`$ in the four cases. The drawn axes are $`u_1`$ ($`=h_1^{}`$) and $`u_2`$ ($`=h_2^{}`$). Of course, all of these cells lie below the discriminant parabola $`u_2=\frac{1}{4}u_{1}^{}{}_{}{}^{2}`$ and are bounded by its tangent lines.
#### 4.2.4. The product representation
It will be useful to have another description of the possible momentum cells.
Let $`\sigma :^m^m`$ be the standard symmetrizing map, so that $`\sigma =(\sigma _1,\mathrm{},\sigma _m)`$ where $`\sigma _k:^m`$ is the $`k`$-th elementary symmetric function of its arguments. The map $`\sigma `$ is one-to-one on the closed set
$$_{}^m=\{(x_1,\mathrm{},x_m)^mx_1x_2\mathrm{}x_m\}.$$
Moreover, $`\sigma :_{}^m\sigma (_{}^m)^m`$ is a homeomorphism onto its image and a real-analytic diffeomorphism on its interior, which will be denoted $`_>^m_{}^m`$. Denote the inverse of $`\sigma `$ by $`\lambda :\sigma (_{}^m)_{}^m`$.
For any momentum cell $`C(p_D,\mu )`$, the subset $`\lambda \left(C(p_D,\mu )\right)`$ is a product of the form
(4.18)
$$\lambda \left(C(p_D,\mu )\right)=I_1\times I_2\times \mathrm{}\times I_m$$
where $`I_1,I_2,\mathrm{},I_m`$ are (non-empty) intervals in $``$ with non-overlapping interiors. The endpoints of the closure $`\overline{I_i}`$ are roots of $`p_D`$ and $`I_i`$ contains such an endpoint if and only if that endpoint is a simple root of $`p_D`$.
In Case 1, where the distinct roots $`r_1>\mathrm{}>r_m`$ are all simple, $`I_1=[r_1,\mathrm{})`$ and $`I_k=[r_k,r_{k1}]`$ for $`1<km`$.
In Case 2 in which, say, $`r_1`$ is the triple root, $`I_1=(r_1,\mathrm{})`$, $`I_2=[r_2,r_1)`$, and (assuming $`m>2`$$`I_k=[r_k,r_{k1}]`$ for $`3km`$.
The intervals $`(I_1,\mathrm{},I_m)`$ will be referred to as the *spectral bands* associated to $`C(p_D,\mu )`$ and $`I_1\times I_2\times \mathrm{}\times I_m`$ will be known as the *spectral product*. Since
(4.19)
$$\sigma (I_1\times I_2\times \mathrm{}\times I_m)=C(p_D,\mu ),$$
specifying the spectral bands is equivalent to specifying $`C(p_D,\mu )`$. Also, note that $`\sigma `$ maps $`I_1^{}\times I_2^{}\times \mathrm{}\times I_m^{}`$ diffeomorphically onto $`C(p_D,\mu )^{}`$, the interior of $`C(p_D,\mu )`$.
###### Proposition 7.
Suppose that $`p_D(t)`$ is a polynomial of degree $`m+2`$ that falls into one of the Cases of Proposition 6. Suppose that there exists a monic polynomial $`p_C(t)`$ of degree $`n+2`$ with the properties that $`p_C(t)/p_D(t)`$ is a polynomial all of whose roots are real roots of $`p_D(t)`$ and that its $`t^{n+1}`$-coefficient vanishes.
Then for every $`k^{}`$ in a possible momentum cell $`C(p_D,\mu )^m`$, there exists a Bochner-Kähler $`n`$-manifold $`(M,g,\mathrm{\Omega })`$ whose characteristic polynomials are $`p_C(t)`$ and $`p_D(t)`$ and whose reduced momentum mapping $`h^{}:M^m`$ assumes the value $`k^{}`$.
###### Proof.
This will be a matter of checking cases.
First, some generalities. Given polynomials $`p_D(t)`$ and $`p_C(t)`$ satisfying the hypotheses of the proposition and a $`k^{}^m`$ lying in a possible momentum cell $`C(p_D,\mu )`$ for $`p_D`$, define $`p_{h^{\prime \prime }}(t)=p_C(t)/p_D(t)`$. By hypothesis, all the roots of $`p_{h^{\prime \prime }}(t)`$ are real and are roots of $`p_D(t)`$ as well.
Define
$$p_k^{}(t)=t^mk_1^{}t^{m1}+\mathrm{}+(1)^mk_m^{}.$$
Let $`\lambda \left(C(p_D,\mu )\right)=I_1\times I_2\times \mathrm{}\times I_m`$, and let $`\lambda (k^{})=(s_1,s_2,\mathrm{},s_m)`$, so that there exists a real factorization of the form
$$p_k^{}(t)=(ts_1)(ts_2)\mathrm{}(ts_m)$$
with $`s_kI_k`$ for $`1km`$.
First, assume that $`k^{}`$ lies in $`C(p_D,\mu )^{}`$, the interior of $`C(p_D,\mu )`$. Then $`s_k`$ lies in $`I_k^{}`$ for $`1km`$ and, in particular, the $`s_k`$ are all distinct and are not roots of $`p_D(t)`$. It follows that the rational function $`p_D(t)/p_k^{}(t)`$ has a simple pole at $`t=s_k`$ for $`1km`$. Since $`p_D(t)`$ has degree $`m+2`$ and is monic, there is a partial fractions expansion of the form
$$\frac{p_D(t)}{p_k^{}(t)}=t^2+b_1t+b_2+\underset{k=1}{\overset{m}{}}\frac{q_k}{(ts_k)}.$$
Because of the way that the possible momentum cells were defined, the inequality $`(1)^{k1}p_D(s_k)>0`$ holds for $`s_kI_k^{}`$, so it follows that $`q_k>0`$ for $`1km`$.
If $`n>m`$, define $`s_{m+1}s_{m+2}\mathrm{}s_n`$ so that
$$p_{h^{\prime \prime }}(t)=(ts_{m+1})(ts_{m+2})\mathrm{}(ts_m).$$
By hypothesis, each root of $`p_{h^{\prime \prime }}(t)`$ is a real root of $`p_D`$ and so is not equal to any of the roots of $`p_k^{}(t)`$.
Consider the element $`(s,t,v)i𝔲(n)^n`$ defined by letting $`s`$ be the diagonal matrix with entries $`s_{i\overline{ı}}=s_i`$ for $`1in`$; letting $`t_i=\sqrt{q_i}`$ for $`1im`$ and $`t_i=0`$ for $`m<in`$ (if $`n>m`$); and letting $`v=b_2`$. The hypothesis that $`p_C`$ have no $`t^{n+1}`$-term is then seen to be equivalent to the condition that $`b_1=\mathrm{tr}s`$, while the condition that each $`s_a`$ for $`a>m`$ be a root of $`p_D`$ is then equivalent to the condition that
$$v_a=s_{a}^{}{}_{}{}^{2}+b_1s_a+v+\underset{i=1}{\overset{m}{}}\frac{t_{i}^{}{}_{}{}^{2}}{(s_as_i)}=0.$$
By Theorem 6, the Bochner-Kähler structure on a neighborhood $`M`$ of $`0^n`$ that has a unitary coframe $`u_0:T_0M^n`$ with $`(H(u_0),T(u_0),V(u_0))=(s,t,v)`$ has $`p_C(t)`$ and $`p_D(t)`$ as its characteristic polynomials and satisfies $`h^{}(0)=k^{}`$. This establishes existence for the interior points of $`C(p_D,\mu )`$.
It remains to treat the boundary cases, i.e., cases in which one or more of the $`s_i`$ are actually roots of $`p_D(t)`$.
Now, if $`s_j=s_{j+1}`$ for any $`j`$, then $`\{s_j\}=I_jI_{j+1}`$, so that $`s_j`$ is a simple root of $`p_D`$. In such a case, necessarily, $`s_{j1}>s_j`$ (if $`j>1`$) since $`I_{j1}I_{j+1}=\mathrm{}`$ and $`s_{j+1}>s_{j+2}`$ (if $`j<m1`$) since $`I_jI_{j+2}=\mathrm{}`$. Consequently, each $`s_j`$ is at most a double root of $`p_k^{}(t)`$ and, if so, it must also be a root of $`p_D(t)`$. It follows that the rational function $`p_D(t)/p_k^{}(t)`$ has a simple pole at $`t=s_j`$ for $`1jm`$. Since $`p_D(t)`$ has degree $`m+2`$ and is monic, there is a partial fractions expansion
$$\frac{p_D(t)}{p_k^{}(t)}=t^2+b_1t+b_2+\underset{j=1}{\overset{m}{}}\frac{q_j}{(ts_j)},$$
where, in order to make the $`q_j`$ unique, it is now necessary to add the condition that $`q_{j+1}=0`$ if $`s_{j+1}=s_j`$. If $`j`$ is such that $`s_j`$ is not a root of $`p_D(t)`$, then the inequality $`(1)^{j1}p_D(s_j)>0`$ holds so that $`q_j>0`$. If $`j`$ is such that $`s_j`$ is a simple root of both $`p_k^{}(t)`$ and $`p_D(t)`$, then $`q_j=0`$. If $`j`$ is such that $`s_j=s_{j+1}`$, then $`(1)^{j1}p_D(t)`$ and $`(1)^{j1}p_k^{}(t)`$ are both positive on $`I_j^{}`$. Since $`s_j`$ is a double root of $`p_k^{}(t)`$ and a simple root of $`p_D(t)`$, it follows that
$$\underset{ts_j^+}{lim}\frac{p_D(t)}{p_k^{}(t)}=+\mathrm{},$$
which can only hold if $`q_j>0`$. In particular, $`q_j0`$ has been defined for $`1jm`$ so that the above partial fractions expansion is valid.
If $`n>m`$, again define $`s_{m+1}s_{m+2}\mathrm{}s_n`$ so that
$$p_{h^{\prime \prime }}(t)=(ts_{m+1})(ts_{m+2})\mathrm{}(ts_m).$$
Again, each root of $`p_{h^{\prime \prime }}(t)`$ is a real root of $`p_D`$ but now it may also be a root of $`p_k^{}(t)`$.
Define an element $`(s,t,v)i𝔲(n)^n`$ by letting $`s`$ be the diagonal matrix with entries $`s_{i\overline{ı}}=s_i`$ for $`1in`$; letting $`t_i=\sqrt{q_i}`$ for $`1im`$ and $`t_i=0`$ for $`m<in`$ (if $`n>m`$); and letting $`v=b_2`$. It must now be verified that the element $`(s,t,v)`$ does indeed have $`p_C(t)`$ and $`p_D(t)`$ as its characteristic polynomials.
Now, the hypothesis that $`p_C`$ have no $`t^{n+1}`$-term is again seen to be equivalent to the condition that $`b_1=\mathrm{tr}s`$, and the condition that $`t_i=0`$ for $`i>m`$ or when $`s_i=s_{i1}`$ implies that
$$\begin{array}{cc}\hfill t^2+(\mathrm{tr}s)t+v+\underset{j=1}{\overset{n}{}}\frac{t_{j}^{}{}_{}{}^{2}}{(ts_j)}& =t^2+b_1t+b_2+\underset{j=1}{\overset{m}{}}\frac{q_j}{(ts_j)}\hfill \\ & =\frac{p_D(t)}{p_k^{}(t)}=\frac{p_C(t)}{p_{h^{\prime \prime }}(t)p_k^{}(t)},\hfill \end{array}$$
so that
$$p_C(t)=\underset{i=1}{\overset{n}{}}(ts_i)\left[t^2+(\mathrm{tr}s)t+v+\underset{j=1}{\overset{n}{}}\frac{t_{j}^{}{}_{}{}^{2}}{(ts_j)}\right],$$
as desired. It remains to verify that $`p_D(t)`$ is the reduced characteristic polynomial associated to $`(s,t,v)`$, i.e., to compute the numbers $`n_i`$ and $`m_i`$ for each eigenvalue $`s_i`$ of $`s`$ according to the recipe of §3.3 and show that $`s_i`$ is a root of $`p_{h^{\prime \prime }}(t)`$ of multiplicity exactly equal to $`n_im_i`$. This will be done by breaking it down into a number of cases.
If $`s_a`$ is not $`s_i`$ for any $`1im`$, then $`p_D(s_a)=0`$ is equivalent to
$$v_a=s_{a}^{}{}_{}{}^{2}+(\mathrm{tr}s)s_a+v+\underset{i=1}{\overset{m}{}}\frac{t_{i}^{}{}_{}{}^{2}}{(s_as_i)}=0,$$
and this implies that $`s_a`$ is an eigenvalue of $`s`$ of some multiplicity $`n_a1`$ that satisfies $`t_a=v_a=0`$, so that $`m_a=0`$. Thus, $`s_a`$ is a root of $`p_{h^{\prime \prime }}(t)`$ and has multiplicity $`n_am_a=n_a`$, as desired.
If $`s_i`$ is not a root of $`p_D(t)`$, then, by construction, it is a simple eigenvalue of $`s`$ and also satisfies $`t_i=\sqrt{q_i}>0`$, so $`m_i=1`$ and $`n_im_i=0`$, so that $`(ts_i)`$ is not a factor of $`p_{h^{\prime \prime }}(t)`$, again, as desired.
If $`s_i`$ is a simple root of $`p_D(t)`$ and a simple root of $`p_k^{}(t)`$, then, by construction, $`t_i=0`$. The quantity $`v_i`$ is calculated to be
$$\begin{array}{cc}\hfill v_i& =s_{i}^{}{}_{}{}^{2}+(\mathrm{tr}s)s_i+v+\underset{ji}{}\frac{t_{j}^{}{}_{}{}^{2}}{(s_is_j)}\hfill \\ & =\underset{ts_i}{lim}\left(t^2+(\mathrm{tr}s)t+v+\underset{j=1}{\overset{m}{}}\frac{t_{j}^{}{}_{}{}^{2}}{(ts_j)}\right)\hfill \\ & =\underset{ts_i}{lim}\frac{p_D(t)}{p_k^{}(t)}0.\hfill \end{array}$$
Thus, the recipe gives $`m_i=1`$, again as desired.
If $`s_i`$ is a simple root of $`p_D(t)`$ and a double root of $`p_k^{}(t)`$, then it can be assumed that $`s_i=s_{i+1}`$, so that $`t_i>0`$ (and $`t_{i+1}=0`$). Since $`n_i2`$, the recipe gives $`m_i=2`$, again, as desired.
Finally, if $`s_i`$ is a multiple root of $`p_D(t)`$, then it cannot be a root of $`p_k^{}(t)`$ at all, by the definition of the momentum cell $`C(p_D,\mu )`$. Consequently, $`t_i=0`$ by definition and calculation shows that $`v_i=0`$ as well. Thus $`m_i=0`$, as desired.
By Theorem 6, the Bochner-Kähler structure on a neighborhood $`M`$ of $`0^n`$ that has a unitary coframe $`u_0:T_0M^n`$ with $`(H(u_0),T(u_0),V(u_0))=(s,t,v)`$ has $`p_C(t)`$ and $`p_D(t)`$ as its characteristic polynomials and satisfies $`h^{}(0)=k^{}`$. This establishes existence for the boundary points of $`C(p_D,\mu )`$. ∎
The way is now paved for the following result, which, together with the previous proposition, classifies the analytically connected equivalence classes of Bochner-Kähler structures.
###### Theorem 7.
The analytically connected class of a Bochner-Kähler structure is determined by $`p_C`$, $`p_D`$, and the momentum cell $`C(p_D,\mu )`$ that contains the reduced momentum image. Moreover, for any Bochner-Kähler structure with data $`(p_C,p_D,\mu )`$, the union of the reduced momentum images of the Bochner-Kähler structures that are analytically connected to it is the entire momentum cell $`C(p_D,\mu )`$.
###### Proof.
It has been established that $`p_C`$ and $`p_D`$ and the momentum cell $`C(p_D,\mu )`$ are invariants of the analytically connected equivalence class. Moreover, by Proposition 7, every point of $`C(p_D,\mu )`$ lies in the image of the reduced momentum mapping of some Bochner-Kähler structure.
To prove Theorem 7, it will thus suffice to show that any two Bochner-Kähler structures with the same data $`(p_C,p_D,\mu )`$ are analytically connected.
Now, if $`(M,g,\mathrm{\Omega })`$ and $`(\stackrel{~}{M},\stackrel{~}{g},\stackrel{~}{\mathrm{\Omega }})`$ are connected Bochner-Kähler manifolds with the same data $`(p_C,p_D,\mu )`$ and their reduced momentum images $`h^{}(M)`$ and $`h^{}(\stackrel{~}{M})`$ have nontrivial intersection, then they contain points $`xM`$ and $`\stackrel{~}{x}\stackrel{~}{M}`$ so that $`f(x)=\stackrel{~}{f}(\stackrel{~}{x})`$ where $`f:M^{2n+1}`$ and $`\stackrel{~}{f}:\stackrel{~}{M}^{2n+1}`$ are the corresponding moduli maps. By Theorem 1 and Corollary 1, the germs of Bochner-Kähler structures around $`xM`$ and $`\stackrel{~}{x}\stackrel{~}{M}`$ are isomorphic. Since $`M`$ and $`\stackrel{~}{M}`$ are connected, the germ of the Bochner-Kähler structure around any $`yM`$ is analytically connected to the germ of the Bochner-Kähler structure around any $`\stackrel{~}{y}M`$.
Now, from Theorem 5, it follows that $`h^{}(M^{})`$ lies in the interior of $`C(p_D,\mu )`$ and that $`h^{}:M^{}C(p_D,\mu )^{}`$ is a submersion onto its image, which is therefore open.
The union of the open sets $`h^{}(\stackrel{~}{M}^{})`$ as $`(\stackrel{~}{M},\stackrel{~}{g},\stackrel{~}{\mathrm{\Omega }})`$ ranges over the Bochner-Kähler structures that are analytically connected to any given $`(M,g,\mathrm{\Omega })`$ is a connected component of $`C(p_D,\mu )^{}`$. Since $`C(p_D,\mu )^{}`$ is convex and hence connected, this union must be all of $`C(p_D,\mu )^{}`$.
By Proposition 7, the union of all the sets $`h^{}(M)`$ as $`(M,g,\mathrm{\Omega })`$ ranges over the Bochner-Kähler structures with data $`(p_C,p_D,\mu )`$ is equal to the entire cell $`C(p_D,\mu )`$. Since $`h^{}(M^{})`$ is a nonempty subset of $`C(p_D,\mu )^{}`$ for any such $`(M,g,\mathrm{\Omega })`$, it follows that all of these are analytically connected, as desired. ∎
###### Remark 8 (Coarse moduli and polytope embeddings).
By Theorem 7, the analytically connected equivalence classes in $`F_n`$ correspond to the data $`(p_C,p_D,\mu )`$ that satisfy the conditions of Proposition 7. Note that, for any given $`p_C(t)`$, there are at most a finite number of choices of $`(p_D,\mu )`$ that will satisfy these constraints. Thus, each value of $`C=(C_2,\mathrm{},C_{n+2})`$ corresponds to only a finite number of equivalence classes. It is in this sense that the functions $`C_i:F_n`$ furnish the complete set of ‘coarse moduli’ for Bochner-Kähler structures in dimension $`n`$.
Moreover, the mapping $`\mathrm{\Delta }:F_n^{2n+1}`$ of §3.5.3 embeds each analytically connected equivalence class as a (not necessarily closed) convex polytope of some dimension $`mn`$. In particular, each of these equivalence classes is contractible.
###### Corollary 4.
If $`(M,g,\mathrm{\Omega })`$ is a complete, connected Bochner-Kähler structure, then the reduced momentum mapping $`h^{}:MC(p_D,\mu )`$ is surjective; the submersion $`h^{}:M^{}C(p_D,\mu )^{}`$ is a fibration; and the fibers of $`h^{}`$ in $`M`$ are connected.
###### Proof.
Fix $`xM`$. By Theorem 7, to prove that $`h^{}`$ is surjective it suffices to show that if $`(\stackrel{~}{M},\stackrel{~}{g},\stackrel{~}{\mathrm{\Omega }})`$ is any Bochner-Kähler structure containing an $`\stackrel{~}{x}`$ with $`f(x)=\stackrel{~}{f}(\stackrel{~}{x})`$, then $`\stackrel{~}{h}^{}(\stackrel{~}{M})`$ is a subset of $`h^{}(M)`$.
Consider any $`\stackrel{~}{y}\stackrel{~}{M}`$ and choose a smooth path $`\stackrel{~}{c}:[0,1]\stackrel{~}{M}`$ with $`\stackrel{~}{c}(0)=\stackrel{~}{x}`$ and $`\stackrel{~}{c}(1)=\stackrel{~}{y}`$. Choose a $`\stackrel{~}{u}_0\stackrel{~}{P}_{\stackrel{~}{x}}`$ and let $`\stackrel{~}{u}:[0,1]\stackrel{~}{P}`$ be the parallel transport of $`\stackrel{~}{u}_0`$ along $`\stackrel{~}{c}`$. Thus $`(\stackrel{~}{u})^{}(\stackrel{~}{\varphi })=0`$ while $`(\stackrel{~}{u})^{}(\stackrel{~}{\omega })=v(s)ds`$ for some $`v:[0,1]^n`$.
Since $`f(x)=\stackrel{~}{f}(\stackrel{~}{x})`$ by hypothesis, there exists a $`u_0P_x`$ so that $`H(u_0)=\stackrel{~}{H}(\stackrel{~}{u}_0)`$, $`T(u_0)=\stackrel{~}{T}(\stackrel{~}{u}_0)`$, and $`V(u_0)=\stackrel{~}{V}(\stackrel{~}{u}_0)`$. Since the metric on $`M`$ is complete, there will exist a unique curve $`u:[0,1]P`$ satisfying the initial condition $`u(0)=u_0`$ and the ordinary differential equations
$$u^{}(\omega )=v(s)ds,u^{}(\varphi )=0.$$
I.e., $`u`$ is the parallel transport of $`u_0`$ along the curve $`c=\pi u`$ in $`M`$. The structure equations (2.14) and the Chain Rule now imply that the two curves defined on $`[0,1]`$
$$(Hu,Tu,Vu)\text{and}(\stackrel{~}{H}\stackrel{~}{u},\stackrel{~}{T}\stackrel{~}{u},\stackrel{~}{V}\stackrel{~}{u})$$
in $`i𝔲(n)^n`$ satisfy the same initial conditions and system of ordinary differential equations, so they are equal on $`[0,1]`$. Now, setting $`s=1`$ yields $`\stackrel{~}{f}(\stackrel{~}{y})=f\left(c(1)\right)h^{}(M)`$. Since $`\stackrel{~}{y}\stackrel{~}{M}`$ was arbitrary, $`\stackrel{~}{h}^{}(\stackrel{~}{M})`$ is a subset of $`h^{}(M)`$.
To prove the second part, suppose first that $`M`$ is simply-connected. Then any local isometry $`\psi :UM`$ defined on an open subset $`UM`$ extends uniquely to a global isometry of $`M`$.<sup>15</sup><sup>15</sup>15Simply choose $`xU`$ and extend $`\psi `$ so that it commutes with the exponential map at $`x`$, i.e., so that $`\psi \left(\mathrm{exp}_x(v)\right)=\mathrm{exp}_{\psi (x)}\left(\psi ^{}(x)(v)\right)`$. The completeness and analyticity of the metric and the simple connectivity of $`M`$ imply that such an extension of $`\psi `$ to all of $`M`$ exists and is an isometry, as desired. Thus $`I(M)`$, the global holomorphic isometry group of $`M`$, acts transitively on the fibers of $`h^{}(M)`$.
Now $`(h^{})^1\left(C(p_D,\mu )^{}\right)=M^{}`$ and, by the first part of the proof, $`h^{}:M^{}C(p_D,\mu )^{}`$ is surjective. Since $`h^{}:M^{}C(p_D,\mu )^{}`$ is also a submersion whose fibers are $`I(M)`$-orbits, it follows that $`h^{}:M^{}C(p_D,\mu )^{}`$ is a fibration.
To treat the case where $`M`$ is not simply-connected, pass to the universal cover $`\stackrel{~}{M}`$ and note that $`\stackrel{~}{h}^{}`$ is invariant under the deck transformations $`\mathrm{\Delta }`$ of the cover $`\stackrel{~}{M}M`$ (which form a discrete subgroup of $`I(\stackrel{~}{M})`$). Since $`\stackrel{~}{h}^{}:\stackrel{~}{M}^{}C(p_D,\mu )^{}`$ is a fibration, dividing by the (free) action of $`\mathrm{\Delta }`$ yields that $`h^{}:M^{}C(p_D,\mu )^{}`$ is also a fibration.
To prove the connectedness of the fibers of $`h^{}`$ it suffices to treat the case where $`M`$ is simply-connected, so assume this. Again, $`I(M)`$ acts transitively on the fibers of $`h^{}`$. Since $`M^{}`$ is the complement of a codimension 1 complex subvariety in $`M`$, it is connected. The exact sequence of the fibration $`h^{}:M^{}C(p_D,\mu )^{}`$ and the contractibility of $`C(p_D,\mu )^{}`$ then imply that the fibers of $`h^{}:M^{}C(p_D,\mu )^{}`$ are connected. By Corollary 2, the $`I(M)`$-stabilizer of any point in $`M`$ is a product of unitary groups and hence is connected. Since the $`I(M)`$-orbit of any $`xM^{}`$ has been shown to be connected, it follows that $`I(M)`$ must be connected as well. Consequently, all of its orbits in $`M`$ are connected and these are the fibers of $`h^{}`$. ∎
### 4.3. A Riemannian submersion
The mapping $`h^{}:MC(p_D,\mu )^m`$ can be used to give more detailed information about the Bochner-Kähler metric.
#### 4.3.1. The cell metric
The proof of Corollary 4 shows that, at least when $`M`$ is complete, there is a metric on $`C(p_D,\mu )^{}`$ for which $`h^{}:M^{}C(p_D,\mu )^{}`$ is a Riemannian submersion. It turns out that this metric exists even when $`M`$ is not complete and can be identified explicitly.
###### Theorem 8.
Given a monic polynomial $`p_D(t)`$ of degree $`m+2`$ that falls into one of the cases of Proposition 6, there exist rational functions $`R_D^{ij}=R_D^{ji}`$ on $`^m`$ with the property that the quadratic form
(4.20)
$$R_D=R_D^{ij}(u)du_idu_j$$
restricts to be positive definite on the interior of each possible momentum cell for $`p_D(t)`$ and moreover, so that for any Bochner-Kähler manifold with reduced momentum polynomial $`p_D(t)`$, the reduced momentum mapping $`h^{}:M^m`$ is a Riemannian submersion when restricted to $`M^{}`$.
###### Proof.
Recall the bundle $`P_2P_1`$ that was introduced in the proof of Proposition 5 and the notation introduced there. On $`P_2`$, the matrix $`H`$ is diagonal and $`H_{i\overline{ı}}=\lambda _i`$ for $`1im`$. The structure equation for $`dH_{i\overline{ı}}`$ then becomes
(4.21)
$$d\lambda _i=T_i(\omega _i+\overline{\omega _i}).$$
By equation (4.2),
$$T_{i}^{}{}_{}{}^{2}=\frac{p_D(\lambda _i)}{_{ji}(\lambda _i\lambda _j)},$$
so (4.21) can be written in the form
$$\mathrm{Re}(\omega _i)=\frac{1}{2}\sqrt{\frac{_{ji}(\lambda _i\lambda _j)}{p_D(\lambda _i)}}d\lambda _i.$$
In other words,
$$\underset{i=1}{\overset{m}{}}\mathrm{Re}(\omega _i)^2=\frac{1}{4}\underset{i=1}{\overset{m}{}}\frac{_{ji}(\lambda _i\lambda _j)}{p_D(\lambda _i)}d\lambda _{i}^{}{}_{}{}^{2}.$$
Now suppose that $`h^{}(M)`$ lies in $`C(p_D,\mu )`$ and that
$$\lambda \left(C(p_D,\mu )\right)=I_1\times I_2\times \mathrm{}\times I_m_{}^m$$
as in §4.2.4. Since $`(1)^{i1}p_D(y_i)>0`$ for $`y_iI_i^{}`$, the quadratic form
(4.22)
$$S=\frac{1}{4}\underset{i=1}{\overset{m}{}}\frac{_{ji}(y_iy_j)}{p_D(y_i)}dy_{i}^{}{}_{}{}^{2}$$
is positive definite on $`I_1^{}\times I_2^{}\times \mathrm{}\times I_m^{}_>^m`$. Since $`S`$ has rational coefficients, is well-defined on $`^m`$ minus the union of the hyperplanes $`p_D(y_i)=0`$, and is invariant under permutations of the coordinates $`y_i`$, it follows that there are unique rational functions $`R_D^{ij}=R_D^{ji}`$ on $`^m`$ so that
$$S=R_D^{ij}\left(\sigma (y)\right)d\left(\sigma _i(y)\right)d\left(\sigma _j(y)\right).$$
I.e., setting $`R_D=R_D^{ij}(u)du_idu_j`$, the positive definite quadratic form $`S`$ defined on $`I_1^{}\times I_2^{}\times \mathrm{}\times I_m^{}`$ is of the form $`\sigma ^{}(R_D)`$. Since $`\sigma :I_1^{}\times I_2^{}\times \mathrm{}\times I_m^{}C(p_D,\mu )^{}`$ is a diffeomorphism, $`R_D`$ is positive definite on $`C(p_D,\mu )^{}`$. Moreover, the above formula on $`M^{}`$ can now be written as
$$\underset{i=1}{\overset{m}{}}\mathrm{Re}(\omega _i)^2=(h^{})^{}\left(R_D\right),$$
showing that $`h^{}:M^{}C(p_D,\mu )^{}`$ is a Riemannian submersion when the target is given the Riemannian metric $`R_D`$. ∎
#### 4.3.2. Explicit formulae
When $`m=2`$, with $`u_1=y_1+y_2`$ and $`u_2=y_1y_2`$, the relation between $`S`$ and $`R`$ is expressible in the intermediate form
$$\begin{array}{cc}\hfill 4S=\frac{y_1y_2}{p_D(y_1)}dy_{1}^{}{}_{}{}^{2}+\frac{y_2y_1}{p_D(y_2)}dy_{2}^{}{}_{}{}^{2}& =\frac{\left(y_{1}^{}{}_{}{}^{2}p_D(y_2)y_{2}^{}{}_{}{}^{2}p_D(y_1)\right)}{(y_1y_2)p_D(y_1)p_D(y_2)}du_{1}^{}{}_{}{}^{2}\hfill \\ & 2\frac{\left(y_1p_D(y_2)y_2p_D(y_1)\right)}{(y_1y_2)p_D(y_1)p_D(y_2)}du_1du_2\hfill \\ & +\frac{p_D(y_2)p_D(y_1)}{(y_1y_2)p_D(y_1)p_D(y_2)}du_{2}^{}{}_{}{}^{2}.\hfill \end{array}$$
Each of the coefficients on the right is visibly a symmetric rational function of $`y_1`$ and $`y_2`$ and so can be written as a rational function of $`u_1`$ and $`u_2`$.
Notice that the expression $`p_D(y_1)p_D(y_2)`$ is a common denominator of all these rational expressions. In the case where $`p_D(t)`$ has all four of its roots real, this can be written in the form
$$p_D(y_1)p_D(y_2)=\underset{\alpha =0}{\overset{3}{}}(y_1r_\alpha )(y_2r_\alpha )=\underset{\alpha =0}{\overset{3}{}}(u_2r_\alpha u_1+r_{\alpha }^{}{}_{}{}^{2}),$$
so that the denominator is actually a product of linear functions on $`^2`$, indeed, the very linear functions whose vanishing defines the faces of possible momentum cells.
For any $`m`$, in Case 4, in which all of the roots of $`p_D`$ are real and distinct, this generalizes, leading to an expression for the metric $`R_D`$ that will turn out to be very useful. As usual, let
$$r_0>r_1>\mathrm{}>r_m>r_{m+1}$$
be the real roots of $`p_D`$. Since the roots are real and distinct, $`(1)^\alpha p_D^{}(r_\alpha )>0`$ for $`0\alpha m+1`$. Note the identity
$$p_D^{}(r_\alpha )=\underset{\beta \alpha }{}(r_\alpha r_\beta ),$$
as well as the classical identities
$$\underset{\alpha =0}{\overset{m+1}{}}\frac{r_{\alpha }^{}{}_{}{}^{k}}{p_D^{}(r_\alpha )}=\{\begin{array}{cc}\frac{(1)^{m+1}}{r_0r_1\mathrm{}r_{m+1}}\hfill & \text{when }k=1;\hfill \\ 0\hfill & \text{when }0km;\hfill \\ 1\hfill & \text{when }k=m+1\text{;}\hfill \\ r_0+\mathrm{}+r_{m+1}\hfill & \text{when }k=m+2\text{.}\hfill \end{array}$$
(The cases $`k=1`$ and $`k=m+2`$ will be used in a later section.) Using coordinates $`(u_1,\mathrm{},u_m)`$ on $`^m`$ as above, define linear functions<sup>16</sup><sup>16</sup>16The reader will note that the formula for $`l_\alpha `$ makes sense in general as long as $`r_\alpha `$ is a simple root of $`p_D(t)`$. Accordingly, $`l_\alpha `$ will taken to be defined by (4.23) in this more general case.
(4.23)
$$l_\alpha =\frac{(r_\alpha ^mr_\alpha ^{m1}u_1+\mathrm{}+(1)^mu_m)}{p_D^{}(r_\alpha )}.$$
for $`0\alpha m+1`$, so that the equations $`l_\alpha =0`$ define the hyperplanes that are the faces of the various possible momentum cells for $`p_D`$. Note that the above classical identities in the range $`0km+1`$ are equivalent to the equations
(4.24)
$$\underset{\alpha =0}{\overset{m+1}{}}l_\alpha =0\text{and}\underset{\alpha =0}{\overset{m+1}{}}r_\alpha l_\alpha =1,$$
which are the only linear relations among the $`l_\alpha `$. Note also that
$$\sigma ^{}(l_\alpha )=\frac{_{i=1}^m(y_ir_\alpha )}{_{\beta \alpha }(r_\beta r_\alpha )}.$$
The metric $`R_D`$ then has the simple expression
(4.25)
$$R_D=\underset{\alpha =0}{\overset{m+1}{}}\frac{dl_{\alpha }^{}{}_{}{}^{2}}{4l_\alpha }.$$
Indeed,
$$\begin{array}{cc}\hfill \sigma ^{}\left(\underset{\alpha =0}{\overset{m+1}{}}l_\alpha \left(\frac{dl_\alpha }{l_\alpha }\right)^2\right)& =\underset{\alpha =0}{\overset{m+1}{}}\frac{_{i=1}^m(y_ir_\alpha )}{_{\beta \alpha }(r_\beta r_\alpha )}\left(\underset{j=1}{\overset{m}{}}\frac{dy_j}{(y_jr_\alpha )}\right)^2\hfill \\ & =\underset{j,k=1}{\overset{m}{}}\left(\underset{\alpha =0}{\overset{m+1}{}}\frac{_{i=1}^m(y_ir_\alpha )}{(y_jr_\alpha )(y_kr_\alpha )_{\beta \alpha }(r_\beta r_\alpha )}\right)dy_jdy_k\hfill \\ & =\underset{i=1}{\overset{m}{}}\frac{_{ji}(y_iy_j)}{p_D(y_i)}dy_{i}^{}{}_{}{}^{2},\hfill \end{array}$$
where the last equality follows from the classical identities above and the Lagrange interpolation identity.
The expression (4.25) can also be written in Hessian form as
$$R_D=R_D^{ij}du_idu_j=\frac{^2G}{u_iu_j}du_idu_j$$
where the potential function $`G`$ has the form
$$G=\frac{1}{4}\underset{\alpha =0}{\overset{m+1}{}}l_\alpha \left(\mathrm{log}|l_\alpha |1\right),$$
a fact that will be useful below.
The formula for $`R_D`$ is evidently singular along the hyperplanes $`l_\alpha =0`$, but this singularity is mild and can be ‘resolved’ with little difficulty. For simplicity, and since this case will be useful in the analysis below, I will illustrate this for the ‘lowest’ cell, i.e., Subcase 4-0. The spectral intervals in this subcase are $`I_i=[r_i,r_{i+1}]`$ and the functions $`l_1,\mathrm{}l_{m+1}`$ are all nonnegative on this cell $`C(p_D,\mu )`$, which is an $`m`$-simplex. In fact, (4.24) shows that $`l_0=(l_1+\mathrm{}+l_{m+1})`$ and that
(4.26)
$$1=\underset{\alpha =1}{\overset{m+1}{}}(r_0r_\alpha )l_\alpha ,$$
so that the functions $`(r_0r_\alpha )l_\alpha `$ for $`1\alpha m+1`$ can be regarded as homogeneous affine coordinates on this simplex. Now let $`E^{m+1}`$ be the $`m`$-dimensional ellipsoid defined by
(4.27)
$$1=\underset{\alpha =1}{\overset{m+1}{}}(r_0r_\alpha )p_{\alpha }^{}{}_{}{}^{2}.$$
There is then a unique smooth map $`s:EC(p_D,\mu )`$ defined by $`s^{}(l_\alpha )=p_\alpha ^2`$ for $`1\alpha m+1`$. Since $`s^{}(l_0)=(p_{1}^{}{}_{}{}^{2}+\mathrm{}+p_{m+1}^{}{}_{}{}^{2})`$, the $`s`$-pullback metric is
(4.28)
$$s^{}(R_D)=s^{}\left(\underset{\alpha =0}{\overset{m+1}{}}\frac{dl_{\alpha }^{}{}_{}{}^{2}}{4l_\alpha }\right)=\underset{\alpha =1}{\overset{m+1}{}}dp_{\alpha }^{}{}_{}{}^{2}\frac{(p_1dp_1+\mathrm{}+p_{m+1}dp_{m+1})^2}{(p_{1}^{}{}_{}{}^{2}+\mathrm{}+p_{m+1}^{}{}_{}{}^{2})}.$$
The quadratic form on the right hand side is well-defined on $`^{m+1}`$ minus the origin and is positive semidefinite there, with the null space of the quadratic form being spanned by the radial vector at each point.<sup>17</sup><sup>17</sup>17In fact, this metric is just the tangential part $`r^2d\sigma _m^2`$ of the expression for the standard metric in polar coordinates $`dr^2+r^2d\sigma _m^2`$, where $`d\sigma _m^2`$ is the standard metric on the $`m`$-sphere. A curious consequence of this fact is that the metric $`R_D`$ is conformally flat. Thus, this quadratic form is positive definite and smooth on $`E`$, thereby providing the desired ‘resolution’ of the singularities of $`R_D`$ on $`C(p_D,\mu )`$. Note that the rank of the mapping $`s`$ at $`p=(p_\alpha )E`$ is equal to one less than the number of nonzero entries $`p_\alpha `$. This will be useful below.
The analysis of $`R_D`$ in Cases 1, 2, and 3 can be derived from the Case 4 analysis by either regarding two of the roots as complex conjugates and combining the corresponding terms in the above sums to obtain real expressions (Case 1) or collecting two or three of the terms and taking the limit as the corresponding roots come together (Cases 2 and 3). This will only be needed in Case 3-1$`b`$ below, so I will do this case and leave the others to the interested reader.
Case 3-1 can be regarded as the limit of Case 4 as the root $`r_0`$ approaches $`r_1`$ while the roots $`r_1`$ through $`r_{m+1}`$ remain fixed. Thus, the relations (4.24) can be solved for $`l_0`$ and $`l_1`$ in the form
$$(r_1r_0)l_0=1\underset{\alpha =2}{\overset{m+1}{}}(r_1r_\alpha )l_\alpha (r_0r_1)l_1=1\underset{\alpha =2}{\overset{m+1}{}}(r_0r_\alpha )l_\alpha $$
Using these formulae, one computes the limit
$$\underset{r_0r_1}{lim}\left(\frac{dl_{0}^{}{}_{}{}^{2}}{l_0}+\frac{dl_{1}^{}{}_{}{}^{2}}{l_1}\right)=\frac{tda^2}{a^2}\frac{2dadt}{a},$$
where
(4.29)
$$a=1\underset{\alpha =2}{\overset{m+1}{}}(r_1r_\alpha )l_\alpha \text{and}t=l_2+l_3+\mathrm{}+l_{m+1},$$
and where the formulae (4.23) for $`l_\alpha `$ for $`2\alpha m+1`$ remain valid. Thus, the formula for $`R_D`$ in Case 3-1 is
(4.30)
$$R_D=\frac{tda^2}{4a^2}\frac{dadt}{2a}+\underset{\alpha =2}{\overset{m+1}{}}\frac{dl_{\alpha }^{}{}_{}{}^{2}}{4l_\alpha }.$$
In Case 3-1$`b`$, all of the quantities $`a,l_2,\mathrm{},l_{m+1}`$ are non-negative. In fact, the quantity $`a>0`$ together with the quantities $`(r_1r_\alpha )l_\alpha `$ for $`2\alpha m+1`$ can be regarded as affine homogeneous coordinates on the momentum cell $`C(p_D,\mu )`$.
Set $`\rho _\alpha =r_1r_\alpha >0`$ and $`\rho =(\rho _2,\mathrm{},\rho _{m+1})`$. Let $`E_\rho ^m`$ be the ellipsoidal domain defined by
$$\underset{\alpha =2}{\overset{m+1}{}}\rho _\alpha p_{\alpha }^{}{}_{}{}^{2}<1.$$
Define a surjective map $`s:E_\rho C(p_D,\mu )`$ by $`s^{}l_\alpha =p_\alpha ^2`$ for $`2\alpha m+1`$. This map $`s`$ satisfies
$$\overline{a}=s^{}(a)=1\underset{\alpha =2}{\overset{m+1}{}}\rho _\alpha p_{\alpha }^{}{}_{}{}^{2}\text{and}\overline{t}=s^{}(t)=p_{2}^{}{}_{}{}^{2}+\mathrm{}+p_{m+1}^{}{}_{}{}^{2}.$$
Thus $`R_\rho =s^{}(R_D)`$ has the form
$$R_\rho =\frac{\overline{t}d\overline{a}^2}{4\overline{a}^2}\frac{d\overline{a}d\overline{t}}{2\overline{a}}+\underset{\alpha =2}{\overset{m+1}{}}dp_{\alpha }^{}{}_{}{}^{2}.$$
Since $`\overline{t}`$ vanishes to second order at $`p=0`$ (the center of $`E_\rho `$), the quadratic form $`R_\rho `$ is visibly positive definite and smooth on a neighborhood of $`p=0`$. Let $`\delta >0`$ be less than any $`1/\sqrt{\rho _\alpha }`$ and consider the annular region $`A_\delta E_\rho `$ defined by $`\overline{t}\delta ^2`$. On this region, $`\overline{a}`$ and $`\overline{t}`$ are both positive and $`R_\rho `$ can be written in the form
$$R_\rho =\frac{\overline{t}}{4}\left(\frac{d\overline{a}}{\overline{a}}\frac{d\overline{t}}{\overline{t}}\right)^2\frac{d\overline{t}^2}{4\overline{t}^2}+\underset{\alpha =2}{\overset{m+1}{}}dp_{\alpha }^{}{}_{}{}^{2}=\frac{\overline{t}}{4}\left(\frac{d\overline{a}}{\overline{a}}\frac{d\overline{t}}{\overline{t}}\right)^2+R_\rho ^{},$$
where $`R_\rho ^{}`$ is defined by this last equality. Since $`\overline{t}=|p|^2`$, it follows without difficulty that $`R_\rho ^{}`$ is positive semidefinite on $`^m`$ minus the origin (where it is singular) and that its null space at each point is one dimensional and is spanned by the radial vector. Since the 1-form $`\rho =d\overline{a}/\overline{a}d\overline{t}/\overline{t}`$ is evidently nonvanishing on the radial vector field, it follows that $`R_\rho `$ is positive definite (and smooth) everywhere on $`E_\rho `$. Thus, $`R_\rho `$ on $`E_\rho `$ provides the desired resolution of the boundary singularities of $`R_D`$ on the momentum cell $`C(p_D,\mu )`$.
Moreover, on $`A_\delta `$, whose outer boundary is defined by $`\overline{a}=0`$, the inequality
$$R_\rho \frac{\delta ^2}{4}\left(\frac{d\overline{a}}{\overline{a}}\frac{d\overline{t}}{\overline{t}}\right)^2=\left(\frac{\delta }{2}d\left(\mathrm{log}\left(\frac{\overline{t}}{\overline{a}}\right)\right)\right)^2$$
holds. Since $`\mathrm{log}(\overline{t}/\overline{a})`$ is proper on $`A_\delta `$, it follows that $`R_\rho `$ is complete on $`E_\rho `$.
This result will be needed in §4.4.3, when completeness is being discussed. For use in that section, I will point out that the above formulae define a convex domain $`E_\rho ^m`$ and a complete metric $`R_\rho `$ on $`E_\rho `$ for any $`\rho =(\rho _2,\mathrm{},\rho _{m+1})`$ satisfying $`\rho _\alpha 0`$ for all $`\alpha `$. (Recall that the metrics that arise as resolutions of singular metrics on $`C(p_D,\mu )`$ satisfy $`0<\rho _2<\mathrm{}<\rho _{m+1}`$.)
The metric $`R_\rho `$ is flat only when $`\rho _2=\mathrm{}=\rho _{m+1}=0`$, in which case $`E_0=^m`$ and $`R_0`$ is the standard flat metric. Moreover, the above formulae show that $`R_\rho `$ is always conformally flat, with $`(E_\rho ,R_\rho )`$ being globally conformal to $`(E_0,R_0)`$.
#### 4.3.3. Necessary conditions for completeness
It turns out that most of the possible momentum cells cannot be the reduced momentum image of a complete Bochner-Kähler manifold.
###### Proposition 8.
If there is a complete Bochner-Kähler $`(M,g,\mathrm{\Omega })`$ whose reduced momentum mapping has image in $`C(p_D,\mu )`$, then $`C(p_D,\mu )`$ is bounded.
###### Proof.
Suppose that $`(M,g,\mathrm{\Omega })`$ is connected and complete, with characteristic polynomials $`p_C`$ and $`p_D`$ but that its reduced momentum mapping takes values in an unbounded momentum cell $`C(p_D,\mu )`$. Let $`I_1\times \mathrm{}\times I_m`$ be the corresponding spectral product. The unboundedness of $`C(p_D,\mu )`$ implies that $`I_1`$ is either $`[r,\mathrm{})`$ or $`(r,\mathrm{})`$ where $`r`$ is the largest real root of $`p_D`$.
Again, let $`P_2P_1`$ be the bundle over $`M^{}`$ constructed in the course of the proof of Proposition 5. Because the structure group of $`P_2`$ is $`\mathrm{I}_m\times G_\mathrm{\Lambda }`$, it follows that the 1-forms $`\omega _i`$ for $`1im`$ are actually well-defined on $`M^{}`$. Let $`E_1`$ be the vector field on $`M^{}`$ that is $`g`$-dual to $`\mathrm{Re}(\omega _1)`$. Then, by the relation $`d\lambda _i=T_i(\omega _i+\overline{\omega _i})`$, it follows that $`d\lambda _i(E_1)=0`$ for $`1<im`$ and that
$$d\lambda _1(E_1)=2T_1=2\sqrt{\frac{p_D(\lambda _1)}{_{j1}(\lambda _1\lambda _j)}}>0.$$
In particular, along an integral curve of $`E_1`$ the functions $`\lambda _j`$ for $`1<jm`$ are constant while $`\lambda _1`$ is strictly increasing.
Fix $`xM^{}`$ and let $`a:[0,T)M`$ be the maximal forward integral curve of $`E_1`$ with $`a(0)=x`$. I claim that $`T`$ cannot be finite. If it were, the fact that $`E_1`$ is a unit speed vector field and that $`M`$ is complete would imply that $`a(t)`$ approaches a limit $`yM`$ as $`t`$ approaches $`T`$ (after all, $`d(a(t),a(s))|ts|`$). The limit point $`y`$ could not lie in $`M^{}`$ since then $`[0,T)`$ would not be maximal. By continuity, $`\lambda (y)`$ must not lie in $`I_1^{}\times \mathrm{}\times I_m^{}`$. However, $`\lambda _i(y)=\lambda _i(x)`$ for $`1<im`$ while $`\lambda _1(y)>\lambda _1(x)`$. Since $`I_1^{}=(r,\mathrm{})`$ this forces $`\lambda (y)`$ to lie in $`I_1^{}\times \mathrm{}\times I_m^{}`$, a contradiction since $`h^{}(M^{})`$ lies in $`C(p_D,\mu )^{}`$. Thus $`T=\mathrm{}`$, as claimed. In particular, the forward flow of $`E_1`$ exists for all time on $`M^{}`$.
However, this leads to a contradiction: Along $`a`$, the element of arc is given by
$$ds=\frac{1}{2}\sqrt{\frac{_{j=2}^m\left(\lambda _1\lambda _j(x)\right)}{p_D(\lambda _1)}}d\lambda _1.$$
Let $`lim_t\mathrm{}\lambda _1\left(a(t)\right)=\lambda _{\mathrm{}}\mathrm{}`$. Since $`a:[0,\mathrm{})M`$ has unit speed, the integral
$$_{\lambda _1(x)}^\lambda _{\mathrm{}}\sqrt{\frac{_{j=2}^m\left(\xi \lambda _j(x)\right)}{p_D(\xi )}}𝑑\xi $$
must be infinite. However, this integral is bounded by
$$_{\lambda _1(x)}^{\mathrm{}}\sqrt{\frac{_{j=2}^m\left(\xi \lambda _j(x)\right)}{p_D(\xi )}}𝑑\xi $$
which converges, since $`p_D(t)`$ has degree $`m+2`$.
This contradiction implies that $`(M,g)`$ could not have been complete. ∎
###### Remark 9 (Bounded momentum cells).
The discussion in §4.2.3 shows that there are only two cases in which the momentum cell is bounded:
The first case is SubCase 3-1$`b`$, i.e., $`p_D`$ has $`r_1`$ as a double root and $`\mu _1=0`$. The spectral bands are $`I_1=[r_2,r_1)`$ and $`I_j=[r_{j+1},r_j]`$ for $`1<jm`$. This cell is bounded but not compact.
The second case is SubCase 4-0, i.e., $`p_D(t)`$ has $`m+2`$ simple roots $`r_0>\mathrm{}>r_{m+1}`$ and the spectral bands are $`I_j=[r_{j+1},r_j]`$ for $`1jm`$. This cell is compact. However, as the next proposition shows, Subcase 4-0 never contains a complete example when $`m>0`$.
###### Proposition 9.
When $`m>0`$, there is no complete Bochner-Kähler manifold whose reduced characteristic polynomial $`p_D`$ has $`m+2`$ distinct roots.
###### Proof.
In view of Proposition 8 and the remark above, what has to be shown is that SubCase 4-0 cannot occur for a complete Bochner-Kähler manifold when $`m>0`$. This will involve an interesting examination of the fixed points of the flow of the canonical torus action.
Thus, suppose, to the contrary, that $`(M,g,\mathrm{\Omega })`$ is a complete Bochner-Kähler structure with $`m>0`$ and and that
$$p_D(t)=(tr_0)(tr_1)\mathrm{}(tr_{m+1})$$
where $`r_0>\mathrm{}>r_{m+1}`$. By Proposition 8, the momentum cell $`C(p_D,\mu )`$ must be bounded, which implies that SubCase 4-0 obtains, namely $`(1)^{i1}p_h^{}(r_i)0`$ for $`1im+1`$.
For $`1\alpha m+1`$, let $`F_\alpha C(p_D,\mu )`$ be the $`\alpha `$-th face of this $`m`$-simplex, i.e., the intersection of $`C(p_D,\mu )`$ with the hyperplane $`l_\alpha =0`$ (where the functions $`l_\alpha `$ are as defined in (4.23)). Let $`N_\alpha =(h^{})^1(F_\alpha )`$ be the preimage of $`F_\alpha `$. Evidently, each $`N_\alpha `$ is a closed, analytic subset of $`M`$ and the union of the $`N_\alpha `$ is the complex subvariety $`NM`$. Thus, $`N_\alpha `$ is a (non-empty) complex subvariety of $`M`$.
For $`0\alpha m+1`$, define functions $`w_\alpha =(h^{})^{}(l_\alpha )0`$ on $`M`$ and then define vector fields $`W_\alpha 𝔷`$ by $`W_\alpha \text{ }\text{ }\mathrm{\Omega }=dw_\alpha `$. By (4.24), the $`W_\alpha `$ satisfy
(4.31)
$$\underset{\alpha =0}{\overset{m+1}{}}W_\alpha =\underset{\alpha =0}{\overset{m+1}{}}r_\alpha W_\alpha =0.$$
Moreover, any $`m`$ of these vector fields are linearly independent on $`M^{}`$. Note that since $`w_\alpha `$ reaches its minimum of $`0`$ along $`N_\alpha `$, the vector field $`W_\alpha `$ vanishes along $`N_\alpha `$. Since $`M`$ is complete, the flows of the vector fields $`W_\alpha `$ are complete.
I am going to show that the flow of each vector field $`W_\alpha `$ is periodic of period $`\pi `$ by examining the rotation $`W_\alpha `$ along the fixed hypersurface $`N_\alpha `$.
Now, equation (3.15) can be written as
(4.32)
$$t^nd\left(p_h(t^1)\right)=t^np_h(t^1)\left(t\alpha _1+t^2\alpha _2+\mathrm{}\right),$$
where I have replaced $`t`$ by $`t`$ and am regarding $`t`$ as a parameter, taken to be sufficiently small so that the series converges in a neighborhood of any given compact domain in $`M`$. Using (3.11), this can be written in the form
(4.33)
$$d\left(p_h(t^1)\right)=tp_h(t^1)\underset{k=0}{\overset{\mathrm{}}{}}\left(T^{}(tH)^k\omega +\omega ^{}(tH)^kT\right)$$
and the series can then be summed, yielding the equation
$$d\left(p_h(t^1)\right)=tp_h(t^1)\left(T^{}(\mathrm{I}_ntH)^1\omega +\omega ^{}(\mathrm{I}_ntH)^1T\right).$$
Replacing $`t`$ by $`t^1`$, this becomes
(4.34)
$$\begin{array}{cc}\hfill d\left(p_h(t)\right)& =p_h(t)\left(T^{}(t\mathrm{I}_nH)^1\omega +\omega ^{}(t\mathrm{I}_nH)^1T\right)\hfill \\ & =\left(T^{}\mathrm{Cof}(t\mathrm{I}_nH)\omega +\omega ^{}\mathrm{Cof}(t\mathrm{I}_nH)T\right).\hfill \end{array}$$
The final expression is valid for all $`t`$, while the middle expression is valid away from the locus $`p_h(t)=0`$ in $`P\times `$.
Since $`p_h(t)=p_{h^{\prime \prime }}(t)p_h^{}(t)`$, and since $`p_{h^{\prime \prime }}(t)`$ has constant coefficients, (4.34) implies
(4.35)
$$d\left(p_h^{}(t)\right)=p_h^{}(t)\left(T^{}(t\mathrm{I}_nH)^1\omega +\omega ^{}(t\mathrm{I}_nH)^1T\right)$$
away from the locus $`p_h(t)0`$.
Define a vector field $`W(t)`$ on $`M`$ by $`W(t)\text{ }\text{ }\mathrm{\Omega }=d\left(p_h^{}(t)\right)`$. This vector field depends polynomially on $`t`$ and lies in $`𝔷`$ for all $`t`$. In fact, comparison with (4.23), the definition of $`l_\alpha `$, shows that $`p_h^{}(r_\alpha )=p_D^{}(r_\alpha )w_\alpha `$, so it follows that
(4.36)
$$W(r_\alpha )=p_D^{}(r_\alpha )W_\alpha ,0\alpha m+1.$$
By (4.35), the vector field $`W(t)`$ has representative function $`w(t):P^n`$ given by
(4.37)
$$w(t)=2ip_h^{}(t)(t\mathrm{I}_nH)^1T.$$
The expression on the left is polynomial in $`t`$, so the expression on the right must be also. Since the flow of $`W(t)`$ is a holomorphic isometry, it follows that
$$d\left(w(t)\right)+\varphi w(t)=w^{}(t)\omega ,$$
where $`w^{}(t)`$ takes values in $`𝔲(n)`$. In fact, by (2.14),
(4.38)
$$\begin{array}{cc}\hfill w^{}(t)& =2ip_h^{}(t)(t\mathrm{I}_nH)^1[TT^{}(t\mathrm{I}_nH)^1T^{}(t\mathrm{I}_nH)^1T\mathrm{I}_n\hfill \\ & H^2h_1HVI_n]\hfill \end{array}$$
and the matrix on the right is visibly skew-Hermitian. When $`T(u)=0`$, formula (4.38) simplifies to the form in which it will be the most useful:
(4.39)
$$w^{}(t)(u)=2ip_{h^{}(u)}(t)\left(t\mathrm{I}_nH(u)\right)^1\left[H(u)^2+h_1(u)H(u)+V(u)I_n\right].$$
Now fix $`\beta `$ in the range $`1\beta m+1`$ and let $`k_\beta C(p_D,\mu )`$ be the vertex that lies on the intersection of the faces $`F_\alpha `$ for $`\alpha 0,\beta `$, i.e., $`k_\beta `$ is the vertex that lies *opposite* the face $`F_\beta `$. Applying Corollary 4, choose $`x_\beta M`$ to satisfy $`h^{}(x_\beta )=k_\beta `$ and then let $`u_\beta P`$ satisfy $`\pi (u_\beta )=x_\beta `$. Then $`T(u_\beta )=0`$ since the differential of $`h^{}`$ vanishes at $`x_\beta `$. In particular, $`x_\beta `$ is a zero of $`W_\alpha `$ for all $`\alpha `$.
Now, $`r_\alpha `$ is a root of $`p_{h^{}(u_\beta )}(t)`$ for all $`\alpha 0,\beta `$ since $`h^{}(u_\beta )`$ lies on each $`F_\alpha `$ with $`\alpha \beta `$. Thus, the set $`\{\lambda _1(x_\beta ),\mathrm{},\lambda _m(x_\beta )\}`$ consists of the $`r_\alpha `$ where $`\alpha 0,\beta `$. Consequently, since (4.2) now simplifies to
$$\begin{array}{cc}\hfill \underset{\alpha =0}{\overset{m+1}{}}(tr_\alpha )=p_D(t)& =p_{h^{}(u_\beta )}(t)\left(t^2+h_1(u_\beta )t+V(u_\beta )\right)\hfill \\ & =\left(\underset{\alpha 0,\beta }{\overset{m+1}{}}(tr_\alpha )\right)\left(t^2+h_1(u_\beta )t+V(u_\beta )\right),\hfill \end{array}$$
it follows that $`\left(t^2+h_1(u_\beta )t+V(u_\beta )\right)=(tr_0)(tr_\beta )`$. In particular, (4.39) becomes
$$w^{}(t)(u_\beta )=2i\left(\underset{\alpha 0,\beta }{}(tr_\alpha )\right)\left[H(u_\beta )r_0\mathrm{I}_n\right]\left[H(u_\beta )r_\beta I_n\right]\left[t\mathrm{I}_nH(u_\beta )\right]^1.$$
Now, any eigenvalue of $`H(u_\beta )`$ is a root of $`p_{h(u_\beta )}(t)=p_{h^{\prime \prime }}(t)p_{h^{}(u_\beta )}(t)`$ and so, by Proposition 5, must be of the form $`r_\gamma `$ for some $`\gamma =0,\mathrm{},m+1`$. Let $`V_{\beta ,\gamma }^n`$ denote the eigenspace of $`H(u_\beta )`$ belonging to the eigenvalue $`r_\gamma `$. Then the above formula implies that $`w^{}(t)(u_\beta )`$ annihilates $`V_{\beta ,\beta }`$ and $`V_{\beta ,0}`$ and that, for $`vV_{\beta ,\gamma }`$ with $`\gamma 0,\beta `$,
(4.40)
$$w^{}(t)(u_\beta )v=2i(r_\gamma r_0)(r_\gamma r_\beta )\left(\underset{\alpha 0,\beta ,\gamma }{}(tr_\alpha )\right)v.$$
Since the right hand side of (4.40) is a polynomial in $`t`$, it now makes sense to substitute $`t=r_\alpha `$ for any $`\alpha `$. When $`\alpha 0,\beta ,\gamma `$, this gives $`w^{}(r_\alpha )(u_\beta )v=0`$ for $`vV_{\beta ,\gamma }`$, while, if $`\gamma 0,\beta `$, this gives
$$w^{}(r_\gamma )(u_\beta )v=2i(r_\gamma r_0)(r_\gamma r_\beta )\left(\underset{\alpha 0,\beta ,\gamma }{}(r_\gamma r_\alpha )\right)v=2ip_D^{}(r_\gamma )v.$$
In other words, for $`\beta \gamma `$ in the range $`1\beta ,\gamma m+1`$,
(4.41)
$$w^{}(r_\gamma )(u_\beta )=2ip_D^{}(r_\gamma )E_{\beta ,\gamma }$$
where $`E_{\beta ,\gamma }:^nV_{\beta ,\gamma }`$ is the orthogonal projection onto this eigenspace. Thus, the flow of $`W(r_\gamma )`$ is periodic of period $`\pi /p_D^{}(r_\gamma )`$ and so, by (4.36), the flow of $`W_\gamma `$ is periodic of period $`\pi `$, as claimed.
Now, further information can be got by evaluating $`w^{}(r_\beta )`$ at $`u_\beta `$ itself. Indeed, in the above formula, if $`v`$ lies in $`V_{\beta ,\gamma }`$ with $`\gamma 0,\beta `$, then putting $`t=r_\beta `$ gives
$$\begin{array}{cc}\hfill w^{}(r_\beta )(u_\beta )(v)& =2i(r_\gamma r_0)(r_\gamma r_\beta )\left(\underset{\alpha 0,\beta ,\gamma }{}(r_\beta r_\alpha )\right)v\hfill \\ & =2i\frac{(r_\gamma r_0)}{(r_\beta r_0)}\left(\underset{\alpha \beta }{}(r_\beta r_\alpha )\right)v=2ip_D^{}(r_\beta )\frac{(r_\gamma r_0)}{(r_\beta r_0)}v,\hfill \end{array}$$
In other words, using the projection notation already introduced,
$$w^{}(r_\beta )(u_\beta )=2ip_D^{}(r_\beta )\underset{\gamma 0,\beta }{}\frac{(r_\gamma r_0)}{(r_\beta r_0)}E_{\beta ,\gamma }.$$
Since the flow of $`W_\beta `$ has period $`\pi `$, each of the ratios $`(r_\gamma r_0)/(r_\beta r_0)`$ must be an integer for $`1\beta \gamma m+1`$. Since $`r_\beta r_\gamma `$ when $`\beta \gamma `$, these ratios cannot be $`+1`$. Thus, as the inverse of each such ratio is another such ratio, these integers must all be $`1`$. However, this is equivalent to $`\frac{1}{2}(r_\beta +r_\gamma )=r_0`$, which is impossible, since $`r_0`$ is greater than either $`r_\beta `$ or $`r_\gamma `$. This contradiction establishes the proposition. ∎
###### Corollary 5.
The only connected compact Bochner-Kähler manifolds are the compact quotients of the symmetric Bochner-Kähler manifolds $`M_c^p\times M_c^{np}`$.
###### Proof.
A compact Bochner-Kähler manifold is necessarily complete and its reduced momentum image is necessarily compact. Proposition 9, Corollary 4, and the fact that only SubCase 4-0 has a compact momentum cell imply that $`m>0`$ is impossible. When $`m=0`$, the momentum mapping is constant and the metric is therefore locally homogeneous, so that, by Proposition 1, its simply-connected cover (which is complete) must be isometric to $`M_c^p\times M_c^{np}`$, as claimed. ∎
###### Remark 10 (Orbifolds).
While Proposition 9 rules out the existence of a Bochner-Kähler manifold in Subcase 4-0, it does not rule out the existence of orbifolds. In fact, the argument in Proposition 9 implies that, if there is a complete orbifold with reduced characteristic polynomial $`p_D(t)`$ as in the proof, then the ratios $`(r_\gamma r_0)/(r_\beta r_0)`$ must all be rational for $`1\beta ,\gamma m+1`$. A little algebra then leads to the formulae
$$\begin{array}{cc}\hfill p_D(t)& =(tr_0)(tr_1)\mathrm{}(tr_{m+1})\hfill \\ \hfill p_C(t)& =(tr_0)^{\nu _0+1}(tr_1)^{\nu _1+1}\mathrm{}(tr_{m+1})^{\nu _{m+1}+1}\hfill \end{array}$$
with
$$r_\beta =r\underset{\alpha =0}{\overset{m+1}{}}(\nu _\alpha +1)(p_\alpha p_\beta ),0\beta m+1$$
where $`r>0`$ is real, $`0=p_0<p_1<p_2<\mathrm{}<p_{m+1}`$ is a strictly increasing sequence of integers with no common divisor, and $`\nu _0,\mathrm{},\nu _{m+1}`$ are nonnegative integers satisfying
$$n=m+\nu _0+\mathrm{}+\nu _{m+1}.$$
While I have not done all of the necessary calculations, it appears that, for each choice of $`r`$, $`p=(p_1,\mathrm{},p_{m+1})`$, and $`\nu =(\nu _0,\mathrm{},\nu _{m+1})`$ satisfying the above conditions, there exists a complete orbifold with characteristic polynomials $`p_C`$ and $`p_D`$ as above that fits into SubCase 4-0. The case $`n=1`$ has already been verified in §3.2.5, and the cases with $`n=m`$ will be verified in §4.4.6.
The parameter $`r`$ can be normalized to $`1`$ by scaling the metric. Thus, up to scaling, there exists a countable family of complete Bochner-Kähler orbifolds in each dimension whose momentum cells are compact. It appears that these orbifolds are weighted projective space in most cases. In fact, it will be seen that every weighted projective space carries a Bochner-Kähler metric.
By the same methods as employed in the proof of Proposition 9, one can prove the following more general periodicity result. Details will be left to the reader.
###### Proposition 10.
Let $`r_\alpha `$ be a simple root of $`p_D`$, let $`w_\alpha =(h^{})^{}(l_\alpha )`$, and let $`W_\alpha `$ be the vector field in $`𝔷`$ defined by $`W_\alpha \text{ }\text{ }\mathrm{\Omega }=dw_\alpha `$. Then on a neighborhood of any zero of $`W_\alpha `$, the flow of $`W_\alpha `$ is periodic, with period $`\pi `$.
###### Remark 11 (Locality).
One must restrict to a neighborhood of a fixed point for the conclusion of Proposition 10. In the first place, without some completeness assumptions, there is no reason to believe that the flow of $`W_\alpha `$ is even defined for all time except near a fixed point. In the second place, even if the flow is defined for all time, by removing the zero locus of $`W_\alpha `$ and passing to a covering space, one could conceivably arrange that $`W_\alpha `$ have no closed orbits at all.
### 4.4. A geodesic foliation
By Theorem 4, the vector fields $`Z_2,\mathrm{},Z_{m+1}`$ are linearly independent (over $``$) on $`M^{}`$ and satisfy $`[Z_i,Z_j]=0`$ for $`2i,jm+1`$. Moreover, since these vector fields are the real parts of holomorphic vector fields, they satisfy
$$[Z_i,Z_j]=[Z_i,JZ_j]=[JZ_i,JZ_j]=0.$$
Since the $`2m`$ vector fields $`Z_2,\mathrm{},Z_{m+1},JZ_2,\mathrm{},JZ_{m+1}`$ Lie-commute and are linearly independent on $`M^{}`$, they are tangent to a foliation $``$ of $`M^{}`$ whose leaves are complex submanifolds of $`M`$ (of complex dimension $`m`$). Moreover, the vector fields $`JZ_2,\mathrm{},JZ_{m+1}`$ are tangent to a foliation $``$ of $`M^{}`$ whose tangent spaces are the orthogonal complements to fibers of $`h^{}:M^{}C(p_D,\mu )^{}`$.
#### 4.4.1. Geometry of the leaves
It turns out that the $``$-leaves are themselves rather interesting objects.
###### Proposition 11.
The leaves of the foliation $``$ are totally geodesic in $`M^{}`$ and the induced Kähler structure on each $``$-leaf is Bochner-Kähler of cohomogeneity $`m`$. The characteristic and momentum polynomials of any $``$-leaf are
$$p_C^L(t)=p_D^L(t)=p_D(t\lambda ),\text{and}p_h^L(t)=p_h^{}^L(t)=p_h^{}(t\lambda )$$
where the constant $`\lambda `$ is defined so that $`p_{h^{\prime \prime }}(t)=t^{nm}(m+2)\lambda t^{nm1}+\mathrm{}`$.
###### Proof.
Return to the structure equations on $`P_2`$ that were introduced in the proof of Proposition 5. Since $`(1)^{i1}p_D(\lambda _i)>0`$ for $`1im`$ while $`p_D(\lambda _a)=0`$ for $`a>m`$, it follows that $`\lambda _i\lambda _a`$ is nonvanishing on $`M^{}`$ and hence on $`P_2`$. Equation (4.4) can thus be written as
$$\varphi _{a\overline{ı}}=\frac{T_i}{\lambda _i\lambda _a}\omega _a.$$
Let $`L^{}M^{}`$ be an $``$-leaf, and let $`P_2^LP_2`$ be the bundle $`\pi ^1(L^{})P_2`$, which is a $`G_\mathrm{\Lambda }`$-bundle over $`L^{}`$. By the definition of the bundle $`P_2`$ and the foliation $``$, the forms $`\omega _a`$ vanish when pulled back to $`P_2^L`$, so, by the above equations, so do the forms $`\varphi _{a\overline{ı}}`$. Consequently, the complex $`m`$-manifold $`L^{}`$ is totally geodesic in $`M^{}`$, as claimed.
Denoting pullback to $`P_2^L`$ by a superscript $`L`$, the formulae
$$\omega ^L=\left(\begin{array}{c}\stackrel{~}{\omega }\\ 0\end{array}\right),\varphi ^L=\left(\begin{array}{cc}\stackrel{~}{\varphi }& 0\\ 0& \varphi ^{\prime \prime }\end{array}\right),H^L=\left(\begin{array}{cc}H^{}& 0\\ 0& \mathrm{\Lambda }\end{array}\right),T^L=\left(\begin{array}{c}T^{}\\ 0\end{array}\right)$$
hold, where $`\stackrel{~}{\omega }`$ takes values in $`^m`$ and $`\stackrel{~}{\varphi }`$ takes values in $`𝔲(m)`$. The notations $`H^{}`$, $`T^{}`$, and $`\mathrm{\Lambda }`$ are as previously established. The Kähler form $`\mathrm{{\rm Y}}`$ induced on $`L^{}`$ by pullback from $`\mathrm{\Omega }`$ then satisfies $`\pi ^{}(\mathrm{{\rm Y}})=\frac{i}{2}\stackrel{~}{\omega }^{}\stackrel{~}{\omega }`$.
The pullbacks of the structure equations to $`P_2^L`$ then imply $`d\stackrel{~}{\omega }=\stackrel{~}{\varphi }\stackrel{~}{\omega }`$, so that $`\stackrel{~}{\varphi }`$ is the connection matrix of the torsion-free Kähler connection of the induced Kähler structure. The pullbacks further imply
$$\begin{array}{cc}\hfill d\stackrel{~}{\varphi }+\stackrel{~}{\varphi }\stackrel{~}{\varphi }& =H^{}\stackrel{~}{\omega }^{}\stackrel{~}{\omega }H^{}\stackrel{~}{\omega }\stackrel{~}{\omega }^{}\stackrel{~}{\omega }\stackrel{~}{\omega }^{}H^{}+\stackrel{~}{\omega }^{}H^{}\stackrel{~}{\omega }\mathrm{I}_m\hfill \\ & +(\mathrm{tr}H^{}+\mathrm{tr}\mathrm{\Lambda })\left(\stackrel{~}{\omega }^{}\stackrel{~}{\omega }\mathrm{I}_m\stackrel{~}{\omega }\stackrel{~}{\omega }^{}\right)\hfill \\ & =\stackrel{~}{H}\stackrel{~}{\omega }^{}\stackrel{~}{\omega }\stackrel{~}{H}\stackrel{~}{\omega }\stackrel{~}{\omega }^{}\stackrel{~}{\omega }\stackrel{~}{\omega }^{}\stackrel{~}{H}+\stackrel{~}{\omega }^{}\stackrel{~}{H}\stackrel{~}{\omega }\mathrm{I}_m\hfill \\ & +(\mathrm{tr}\stackrel{~}{H})\left(\stackrel{~}{\omega }^{}\stackrel{~}{\omega }\mathrm{I}_m\stackrel{~}{\omega }\stackrel{~}{\omega }^{}\right)\hfill \end{array}$$
where $`\stackrel{~}{H}=H^{}+\lambda \mathrm{I}_m`$ and $`(m+2)\lambda =\mathrm{tr}\mathrm{\Lambda }`$. Since $`p_{h^{\prime \prime }}(t)=det\left(tI_{nm}\mathrm{\Lambda }\right)`$, this defines $`\lambda `$ as in the statement of the proposition.
Thus, by definition, the induced metric on $`L^{}`$ is Bochner-Kähler and has momentum polynomial
$$p_h^L(t)=det\left(tI_m\stackrel{~}{H}\right)=det\left((t\lambda )I_mH^{}\right)=p_h^{}(t\lambda ),$$
as claimed. Moreover, the pullback of the $`dH`$ equation implies
$$d\stackrel{~}{H}=\stackrel{~}{\varphi }\stackrel{~}{H}+\stackrel{~}{H}\stackrel{~}{\varphi }+T^{}\stackrel{~}{\omega }^{}+\stackrel{~}{\omega }(T^{})^{}$$
so that $`\stackrel{~}{T}=T^{}`$ is the $`^m`$-valued function defined by the structure equations for $`\mathrm{{\rm Y}}`$.
Since the entries of $`\stackrel{~}{T}=T^{}`$ are all nonzero and the eigenvalues of $`\stackrel{~}{H}`$ are distinct, the rank of the momentum mapping for $`L^{}`$ is $`m`$, implying that $`p_h^{}^L(t)=p_h^L(t)`$.
Finally, the pullback of the identity for $`dT`$ becomes
$$d\stackrel{~}{T}=\stackrel{~}{\varphi }\stackrel{~}{T}+\left((\stackrel{~}{H})^2+\mathrm{tr}(\stackrel{~}{H})\stackrel{~}{H}+(V\lambda \mathrm{tr}(H^{})(m+1)\lambda ^2)\right)\stackrel{~}{\omega },$$
so that, setting $`\stackrel{~}{V}=V\lambda \mathrm{tr}(H^{})(m+1)\lambda ^2`$, the structure function for the metric on $`L^{}`$ takes the form $`(\stackrel{~}{H},\stackrel{~}{T},\stackrel{~}{V})`$.
The formula for $`p_C^L(t)`$ then becomes
$$\begin{array}{cc}\hfill p_C^L(t)& =det(tI_m\stackrel{~}{H})\left(t^2+\mathrm{tr}(\stackrel{~}{H})t+\stackrel{~}{V}\right)+(\stackrel{~}{T})^{}\mathrm{Cof}(tI_m\stackrel{~}{H})\stackrel{~}{T}\hfill \\ & =det((t\lambda )I_mH^{})\left((t\lambda )^2+\mathrm{tr}(H)(t\lambda )+V\right)\hfill \\ & +(\stackrel{~}{T})^{}\mathrm{Cof}\left((t\lambda )I_mH^{}\right)\stackrel{~}{T}\hfill \\ & =\frac{p_C(t\lambda )}{p_{h^{\prime \prime }}(t\lambda )}=p_D(t\lambda )\hfill \end{array}$$
where the last line uses the definition of $`p_C(t)`$, the identity $`p_h(t)=p_h^{}(t)p_{h^{\prime \prime }}(t)`$, and the identity
$$\mathrm{Cof}(tI_nH)=\left(\begin{array}{cc}p_{h^{\prime \prime }}(t)\mathrm{Cof}(tI_mH^{})& 0\\ 0& p_h^{}(t)\mathrm{Cof}(tI_{nm}\mathrm{\Lambda })\end{array}\right).$$
These formulae establish the proposition. ∎
###### Corollary 6.
The momentum mapping $`h^L:L^{}^m`$ is equal to the restriction of $`h^{}:M^{}^m`$ to $`L^{}`$ followed by an invertible linear map $`\mathrm{\Phi }_\lambda :^m^m`$. The corresponding momentum cells satisfy $`C(p_D^L,\mu ^L)=\mathrm{\Phi }_\lambda \left(C(p_D,\mu )\right)`$.
#### 4.4.2. Completion and real slices
It will now be shown that the $``$\- and $``$-leaves can be extended through the locus $`N`$ where the $`Z_k`$ become dependent.
###### Proposition 12.
If the metric on $`M`$ is complete, then the closure of any $``$-leaf $`L^{}`$ is a complete, totally geodesic complex $`m`$-manifold $`LM`$. Moreover, the geodesic completion of any $``$-leaf $`R^{}L^{}`$ is a totally geodesic real $`m`$-manifold $`R`$ and the mapping $`h^{}:RC(p_D,\mu )`$ is surjective.
###### Proof.
Before beginning the proof, it will be useful to establish the following fact. If $`g`$ is any real-analytic, complete metric on a manifold $`M`$ and $`SM`$ is a connected, totally geodesic submanifold of some dimension $`s`$, then $`S`$ can be ‘completed’: There exists an $`s`$-manifold $`\overline{S}`$ and a totally geodesic immersion $`\iota :\overline{S}M`$ whose image contains $`S`$ and for which the induced metric $`\overline{g}=\iota ^{}g`$ on $`\overline{S}`$ is complete. This completion $`(\overline{S},\iota )`$ is unique up to diffeomorphism.
Here is a sketch of the proof: Fix $`xS`$ and consider, for every $`vT_xS`$, the constant speed geodesic $`\gamma _v:M`$ defined by $`\gamma _v(t)=\mathrm{exp}_x(tv)`$. Let $`E(v)T_{\gamma _v(1)}M`$ be the parallel translation of $`E(0)=T_xS`$ along $`\gamma _v`$ from $`t=0`$ to $`t=1`$, and let $`\overline{S}\mathrm{Gr}_s(TM)`$ be the set of all such $`E(v)`$. Since $`S`$ is totally geodesic, when $`|v|`$ is sufficiently small $`E(v)`$ is equal to $`T_{\gamma _v(1)}S`$. It follows, by the real-analyticity of $`g`$, that $`\mathrm{exp}_{\gamma _v(1)}:E(v)M`$ embeds $`B_\delta (0)E(v)`$ into $`M`$ as a totally geodesic submanifold of $`M`$ as long as $`\delta >0`$ is less than the injectivity radius at $`\gamma _v(1)`$. From this, it is not hard to prove that $`\overline{S}`$ is an embedded submanifold of $`\mathrm{Gr}_s(TM)`$. Moreover, the basepoint projection $`\iota :\overline{S}M`$ defined by $`\iota \left(E(v)\right)=\gamma _v(1)`$ is a totally geodesic immersion. Since each of the geodesics $`\gamma _v`$ lifts to a complete geodesic $`tE(tv)`$ in $`\overline{S}`$ for the induced metric $`\overline{g}=\iota ^{}g`$, all of the $`\overline{g}`$-geodesics in $`\overline{S}`$ passing through $`E(0)`$ are complete. Thus, $`\overline{g}`$ is complete. The completeness of the metric then ensures that $`\iota (\overline{S})`$ contains $`S`$, as desired.
Now apply this result to the leaf $`L^{}M^{}`$ and consider $`\overline{L^{}}\mathrm{Gr}_{2m}(TM)`$. Since the induced metric on $`\overline{L^{}}`$ is real-analytic and is Bochner-Kähler on an open set, it is Bochner-Kähler everywhere. Moreover, by Proposition 11, it has cohomogeneity $`m`$, equal to its complex dimension. Let $`h^L:\overline{L^{}}^m`$ denote its momentum mapping. By Corollary 6 and real-analyticity, $`h^L=\mathrm{\Phi }_\lambda h^{}\iota `$, since this holds on $`L^{}`$. Since the rank of $`𝔷`$ is $`m`$, the proof of Theorem 4 coupled with the remarks of §3.3.4 show that $`𝔷`$ accounts for all of the infinitesimal symmetries of $`\overline{L^{}}`$, i.e., that the full symmetry group of $`\overline{L^{}}`$ is generated by the canonical torus action, even if one were to pass to its simply-connected cover. In particular, by Corollary 4, the fibers of $`h^L`$ are connected and are the orbits of the canonical torus action.
Now, $`h^L`$ is a submersion outside some closed complex submanifold $`K\overline{L^{}}`$. Since $`h^{}`$ is a submersion only when it is restricted to $`M^{}`$, it follows that that $`\iota \left(\overline{L^{}}K\right)`$ must lie in $`M^{}`$. Since $`\overline{L^{}}K`$ is connected, and since it contains the $``$-leaf $`L^{}`$, it follows from analyticity that $`\iota (\overline{L^{}}K)`$ must be equal to $`L^{}`$. Thus, $`\overline{L^{}}K`$ consists of the tangent planes to $`L^{}`$. It follows that $`\iota `$ is one-to-one on $`\overline{L^{}}K`$.
If $`\iota `$ were not one-to-one on $`K`$, this would violate the connectedness of the fibers of $`h^L`$, so $`\iota `$ is one-to-one everywhere. In other words, $`L=\iota (\overline{L^{}})`$ is a submanifold of $`M`$, as claimed. Obviously, $`L`$ is the closure of $`L^{}`$ in $`M`$.
Now, turning to the geometry of the leaves of the foliation $``$, note that these leaves are defined by the equations $`\omega _a=\mathrm{Im}(\omega _i)=0`$ (since $`H`$ and $`T`$ are real on $`P_2`$). To prove that these leaves are totally geodesic, it would suffice to show that the imaginary part of $`\varphi ^{}`$ vanishes when one restricts to such a leaf. Thus, write $`\stackrel{~}{\omega }=\xi +i\eta `$ and $`\varphi ^{}=\theta +i\psi `$, where $`\xi `$, $`\eta `$, $`\theta `$, and $`\psi `$ take values in $`^m`$, $`^m`$, $`𝔰𝔬(m)`$, and the space of real symmetric matrices, respectively. Since $`H`$ and $`T`$ are real-valued, the imaginary part of the equation for $`dH^{}`$ becomes
$$0=H^{}\psi \psi H^{}T^{}{}_{}{}^{t}\eta +\eta {}_{}{}^{t}T_{}^{}.$$
It then follows by linear algebra that on the open set $`UM^{}`$ where $`H^{}`$ has no two eigenvalues that sum to zero, the components of $`\psi `$ are linear combinations of the components of $`\eta `$. By the structure equation for $`dH^{}`$, the eigenvalues of $`H^{}`$ are independent on each leaf of $``$ since $`d\lambda _i=2T_i,\xi `$. Thus, the open set $`U`$ intersects each $``$-leaf in a dense open set. Consequently, the components of $`\psi `$ vanish on each $``$-leaf, implying that each $``$-leaf is totally geodesic, as desired.
Now, let $`R^{}`$ be an $``$-leaf and let $`\overline{R}\mathrm{Gr}_m(TM)`$ be its geodesic completion. By construction, $`R^{}`$ meets each isometry orbit in $`M^{}`$ orthogonally. Thus, by Theorem 8, the map $`h^{}:R^{}C(p_D,\mu )^{}`$ is an isometry when $`R^{}`$ is given the induced metric. It remains to show that $`h^{}\iota :\overline{R}C(p_D,\mu )`$ is surjective. The completeness and real-analyticity of the induced metric $`\overline{g}`$ on $`\overline{R}`$ coupled with the analysis of the resolution of the singular cell metric in SubCase 3-1$`b`$ done at the end of §4.3.2, shows that $`(\overline{R},\overline{g})`$ must be an isometric quotient of $`(E_\rho ,R_\rho )`$ for some $`\rho `$. The completeness of this mapping and the surjectivity of this resolution imply the desired surjectivity. ∎
#### 4.4.3. The leaf metric
The Bochner-Kähler metric induced on a leaf $`L^{}`$ can now be described rather explicitly in terms of the geometry of the momentum cell associated to $`M`$.
###### Theorem 9.
Let $`\left(R_{ij}^D(u)\right)`$ be the inverse matrix to the coefficient matrix $`\left(R_D^{ij}(u)\right)`$ of the cell metric $`R_D`$ on $`C(p_D,\mu )^{}`$. Then, on the universal cover of $`L^{}`$, there exist functions $`\theta ^1,\mathrm{},\theta ^m`$ for which the induced Kähler form and metric are
$$\mathrm{{\rm Y}}=dh_k^{}d\theta ^k\text{and}ds^2=R_D^{jk}(h^{})dh_j^{}dh_k^{}+R_{jk}^D(h^{})d\theta ^jd\theta ^k.$$
###### Proof.
First, it will be useful to take a different basis for $`𝔷`$. Recall that the functions $`(h_1^{},\mathrm{},h_m^{})`$ are constant linear combinations of the functions $`(h_1,\mathrm{},h_m)`$ and vice versa. By equation (3.19), $`Z_{k+1}\text{ }\text{ }\mathrm{\Omega }=dh_k`$ for $`1km`$, so the vector fields $`Z_2^{},\mathrm{},Z_{m+1}^{}`$ defined by $`Z_{k+1}^{}\text{ }\text{ }\mathrm{\Omega }=dh_k^{}`$ for $`1km`$ are also a basis of $`𝔷`$.
Let $`L^{}`$ be an $``$-leaf. The vector fields $`\{Z_k^{}iJZ_k^{}2km+1\}`$ are a basis for the holomorphic vector fields on $`L^{}`$, so there are unique holomorphic 1-forms $`\zeta ^1,\mathrm{},\zeta ^m`$ on $`L^{}`$ that satisfy
$$\zeta ^j\left(Z_{k+1}^{}iJZ_{k+1}^{}\right)=ı\delta _k^j$$
for $`1j,km`$. (The introduction of the factor of $`ı`$ simplifies formulae to appear below.) Because the vector fields $`Z_k^{}iJZ_k^{}`$ are Lie-commuting, the 1-forms $`\zeta ^j`$ are closed.
Write $`\zeta ^j=\xi ^j+ı\eta ^j`$ where $`\xi ^j`$ and $`\eta ^j`$ are real $`1`$-forms. Then the defining equation above is equivalent to
$$\eta ^j(Z_{k+1}^{})=\xi ^j(JZ_{k+1}^{})=\frac{1}{2}\delta _k^j\text{and}\xi ^j(Z_{k+1}^{})=\eta ^j(JZ_{k+1}^{})=0.$$
Since the $`\zeta ^j`$ are a basis for the holomorphic 1-forms on $`L^{}`$, the metric on $`L^{}`$ can be written in the form
$$ds^2=g_{jk}\zeta ^j\overline{\zeta ^k},$$
where $`g_{jk}=\overline{g_{kj}}`$ and where the pullback of $`\mathrm{\Omega }`$ to $`L^{}`$ is
$$\mathrm{{\rm Y}}=\frac{i}{2}g_{jk}\zeta ^j\overline{\zeta ^k}.$$
Now,
$$dh_k^{}=Z_{k+1}^{}\text{ }\text{ }\mathrm{{\rm Y}}=\frac{1}{2}\left(g_{kj}\overline{\zeta ^j}+g_{jk}\zeta ^j\right),$$
or, equivalently,
$$dh_k^{}=\frac{1}{2}(g_{kj}+g_{jk})\xi ^j+\frac{i}{2}(g_{kj}g_{jk})\eta ^j.$$
Since $`Z_{j+1}^{}`$ for $`1jm`$ is tangent to the fibers of $`h^{}`$, the coefficient of $`\eta ^j`$ in the above equation must vanish, i.e., $`g_{kj}=g_{jk}=\overline{g_{kj}}`$. Thus,
$$dh_k^{}=g_{kj}\xi ^j.$$
Define $`g^{ij}=g^{ji}`$ so that $`g^{ij}g_{jk}=\delta _k^i`$. Note, in particular, that $`\xi ^j=g^{jk}dh_k^{}`$. The metric on $`L^{}`$ can now be written in the form
$$\begin{array}{cc}\hfill ds^2& =g_{jk}\zeta ^j\overline{\zeta ^k}=g_{jk}(\xi ^j+i\eta ^j)(\xi ^ki\eta ^k)\hfill \\ & =g_{jk}\left(\xi ^j\xi ^k+\eta ^j\eta ^k\right)\hfill \\ & =g^{jk}dh_j^{}dh_k^{}+g_{jk}\eta ^j\eta ^k.\hfill \end{array}$$
Since $`L^{}`$ is totally geodesic in $`M^{}`$, it follows from Theorem 8 that
$$g^{ij}dh_i^{}dh_j^{}=(h^{})^{}(R_D)=R_D^{ij}(h^{})dh_i^{}dh_j^{}.$$
Thus, $`g^{ij}=R_D^{ij}(h^{})`$ and so
$$\xi ^j=g^{jk}dh_k^{}=R_D^{jk}(h^{})dh_k^{}=(h^{})^{}\left(R_D^{jk}(u)du_k\right)$$
Now lift everything to the universal cover of $`L^{}`$. Since the $`\eta ^k`$ are closed, there exist functions $`\theta ^1,\mathrm{},\theta ^m`$ on this universal cover so that $`\eta ^k=d\theta ^k`$. Then
$$ds^2=R_D^{jk}(h^{})dh_j^{}dh_k^{}+R_{jk}^D(h^{})d\theta ^jd\theta ^k,$$
where $`(R_{jk}^D)`$ is the inverse matrix to $`(R_D^{jk})`$ and
$$\mathrm{{\rm Y}}=R_{jk}^D(h^{})\xi ^jd\theta ^k=dh_k^{}d\theta ^k.$$
These are the desired formulae. ∎
###### Remark 12 (Kähler metrics of Hessian type).
The reader may find the metric in the above form to be very familiar. In fact, Kähler metrics of this form are well-known in the literature as being of *Hessian type*. Their general form is as follows: Let $`D^m`$ be an open domain in $`^m`$, assumed to be simply-connected for simplicity. Let $`x_1,\mathrm{},x_m`$ be any linear coordinates on $`^m`$ and suppose that $`g`$ is a Riemannian metric on $`D`$, written in the form
$$g=g^{jk}(x)dx_jdx_k.$$
Using the flat affine structure on $`^m`$ restricted to $`D`$, one gets a canonical metric on $`T^{}D`$ as follows: Let $`y^1,\mathrm{},y^m`$ be the coordinates that are linear on the fibers of $`T^{}DD`$ and dual to the coordinates $`x_1,\mathrm{},x_m`$ in the sense that the tautological 1-form on $`T^{}D`$ is $`y^jdx_j`$. Let $`g_{jk}=g_{kj}`$ be the functions on $`D`$ so that $`g^{jk}g_{kl}=\delta _l^j`$ and define the metric
$$\widehat{g}=g^{jk}(x)dx_jdx_k+g_{jk}(x)dy^jdy^k.$$
This metric on $`T^{}D`$ does not depend on the choice of coordinates $`x_i`$, but only on $`g`$ and the flat affine structure that $`D`$ inherits from $`^m`$. Moreover, this metric is compatible with the symplectic form on $`T^{}D`$ given by $`\mathrm{{\rm Y}}=d\left(y^jdx_j\right)=dx_jdy^j`$.
Thus, the metric $`\widehat{g}`$ and 2-form $`\mathrm{{\rm Y}}`$ define an almost complex structure on $`T^{}D`$ for which the 1-forms
$$\zeta ^j=g^{jk}(x)dx_k+ıdy^j$$
give a basis for the $`(1,0)`$-forms. This almost complex structure will be integrable if and only if the forms $`\zeta ^j`$ are closed. In other words, the pair $`(\widehat{g},\mathrm{{\rm Y}})`$ defines a Kähler metric on $`T^{}D`$ if and only if $`d\left(g^{jk}(x)dx_k\right)=0`$ for all $`1jm`$. Since $`D`$ is simply-connected, this closure condition holds if and only if there exists a convex ‘potential’ function $`G:D`$ for which
$$g^{jk}=\frac{^2G}{x_jx_k},$$
i.e., if and only if the metric $`g`$ is of Hessian type. For this reason, metrics of the form $`\widehat{g}`$ as above are often called Kähler metrics of Hessian type. Note that, for such a metric, translation in the $`y`$-variables defines a Hamiltonian torus action that is holomorphic and whose momentum mapping is the basepoint projection $`T^{}DD`$.
For further investigation of these metrics, the reader might consult and .
In the case of Bochner-Kähler metrics, the formula for the potential function $`G:C(p_D,\mu )^{}`$ has been indicated in §4.3.2. The main problem with this representation is that it only describes the leaf metric away from the singular locus $`N`$. More work must now be done to analyze the metric near this locus.
#### 4.4.4. A partial completion
By Theorem 9, the $`2`$-form and Riemannian metric on $`C(p_D,\mu )^{}\times ^m`$ defined by
$$\mathrm{{\rm Y}}=du_kd\theta ^k\text{and}ds^2=R_D^{jk}(u)du_jdu_k+R_{jk}^D(u)d\theta ^jd\theta ^k$$
define a Bochner-Kähler structure on $`C(p_D,\mu )^{}\times ^m`$. The simply-connected cover of any $``$-leaf $`L^{}`$ has an immersion into $`C(p_D,\mu )^{}\times ^m`$, canonical up to a translation in the $`\theta `$-coordinates, that pulls this Bochner-Kähler structure back to the induced one on $`L^{}`$. In this sense, this Bochner-Kähler structure on $`C(p_D,\mu )^{}\times ^m`$ is universal for Bochner-Kähler metrics associated to this reduced momentum cell. Under this immersion, which is a local diffeomorphism, the vector field $`Z_{k+1}^{}`$ is carried into $`/\theta ^k`$.
Suppose now that $`r_\alpha `$ is a simple root of $`p_D(t)`$ such that $`l_\alpha =0`$ defines a face of $`C(p_D,\mu )`$. Then the vector field $`W_\alpha `$ is defined (Proposition 10) and has the expansion
$$W_\alpha =\frac{1}{p_D^{}(r_\alpha )}\left(r_{\alpha }^{}{}_{}{}^{m1}Z_2^{}r_{\alpha }^{}{}_{}{}^{m2}Z_3^{}+\mathrm{}+(1)^{m1}Z_{m+1}^{}\right).$$
It follows that, under the canonical immersion, $`W_\alpha `$ is carried over to the vector field
$$\mathrm{\Theta }_\alpha =\frac{1}{p_D^{}(r_\alpha )}\left(r_{\alpha }^{}{}_{}{}^{m1}\frac{}{\theta ^1}r_{\alpha }^{}{}_{}{}^{m2}\frac{}{\theta ^2}+\mathrm{}+(1)^{m1}\frac{}{\theta ^m}\right).$$
Suppose now that $`M`$ contains points that satisfy $`w_\alpha =0`$, i.e., the image of the reduced momentum mapping contains points that lie on the face $`l_\alpha =0`$. Then by Proposition 10, near such points the flow of the vector field $`W_\alpha `$ is periodic with period $`\pi `$. This suggests considering the vector
$$\tau _\alpha =\frac{\pi }{p_D^{}(r_\alpha )}((r_\alpha )^{m1},(r_\alpha )^{m2},\mathrm{},1)^m.$$
The above Bochner-Kähler structure is well-defined on $`C(p_D,\mu )^{}\times \left(^m/(\tau _\alpha )\right)`$. It is not hard to see that there exists a simply-connected complex $`m`$-manifold $`X_\alpha `$ endowed with a Bochner-Kähler structure and a totally geodesic hypersurface $`Y_\alpha X_\alpha `$ so that, first, the Bochner-Kähler structure on $`X_\alpha Y_\alpha `$ is isomorphic to the above Bochner-Kähler structure on $`C(p_D,\mu )^{}\times \left(^m/(\tau _\alpha )\right)`$ and, second, the image of the momentum mapping on $`X_\alpha `$ is equal to $`C(p_D,\mu )^{}`$ union the interior of the face $`l_\alpha =0`$.
The argument for this ‘face-wise’ extension is based on Theorem 7, which shows that, for any point $`vC(p_D,\mu )`$, there must exist *some* Bochner-Kähler metric in the given analytically connected equivalence class whose reduced momentum mapping assumes the value $`v`$. Taking $`v`$ to lie in the interior of the face $`l_\alpha =0`$ and applying uniqueness, one sees that the extension must exist locally. A simple patching argument then allows one to produce the extension $`X_\alpha `$. Details are left to the reader, but see the next section, where the extension is computed explicitly in a couple of cases of interest.
###### Remark 13 (Guillemin’s completion).
The reader should also compare Guillemin’s description of a Kähler metric constructed from the data of a polytope, since the issue of completion across the facets is much the same. However, one big difference in the present case is that the polytopes involved here are not necessarily closed. Another is that they do not generally satisfy the rationality requirements for the global existence theorems that Guillemin is able to cite. Instead, in the present case the singular loci corresponding to the faces are ‘filled in’ with ‘patches’ whose existence stem from Theorem 7.
This construction generalizes: If $`A=\{\alpha _1,\mathrm{},\alpha _k\}`$ is a set of $`km`$ distinct simple roots of $`p_D(t)`$ for which each hyperplane $`l_{\alpha _j}=0`$ defines a face of $`C(p_D,\mu )`$, then the vectors $`v_{\alpha _j}`$ as defined above generate a discrete subgroup $`\mathrm{\Lambda }_A^m`$ and the Bochner-Kähler structure descends to $`C(p_D,\mu )^{}\times \left(^m/\mathrm{\Lambda }_A\right)`$. Moreover, there is a simply-connected complex $`m`$-manifold $`X_A`$ endowed with a Bochner-Kähler structure and a (reducible) hypersurface $`Y_AX_A`$ (whose irreducible components are totally geodesic) so that, first, $`X_AY_A`$ is isomorphic as a Bochner-Kähler manifold to $`C(p_D,\mu )^{}\times \left(^m/\mathrm{\Lambda }_A\right)`$, and, second, the image of the momentum mapping on $`X_A`$ is equal to $`C(p_D,\mu )^{}`$ union the faces $`l_{\alpha _j}=0`$ ($`1jk`$) and minus any faces omitted from this list.
In cases where $`C(p_D,\mu )`$ has at most $`m`$ simple faces (which includes all the cases except SubCase 4-$`i`$ for $`i<m`$), one can take $`A`$ to be the set of all the $`\alpha `$ for which $`l_\alpha =0`$ is a simple face of $`C(p_D,\mu )`$, and the result will be $`X_A`$, whose momentum image is the entire momentum cell. This is, in some sense, ‘the maximally complete’ Bochner-Kähler structure of dimension $`m`$ with the given momentum cell as momentum image. By Propositions 8 and 11, however, $`X_A`$ cannot be metrically complete unless the cell is bounded and has exactly $`m`$ simple faces. As has been seen, when $`m>0`$ this can only happen in SubCase 3-1$`b`$. As will be seen in the next section, the metric on $`X_A`$ does turn out to be complete in this SubCase.
In SubCase 4-$`i`$ for $`i<m`$, the hyperplane $`l_\alpha =0`$ is a simple face of $`C(p_D,\mu )`$ for all $`\alpha i`$. The relations
$$\underset{\alpha }{}W_\alpha =\underset{\alpha }{}r_\alpha W_\alpha =0,$$
then imply the relation
$$\underset{\alpha i}{}(r_ir_\alpha )W_\alpha =0$$
among the vector fields that would have to be periodic if one were going to be able to complete the metric across all $`m+1`$ of the faces simultaneously. This, in turn implies that
$$\underset{\alpha i}{}(r_ir_\alpha )\tau _\alpha =0$$
and this is the unique linear relation among the $`\{\tau _\alpha |\alpha i\}`$. These $`m+1`$ vectors generate a discrete lattice $`\mathrm{\Lambda }_i`$ in $`^m`$ if and only if the ratios $`(r_ir_\alpha )/(r_ir_\beta )`$ are rational for all $`\alpha ,\beta i`$.
However, this rationality condition is not sufficient for the Bochner-Kähler structure on $`C(p_D,\mu )^{}\times \left(^m/\mathrm{\Lambda }_i\right)`$ to complete to a smooth manifold. In fact, the necessary condition for this is that these ratios all be integers, which an elementary argument shows not to be possible. Instead, the rationality is sufficient to ensure that the metric extends to a smooth orbifold whose momentum mapping is onto $`C(p_D,\mu )`$. By Propositions 8 and 11, this metric is complete only in SubCase 4-0, and this returns to the orbifold discussion at the end of §4.3.3.
#### 4.4.5. Complete examples
I am now going to describe a formula that defines an $`n`$-parameter family of complete Bochner-Kähler metrics on $`^n`$. I will then state a theorem about these metrics and follow this with a discussion that motivates the derivation of this (rather unlikely looking) formula.
First, fix $`\rho =(\rho _1,\mathrm{},\rho _n)^n`$ where each $`\rho _i`$ is a non-negative real number. I claim that there is a real-analytic function $`s:^n[0,\mathrm{})`$ so that
$$s(z)\underset{i=1}{\overset{n}{}}e^{\rho _is(z)}|z_i|^2=0.$$
for all $`z^n`$. (This claim will be justified below.) Of course, the function $`s`$ depends on $`\rho `$, but I will not notate this. When $`\rho =0`$, one has $`s(z)=|z|^2`$, but otherwise this is not an elementary function. By construction, the function $`s`$ is invariant under the standard $`n`$-torus action on $`^n`$ defined by
$$(e^{i\theta _1},\mathrm{},e^{i\theta _n})(z_1,\mathrm{},z_n)=(e^{i\theta _1}z_1,\mathrm{},e^{i\theta _n}z_n).$$
Now set
$$S(z)=1+\underset{i=1}{\overset{n}{}}\rho _ie^{\rho _is(z)}|z_i|^21.$$
Define an Hermitian symmetric positive definite matrix $`G(z)=\left(G^{ij}(z)\right)`$ by
$$G^{ij}(z)=S(z)\left(\delta ^{ij}e^{\rho _is(z)}+\left(\rho _i+\rho _j+\rho _i\rho _js(z)\right)\overline{z_i}z_j\right).$$
Write $`G(z)^1=\left(G_{ij}(z)\right)>0`$ and define the Hermitian metric
$$g_\rho =G_{ij}(z)dz_id\overline{z_j}.$$
This metric is evidently invariant under the standard $`n`$-torus action defined above. Of course, $`g_0`$ is the standard flat metric on $`^n`$.
###### Theorem 10.
For every choice of $`\rho _i0`$, the metric $`g_\rho `$ is Bochner-Kähler and complete on $`^n`$. Conversely, every simply-connected, complete Bochner-Kähler manifold in dimension $`n`$ is either homogeneous or is isometric to $`(^n,g_\rho )`$ for some $`\rho `$ with $`\rho _i0`$. When the $`\rho _i`$ are distinct and positive, the only symmetries of this metric belong to the standard $`n`$-torus action on $`^n`$.
###### Proof.
The structure of the proof will be as follows: I will first assume that I have a complete Bochner-Kähler metric that is not locally homogeneous and consider the induced metric on a completed $``$-leaf. Knowing by earlier discussions that the only possibility for this is in SubCase 3-1$`b`$, I will use knowledge of the form of $`p_D`$ and the momentum cell to choose a particularly good basis for $`𝔷`$, one for which each of the vector fields of the basis has a periodic flow of period $`\pi `$. I will then attempt to find global holomorphic coordinates on the leaf that will carry these vector fields into the vector fields that generate the standard $`m`$-torus action defined above. Using these calculations as a guide and then comparing with the discussion at the end of §4.3.2 of the ‘resolution’ of boundary singularities of the cell metric in SubCase 3-1$`b`$, I will finally arrive at a candidate for the metric in these good coordinates and finish by showing how completeness and real-analyticity give the conclusions of the theorem.
Thus, suppose that $`M^n`$ is simply connected and has a complete Bochner-Kähler metric of cohomogeneity $`m>0`$. As has already been remarked, the momentum cell must fall into SubCase 3-1$`b`$, so that
$$p_D(t)=(tr_1)^2(tr_2)\mathrm{}(tr_{m+1}),(r_1>\mathrm{}>r_{m+1}).$$
For notational simplicity, I will use the index range $`2\alpha ,\beta ,\gamma m+1`$ and the abbreviation $`\rho _\alpha `$ ($`>0`$) for $`r_1r_\alpha `$ in what follows. Use (4.23) as the definition of the linear function $`l_\alpha :^m`$ for $`\alpha 2`$, as before, and set
$$a=1\underset{\alpha }{}\rho _\alpha l_\alpha \text{and}t=l_2+\mathrm{}+l_{m+1},$$
as was done in §4.3.2 during the analysis of the metric $`R_D`$ on this momentum cell $`C(p_D,\mu )^m`$, which is defined by the inequalities $`l_\alpha 0`$ and $`a>0`$. Note that this momentum cell contains only one vertex, namely the point $`k_1`$ where all of the $`l_\alpha `$ vanish. Also as before, let $`F_\alpha C(p_D,\mu )`$ be the face defined by $`l_\alpha =0`$ for $`2\alpha m+1`$.
As was done in the proof of Proposition 9, set $`w_\alpha =(h^{})^{}(l_\alpha )`$ and consider the $`m`$ vector fields $`W_\alpha 𝔷`$ defined by $`W_\alpha \text{ }\text{ }\mathrm{\Omega }=dw_\alpha `$. These vector fields are a basis of $`𝔷`$ and are linearly independent on $`M^{}`$.
Fix $`qM^{}`$, let $`LM`$ be the leaf of the foliation $``$ passing through $`q`$, and let $`EL`$ be the leaf of the foliation $``$ passing through $`q`$. On $`E^{}`$, the map $`h^{}:EC(p_D,\mu )`$ is a local isometry when $`C(p_D,\mu )`$ is given the metric $`R_D`$.
Recall the discussion and notation at the end of §4.3.2 about the metric $`R_\rho `$ on the ellipsoidal domain $`E_\rho ^m`$. The map $`s:E_\rho C(p_D,\mu )`$ is surjective and is isometric and submersive away from the hyperplanes $`p_\alpha =0`$. Letting $`pE_\rho `$ be the point with coordinates $`p_\alpha =\sqrt{w_\alpha (q)}>0`$, it follows that there is a real-analytic map $`\psi `$ from a neighborhood of $`pE_\rho `$ to a neighborhood of $`qE`$ satisfying $`\psi (p)=q`$ and $`h^{}\psi =s`$. This map is an isometry when $`E_\rho `$ is endowed with the metric $`R_\rho `$. Since $`E_\rho `$ is simply-connected and the metric $`R_\rho `$ is both real-analytic and complete, it follows that $`\psi `$ can be extended uniquely as a global isometry $`\psi :E_\rho E`$ and that it satisfies $`h^{}\psi =s`$. Since the rank of the differential of $`s:E_\rho C(p_D,\mu )`$ at any $`p=(p_\alpha )`$ is equal to the number of nonzero entries $`p_\alpha `$, it follows that the rank of the differential of $`h^{}`$ at any $`xE`$ is equal to $`m`$ minus the number of faces $`F_\alpha `$ on which $`h^{}(x)`$ lies. Since the rank of the differential of $`h^{}`$ at $`x`$ is equal to the dimension of the span of $`\{W_\alpha (x)|2\alpha m+1\}`$, it follows from this discussion that for any $`x`$, the nonzero elements in the list $`(W_2(x),\mathrm{},W_{m+1}(x))`$ are linearly independent. This observation will be useful below.
Each $`W_\alpha `$ vanishes on $`N_\alpha =(h^{})^1(F_\alpha )`$, which, since the flow of $`W_\alpha `$ is isometric, is a totally geodesic submanifold of $`M`$ and, moreover, is a complex submanifold of $`M`$ as well (since $`W_\alpha `$ is the real part of a holomorphic vector field on $`M`$). It also follows from the discussion in the previous paragraph that $`W_\alpha `$ is nonzero off of $`N_\alpha `$. Let $`L_\alpha =N_\alpha L`$. Then $`L_\alpha `$ is a totally geodesic complex hypersurface in $`L`$.
One of the goals of this argument is to show that there are holomorphic coordinates $`z_2,\mathrm{}z_{m+1}`$ on $`L`$ for which
$$W_\alpha iJW_\alpha =4iz_\alpha \frac{}{z_\alpha },2\alpha m+1$$
and to find the expression for the induced Kähler metric on $`L`$ in these coordinates. (The choice of the coefficient $`4`$ is dictated by the fact that the flow of the vector field $`W_\alpha `$ has period $`\pi `$. The proof of this periodicity follows the same lines as the corresponding proof in the SubCase 4-0 situation analyzed in Proposition 9. Since it only differs in details from that proof, the argument will be left to the reader.)
Accordingly, let $`\zeta ^2,\mathrm{},\zeta ^{m+1}`$ be the holomorphic 1-forms on $`L^{}`$ that satisfy
$$\zeta ^\alpha (W_\beta iJW_\beta )=4i\delta _\beta ^\alpha $$
These forms extend meromorphically to $`L`$, with simple poles along the hypersurfaces $`L_\alpha `$. Since the vector fields $`W_\alpha `$ Lie-commute, it follows that each $`\zeta ^\alpha `$ is closed. Note that, if the coordinates $`z_\alpha `$ are to exist as claimed, it will have to be true that $`\zeta ^\alpha =dz_\alpha /z_\alpha `$.
Writing $`\zeta ^\alpha =\xi ^\alpha +i\eta ^\alpha `$, the above equations are equivalent to
$$\xi ^\alpha (W_\beta )=\eta ^\alpha (JW_\beta )=0,\xi ^\alpha (JW_\beta )=\eta ^\alpha (W_\beta )=2\delta _\beta ^\alpha ,$$
Again, if the coordinates $`z_\alpha `$ exist as claimed, it will follow that $`2\xi ^\alpha =d\left(\mathrm{log}|z_\alpha |^2\right)`$.
Since the $`\zeta ^\alpha `$ are a basis for the holomorphic 1-forms on $`L^{}`$, the metric on $`L^{}`$ can be written in the form
$$ds^2=g_{\alpha \beta }\zeta ^\alpha \overline{\zeta ^\beta },$$
where $`g_{\alpha \beta }=\overline{g_{\beta \alpha }}`$ and where the pullback of $`\mathrm{\Omega }`$ to $`L^{}`$ is
$$\mathrm{{\rm Y}}=\frac{i}{2}g_{\alpha \beta }\zeta ^\alpha \overline{\zeta ^\beta }.$$
Now, the identity $`\zeta ^\alpha (W_\beta )=2i\delta _\beta ^\alpha `$ implies
$$dw_\alpha =W_\alpha \text{ }\text{ }\mathrm{{\rm Y}}=g_{\alpha \beta }\overline{\zeta ^\beta }+g_{\beta \alpha }\zeta ^\beta ,$$
or, equivalently,
$$dw_\alpha =(g_{\alpha \beta }+g_{\beta \alpha })\xi ^\beta i(g_{\alpha \beta }g_{\beta \alpha })\eta ^\beta .$$
Since $`W_\alpha `$ is tangent to the fibers of $`h^{}`$, and since $`w_\alpha =(h^{})^{}(l_\alpha )`$ is constant on those fibers, it follows that the coefficient of $`\eta ^\beta `$ in the above equation must vanish, i.e., $`g_{\alpha \beta }=g_{\beta \alpha }=\overline{g_{\alpha \beta }}`$. Thus,
$$dw_\alpha =2g_{\alpha \beta }\xi ^\beta .$$
Define $`g^{\alpha \beta }=g^{\beta \alpha }`$ so that $`g^{\alpha \beta }g_{\beta \gamma }=\delta _\gamma ^\alpha `$. Note, in particular, that $`\xi ^\alpha =\frac{1}{2}g^{\alpha \beta }dw_\beta `$. The metric on $`L^{}`$ can now be written in the form
$$\begin{array}{cc}\hfill ds^2& =g_{\alpha \beta }\zeta ^\alpha \overline{\zeta ^\beta }=g_{\alpha \beta }(\xi ^\alpha +i\eta ^\alpha )(\xi ^\beta i\eta ^\beta )\hfill \\ & =g_{\alpha \beta }\left(\xi ^\alpha \xi ^\beta +\eta ^\alpha \eta ^\beta \right)\hfill \\ & =\frac{1}{4}g^{\alpha \beta }dw_\alpha dw_\beta +g_{\alpha \beta }\eta ^\alpha \eta ^\beta .\hfill \end{array}$$
Since $`L^{}`$ is totally geodesic in $`M^{}`$, it follows from Theorem 8 that
$$\frac{1}{4}g^{\alpha \beta }dw_\alpha dw_\beta =(h^{})^{}(R_D)=(h^{})^{}\left(T^{\alpha \beta }\right)dw_\alpha dw_\beta ,$$
where $`T^{\alpha \beta }=T^{\beta \alpha }`$ for $`2\alpha ,\beta m+1`$ is defined on $`C(p_D,\mu )^{}`$ so that the formula
$$\underset{\alpha ,\beta =2}{\overset{m+1}{}}T^{\alpha \beta }dl_\alpha dl_\beta =R_D=\frac{tda^2}{4a^2}\frac{dadt}{2a}+\underset{\alpha =2}{\overset{m+1}{}}\frac{dl_{\alpha }^{}{}_{}{}^{2}}{4l_\alpha }$$
holds. Thus, $`g^{\alpha \beta }=4(h^{})^{}\left(T^{\alpha \beta }\right)`$.
Using the definitions of $`a`$ and $`t`$, it follows from the formula for $`R_D`$ that
$$4T^{\alpha \beta }=\frac{\delta _{\alpha \beta }}{l_\alpha }+\frac{(\rho _\alpha +\rho _\beta )}{a}+\frac{\rho _\alpha \rho _\beta t}{a^2},$$
so that, in particular,
$$\begin{array}{cc}\hfill 4T^{\alpha \beta }dl_\beta & =\frac{dl_\alpha }{l_\alpha }\frac{da}{a}+\rho _\alpha \left(\frac{dt}{a}\frac{tda}{a^2}\right)\hfill \\ & =d\left(\mathrm{log}\frac{l_\alpha }{a}+\rho _\alpha \frac{t}{a}\right).\hfill \end{array}$$
Meanwhile, if the coordinates $`z_\alpha `$ exist as claimed, this will imply that
$$\frac{d|z_\alpha |^2}{|z_\alpha |^2}=2\xi ^\alpha =g^{\alpha \beta }dw_\beta =(h^{})^{}\left(4T^{\alpha \beta }dl_\beta \right)=d\left((h^{})^{}\left(\mathrm{log}\frac{l_\alpha }{a}+\rho _\alpha \frac{t}{a}\right)\right),$$
i.e., there will exist constants $`c_\alpha >0`$ so that
$$|z_\alpha |^2=c_\alpha (h^{})^{}\left(\frac{l_\alpha }{a}e^{\rho _\alpha t/a}\right).$$
Since $`z_\alpha `$ would only be determined up to a multiplicative constant anyway by the above normalizations, one might as well take $`c_\alpha =1`$, which will normalize the $`z_\alpha `$ up to a phase.
These calculations suggest the following construction of a candidate for the leaf metric: Consider the system of equations
$$p_\alpha =\frac{y_\alpha }{b}e^{\rho _\alpha s/b},\text{where}b=1\underset{\beta =2}{\overset{m+1}{}}\rho _\beta y_\beta \text{and}s=\underset{\beta =2}{\overset{m+1}{}}y_\beta ,$$
relating the $`m`$ variables $`y_2,\mathrm{},y_{m+1}`$ to the variables $`p_2,\mathrm{},p_{m+1}`$. These formulae define a real-analytic mapping $`𝐩`$ from the open halfspace $`H_y^m`$ defined by $`b>0`$ in $`y`$-space into $`p`$-space.
I claim that the mapping $`𝐩`$ is a diffeomorphism from $`H_y`$ onto its image $`D_p^m`$ and that this open image contains the primary orthant $`O_p^m`$, i.e., the closed domain defined by $`p_\alpha 0`$. Consequently, $`𝐩`$ has an inverse $`𝐲:D_pH_y`$, i.e., the above equations can be solved real-analytically in the form
$$y_\alpha =𝐲_\alpha (p_2,\mathrm{},p_{m+1}).$$
Moreover, this inverse $`𝐲`$ takes $`O_p`$ diffeomorphically onto the partially open simplex $`\mathrm{\Sigma }H_y`$ defined by the inequalities $`y_\alpha 0`$ and $`_\beta \rho _\beta y_\beta <1`$.
To prove this claim, it is helpful to introduce the intermediate quantities
$$u_\alpha =\frac{y_\alpha }{(1_\beta \rho _\beta y_\beta )}.$$
These equations can be inverted in the form
$$y_\alpha =\frac{u_\alpha }{(1+_\beta \rho _\beta u_\beta )},$$
thus showing that they define a diffeomorphism from $`H_y`$ to the half-space $`H_u^m`$ defined by $`1+_\beta \rho _\beta u_\beta >0`$. Then the claim above amounts to showing that the mapping defined by
$$p_\alpha =u_\alpha e^{\rho _\alpha (u_2+\mathrm{}+u_{m+1})}$$
is invertible on the domain $`H_u`$ and that its image has the desired properties.
Consider the function $`f`$ on $`\times ^m`$ defined by
$$f(r,p)=r\underset{\alpha }{}e^{\rho _\alpha r}p_\alpha .$$
Now, $`f/r=1+_\alpha \rho _\alpha e^{\rho _\alpha r}p_\alpha `$ is positive on $`\times O_p`$, so that $`rf(r,\overline{p})`$ is a strictly increasing function on $``$ for every $`\overline{p}O_p`$. Note that $`f(0,\overline{p})<0`$ for $`\overline{p}O_p`$ and that $`lim_r\mathrm{}f(r,\overline{p})=\mathrm{}`$ for $`\overline{p}O_p`$ (since each of the $`\rho _\alpha `$ is positive). It then follows by the intermediate value theorem and the implicit function theorem that the equation $`f(r,p)=0`$ can be solved uniquely and real-analytically for $`r0`$ on an open set $`O_p^{}^m`$ containing the domain $`O_p`$.
Thus, let $`𝐫:O_p^{}`$ satisfy $`f(𝐫(p),p)=0`$ and set
$$u_\alpha =p_\alpha e^{\rho _\alpha 𝐫(p)}.$$
Then, by construction,
$$\underset{\alpha }{}u_\alpha =\underset{\alpha }{}p_\alpha e^{\rho _\alpha 𝐫(p)}=𝐫(p),$$
so that $`p_\alpha =u_\alpha e^{\rho _\alpha (u_2+\mathrm{}+u_{m+1})}`$. Moreover, when $`p`$ lies in $`O_p`$,
$$1+\underset{\beta }{}\rho _\alpha u_\alpha =1+\underset{\alpha }{}\rho _\alpha e^{\rho _\alpha 𝐫(p)}p_\alpha =\frac{f}{r}(𝐫(p),p)>0,$$
so the image point lies in $`H_u`$. The inversion of the original system is therefore
$$y_\alpha =\frac{p_\alpha e^{\rho _\alpha 𝐫(p)}}{1+_\beta \rho _\beta p_\beta e^{\rho _\beta 𝐫(p)}}=𝐲_\alpha (p_2,\mathrm{},p_{m+1}),$$
as was desired.
Now define a metric on $`^m`$ as follows: First, define functions on $`^m`$ by
$$G^{\alpha \beta }(𝐳)=\overline{z_\alpha }z_\beta \left(\frac{\delta _{\alpha \beta }}{𝐲_\alpha (|z_2|^2,\mathrm{},|z_{m+1}|^2)}+\frac{(\rho _\alpha +\rho _\beta )}{𝐚}+\frac{\rho _\alpha \rho _\beta 𝐭}{𝐚^2}\right),$$
where
$$𝐚=1\underset{\beta }{}\rho _\beta 𝐲_\beta (|z_2|^2,\mathrm{},|z_{m+1}|^2)\text{and}𝐭=\underset{\beta }{}𝐲_\beta (|z_2|^2,\mathrm{},|z_{m+1}|^2).$$
Note that $`𝐚`$ is strictly positive on $`^m`$. Moreover, since $`𝐲_\alpha =p_\alpha 𝐲_\alpha ^{}`$ where $`𝐲_\alpha ^{}`$ is strictly positive on $`^m`$, the formula for $`G^{\alpha \beta }=\overline{G^{\beta \alpha }}`$ defines a smooth function on $`^m`$ for all $`\alpha `$ and $`\beta `$. Moreover, the inequalities satisfied by the $`𝐲_\alpha `$ show that the Hermitian matrix $`G(𝐳)=\left(G^{\alpha \beta }(𝐳)\right)`$ is positive definite for all $`𝐳^m`$. Let $`G_{\alpha \beta }(𝐳)`$ denote the components of the inverse matrix and define
$$d𝐬^2=G_{\alpha \beta }(𝐳)dz_\alpha d\overline{z_\beta }.$$
This is an Hermitian metric on $`^m`$. It is visibly invariant under the torus action generated by the real parts of the holomorphic vector fields
$$Z_\alpha =4iz_\alpha \frac{}{z_\alpha }.$$
Setting $`\zeta ^\alpha =dz_\alpha /z_\alpha `$ yields $`\zeta ^\alpha (Z_\beta )=4i\delta _\beta ^\alpha `$. Tracing through the construction above, one sees that, away from the complex hyperplanes $`z_\alpha =0`$, the metric can be written in the form
$$d𝐬^2=f_{\alpha \beta }(𝐳)\zeta ^\alpha \overline{\zeta ^\beta },$$
where the inverse matrix $`f^{\alpha \beta }`$ has the form
$$f^{\alpha \beta }(𝐳)=\frac{\delta _{\alpha \beta }}{𝐲_\alpha (|z_2|^2,\mathrm{},|z_{m+1}|^2)}+\frac{(\rho _\alpha +\rho _\beta )}{𝐚}+\frac{\rho _\alpha \rho _\beta 𝐭}{𝐚^2},$$
with
$$𝐚=1\underset{\beta }{}\rho _\beta 𝐲_\beta (|z_2|^2,\mathrm{},|z_{m+1}|^2)\text{and}𝐭=\underset{\beta }{}𝐲_\beta (|z_2|^2,\mathrm{},|z_{m+1}|^2).$$
Thus, the map from $`^m`$ to $`C(p_D,\mu )`$ defined by $`l_\alpha =𝐲_\alpha (|z_2|^2,\mathrm{},|z_{m+1}|^2)`$ is a Riemannian submersion from the complement of the hyperplanes $`z_\alpha =0`$ onto $`C(p_D,\mu )`$.
It not difficult now to trace through the construction and see that the restriction of the metric $`d𝐬^2`$ to $`^m^m`$ is isometric to the metric $`R_\rho `$ on $`E_\rho `$ and is hence complete. It now follows without difficulty that $`d𝐬^2`$ is complete on $`^m`$. Note that this completeness is a consequence of the completeness of the metric $`R_\rho `$ on $`E_\rho `$ and so, by the discussion in §4.3.2, is valid for any $`\rho `$ all of whose entries are non-negative, and not just for those whose entries are positive and strictly increasing.
Moreover, looking back at the formula for the metric on $`L`$ and comparing terms, one sees that $`(^m,d𝐬^2)`$ is locally and (therefore, by completeness) globally holomorphically isometric to $`L`$ with its induced metric and that, under this isomorphism, the Kähler form corresponding to $`d𝐬^2`$ is simply
$$\mathrm{{\rm Y}}=\frac{i}{2}f_{\alpha \beta }(𝐳)\zeta ^\alpha \overline{\zeta ^\beta }=\frac{i}{2}G_{\alpha \beta }(𝐳)dz_\alpha \overline{dz_\beta }.$$
This provides the desired ‘explicit’ formula for the metric on the leaf $`L`$. (Bear in mind, though, that the function $`𝐫`$, which is the crucial ingredient in the recipe for the metric, was found by abstractly solving an implicit equation.)
As the reader can verify, the formula given above simplifies (after some trivial changes in notation) to the formula for $`g_\rho `$ given before the statement of Theorem 10.
The argument to this point shows that the metric $`g_\rho `$ defined before the statement of Theorem 10 is Bochner-Kähler for any $`\rho `$ all of whose entries are positive and distinct. However, the property of being Bochner-Kähler is preserved in the limit as any of the entries of $`\rho `$ vanish or become equal. (The curvature tensor is evidently analytic in $`\rho `$.) Consequently, the metric $`g_\rho `$ is Bochner-Kähler and complete for any $`\rho `$ with non-negative entries.
Finally, returning to the notation used before the statement of Theorem 10, suppose $`\rho =(\rho _1,\mathrm{},\rho ^n)`$ with
$$0\rho _1\rho _2\mathrm{}\rho _n.$$
Suppose first that these inequalities are strict and set
$$r_1=\frac{1}{(n+2)}\left(\rho _1+\mathrm{}+\rho _n\right)$$
and then $`r_\alpha =r_1\rho _{\alpha 1}`$ for $`2\alpha n+1`$, so that
$$2r_1+r_2+\mathrm{}+r_{n+1}=0$$
and $`r_1>r_2>\mathrm{}>r_{n+1}`$. From the construction in the first part of the proof, it follows that the metric $`g_\rho `$ satisfies
$$p_C(t)=p_D(t)=(tr_1)^2(tr_2)\mathrm{}(tr_{n+1})$$
and has cohomogeneity $`n`$. Since the metric is complete, by Propositions 8 and 9, the momentum cell must fall into the SubCase 3-1$`b`$. Moreover, since any strictly decreasing sequence $`(r_1,\mathrm{},r_{n+1})`$ satisfying $`2r_1+r_2+\mathrm{}+r_{n+1}=0`$ can be written in the above form for a unique $`\rho `$ with $`0<\rho _1<\mathrm{}<\rho _n`$, it follows that such parameters account for all of the $`n`$-dimensional reduced momentum cells in SubCase 3-1$`b`$. Thus, by Theorem 7, this formula gives all of the complete, simply-connected cohomogeneity $`n`$ Bochner-Kähler metrics of dimension $`n`$. Note that the origin is the unique fixed point of all of the vector fields in $`𝔷`$, and it follows from (4.1) that
$$p_{h(0)}(t)=(tr_2)\mathrm{}(tr_{n+1})\text{and}\left(t^2+h_1(0)t+V(0)\right)=(tr_1)^2.$$
From this, it follows from Proposition 4 that the Lie algebra of the symmetry group is $`𝔷`$. Since the group of symmetries is necessarily connected, it follows that the flows in $`𝔷`$ generate the entire symmetry group.
Now consider what happens as the $`\rho _i`$ vary. The metric $`g_\rho `$ depends analytically on $`\rho `$, so the formulae for $`p_{h(0)}(t)`$ and $`V(0)`$ must remain true for all values of $`\rho `$. The vector fields in $`𝔷`$ all vanish at $`0`$, so it follows that $`B_3=|T|^2`$ must vanish at $`0`$. Now, applying Theorem 6, one sees that, as $`\rho `$ varies through $`^n`$ satisfying $`0\rho _1\mathrm{}\rho _n`$, the values of the moduli pass through all of the values that can give rise to momentum cells in SubCase 3-1$`b`$, with the one exception of $`\rho =0`$, since, in this case, there is no such cell. Consequently, as $`\rho `$ varies in the primary orthant, the $`g_\rho `$ account for all of the possible analytically connected equivalence classes that can contain a complete metric. Since these metrics are all complete, it follows from Theorem 7, and Propositions 8 and 9, that these contain all of the inhomogeneous complete Bochner-Kähler metrics on simply connected manifolds. ∎
###### Remark 14 (Existence).
Interestingly, the argument above justifies the original assumption that there exists a complete Bochner-Kähler metric that is not locally homogeneous by producing such examples at the end.
#### 4.4.6. Weighted projective spaces
A construction similar to that in the SubCase 3-1$`b`$ can be used to express the leaf metric in complex coordinates in SubCase 4-0. Since the details are similar to those in the proof of Theorem 10, I will be brief.
Consider a momentum cell $`C(p_D,\mu )`$ in SubCase 4-0, with
$$p_D(t)=(tr_0)(tr_1)\mathrm{}(tr_{m+1}),(r_0>\mathrm{}>r_{m+1}).$$
The cell $`C(p_D,\mu )`$ is defined by the inequalities $`l_\alpha 0`$ for $`1\alpha m+1`$.
The first task is to produce holomorphic coordinates on the completion $`X_A`$ when $`A=\{2,\mathrm{},m+1\}`$. In fact, the argument to follow will show that $`X_A`$ is biholomorphic to $`^m`$. By Proposition 11, it suffices consider the case $`n=m`$, for one can always reduce to this case by simultaneously translating all of the $`r_\alpha `$ until $`r_0+\mathrm{}+r_{m+1}=0`$. So assume that this has been done.
For simplicity, use the abbreviation $`\rho _\alpha `$ ($`>0`$) for $`r_0r_\alpha `$ when $`\alpha 1`$. Use (4.23) as the definition of the linear function $`l_\alpha :^m`$ as before, and note that the relations (4.24) can be written as
$$\rho _1l_1=1\underset{\alpha >1}{}\rho _\alpha l_\alpha \text{and}\rho _1l_0=1\underset{\alpha >1}{}(\rho _\alpha \rho _1)l_\alpha .$$
The functions $`l_2,\mathrm{},l_{m+1}`$ are nonnegative coordinates on the cell, the function $`l_1`$ is nonnegative, and the function $`l_0`$ is strictly negative. Of course, the function $`l_1`$ is strictly positive on the cell minus the face $`l_1=0`$, and this will be important below. In what follows, whenever repeated indices invoke the summation convention, the range will be assumed to be $`2\alpha ,\beta m+1`$ unless stated otherwise.
As was done in the proof of Proposition 9, set $`w_\alpha =(h^{})^{}(l_\alpha )`$ and consider the vector fields $`W_\alpha 𝔷`$ defined by $`W_\alpha \text{ }\text{ }\mathrm{\Omega }=dw_\alpha `$. The vector fields $`W_2,\mathrm{},W_{m+1}`$ are a basis of $`𝔷`$ and are linearly independent on $`X_A^{}`$. The map $`h^{}:X_AC(p_D,\mu )`$ is a Riemannian submersion on $`X_A^{}`$ when $`C(p_D,\mu )`$ is given the metric $`R_D`$. The image $`h^{}(X_A)`$ is equal to $`C(p_D,\mu )`$ minus the face $`l_1=0`$.
Each $`W_\alpha `$ vanishes on $`N_\alpha =(h^{})^1(F_\alpha )`$, which, since the flow of $`W_\alpha `$ is isometric, is a totally geodesic complex hypersurface in $`X_A`$. Moreover, $`W_\alpha `$ is nonzero off of $`N_\alpha `$ for $`\alpha 2`$.
As before, I will show that there are holomorphic coordinates $`z_2,\mathrm{}z_{m+1}`$ on $`X_A`$ for which
$$W_\alpha iJW_\alpha =4iz_\alpha \frac{}{z_\alpha },2\alpha m+1,$$
and find the expression for the Bochner-Kähler metric on $`X_A`$ in these coordinates.
Accordingly, let $`\zeta ^2,\mathrm{},\zeta ^{m+1}`$ be the holomorphic 1-forms on $`X_A^{}`$ that satisfy
$$\zeta ^\alpha (W_\beta iJW_\beta )=4i\delta _\beta ^\alpha $$
These forms extend meromorphically to $`X_A`$, with simple poles along the hypersurfaces $`N_\alpha `$. Since the vector fields $`W_\alpha `$ Lie-commute, it follows that each $`\zeta ^\alpha `$ is closed. As before, if the coordinates $`z_\alpha `$ are to exist as claimed, it will have to be true that $`\zeta ^\alpha =dz_\alpha /z_\alpha `$.
Writing $`\zeta ^\alpha =\xi ^\alpha +i\eta ^\alpha `$, the above equations are equivalent to
$$\xi ^\alpha (W_\beta )=\eta ^\alpha (JW_\beta )=0,\xi ^\alpha (JW_\beta )=\eta ^\alpha (W_\beta )=2\delta _\beta ^\alpha ,$$
Again, if the coordinates $`z_\alpha `$ exist as claimed, it will follow that $`2\xi ^\alpha =d\left(\mathrm{log}|z_\alpha |^2\right)`$.
Since the $`\zeta ^\alpha `$ are a basis for the holomorphic 1-forms on $`X_A^{}`$, the metric on $`X_A^{}`$ can be written in the form
$$ds^2=g_{\alpha \beta }\zeta ^\alpha \overline{\zeta ^\beta },$$
where $`g_{\alpha \beta }=\overline{g_{\beta \alpha }}`$ and where the pullback of $`\mathrm{\Omega }`$ to $`X_A^{}`$ is
$$\mathrm{{\rm Y}}=\frac{i}{2}g_{\alpha \beta }\zeta ^\alpha \overline{\zeta ^\beta }.$$
Now, the identity $`\zeta ^\alpha (W_\beta )=2i\delta _\beta ^\alpha `$ implies
$$dw_\alpha =W_\alpha \text{ }\text{ }\mathrm{{\rm Y}}=g_{\alpha \beta }\overline{\zeta ^\beta }+g_{\beta \alpha }\zeta ^\beta ,$$
or, equivalently,
$$dw_\alpha =(g_{\alpha \beta }+g_{\beta \alpha })\xi ^\beta i(g_{\alpha \beta }g_{\beta \alpha })\eta ^\beta .$$
Since $`W_\alpha `$ is tangent to the fibers of $`h^{}`$, and since $`w_\alpha =(h^{})^{}(l_\alpha )`$ is constant on those fibers, it follows that the coefficient of $`\eta ^\beta `$ in the above equation must vanish, i.e., $`g_{\alpha \beta }=g_{\beta \alpha }=\overline{g_{\alpha \beta }}`$. Thus,
$$dw_\alpha =2g_{\alpha \beta }\xi ^\beta .$$
Define $`g^{\alpha \beta }=g^{\beta \alpha }`$ so that $`g^{\alpha \beta }g_{\beta \gamma }=\delta _\gamma ^\alpha `$. Note, in particular, that $`\xi ^\alpha =\frac{1}{2}g^{\alpha \beta }dw_\beta `$. The metric on $`X_A^{}`$ can now be written in the form
$$\begin{array}{cc}\hfill ds^2& =g_{\alpha \beta }\zeta ^\alpha \overline{\zeta ^\beta }=g_{\alpha \beta }(\xi ^\alpha +i\eta ^\alpha )(\xi ^\beta i\eta ^\beta )\hfill \\ & =g_{\alpha \beta }\left(\xi ^\alpha \xi ^\beta +\eta ^\alpha \eta ^\beta \right)\hfill \\ & =\frac{1}{4}g^{\alpha \beta }dw_\alpha dw_\beta +g_{\alpha \beta }\eta ^\alpha \eta ^\beta .\hfill \end{array}$$
Since $`h^{}`$ is a Riemannian submersion on $`X_A^{}`$, it follows that
$$\frac{1}{4}g^{\alpha \beta }dw_\alpha dw_\beta =(h^{})^{}(R_D)=(h^{})^{}\left(T^{\alpha \beta }\right)dw_\alpha dw_\beta ,$$
where $`T^{\alpha \beta }=T^{\beta \alpha }`$ for $`2\alpha ,\beta m+1`$ is defined on $`C(p_D,\mu )^{}`$ so that the formula
$$\underset{\alpha ,\beta =2}{\overset{m+1}{}}T^{\alpha \beta }dl_\alpha dl_\beta =R_D=\underset{\alpha =0}{\overset{m+1}{}}\frac{dl_{\alpha }^{}{}_{}{}^{2}}{4l_\alpha }$$
holds. Thus, $`g^{\alpha \beta }=4(h^{})^{}\left(T^{\alpha \beta }\right)`$.
Using the relations above that express $`l_0`$ and $`l_1`$ in terms of $`l_2,\mathrm{},l_{m+1}`$, it follows from the formula for $`R_D`$ that
$$4T^{\alpha \beta }=\frac{\delta _{\alpha \beta }}{l_\alpha }+\frac{(\rho _\alpha \rho _1)(\rho _\beta \rho _1)}{\rho _{1}^{}{}_{}{}^{2}l_0}+\frac{\rho _\alpha \rho _\beta }{\rho _{1}^{}{}_{}{}^{2}l_1},$$
so that, in particular,
$$\begin{array}{cc}\hfill 4T^{\alpha \beta }dl_\beta & =\frac{dl_\alpha }{l_\alpha }+\frac{\rho _\alpha \rho _1}{\rho _1}\frac{dl_0}{l_0}\frac{\rho _\alpha }{\rho _1}\frac{dl_1}{l_1}\hfill \\ & =d\left(\mathrm{log}\left(l_\alpha (l_0)^{(\rho _\alpha \rho _1)/\rho _1}(l_1)^{\rho _\alpha /\rho _1}\right)\right).\hfill \end{array}$$
Meanwhile, if the coordinates $`z_\alpha `$ exist as claimed, this will imply that
$$\frac{d|z_\alpha |^2}{|z_\alpha |^2}=2\xi ^\alpha =g^{\alpha \beta }dw_\beta =(h^{})^{}\left(4T^{\alpha \beta }dl_\beta \right)=d\left((h^{})^{}\left(\mathrm{log}\frac{l_\alpha (l_0)^{(\rho _\alpha \rho _1)/\rho _1}}{(l_1)^{\rho _\alpha /\rho _1}}\right)\right),$$
i.e., there will exist constants $`c_\alpha >0`$ so that
$$|z_\alpha |^2=c_\alpha (h^{})^{}\left(\frac{l_\alpha (l_0)^{(\rho _\alpha \rho _1)/\rho _1}}{(l_1)^{\rho _\alpha /\rho _1}}\right).$$
Since $`z_\alpha `$ would only be determined up to a multiplicative constant anyway by the above normalizations, one might as well take $`c_\alpha =1`$, which will normalize the $`z_\alpha `$ up to a phase. Writing $`x_\alpha =l_\alpha /l_00`$ for $`\alpha >0`$, this equation can be written more simply as
$$|z_\alpha |^2=(h^{})^{}\left(\frac{x_\alpha }{(x_1)^{\rho _\alpha /\rho _1}}\right),\alpha =2,3,\mathrm{},m+1,$$
where the $`x_\alpha 0`$ satisfy the relation $`x_1+\mathrm{}+x_{m+1}=1`$.
Consider the equation
$$s+\underset{\alpha =2}{\overset{m+1}{}}|z_\alpha |^2s^{\rho _\alpha /\rho _1}=1$$
on $`\times ^m`$. An analysis similar to the one performed in the proof of Theorem 10 shows that when $`\rho _\alpha /\rho _10`$ for $`\alpha 2`$ there is a unique real-analytic function $`𝐬:^m(0,\mathrm{})`$ that satisfies
$$𝐬(z)+\underset{\alpha =2}{\overset{m+1}{}}|z_\alpha |^2\left(𝐬(z)\right)^{\rho _\alpha /\rho _1}=1$$
for all $`z^m`$. Note that the function $`𝐬`$ is invariant under the standard $`m`$-torus action on $`^m`$ and is algebraic if and only if all of the ratios $`\rho _\alpha /\rho _1`$ are rational. Using the function $`𝐬`$, the equations above can be solved in the form $`(h^{})^{}(x_1)=𝐬(z)`$ and
$$(h^{})^{}(x_\alpha )=|z_\alpha |^2\left(𝐬(z)\right)^{\rho _\alpha /\rho _1},(\alpha >1),$$
whence, by algebra, follows the formula
$$w_\alpha =(h^{})^{}(l_\alpha )=\frac{|z_\alpha |^2\left(𝐬(z)\right)^{\rho _\alpha /\rho _1}}{\rho _1+_{\beta =2}^{m+1}(\rho _\beta \rho _1)|z_\beta |^2\left(𝐬(z)\right)^{\rho _\beta /\rho _1}},(2\alpha m+1).$$
This motivates defining a metric on $`^m`$ as follows: Set
$$𝐰_\alpha (z)=\frac{|z_\alpha |^2\left(𝐬(z)\right)^{\rho _\alpha /\rho _1}}{\rho _1+_{\beta =2}^{m+1}(\rho _\beta \rho _1)|z_\beta |^2\left(𝐬(z)\right)^{\rho _\beta /\rho _1}}$$
for $`2\alpha m+1`$ and define functions $`𝐰_1`$ and $`𝐰_0`$ on $`^m`$ by
$$\rho _1𝐰_1(z)=1\underset{\alpha >1}{}\rho _\alpha 𝐰_\alpha (z)\text{and}\rho _1𝐰_0(z)=1\underset{\alpha >1}{}(\rho _\alpha \rho _1)𝐰_\alpha (z).$$
Then $`𝐰_1`$ and $`𝐰_0`$ are strictly positive on $`^m`$. For $`2\alpha ,\beta m+1`$, define functions on $`^m`$ by
$$G^{\alpha \beta }(z)=\overline{z_\alpha }z_\beta \left(\frac{\delta _{\alpha \beta }}{𝐰_\alpha (z)}+\frac{(\rho _\alpha \rho _1)(\rho _\beta \rho _1)}{\rho _{1}^{}{}_{}{}^{2}𝐰_0(z)}+\frac{\rho _\alpha \rho _\beta }{\rho _{1}^{}{}_{}{}^{2}𝐰_1(z)}\right).$$
It is not difficult to show that the Hermitian matrix $`G(z)=\left(G^{\alpha \beta }(𝐳)\right)`$ is positive definite for all $`z^m`$. Let $`G_{\alpha \beta }(z)`$ denote the components of the inverse matrix and define
$$d𝐬^2=G_{\alpha \beta }(z)dz_\alpha d\overline{z_\beta }.$$
This is an Hermitian metric on $`^m`$. It is visibly invariant under the torus action generated by the real parts of the holomorphic vector fields
$$Z_\alpha =4iz_\alpha \frac{}{z_\alpha }.$$
Setting $`\zeta ^\alpha =dz_\alpha /z_\alpha `$ yields $`\zeta ^\alpha (Z_\beta )=4i\delta _\beta ^\alpha `$. Tracing through the construction above, one sees that, away from the complex hyperplanes $`z_\alpha =0`$, the metric can be written in the form
$$d𝐬^2=f_{\alpha \beta }(z)\zeta ^\alpha \overline{\zeta ^\beta },$$
where the inverse matrix $`f^{\alpha \beta }`$ has the form
$$f^{\alpha \beta }(z)=\frac{\delta _{\alpha \beta }}{𝐰_\alpha (z)}+\frac{(\rho _\alpha \rho _1)(\rho _\beta \rho _1)}{\rho _{1}^{}{}_{}{}^{2}𝐰_0(z)}+\frac{\rho _\alpha \rho _\beta }{\rho _{1}^{}{}_{}{}^{2}𝐰_1(z)}.$$
In particular, the map from $`^m`$ to $`C(p_D,\mu )`$ defined by $`l_\alpha =𝐰_\alpha (|z_2|^2,\mathrm{},|z_{m+1}|^2)`$ is a Riemannian submersion from the complement of the hyperplanes $`z_\alpha =0`$ onto $`C(p_D,\mu )^{}`$ endowed with the metric $`R_D`$.
It not difficult now to trace through the construction and see that the restriction of the metric $`d𝐬^2`$ to $`^m^m`$ is isometric to the metric $`R_D`$ on $`E`$ as defined in §4.3.2.
Moreover, looking back at the formulae for the metric on $`X_A^{}`$ and comparing terms, one sees that $`(^m,d𝐬^2)`$ must be globally holomorphically isometric to $`X_A`$ with its Bochner-Kähler metric and that, under this isomorphism, the Kähler form corresponding to $`d𝐬^2`$ is simply
$$\mathrm{{\rm Y}}=\frac{i}{2}f_{\alpha \beta }(𝐳)\zeta ^\alpha \overline{\zeta ^\beta }=\frac{i}{2}G_{\alpha \beta }(𝐳)dz_\alpha \overline{dz_\beta }.$$
This provides the desired explicit formula for the metric on the leaf $`X_A`$.
Although the derivation provided the inequalities $`0<\rho _1<\mathrm{}<\rho _{m+1}`$, the recipe given for the metric only needs the assumption $`\rho _\alpha >0`$ for $`1\alpha m+1`$. This means, for example, that the metric makes sense when all of the $`\rho _\alpha `$ are equal. In this case, the reader can verify that the metric $`d𝐬^2`$ on $`^m`$ is simply the Fubini-Study metric on $`^m`$ restricted to the complement of a hyperplane.
Suppose now that all of the ratios $`\rho _\alpha /\rho _1`$ are rational and let $`r>0`$ be such that $`\rho _\alpha =(m+2)rp_\alpha `$ where the numbers $`0<p_1<\mathrm{}<p_{m+1})`$ are integers with no common divisor. This uniquely defines $`r`$ and the integers $`p_\alpha `$. Moreover, the equations $`r_0r_\alpha =\rho _\alpha =rp_\alpha `$ and $`r_0+r_1+\mathrm{}+r_{m+1}=0`$ imply
$$r_\alpha =r\left(\underset{\beta =0}{\overset{m+1}{}}p_\beta (m+2)p_\alpha \right)$$
where, for notational symmetry, I have set $`p_0=0`$.
Recalling that $`\rho _1W_1+\rho _2W_2+\mathrm{}+\rho _{m+1}W_{m+1}=0`$, it follows that
$$p_1\left(W_1iJW_1\right)=4i\left(p_2z_2\frac{}{z_2}+\mathrm{}+p_{m+1}z_{m+1}\frac{}{z_{m+1}}\right).$$
Now, set $`[p]=[p_1,\mathrm{},p_{m+1}]`$ and consider the weighted projective space $`^{[p]}`$ one gets by taking the quotient of $`^{m+1}`$ minus the origin by the $`^{}`$-action
$$\lambda (Z_1,Z_2,\mathrm{},Z_{m+1})=(\lambda ^{p_1}Z_1,\lambda ^{p_2}Z_2,\mathrm{},\lambda ^{p_{m+1}}Z_{m+1}).$$
This is an orbifold and not a manifold except when $`p_1=\mathrm{}=p_{m+1}`$ (in which case, this is $`^m`$). Let $`[Z_1,\mathrm{},Z_{m+1}]^{[p]}`$ denote the orbit of $`(Z_1,\mathrm{},Z_{m+1})^m`$. Consider the holomorphic mapping $`\mathrm{\Phi }_1:^m^{[p]}`$ defined by
$$\mathrm{\Phi }_1(Z_2,\mathrm{},Z_{m+1})=[1,Z_2,\mathrm{},Z_{m+1}].$$
This mapping is a $`p_1`$-fold branched covering of its image and the above considerations show that the metric $`d𝐬^2`$ extends to be a smooth orbifold metric on $`^{[p]}`$. The end result is the following:
###### Theorem 11.
Every weighted projective space $`^{[p]}`$ supports a Bochner-Kähler metric.
###### Remark 15 (Uniqueness).
Of course, in the classical case of projective space, the Bochner-Kähler metric so constructed is a constant multiple of the Fubini-Study metric. By Corollary 5, this is the unique Bochner-Kähler metric on $`^m`$, up to a constant multiple. I suspect, though I have not checked all of the details, that this uniqueness holds for all of the weighted projective spaces.
###### Remark 16 (Reduction).
The reader will recall that one way of constructing the Fubini-Study metric is to apply reduction to the flat Kähler metric under the diagonal $`S^1`$-action. Given this, one might suspect that the Bochner-Kähler metric on $`^{[p]}`$ is got from the flat Kähler metric by applying reduction to the weighted $`S^1`$-action described above. However, calculation shows that, except in the classical case, the reduction metric is *not* Bochner-Kähler.
### 4.5. Reduction and the full metric
Theorem 9 provides a formula for the induced metric on the $``$-leaves of a Bochner-Kähler metric. In the case of maximal cohomogeneity, i.e., $`m=n`$, the regular set $`M^{}`$ constitutes a single $``$-leaf, so this formula determines the metric completely. In this section, I will indicate how one can reconstruct the full metric from the knowledge of the metric on the $``$-leaves. Thus, for the rest of this section, I will assume that $`M^n`$ is endowed with a Bochner-Kähler metric of cohomogeneity $`m`$ satisfying $`0<m<n`$, since otherwise, there is nothing to do.
Let $`p_C(t)`$ and $`p_D(t)`$ be the characteristic polynomials of the Bochner-Kähler structure. Write
$$p_{h^{\prime \prime }}(t)=(t\lambda _{m+1})\mathrm{}(t\lambda _n)$$
where, by Proposition 5, the roots $`\lambda _{m+1}\mathrm{}\lambda _n`$ are also roots of $`p_D(t)`$. Let $`\pi :P_2M^{}`$ be the $`G_\mathrm{\Lambda }`$-bundle as described in the proof of Proposition 5 and return to that notation, particularly the index ranges. Recall that the $`\lambda _i`$ are distinct and not equal to any of the $`\lambda _a`$, and that the $`T_i`$ are positive and real and satisfy
$$T_{i}^{}{}_{}{}^{2}=\frac{p_D(\lambda _i)}{_{ji}(\lambda _i\lambda _j)}.$$
Also, recall the relations
$$\varphi _{a\overline{ı}}=\frac{T_i\omega _a}{\lambda _i\lambda _a},$$
which followed from the structure equations applied to $`h_{a\overline{ı}}=0`$. The structure equations applied to $`h_{a\overline{b}}=\delta _{ab}\lambda _a`$ yield
$$(\lambda _a\lambda _b)\varphi _{a\overline{b}}=0,$$
so that $`\varphi _{a\overline{b}}=0`$ when $`\lambda _a\lambda _b`$. The structure equations applied to $`h_{i\overline{ȷ}}=0`$ for $`ij`$ yield the relations
$$\varphi _{i\overline{ȷ}}=\frac{T_i\overline{\omega _j}+T_j\omega _i}{\lambda _i\lambda _j},ij,$$
while the structure equations applied to $`h_{i\overline{ı}}=\lambda _i`$ yield
$$d\lambda _i=T_i(\omega _i+\overline{\omega _i}).$$
Meanwhile, the structure equations applied to $`T_i`$ yield
$$\begin{array}{cc}\hfill dT_i& =\varphi _{i\overline{ȷ}}T_j+(\lambda _{i}^{}{}_{}{}^{2}+h_1\lambda _i+V)\omega _i\hfill \\ & =\varphi _{i\overline{ı}}T_i+\underset{ji}{}\left(\frac{T_i\overline{\omega _j}+T_j\omega _i}{\lambda _i\lambda _j}\right)T_j+(\lambda _{i}^{}{}_{}{}^{2}+h_1\lambda _i+V)\omega _i\hfill \end{array}$$
which can be rearranged to give
$$\frac{dT_i}{T_i}=\varphi _{i\overline{ı}}+\underset{ji}{}\frac{T_j\overline{\omega _j}}{\lambda _i\lambda _j}+\left((\lambda _{i}^{}{}_{}{}^{2}+h_1\lambda _i+V)+\underset{ji}{}\frac{T_{j}^{}{}_{}{}^{2}}{\lambda _i\lambda _j}\right)\frac{\omega _i}{T_i}.$$
Now, the structure equations for $`\omega _a`$ are (summation over $`i`$ and $`b`$)
$$\begin{array}{cc}\hfill d\omega _a& =\varphi _{a\overline{ı}}\omega _i\varphi _{a\overline{b}}\omega _b=\frac{T_i\omega _a}{\lambda _a\lambda _i}\omega _i\varphi _{a\overline{b}}\omega _b\hfill \\ & =\left(\varphi _{a\overline{b}}+\delta _{a\overline{b}}\frac{T_i\omega _i}{\lambda _a\lambda _i}\right)\omega _b=\left(\phi _{a\overline{b}}+\frac{1}{2}\delta _{a\overline{b}}\frac{d\lambda _i}{\lambda _a\lambda _i}\right)\omega _b,\hfill \end{array}$$
where
$$\phi _{a\overline{b}}=\overline{\phi _{b\overline{a}}}=\varphi _{a\overline{b}}+\frac{1}{2}\delta _{a\overline{b}}\underset{i}{}\frac{T_i(\omega _i\overline{\omega _i})}{\lambda _a\lambda _i}.$$
Since $`p_h^{}(\lambda _a)=_i(\lambda _a\lambda _i)`$ and since $`\varphi _{a\overline{b}}=\phi _{a\overline{b}}=0`$ when $`\lambda _a\lambda _b`$, setting
$$\eta _a=|p_h^{}(\lambda _a)|^{1/2}\omega _a$$
yields $`d\eta _a=\phi _{a\overline{b}}\eta _b`$. This implies that, for each root $`r`$ of $`p_{h^{\prime \prime }}(t)`$, the quadratic form and 2-form
$$g_r=\underset{\{a:\lambda _a=r\}}{}\eta _a\overline{\eta _a}\text{and}\mathrm{\Omega }_r=\frac{ı}{2}\underset{\{a:\lambda _a=r\}}{}\eta _a\overline{\eta _a}$$
define a Kähler structure on the space of leaves of the system $`\{\eta _a=0\text{ }\lambda _a=r\}`$ in any open set in $`M^{}`$ where this leaf space is Hausdorff. If $`r`$ has multiplicity $`\nu >0`$, this leaf space has complex dimension $`\nu `$.
To compute the curvature of this leaf space, one needs to compute the 2-forms
$$\mathrm{\Phi }_{a\overline{b}}=d\phi _{a\overline{b}}+\phi _{a\overline{c}}\phi _{c\overline{b}},$$
so I now turn to this task. Since $`\phi _{a\overline{b}}=\varphi _{a\overline{b}}`$ when $`ab`$, the structure equations for $`ab`$ yield (summation on $`i`$ and $`c`$)
$$\begin{array}{cc}\hfill \mathrm{\Phi }_{a\overline{b}}& =d\varphi _{a\overline{b}}+\varphi _{a\overline{c}}\varphi _{c\overline{b}}=\varphi _{a\overline{i}}\varphi _{i\overline{b}}(\lambda _a+\lambda _b+h_1)\omega _a\overline{\omega _b}\hfill \\ & =\left[\frac{T_{i}^{}{}_{}{}^{2}}{(\lambda _a\lambda _i)(\lambda _b\lambda _i)}(\lambda _a+\lambda _b+h_1)\right]\omega _a\overline{\omega _b}\hfill \\ & =\left[\frac{p_D(\lambda _i)}{(\lambda _a\lambda _i)(\lambda _b\lambda _i)_{ji}(\lambda _i\lambda _j)}(\lambda _a+\lambda _b+h_1)\right]\omega _a\overline{\omega _b}\hfill \end{array}$$
Rather miraculously, when $`\lambda _a\lambda _b`$, the classical identities of §4.3.2 show that this expression is zero, as should have been expected. On the other hand, if $`\lambda _a=\lambda _b=r_i`$ (but still $`ab`$), the same classical identities show that this expression simplifies to
$$\mathrm{\Phi }_{a\overline{b}}=\frac{p_D^{}(r_i)}{p_h^{}(r_i)}\omega _a\overline{\omega _b}=\frac{p_D^{}(r_i)|p_h^{}(r_i)|}{p_h^{}(r_i)}\eta _a\overline{\eta _b}=(1)^{\mu _i}p_D^{}(r_i)\eta _a\overline{\eta _b}.$$
(Recall that $`(1)^{\mu _i}p_h^{}(r_i)>0`$ on $`M^{}`$.) It remains to compute the quantities $`\mathrm{\Phi }_{a\overline{a}}`$. This computation is greatly simplified by first observing that $`\mathrm{\Phi }_{a\overline{a}}`$ must be a sum of terms of the form $`\omega _b\overline{\omega _c}`$ where $`\lambda _a=\lambda _b=\lambda _c`$. Thus, in carrying out the expansion from the definitions, all other terms can be ignored. For simplicity, I will use $``$ to denote equality modulo the ideal generated by the 1-forms $`\omega _i`$ and $`\overline{\omega _i}`$ for $`1im`$ and the 1-forms $`\varphi _{a\overline{b}}`$ for $`m<a,b,n`$. Then, first of all (summation over $`j`$ and $`b`$),
$$d\omega _i=\varphi _{i\overline{j}}\omega _j\varphi _{i\overline{b}}\omega _b\frac{T_i\omega _b\overline{\omega _b}}{\lambda _b\lambda _i}.$$
Using this and the identities quoted above, the calculation of $`\mathrm{\Phi }_{a\overline{a}}`$ follows from the structure equations goes as (summation over $`j`$ and $`b`$)
$$\begin{array}{cc}\hfill \mathrm{\Phi }_{a\overline{a}}& =d\phi _{a\overline{a}}+\phi _{a\overline{b}}\phi _{b\overline{a}}d\left(\varphi _{a\overline{a}}+\frac{1}{2}\frac{T_j(\omega _j\overline{\omega _j})}{\lambda _a\lambda _j}\right)\hfill \\ & \varphi _{a\overline{ȷ}}\varphi _{j\overline{a}}(2\lambda _a+h_1)\omega _a\overline{\omega _a}(\lambda _a+\lambda _b+h_1)\omega _b\overline{\omega _b}+\frac{T_{j}^{}{}_{}{}^{2}\omega _b\overline{\omega _b}}{(\lambda _a\lambda _j)(\lambda _b\lambda _j)}\hfill \\ & =\frac{p_D^{}(r_i)}{p_h^{}(r_i)}\left(\omega _a\overline{\omega _a}+\underset{\{b:\lambda _a=r_i\}}{}\omega _b\overline{\omega _b}\right)\hfill \\ & =(1)^{\mu _i}p_D^{}(r_i)\left(\eta _a\overline{\eta _a}+\underset{\{b:\lambda _a=r_i\}}{}\eta _b\overline{\eta _b}\right).\hfill \end{array}$$
These formulae imply that the Kähler structure defined by $`g_{r_i}`$ and $`\mathrm{\Omega }_{r_i}`$ actually has constant holomorphic sectional curvature equal to $`(1)^{\mu _i}\mathrm{\hspace{0.17em}4}p_D^{}(r_i)`$.
#### 4.5.1. Reduction
Since $`h^{}:MC(p_D,\mu )`$ is the momentum mapping of the infinitesimal torus action defined by the basis $`Z_1^{},\mathrm{},Z_m^{}`$ of $`𝔷`$, it is natural to consider the effect of applying symplectic reduction. Since the torus action is not assumed to be globally defined (because no completeness assumptions have been made about the metric), this can only be done locally.
For simplicity, I will only consider reduction at a regular value of the reduced momentum mapping $`h^{}:MC(p_D,\mu )`$. Recall that $`h^{}:M^{}C(p_D,\mu )^{}`$ is a submersion, let $`xM^{}`$ be fixed and let $`\kappa =h^{}(x)C(p_D,\mu )^{}`$. The method of symplectic reduction then consists of the following: Consider the codimension $`m`$ submanifold $`(h^{})^1(\kappa )M^{}`$. This submanifold is foliated by $`m`$-dimensional leaves whose tangent spaces are spanned by the vector fields $`Z_2,\mathrm{},Z_{m+1}`$. Suppose that this foliation is simple, i.e., its leaf space $`M_\kappa `$ is Hausdorff. (This can always be arranged by restricting to a suitable open neighborhood of $`x`$.) Then the pullback of $`\mathrm{\Omega }`$ to $`(h^{})^1(\kappa )M^{}`$ is a closed 2-form that is the pullback to $`(h^{})^1(\kappa )`$ of a symplectic form $`\mathrm{\Omega }_\kappa `$ on $`M_\kappa `$. The symplectic manifold $`(M_\kappa ,\mathrm{\Omega }_\kappa )`$ is then called the *symplectic reduction* of $`(M,\mathrm{\Omega })`$ at $`\kappa `$.
###### Proposition 13.
Fix $`xM^{}`$ and let $`\kappa =h^{}(x)C(p_D,\mu )^{}`$. There is a unique metric $`g_\kappa `$ on $`M_\kappa `$ for which the leaf projection $`(h^{})^1(\kappa )M_\kappa `$ is a Riemannian submersion.
The data $`(M_\kappa ,g_\kappa ,\mathrm{\Omega }_\kappa )`$ defines a Kähler structure on $`M_\kappa `$ that is locally isomorphic to a product of complex space forms. Specifically, for each root $`r`$ of $`p_{h^{\prime \prime }}(t)`$ of multiplicity $`\nu `$, the local product contains a $`\nu `$-dimensional complex space form of constant holomorphic sectional curvature
$$c(r,\kappa )=\frac{4p_D^{}(r)}{p_{h(x)}(r)}$$
and these are all of the factors.
###### Proof.
Let $`P_2(\kappa )P_2`$ be the part of $`P_2`$ that lies over $`(h^{})^1(\kappa )M^{}`$. The structure equations on $`P_2(\kappa )`$ are the same as those on $`P_2`$ with the difference that, after restriction to $`P_2(\kappa )`$ the functions $`\lambda _i`$ and $`T_i`$ become constant and the 1-forms $`\omega _i`$ become purely imaginary. Note that the equations $`\omega _a=0`$ define the foliation by the torus leaves.
Now going back to the structure equations, just derived above, one sees that, on $`P_2(\kappa )`$, the equations
$$d\omega _a=\phi _{a\overline{b}}\omega _b$$
hold, where $`\phi =(\phi _{a\overline{b}})=\phi ^{}`$ is blocked according to the multiplicities in the descending string of eigenvalues
$$\lambda _{m+1}\lambda _{m+2}\mathrm{}\lambda _n.$$
It follows, of course, that quadratic form $`g_\kappa =\omega _a\overline{\omega _a}`$ is well-defined on the leaf space $`M_\kappa `$ and that this metric and the symplectically reduced $`2`$-form $`\mathrm{\Omega }_\kappa =\frac{ı}{2}\omega _a\overline{\omega _a}`$ define a Kähler structure on $`M_\kappa `$.
Finally, the computation of the curvature forms above shows that this Kähler structure is a product of the type described in the proposition. ∎
#### 4.5.2. The general metric
As the preceding formulae and Proposition 13 now make clear, a recipe for any Bochner-Kähler metric on its regular locus $`M^{}`$ can be constructed as a generalized warped product over a momentum cell, where the fibers are products of so-called Sasakian space forms, i.e., the canonical circle bundles over complex space forms of constant holomorphic sectional curvature. In other words, once the leaf metric has been found, as in Theorem 9, the full metric can be constructed by group theoretic means. This is to be expected, since, after all, the pseudo-group of local symmetries of a connected Bochner-Kähler metric acts transitively on the fibers of the momentum mapping.
The explicit formula does not appear to be of great interest. For brevity, I will not go into details.
## 5. Final remarks
In this last section, I will make some remarks about related geometries.
### 5.1. Pseudo-Kähler geometry
When a complex $`n`$-manifold $`M`$ is endowed with a pseudo-Kähler structure, i.e., a pseudo-Riemannian metric $`g`$ that is invariant under the complex structure and whose associated 2-form $`\mathrm{\Omega }`$ is closed, the structure group of the geometry is $`\mathrm{U}(p,q)`$ for some $`p,q>0`$ with $`p+q=n`$. Since this group is simply a different real form of the group $`\mathrm{U}(n)`$, one would expect a similar decomposition of the curvature tensor. Indeed, this is what happens, the curvature tensor again breaking into the sum of three irreducible tensors. For simplicity of terminology, I will still refer to these as the scalar curvature, the traceless Ricci curvature, and the Bochner curvature and will refer to pseudo-Kähler structures for which the Bochner curvature vanishes as Bochner-Kähler.
The differential analysis of §2.3 extends essentially without change to the pseudo-Kähler case; it is just a matter of changing a few signs. Theorems 1 through 4 generalize with essentially no change as well. However, past this point, the analysis becomes somewhat more complicated because the orbit structure of the action of $`\mathrm{U}(p,q)`$ on $`𝔲(p,q)^n`$ is considerably more complicated than before. One must deal with non-diagonalizable elements, nilpotent orbits, and a host of other problems. It seems unlikely that the simple description of the analytically connected equivalence classes found in the Kähler case can be carried through in the pseudo-Kähler case.
### 5.2. A split-form analog
There is another ‘real form’ of Kähler geometry that has an analog of the Bochner-Kähler condition.
A Kähler structure can be thought of as a symplectic manifold $`(M^{2n},\mathrm{\Omega })`$ endowed with an $`\mathrm{\Omega }`$-skew endomorphism $`J:TMTM`$ that satisfies $`J^2=\mathrm{I}`$ and a torsion-free connection $``$ with respect to which both $`\mathrm{\Omega }`$ and $`J`$ are parallel.
A different geometry results if one starts with a symplectic manifold as above and considers an $`\mathrm{\Omega }`$-skew endomorphism $`K:TMTM`$ that satisfies $`K^2=+\mathrm{I}`$ together with a torsion-free connection $``$ with respect to which both $`\mathrm{\Omega }`$ and $`K`$ are parallel. Some authors call the data $`(M,\mathrm{\Omega },K,)`$ a *hyperbolic Kähler structure*, though this terminology seems likely to invite confusion.
Since the null plane fields of $`K\pm \mathrm{I}`$ are necessarily $`\mathrm{\Omega }`$-Lagrangian plane fields and since the hypothesis that there be a torsion-free connection with respect to which they are parallel implies that these two plane fields are integrable, such a structure endows the symplectic manifold with a pair of transverse, $`\mathrm{\Omega }`$-Lagrangian foliations $`_\pm `$.
Conversely, any symplectic manifold $`(M^{2n},\mathrm{\Omega })`$ endowed with a pair of transverse, $`\mathrm{\Omega }`$-Lagrangian foliations $`_\pm `$ has an $`\mathrm{\Omega }`$-skew endomorphism $`K:TMTM`$ so that the tangent plane fields to the two foliations are the kernels of $`K\pm \mathrm{I}`$ and a unique torsion-free connection $``$ with respect to which both $`\mathrm{\Omega }`$ and $`K`$ are parallel. Thus, it makes sense to call such a structure a *bi-Lagrangian* structure, which I will do for the rest of this subsection.
Let $`_n`$ denote the space of *row* vectors of length $`n`$ whose entries are real numbers, so that the natural matrix multiplication $`_n\times ^n`$ is a non-degenerate pairing and $`_n`$ is thus identified as the dual space of $`^n`$. Endow $`_n^n`$ with its natural induced symplectic structure. Let $`\mathrm{GL}(n,)`$ act on $`_n^n`$ on the left by
$$A(\xi ,x)=(\xi A^1,Ax).$$
This action preserves the symplectic structure on $`_n^n`$ and its bi-Lagrangian splitting into $`L_{}=_n0`$ and $`L_+=0^n`$. In fact, $`\mathrm{GL}(n,)`$ is the largest subgroup of $`\mathrm{Aut}(_n^n)`$ that preserves these structures.
Now let $`(M^{2n},\mathrm{\Omega },_\pm )`$ be a bi-Lagrangian manifold. Say that a coframe $`u:T_xM_n^n`$ is *adapted* if $`u`$ is a symplectic isomorphism, identifies $`T_x_{}`$ with $`_n0`$, and identifies $`T_x_+`$ with $`0^n`$. The bundle $`\pi :PM`$ of adapted coframes is then naturally a right $`\mathrm{GL}(n,)`$-bundle with the action defined by
$$(uA)(v)=A^1u(v).$$
The tautological 1-form of this $`\mathrm{GL}(n,)`$-structure can be written in the form $`(\eta ,\omega )`$ where $`\eta `$ takes values in $`_n`$ and $`\omega `$ takes values in $`^n`$. One then has the formula $`\pi ^{}(\mathrm{\Omega })=\eta \omega `$.
The existence of a torsion-free connection with respect to which $`\mathrm{\Omega }`$ and $`K`$ are parallel is equivalent to the existence of a $`𝔤𝔩(n,)`$-valued 1-form $`\varphi `$ on $`P`$ satisfying the equations
(5.1)
$$d\eta =\eta \varphi ,d\omega =\varphi \omega .$$
This is the *first structure equation* of Cartan. The 2-form $`\mathrm{\Phi }=d\varphi +\varphi \varphi `$ then satisfies the *first Bianchi identities*
(5.2)
$$\eta \mathrm{\Phi }=\mathrm{\Phi }\omega =0.$$
These identities imply that there is a function $`R:P\mathrm{Hom}(_n^n,𝔤𝔩(n,))`$ so that the *second structure equation* holds:
$$\mathrm{\Phi }=d\varphi +\varphi \varphi =R(\eta \omega )$$
and, moreover, that $`R`$ can be interpreted as taking values in a ‘curvature space’ $`𝒦`$ that is isomorphic as a $`\mathrm{GL}(n,)`$-module to $`S^2(_n)S^2(^n)`$. Applying the trace (or ‘contraction’) maps
$$S^2(_n)S^2(^n)_n^n,$$
then yields, as in the Kähler case, a decomposition of $`𝒦`$ into three irreducible, inequivalent $`\mathrm{GL}(n,)`$-modules and a corresponding decomposition of the curvature tensor of any bi-Lagrangian structure into three parts. For simplicity, I will refer to these three parts as the scalar curvature, the traceless Ricci curvature and the Bochner curvature. (In , the latter curvature is called the “$`HB`$-tensor”.)
When the Bochner curvature vanishes, the bi-Lagrangian structure will be said to be *Bochner-bi-Lagrangian*. This vanishing condition is equivalent to the existence of a function $`S:P𝔤𝔩(n,)^n_n`$ that satisfies
(5.3)
$$d\varphi =\varphi \varphi +S\eta \omega S\omega \eta \omega \eta S+\eta S\omega \mathrm{I}_n.$$
The reader will note the analogy with the second structure equation for Bochner-Kähler structures.
The same sort of analysis as in §2.3 shows that there exist functions $`F:P^n`$ and $`G:P_n`$ so that
(5.4)
$$dS=\varphi S+S\varphi +F\eta +\omega G+\frac{1}{2}(G\omega +\eta F)I_n;$$
that there exists a function $`Q:P`$ so that
(5.5)
$$dF=\varphi F+\left(Q\mathrm{I}_n+S^2\right)\omega ,dG=G\varphi +\eta \left(Q\mathrm{I}_n+S^2\right);$$
and that
(5.6)
$$dQ=GS\omega +\eta SF.$$
Moreover, the exterior derivatives of equations (5.15.6) are identities.
Thus, the system of structure equations (5.15.6) satisfies the conditions for Cartan’s Theorem A.1 to apply (see the appendix). In particular, the analog of Theorem 1 holds for Bochner-bi-Lagrangian structures and there is a finite-dimensional moduli space of germs of such structures. The analog of Theorem 3 will hold as well, in that there will be $`n+1`$ polynomials in the functions $`S`$, $`F`$, $`G`$, and $`Q`$ that are constant on each connected Bochner-bi-Lagrangian structure bundle and the rank of the mapping $`(S,F,G,Q):P𝔤𝔩(n,)^n_n`$ is never more than $`n`$, implying that the ‘group’ of local isometries of the structure always acts with local cohomogeneity at most $`n`$.
In principle, one could describe the analytically connected equivalence classes for this type of structure and examine completeness questions, and so on. This project is made much more complicated than its Kähler analog by the fact that the $`\mathrm{GL}(n,)`$-invariant polynomials on $`𝔤𝔩(n,)^n_n`$ do not separate the $`\mathrm{GL}(n,)`$-orbits. This is potentially interesting, since it means that one could possibly have continuous families of Bochner-bi-Lagrangian structures all with the same coarse moduli. Whether this really does happen is an interesting question.
### 5.3. Self-dual Kähler metrics
*This section was added after P. Gauduchon sent me the preprint . I thank the authors for bringing it to my attention.*
The reader will recall that, when $`n=2`$, the Bochner tensor is the same as the anti-self-dual part of the Weyl tensor. I.e., when $`n=2`$, Bochner-Kähler metrics are the same as self-dual Kähler metrics. The self-dual part of the Weyl curvature in this case is essentially the scalar curvature $`s`$. In particular, the squared norm of the Weyl curvature is the same as $`s^2`$, up to a universal constant factor.
From this point of view, some of the results in this article in the case of dimension 2 had already been obtained. For example, in \[9, Theorem 1\] (which also follows from earlier work by B.-Y. Chen ) asserts that there are no compact self-dual Kähler manifolds other than the locally symmetric ones. Of course, this is the $`n=2`$ case of Corollary 5 of the present article. Their proofs use non-trivial global results about complex surfaces, while the proof in the present article is essentially self-contained. It is also interesting to note that, in view of Theorem 11, their proofs must make essential use of the hypothesis that the domain of definition of the metric is a compact manifold, rather than just a compact orbifold.
After the initial version of this article was posted to the arXiv, I was contacted by P. Gauduchon, who explained that he and V. Apostolov had recently obtained a local classification of self-dual Hermitian-Einstein metrics and that this implied a local classification of self-dual Kähler metrics. In particular, they had also proved that such metrics always have local cohomogeneity at most 2. For more information about their version of the local classification, the reader should consult their preprint . In particular, their work provides an independent alternative to the classification derived in this article when $`n=2`$.
In fact, a remarkable relation between self-dual Kähler metrics and Einstein metrics follows from the work of Derdzinski and Apostolov and Gauduchon . The interested reader should consult for details, but I will summarize some of their results here as preparation for the remarks I want to make at the end of this subsection.
If $`g`$ is a self-dual Kähler metric on a complex 2-manifold $`M`$ with scalar curvature $`s`$ not identically zero, then $`g`$ is not conformally flat. Apostolov and Gauduchon show that on the open set $`M^{}`$ where $`s`$ is nonzero, the Hermitian metric $`g^{}=s^2g`$ is Einstein (as well as being self-dual). Of course, unless $`s`$ is constant (which only happens when $`g`$ is locally symmetric), $`g^{}`$ will not be Kähler.
Conversely, Apostolov and Gauduchon show that any self-dual Hermitian Einstein metric that is not conformally flat is of the form $`g^{}`$ for a unique self-dual Kähler metric $`g`$ with non-zero scalar curvature.
However, from the point of view in , completeness issues for either self-dual Hermitian Einstein metrics or self-dual Kähler metrics appear not to be easily resolvable. For example, they did not know<sup>18</sup><sup>18</sup>18P. Gauduchon, private communication. whether or not there were any complete examples of cohomogeneity 2. Using the description in this article, however, it is easy to see that there are many complete examples of self-dual Hermitian Einstein metrics with cohomogeneity 2.
Before discussing these examples, here are three general observations that will be useful: Let $`M`$ be a connected complex surface endowed with a Bochner-Kähler metric $`g`$ and characteristic polynomial $`p_C(t)=t^4+C_2t^2+C_3t+C_4`$ and momentum mapping $`h=(h_1,h_2):M^2`$. First, the scalar curvature of $`g`$ is $`s=24h_1`$. Second, the Einstein constant of $`g^{}`$ is $`6912C_3`$. Third, the squared norm of the self-dual part of the Weyl curvature of $`g^{}`$ is $`cs^6>0`$ for some universal constant $`c>0`$.
Now, consider the complete cohomogeneity 2 metrics on $`^{\mathrm{\hspace{0.17em}2}}`$ provided by Theorem 10, where the parameters $`\rho _1`$ and $`\rho _2`$ satisfy $`0<\rho _1<\rho _2`$. The characteristic polynomials are
$$p_C(t)=p_D(t)=(tr_1)^2(tr_2)(tr_3)$$
where
$$r_1=\frac{1}{4}(\rho _1+\rho _2),r_2=\frac{1}{4}(\rho _23\rho _1),r_3=\frac{1}{4}(\rho _13\rho _2).$$
The momentum cell $`C(p_D,\mu )`$ is the bounded cell of SubCase 3-1$`b`$ (see Figure 2). Since the momentum mapping $`h=h^{}:^{\mathrm{\hspace{0.17em}2}}C(p_D,\mu )`$ is surjective and since the eigenvalues of $`H`$ satisfy $`r_2\lambda _1<r_1`$ and $`r_3\lambda _2r_2`$, it follows that $`h_1=\mathrm{tr}H=\lambda _1+\lambda _2`$ varies between an infimum of $`r_2+r_3`$ (achieved only at $`0^{\mathrm{\hspace{0.17em}2}}`$) and a supremum of $`r_1+r_2`$ (not achieved). Thus, since $`s=24h_1`$, the scalar curvature satisfies the bounds
$$12(\rho _2\rho _1)<s12(\rho _2+\rho _1)=s(0).$$
Moreover, since $`C(p_D,\mu )`$ has only one vertex and neither of its two faces is vertical, it follows that $`dh_1`$ vanishes only at $`0`$. Consequently, the equation $`s=0`$ defines a smooth hypersurface $`S^{\mathrm{\hspace{0.17em}2}}`$. This hypersurface is unbounded because the $`u_2`$-axis (i.e., $`u_1=0`$) cuts through the omitted face of $`C(p_D,\mu )`$.
Since $`s`$ is bounded, it follows that $`g_\rho ^{}=s^2g_\rho `$ is complete on each of the two domains $`D_+`$ (where $`s>0`$) and $`D_{}`$ (where $`s<0`$). The domain $`D_+`$ is contractible, while $`D_{}`$ has the homotopy type of a circle. Thus, this one example of a self-dual Kähler metric gives rise to two *distinct* complete, self-dual Hermitian Einstein manifolds.
As another interesting example, consider the Bochner-Kähler metric of SubCase 4-1, where $`r_0>r_1>r_2>r_3`$ are chosen so that $`r_0+r_3<0`$. Choosing the ‘completion’ $`X_2^{\mathrm{\hspace{0.17em}2}}`$ obtained by omitting the face $`l_2=0`$, one sees that the domain $`D_+^{\mathrm{\hspace{0.17em}2}}`$ defined by $`s>0`$ is bounded, with boundary a smooth compact hypersurface $`S^{\mathrm{\hspace{0.17em}2}}`$. Again, the corresponding self-dual Hermitian-Einstein metric $`g^{}`$ is complete on $`D_+`$. This case is interesting because $`(D_+,g^{})`$ exists even when the ratios of the roots $`r_i`$ are not rational, so that any attempt to ‘complete’ the corresponding self-dual Kähler metric $`(D_+,g)`$ to a maximal domain leads inevitably to worse than orbifold singularities, i.e., to a non-Hausdorff complex space.
By considering Case 1, one can construct an example of a self-dual Hermitian-Einstein manifold $`(M,g^{})`$ that is maximally extended and the corresponding self-dual Kähler metrics $`(M,g)`$ is maximally extended, but such that neither $`g`$ nor $`g^{}`$ is complete. Neither can be extended because the scalar curvature of $`g`$ is proper on $`M`$ and tends to $`\mathrm{}`$ while the squared norm of the Weyl curvature of $`g^{}`$ is proper on $`M`$ and tends to $`+\mathrm{}`$.
What is perhaps more interesting are the Case 4-0 examples, which include the weighted projective planes $`^{[p_1,p_2,p_3]}`$ where $`0<p_1<p_2<p_3`$ are integers with greatest common divisor equal to 1. For the Bochner-Kähler metric $`g`$ on this orbifold, the scalar curvature is everywhere positive as long as $`p_3<p_1+p_2`$ and the corresponding Hermitian Einstein metric has positive Einstein constant. When $`p_3=p_1+p_2`$, the scalar curvature is positive except at one point (a singular orbifold point) and the corresponding Hermitian Einstein metric has vanishing Einstein constant and is complete on the (orbifold) complement of this point. Finally, when $`p_3>p_1+p_2`$, the scalar curvature vanishes along a hypersurface $`S^{[p_1,p_2,p_3]}`$. The complement consists of two open sets $`_\pm ^{[p_1,p_2,p_3]}`$ (labeled according to the sign of $`s`$), each endowed with a complete Hermitian Einstein metric with negative Einstein constant. One of these two pieces, $`_{}^{[p_1,p_2,p_3]}`$, can be ‘unfolded’ to become a smooth, complete, Hermitian Einstein manifold that is biholomorphic to a bounded domain in $`^{\mathrm{\hspace{0.17em}2}}`$, while the other, $`_+^{[p_1,p_2,p_3]}`$, has unremovable orbifold singularities.
## Appendix A Cartan’s Generalization of Lie’s Third Theorem
This appendix is an exposition of the passage \[6, Chapter II, §§17–29\] from Cartan’s work on a generalization of Lie’s Third Fundamental Theorem to the ‘intransitive case’ together with a few comments of an elementary nature designed to extend the applicability of Cartan’s results to the smooth category and to a ‘semi-global’ setting. (In , Cartan worked almost entirely in what would now probably be called the category of real-analytic germs.) These results have, in modern times, been incorporated into the theory of local Lie algebras, Lie algebroids, and Lie groupoids. For references and surveys of this modern work the reader might consult and . The point of view that I take in this appendix is decidedly not modern; instead I follow Cartan’s exposition and development. I do this since Cartan’s version of the result is more suited for the application in this article.
### A.1. Cartan’s Problem
One is given the following data:
1. a nonempty open set $`X^s`$ (with coordinates $`x=(x^a)`$ on $`^s`$),
2. an integer $`n1`$, and
3. functions $`F_i^a`$ and $`C_{jk}^i=C_{kj}^i`$ on $`X`$, for $`1i,j,kn`$ and $`1as`$.
The goal is to describe the solutions to the following ‘realization problem’: Find
1. a manifold $`N^n`$,
2. a coframing $`\eta =(\eta ^i)`$ of $`N`$, and
3. a mapping $`h=(h^a):NX^s`$
satisfying
(A.1)
$$d\eta ^i=\frac{1}{2}C_{jk}^i(h)\eta ^j\eta ^k,dh^a=F_i^a(h)\eta ^i.$$
###### Example 8 (Lie’s Third Fundamental Theorem).
Consider the simple case where the $`F_i^a`$ are all zero. Then the mapping $`h:NX`$ of any realization must be constant, say $`h=\overline{h}`$. A necessary condition on the constants $`\overline{C}_{jk}^i=C_{jk}^i(\overline{h})`$ can then be found by computing the exterior derivatives of the equations
$$d\eta ^i=\frac{1}{2}\overline{C}_{jk}^i\eta ^j\eta ^k.$$
These give $`0=\overline{C}_{jl}^id\eta ^j\eta ^l`$, which, in view of the above relations, can be rewritten (after an index substitution and skewsymmetrization) in the form
$$0=\frac{1}{2}\overline{C}_{pl}^i\overline{C}_{jk}^p\eta ^j\eta ^k\eta ^l=\frac{1}{6}\left(\overline{C}_{pj}^i\overline{C}_{kl}^p+\overline{C}_{pk}^i\overline{C}_{lj}^p+\overline{C}_{pl}^i\overline{C}_{jk}^p\right)\eta ^j\eta ^k\eta ^l.$$
Using the linear independence of the $`\eta ^i`$, one derives the *Jacobi conditions*
$$\overline{C}_{pj}^i\overline{C}_{kl}^p+\overline{C}_{pk}^i\overline{C}_{lj}^p+\overline{C}_{pl}^i\overline{C}_{jk}^p=0$$
as necessary conditions for the existence of a solution to the problem. In other words, any realization $`(N,\eta ,h)`$ must have $`h`$ be constant and take values in the locus $`X^{}X`$ defined by the equations
$$C_{pj}^iC_{kl}^p+C_{pk}^iC_{lj}^p+C_{pl}^iC_{jk}^p=0.$$
Conversely, Lie’s Third Fundamental Theorem asserts that the Jacobi conditions suffice to ensure the existence of a solution to the realization problem. I.e., if $`\overline{h}`$ lies in $`X^{}`$, then there exists a realization $`(N,\eta ,h)`$ with $`h\overline{h}`$. Moreover, any two realizations assuming the same value $`\overline{h}`$ are locally equivalent in the obvious sense.
### A.2. Differential conditions in the general case
Even when the $`F_i^a`$ are not assumed to be zero, exterior differentiation of the equations (A.1) of a realization $`(N,\eta ,h)`$ yields a set of necessary conditions on the map $`h:NX`$. Namely, it must satisfy
$$F_i^b(h)\frac{F_j^a}{x^b}(h)F_j^b(h)\frac{F_i^a}{x^b}(h)=C_{ij}^l(h)F_l^a(h)$$
(which is equivalent to $`d(dh^a)=0`$) and
$$\begin{array}{c}F_j^a(h)\frac{C_{kl}^i}{x^a}(h)+F_k^a(h)\frac{C_{lj}^i}{x^a}(h)+F_l^a(h)\frac{C_{jk}^i}{x^a}(h)\hfill \\ \hfill =\left(C_{mj}^i(h)C_{kl}^m(h)+C_{mk}^i(h)C_{lj}^m(h)+C_{ml}^i(h)C_{jk}^m(h)\right)\end{array}$$
(which is equivalent to $`d(d\eta ^i)=0`$). Unless these equations are identities, they place restrictions on the range of $`h`$.
### A.3. Cartan’s existence theorem
On the other hand, if the above equations *are* identities on the functions $`F_i^a`$ and $`C_{jk}^i`$, then one might hope to find realizations of (A.1) without placing any further restrictions on the range of $`h`$.
In , Cartan proved<sup>19</sup><sup>19</sup>19It would be more accurate to say that Cartan only outlined the proof of this result. However, the reader knowledgeable about Cartan-Kähler theory will have no trouble supplying the details. Also, while Cartan does not always explicitly state the assumption of real-analyticity, it is clear from context that he intended this assumption to be in force. just such a result in the real-analytic category.
###### Theorem A.1 (Cartan).
Suppose that $`X^s`$ is an open set and suppose that $`F_i^a`$ and $`C_{jk}^i=C_{kj}^i`$ for $`1as`$ and $`1i,j,kn`$ are real-analytic functions on $`X`$ that satisfy
(A.2)
$$F_i^b\frac{F_j^a}{x^b}F_j^b\frac{F_i^a}{x^b}=F_l^aC_{ij}^l.$$
and
(A.3)
$$F_j^a\frac{C_{kl}^i}{x^a}+F_k^a\frac{C_{lj}^i}{x^a}+F_l^a\frac{C_{jk}^i}{x^a}=\left(C_{mj}^iC_{kl}^m+C_{mk}^iC_{lj}^m+C_{ml}^iC_{jk}^m\right)$$
Then for every $`h_0X`$, there exists a real-analytic realization $`(N,\eta ,h)`$ satisfying the structure equations
$$d\eta ^i=\frac{1}{2}C_{jk}^i(h)\eta ^j\eta ^k,dh^a=F_i^a(h)\eta ^i.$$
and a $`p_0N`$ for which $`h(p_0)=h_0`$.
Moreover, this realization is locally unique in the following sense: Given any other real-analytic realization $`(\stackrel{~}{N},\stackrel{~}{\eta },\stackrel{~}{h})`$ satisfying the corresponding structure equations that contains a point $`\stackrel{~}{p}_0\stackrel{~}{N}`$ satisfying $`\stackrel{~}{h}(\stackrel{~}{p}_0)=h_0`$, there exists a $`p_0`$-neighborhood $`UN`$, a $`\stackrel{~}{p}_0`$-neighborhood $`\stackrel{~}{U}\stackrel{~}{N}`$, and a real-analytic diffeomorphism $`\varphi :\stackrel{~}{U}U`$ so that
$$\varphi (\stackrel{~}{p}_0)=p_0,\varphi ^{}(\eta )=\stackrel{~}{\eta },\text{and}\varphi ^{}(h)=\stackrel{~}{h}.$$
###### Remark 17 (A Paraphrase).
Informally, one can state Cartan’s result in the following way: There is a ‘solution’ of the structure equations (A.1) provided that the exterior derivatives of these equations are identities, i.e., $`d^2=0`$ is a formal consequence of (A.1). A solution is uniquely specified by choosing the values of the ‘invariants’ $`h=(h^a)`$ at one point in the domain of the solution.
#### A.3.1. Real-analyticity
The full theorem that Cartan proves in the cited passage is more general than Theorem A.1 and has to do with existence of so-called ‘infinite groups’ (nowadays called pseudo-groups) satisfying a given set of structure equations. However, Theorem A.1 is all that is needed in this article. Cartan’s proof is via the Cartan-Kähler Theorem, which is only valid in the real-analytic category. While the general theorem that Cartan proves really does need real-analyticity, the special case being discussed here as Theorem A.1 can be proved without recourse to the Cartan-Kähler Theorem. Indeed, it can be proved using only the Frobenius Theorem, the Poincaré Lemma, and Lie’s Third Fundamental Theorem (the classical one). See the work of Pradines for this development.
Thus, the above theorem (both existence and uniqueness) is actually valid in the smooth category. However, note that, in the case where $`F`$ and $`C`$ actually are real-analytic and satisfy (A.2) and (A.3), it follows from Cartan’s uniqueness result that any sufficiently differentiable realization of (A.1) is real-analytic in suitable coordinates.
###### Example 9 (Application).
In this article, Theorem A.1 will be applied to the equations (2.14). In that case, the functions $`F`$ and $`C`$ are polynomial (in fact, either linear or quadratic) in the linear coordinates on $`X=^s=i𝔲(n)^n`$ (here $`s=n^2+2n+1`$). Thus, the realizations are all real-analytic in this case.
### A.4. A coordinate-free reformulation
Cartan’s conditions can be recast into a somewhat more geometric form as follows: Suppose there are given functions $`F_i^a`$ and $`C_{jk}^i=C_{kj}^i`$ on a domain $`X^s`$. Define $`n`$ vector fields on $`X`$ by
$$F_i=F_i^b\frac{}{x^b}$$
for $`1in`$. Then (A.2) can be written in terms of the Lie bracket as
(A.4)
$$[F_i,F_j]=C_{ij}^kF_k.$$
Also, (A.3) can be written as
(A.5)
$$F_jC_{kl}^i+F_kC_{lj}^i+F_lC_{jk}^i=\left(C_{mj}^iC_{kl}^m+C_{mk}^iC_{lj}^m+C_{ml}^iC_{jk}^m\right).$$
When the vector fields $`F_i`$ are linearly independent, (A.5) follows directly from (A.4); it is simply the Jacobi identity for the Lie bracket. However, when the $`F_i`$ are everywhere linearly dependent (as is the case in the application in this article) the equations (A.5) are not consequences of (A.4).
In (A.4) and (A.5), no explicit reference is made to the coordinates $`x^a`$ on $`X`$. Thus, it makes sense to speak of systems $`(X,F,C)`$ satisfying (A.4) and (A.5) where $`X`$ is any smooth manifold, the $`F_i`$ are smooth vector fields on $`X`$ and the $`C_{ij}^k`$ are smooth functions on $`X`$. Such a system $`(X,F,C)`$ is an example of what has since become known as a *local Lie algebra* or a *Lie algebroid* . The notion of a realization $`(N,\eta ,h)`$ generalizes as well, with the formula for $`d\eta ^i`$ remaining the same but the formula $`dh=F_i\eta ^i`$ now being interpreted as a formula for $`dh:TNTX`$ in the obvious sense. This ‘coordinate free’ formulation of Cartan’s problem will not be needed in this article, so I will not discuss it any further here.
### A.5. The leaves of $`F`$
For each $`xX`$, let $`r(x)`$ be the dimension of the span of the vectors $`\{F_i(x)\}_{1in}`$. When $`r`$ is a constant function on $`X`$ and (A.4) holds, the Frobenius theorem asserts that the vector fields $`F_i`$ are tangent to a foliation of $`X`$ whose leaves have dimension $`r`$.
In most applications, however, the function $`r`$ is not constant on $`X`$. (Indeed, it is not constant for the system (2.14).) Nevertheless, there is a simple generalization of the Frobenius theorem that does hold whenever (A.4) holds.
Say that a smooth curve $`\xi :[a,b]X`$ is an *$`F`$-curve*<sup>20</sup><sup>20</sup>20When $`r`$ is not constant, this condition is *a priori* stronger than the mere condition that $`\xi ^{}(t)`$ lie in the span of $`\{F_i\left(\xi (t)\right)\}_{1in}`$ for all $`t[a,b]`$. if there exist smooth functions $`v^i`$ on $`[a,b]`$ for which $`\xi ^{}(t)=v^i(t)F_i\left(\xi (t)\right)`$ and say that $`x_1`$ and $`x_2`$ in $`X`$ are *$`F`$-equivalent* if they can be joined by a smooth $`F`$-curve.
The generalized Frobenius theorem says that, if the vector field system $`F`$ satisfies (A.4), then the $`F`$-equivalence class $`[x]_F`$ of $`xX`$ is a smooth, connected submanifold of $`X`$ of dimension $`r(x)`$. It is called the *$`F`$-leaf* through $`x`$. This generalized notion of a foliation is sometimes known as a *Stefan foliation* in the literature. For further discussion of this singular leaf structure, which is virtually the same as the sort of singular leaf structure that one encounters in the theory of Poisson manifolds, see .
### A.6. The rank of a realization
Suppose now that $`(X,F,C)`$ satisfies (A.2) and (A.3) (or, equivalently, (A.4) and (A.5)).
Then, for any realization $`(N,\eta ,h)`$ of the structure equations (A.1) with $`N`$ connected, the map $`h:NX`$ will have its image lie in a single $`F`$-leaf $`LX`$, whose dimension will be $`r(N)=r\left(h(p)\right)`$ for some (and hence any) $`pN`$. Moreover, the structure equations (A.1) imply that the map $`h:NL`$ has constant rank $`r(N)`$ and hence is a submersion onto its (open) image in $`L`$.
In , Cartan assumes these results without proof or remark. It is not clear whether he knew these facts (which, even in the real-analytic case, require argument it seems to me) or merely assumed that he was in some ‘generic’ case where they held. In any case, he does not make an issue of it.
The integer $`r(N)`$ will be referred to as the *rank* of the realization $`(N,\eta ,h)`$.
###### Example 10 (Application).
For the system (2.14), the dimension of an $`F`$-leaf can be as low as $`0`$ or as high as $`n(n+1)`$.
### A.7. The symmetry algebra of a leaf
Let $`(X,F,C)`$ satisfy (A.2) and (A.3). Let $`LX`$ be an $`F`$-leaf of rank $`r`$, and let $`(N,\eta ,h)`$ be a realization of the structure equations (A.1) whose image $`h(N)`$ is an open subset of $`L`$. Then by Theorem A.1, given any $`\overline{h}h(N)`$ and any two points $`p_1`$ and $`p_2`$ in the fiber $`h^1(\overline{h})N`$, there is a locally defined ‘symmetry’ of the realization that carries $`p_1`$ to $`p_2`$. This locally defined symmetry is unique in a neighborhood of $`p_1`$.
Cartan might have expressed this fact by saying something like ‘the group of symmetries of the system $`(\eta ,h)`$ acts simply transitively on the fibers of $`h`$’. In the modern literature, this sort of vagueness about the domain of the ‘group’ of ‘local symmetries’ of such data is usually avoided by giving a more precise statement using the language of (finite-dimensional) pseudo-groups. Rather than introduce this sort of terminology, I will give the corresponding infinitesimal formulation, which is simpler.
###### Theorem A.2.
If $`N`$ is connected and simply-connected and $`(N,\eta ,h)`$ is a realization of (A.1) of rank $`r`$, then the subset $`𝔥𝔛(N)`$ consisting of the vector fields on $`N`$ whose (local) flows on $`N`$ preserve $`\eta `$ and $`h`$ is a Lie algebra of dimension $`nr`$. Moreover, for any $`xN`$, the evaluation map $`e_x:𝔥T_xN`$ is a vector space isomorphism onto the kernel of $`h^{}(x):T_xN^s`$.
Up to isomorphism, the Lie algebra $`𝔥`$ depends only on the leaf $`L`$ that contains $`h(N)`$. It will be referred to as the *symmetry algebra* of $`L`$.
It is useful to note that the symmetry algebra of a leaf $`L`$ can be computed without actually having to find a realization $`(N,\eta ,h)`$ with $`h(N)L`$. In fact, for any $`\overline{h}L`$, define a skewsymmetric bilinear pairing $`[,]_{\overline{h}}:^n\times ^n^n`$ by
$$[E_i,E_j]_{\overline{h}}=C_{ij}^k(\overline{h})E_k,$$
where $`E_i`$ is the standard basis of $`^n`$. Let $`𝔥_{\overline{h}}^n`$ be the subspace that is the kernel of the (surjective) linear map $`\lambda _{\overline{h}}:^nT_{\overline{h}}L`$ that satisfies $`\lambda _{\overline{h}}(E_i)=F_i(\overline{h})`$. Then the restriction of $`[,]_{\overline{h}}`$ to $`𝔥_{\overline{h}}`$ defines a Lie algebra structure on $`𝔥_{\overline{h}}`$. One can verify that, up to isomorphism, this Lie algebra does not depend on the choice of $`\overline{h}L`$ and that this is indeed the symmetry algebra of $`L`$.
### A.8. A semi-global realization
With these concepts, a ‘semi-global’ version of Cartan’s existence and uniqueness result can be stated. For lack of space, I will not discuss the (relatively straightforward) proof, which, in any case, can be found in the above cited references.
###### Theorem A.3.
Let $`(X,F,C)`$ satisfy (A.2) and (A.3), let $`LX`$ be an $`F`$-leaf with symmetry algebra $`𝔥`$. Let $`H`$ be a Lie group whose Lie algebra is $`𝔥`$.
Then over any contractible open subset $`UL`$ there exists a principal left $`H`$-bundle $`(h^a)=h:NU`$ together with an $`H`$-invariant coframing $`\eta =(\eta ^i)`$ on $`N`$ so that $`(N,\eta ,h)`$ satisfies (A.1). This realization is unique up to isomorphism.
Simple examples show that existence and/or uniqueness can fail when $`U`$ has nontrivial homotopy groups. In fact, this is the source of the orbifold singularities encountered in §4.3.3.
When $`H`$ is abelian, the obstruction to global existence on a leaf $`L`$ can be formulated as the vanishing of an element of an appropriate cohomology group on $`L`$. When $`H`$ is nonabelian, there is still a cohomological condition, but it takes values in a certain nonabelian cohomology set. Since this refinement will not be needed in this article, it will not be discussed. |
warning/0003/hep-th0003239.html | ar5iv | text | # Geometric interpretation of Schwarzschild instantons
## 1 Introduction
An Abelian instanton is a self-dual solution to Euclidean Maxwell’s equations. In the case of the Taub-NUT metric on $`^4`$ such a non-trivial solution was found by Eguchi-Hanson in 1979. In mathematical terms a self-dual solution to Euclidean Maxwell’s equations with finite energy is a self-dual $`L^2`$ harmonic $`2`$-form with integer cohomology class. In this context the Eguchi-Hanson solution was reinvented by Gibbons in 1996. Motivated by Sen’s S-duality conjecture he constructed a non-topological <sup>1</sup><sup>1</sup>1 In general we call a non-trivial $`L^2`$ harmonic form on a complete Riemannian manifold non-topological if either it is exact or not cohomologous to a compactly supported differential form. Roughly speaking the existence of non-topological $`L^2`$ harmonic forms are not predictable by topological means. (Cf. .) self-dual $`L^2`$ harmonic $`2`$-form in the Taub-NUT metric. A curious feature of this form is that, living on a space with no topology, it is cohomologically trivial, producing a family of Abelian instantons with continuous energy.
Gibbons’ construction is geometric in nature; indeed the $`L^2`$ harmonic $`2`$-form is obtained as the exterior derivative of a $`1`$-form dual to a Killing field of some natural $`U(1)`$-action. In 1999 Hitchin completed the proof of Sen’s S-duality conjecture in the Taub-NUT case by showing that the whole $`L^2`$ harmonic space is spanned by the Eguchi-Hanson-Gibbons $`2`$-form.
In this note we imitate this construction of Gibbons for the case of the Euclidean Schwarzschild metric. It is a Ricci-flat metric on $`^2\times S^2`$ and was constructed by Hawking in 1976 as the Wick rotation of the Schwarzschild space-time.
We show that the rotation on the $`^2`$ part induces a Killing field such that the exterior derivative of the dual $`1`$-form has finite energy. On a Ricci-flat manifold it follows from Killing’s equations that the form obtained this way solves Maxwell’s equations . However, unlike the Taub-NUT case, this form is not self-dual (this fact is related<sup>2</sup><sup>2</sup>2Cf. Theorem 4 of . to the observation that the Euclidean Schwarzschild manifold is not hyperkähler while the Taub-NUT manifold is). Self-dualizing the form produces a self-dual $`L^2`$ harmonic $`2`$-form, which is not trivial<sup>3</sup><sup>3</sup>3Nevertheless it is still non-topological in the sense of footnote 1 above, since on $`M`$ every compactly supported $`2`$-form is exact. cohomologically. Thus in order to obtain Abelian instantons, we have to quantize the form to have integer cohomology class. In this way we get Abelian instantons lying on $`U(1)`$-bundles of first Chern numbers $`n`$ and first Pontryagin numbers $`2n^2`$.
On the other hand $`SU(2)`$-instantons on the Euclidean Schwarzschild manifold were constructed by Charap and Duff in 1977. They considered $`O(3)`$-invariant instantons, where the action of $`O(3)`$ is induced from the symmetry group of $`S^2`$. In this way their ansatz was reduced to a system of three relatively simple partial differential equations. They were able to find three kind of solutions of this system. The first was the trivial flat connection; the second the non-trivial “metric connection” of second Chern number $`1`$ obtained earlier in ; and the third was a family of solutions which gave rise to instantons of second Chern number $`2n^2`$. Apparently they refer to this last family as non-Abelian dyons and give no geometrical interpretation.
Representing $`U(1)`$ as a subgroup of $`SU(2)`$ we obtain $`SU(2)`$ instantons with second Chern numbers (i.e. instanton numbers) $`2n^2`$ from our integer $`L^2`$ harmonic forms. The main result of the present note is that this family coincides with the third group of $`SU(2)`$-instantons found by Charap and Duff. In spite of a few work dealing with or mentioning the Charap–Duff instantons apparently its Abelian character has not been recognized yet.
Using a recent result of Hitchin we finish our paper by showing that there are no other Abelian instantons, i.e. self-dual $`L^2`$ harmonic $`2`$-forms on the Euclidean Schwarzschild manifold. Indeed with the help of a result of Dodziuk we are able to determine the whole $`L^2`$ harmonic space.
### Acknowledgement.
The work in this paper was done when the second author visited the Yukawa Institute of Kyoto University in February 2000. We are grateful for Prof. G.W. Gibbons for insightful discussions and Prof. H. Kodama and the Yukawa Institute for the invitation and hospitality.
## 2 Construction of the Abelian instanton
Hawking invented the Euclidean Schwarzschild manifold to argue for the thermal nature of particle creation at a Schwarzschild black hole.
Mathematically the Euclidean Schwarzschild $`4`$-manifold $`M`$ is a complete solution to the Euclidean Einstein’s equations with zero cosmological constant, and has the non-trivial topology $`M^2\times S^2`$. In other words it is a Ricci flat manifold. It is not a gravitational instanton (such as e.g. the Taub-NUT metric or the Eguchi-Hanson metric) in that its curvature tensor is not self-dual. Thus it is not hyperkähler either, which property will effect our considerations (cf. Theorem 4 of ) in the form of the existence of non-self-dual $`L^2`$ harmonic forms on $`M`$.
According to (14.3.11) of , we have a particularly nice form of the metric $`g`$ on a dense open subset $`(^2\{O\})\times S^2M^2\times S^2`$ of the Euclidean Schwarzschild manifold. It is convenient to use polar coordinates $`(r,\tau )`$ on $`^2\{O\}`$ in the range $`r(2m,\mathrm{})`$ and $`\tau [0,8\pi m)`$, where $`m>0`$ is a fixed constant. The metric then takes the form
$$\mathrm{d}s^2=\left(1\frac{2m}{r}\right)\mathrm{d}\tau ^2+\left(1\frac{2m}{r}\right)^1\mathrm{d}r^2+r^2\mathrm{d}\mathrm{\Omega }^2,$$
where $`\mathrm{d}\mathrm{\Omega }^2`$ stands for the line element of the unit round $`S^2`$. In sphere coordinates $`\mathrm{\Theta }(0,\pi )`$ and $`\varphi [0,2\pi )`$ it is
$$\mathrm{d}\mathrm{\Omega }^2=\mathrm{d}\mathrm{\Theta }^2+\mathrm{sin}^2\mathrm{\Theta }\mathrm{d}\varphi ^2$$
on the open coordinate chart $`(S^2(\{S\}\{N\}))S^2`$. Consequently the above metric takes the following form on the open, dense coordinate chart $`U:=(^2\{O\})\times (S^2(\{S\}\{N\}))M^2\times S^2`$:
$`\mathrm{d}s^2=\left(1{\displaystyle \frac{2m}{r}}\right)\mathrm{d}\tau ^2+\left(1{\displaystyle \frac{2m}{r}}\right)^1\mathrm{d}r^2+r^2(\mathrm{d}\mathrm{\Theta }^2+\mathrm{sin}^2\mathrm{\Theta }\mathrm{d}\varphi ^2).`$ (1)
Despite the apparent singularity of the metric at the origin $`O^2`$, it can be extended analytically to the whole $`^2\times S^2`$ as demonstrated on page 407 of .
The $`U(1)`$-action defined by $`\tau \tau +4m\lambda `$ for $`e^{i\lambda }U(1)`$ leaves this metric invariant, and thus defines the Killing field
$$X:=\frac{1}{4m}\frac{}{\tau },$$
which (together with the $`U(1)`$-action itself) clearly extends to a Killing field on the whole Euclidean Schwarzschild manifold, which we will also denote by $`X`$.
Now consider the differential $`1`$-form $`\xi :=g(X,)`$ dual to $`X`$. In our coordinate chart $`U`$ it takes the form
$$\xi =\frac{1}{4m}\left(1\frac{2m}{r}\right)\mathrm{d}\tau .$$
General considerations about Killing’s equations on a Ricci flat manifold yield that $`\mathrm{d}\xi `$ is a harmonic $`2`$-form, which on a complete manifold is equivalent to saying that it is closed and coclosed. For a proof see page 442-443 of . In our situation we can check it by hand that our form
$$\mathrm{d}\xi =\frac{1}{2r^2}\mathrm{d}\tau \mathrm{d}r$$
is coclosed. For this we need to calculate $`\mathrm{d}\xi `$. Evoking the local coordinate representation of the general Hodge operation (e.g. page 5 of ), the Hodge-operation $`:\mathrm{\Omega }^2(M)\mathrm{\Omega }^2(M)`$ on the Euclidean Schwarzschild manifold $`(M,g)`$ can be written as
$$\mathrm{d}\tau \mathrm{d}r=r^2\mathrm{sin}\mathrm{\Theta }\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi ,\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi =\frac{1}{r^2\mathrm{sin}\mathrm{\Theta }}\mathrm{d}\tau \mathrm{d}r,$$
$$\mathrm{d}\tau \mathrm{d}\mathrm{\Theta }=(1\frac{2m}{r})^1\mathrm{sin}\mathrm{\Theta }\mathrm{d}r\mathrm{d}\varphi ,\mathrm{d}r\mathrm{d}\varphi =(1\frac{2m}{r})\frac{1}{\mathrm{sin}\mathrm{\Theta }}\mathrm{d}\tau \mathrm{d}\mathrm{\Theta },$$
$$\mathrm{d}\tau \mathrm{d}\varphi =(1\frac{2m}{r})^1\frac{1}{\mathrm{sin}\mathrm{\Theta }}\mathrm{d}r\mathrm{d}\mathrm{\Theta },\mathrm{d}r\mathrm{d}\mathrm{\Theta }=(1\frac{2m}{r})\mathrm{sin}\mathrm{\Theta }\mathrm{d}\tau \mathrm{d}\varphi .$$
The orientation is fixed such that $`\epsilon _{\tau r\mathrm{\Theta }\varphi }=1`$. From here we can see that
$$\mathrm{d}\xi =\frac{1}{2}\mathrm{sin}\mathrm{\Theta }\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi $$
is closed. Thus $`\mathrm{d}\xi `$ is indeed harmonic. Now we show that it is $`L^2`$ by calculating the Maxwell action of it: using the parameterization of the Euclidean Schwarzschild manifold given above we find
$`\mathrm{d}\xi _{L^2(M)}^2=\mathrm{d}\xi _{L^2(M)}^2={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{M}{}}\mathrm{d}\xi \mathrm{d}\xi ={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{0}{\overset{2\pi }{}}}{\displaystyle \underset{0}{\overset{\pi }{}}}{\displaystyle \underset{2m}{\overset{\mathrm{}}{}}}{\displaystyle \underset{0}{\overset{8\pi m}{}}}{\displaystyle \frac{\mathrm{sin}\mathrm{\Theta }}{4r^2}}\mathrm{d}\tau \mathrm{d}r\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi ={\displaystyle \frac{1}{2}}.`$ (2)
In this way we have produced a $`2`$-dimensional space of $`L^2`$ harmonic $`2`$-forms on $`M`$ spanned by $`\mathrm{d}\xi `$ and $`\mathrm{d}\xi `$, and a $`1`$-dimensional subspace of (anti)self-dual $`L^2`$ harmonic forms spanned by $`\omega _\pm :=\mathrm{d}\xi \pm \mathrm{d}\xi `$. From now on, without loss of generality we focus on self-dual forms only, i.e we will use the notation $`\omega :=\omega _+`$. Hence the self-dual form looks like
$`\omega ={\displaystyle \frac{1}{2}}\left({\displaystyle \frac{1}{r^2}}\mathrm{d}\tau \mathrm{d}r+\mathrm{sin}\mathrm{\Theta }\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi \right)`$ (3)
on $`U`$. By (2), the Maxwell action or $`L^2`$-norm of the self-dual $`\omega `$ is given by
$`\omega _{L^2(M)}^2={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{M}{}}\omega \omega ={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{M}{}}2\mathrm{d}\xi \mathrm{d}\xi =1.`$ (4)
The self-dual $`2`$-form $`\omega `$ is not trivial topologically; indeed its cohomology class can be easily identified with the first Chern class of the $`U(1)`$-bundle $`H`$ whose restriction $`H|_{S^2}`$ is nothing but the Hopf $`U(1)`$-bundle (i.e. the positive generator of $`H^2(S^2,)`$) through the isomorphism $`H^2(^2\times S^2,)H^2(S^2,)`$ via the integral
$`{\displaystyle \frac{1}{2\pi }}{\displaystyle \underset{S^2}{}}\omega |_{S^2}={\displaystyle \frac{1}{2\pi }}{\displaystyle \underset{S^2}{}}\mathrm{d}\xi ={\displaystyle \frac{1}{4\pi }}{\displaystyle \underset{0}{\overset{2\pi }{}}}{\displaystyle \underset{0}{\overset{\pi }{}}}\mathrm{sin}\mathrm{\Theta }\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi =1,`$ (5)
where we embedded $`S^2`$ into $`M`$ as $`S^2\{p\}\times S^2M^2\times S^2`$, where for the sake of simplicity $`p^2`$ differs from the origin.
According to (5) $`\frac{1}{2\pi }\omega H^2(M,)`$ is an integer form, thus there is a connection $`A_1`$ on $`H`$, whose curvature satisfies $`F_{A_1}=\omega 𝐤`$, where we used the identification $`𝔲(1)𝐤`$. Furthermore it is unique, since $`\pi _1(M)=1`$, consequently any flat connection must be the trivial one. Similarly the $`U(1)`$-bundle $`H^n`$ admits a unique connection $`A_n`$ such that $`F_{A_n}=n\omega 𝐤`$.
Now we write down $`A_n`$ locally on two charts and explain how to glue them together: Let us denote by $`H^\pm `$ the northern and southern hemispheres of $`S^2`$ respectively, in other words $`H^+`$ is the set of points, where $`\mathrm{\Theta }\pi /2`$ and $`H^{}`$ is the set, where $`\mathrm{\Theta }\pi /2`$. Consider the coordinate charts $`U^\pm :=^2\times H^\pm `$ of the space $`M=^2\times S^2`$. Clearly, $`M=U^+U^{}`$ and $`U^+U^{}^2\times S^1`$ is given by the points satisfying $`\mathrm{\Theta }=\pi /2`$. By integrating (3), in our coordinate chart $`U`$ and an appropriate trivialization of $`H^n`$ the connection $`A_n`$ takes the form ($`c_1`$, $`c_2`$ are arbitrary real constants):
$`A_n^\pm ={\displaystyle \frac{n}{2}}\left(\left(c_1{\displaystyle \frac{1}{r}}\right)\mathrm{d}\tau +(c_2+\mathrm{cos}\mathrm{\Theta })\mathrm{d}\varphi \right)𝐤.`$
For this to extend to the North pole ($`\mathrm{\Theta }=0`$) and respectively to the South pole ($`\mathrm{\Theta }=\pi `$), we need to choose $`c_2=1`$ on $`U^+`$ and respectively $`c_2=1`$ on $`U^{}`$. Thus our connection $`A_n`$ takes the following shape on the charts $`U^\pm `$:
$`A_n^\pm ={\displaystyle \frac{n}{2}}\left(\left(c_1{\displaystyle \frac{1}{r}}\right)\mathrm{d}\tau +(1+\mathrm{cos}\mathrm{\Theta })\mathrm{d}\varphi \right)𝐤.`$ (6)
Note that these connection forms are regular along $`U^+U^{}`$ and are related by the Abelian gauge transformation
$$A_n^+A_n^{}=n\mathrm{d}\varphi 𝐤$$
given by $`e^{n\varphi 𝐤}U(1)`$ along $`U^+U^{}`$. We recognize the above connections as the $`L^2`$ harmonic generalizations for the Euclidean Schwarzschild case of the connections appearing in the well-known bundle-theoretic description of the Dirac magnetic monopole, see e.g. page 231-232 of . The extra term $`(c1/r)\mathrm{d}\tau `$ can be interpreted as a scalar potential and will cause that our solutions carry electric charge.
Consider now the associated $`U(2)`$-bundle $`P_{U(2)}H^nH^n`$, via the diagonal embedding of $`U(1)\times U(1)U(2)`$, and the associated connection $`B_n=A_nA_n`$ with curvature form $`F_{A_n}F_{A_n}`$ on it. Since $`H^4(M,)0`$ the principal $`U(2)`$-bundle $`P_{U(2)}`$ of $`H^nH^n`$ is trivial. Moreover its determinant $`U(1)`$-bundle is trivial and thus $`P_{U(2)}`$ reduces to the trivial $`SU(2)`$-bundle which we denote by $`P=M\times SU(2)`$. Furthermore the $`U(2)`$-connection $`B_n`$ induces a trivial connection on the determinantal $`U(1)`$-bundle so it reduces to an $`SU(2)`$-connection on $`P`$. In our coordinate charts $`U^\pm `$ the connection $`B_n`$ is induced by the embedding $`𝐤𝔲(1)𝔰𝔲(2)\mathrm{Im}`$. In other words self-dual $`L^2`$ harmonic $`2`$-forms may be regarded as the curvature $`2`$-forms of (reducible) self-dual Yang-Mills $`SU(2)`$-connections given locally by the formula (6).
Using (4) we find that the second Chern numbers of these self-dual Yang-Mills $`SU(2)`$-connections $`B_n=A_nA_n`$ on the associated $`SU(2)`$-bundles $`H^nH^n`$ satisfy:
$$\frac{1}{8\pi ^2}\underset{M}{}tr\left(F_{A_n}F_{A_n}F_{A_n}F_{A_n}\right)=\frac{1}{8\pi ^2}\underset{M}{}2F_{A_n}F_{A_n}=2n^2$$
since we have $`tr(AB)=2\mathrm{R}\mathrm{e}(x\overline{y})`$ for the Killing-form on the Lie algebra $`𝔰𝔲(2)\mathrm{Im}`$.
Note that if we calculate the first Pontryagin number of the connection $`A_n`$ on the real plane bundle $`H^n`$ (here we made the identification $`U(1)SO(2)`$) we also find
$$\frac{1}{4\pi ^2}\underset{M}{}F_{A_n}F_{A_n}=2n^2.$$
In the following section we prove that the reducible $`SU(2)`$-instantons just derived coincide with the third group of instantons found by Charap and Duff .
## 3 Identification with instantons of Charap and Duff
Now we will follow . In that paper solutions of type (II) of the self-duality equations on $`P`$ are referred to as ”non-Abelian dyons” of Pontryagin numbers $`2n^2`$. Let us denote them $`\stackrel{~}{A}_n`$. In this section we show that they are in fact reducible, i.e. Abelian connections and identify them with the connections $`B_n`$ defined above. To round things off, we finish this section by giving the explicit local gauge transformations which identify our Abelian connections (6) with Charap–Duff’s (8).
Let $`n`$ be an integer and focus our attention to solution (II), more precisely the self-dual one, which means that we choose all the functions of positive sign. Putting solution (II) into the spherical symmetric ansatz (5) of and adjusting notations of to ours via the identification $`𝔰𝔲(2)\mathrm{Im}`$ given by $`\{\sigma ^1/2,\sigma ^2/2,\sigma ^3/2\}\{𝐢/2,𝐣/2,𝐤/2\}`$, the coordinate transformation
$`(\tau ,x^1,x^2,x^3)(n\tau ,r\mathrm{sin}\mathrm{\Theta }\mathrm{cos}(n\varphi ),r\mathrm{sin}\mathrm{\Theta }\mathrm{sin}(n\varphi ),r\mathrm{cos}\mathrm{\Theta })`$ (7)
and the notation
$$𝐪_n:=\mathrm{sin}\mathrm{\Theta }\mathrm{cos}(n\varphi )𝐢+\mathrm{sin}\mathrm{\Theta }\mathrm{sin}(n\varphi )𝐣+\mathrm{cos}\mathrm{\Theta }𝐤,$$
we get the new form for the self-dual connection
$`\stackrel{~}{A}_n={\displaystyle \frac{n}{2}}\left(\left(c{\displaystyle \frac{1}{r}}\right)\mathrm{d}\tau +\mathrm{cos}\mathrm{\Theta }\mathrm{d}\varphi \right)𝐪_n{\displaystyle \frac{n}{2}}\mathrm{d}\varphi 𝐤+{\displaystyle \frac{1}{2}}\mathrm{d}\mathrm{\Theta }(\mathrm{sin}(n\varphi )𝐢\mathrm{cos}(n\varphi )𝐣).`$ (8)
A long but straightforward calculation shows that the curvature takes the form
$`F_{\stackrel{~}{A}_n}=n\omega 𝐪_n.`$ (9)
Consider now the $`U(1)`$-sub-bundle $`H_n`$ of $`P`$ whose smooth sections are given by $`s=\mathrm{exp}(f𝐪_n),`$ where $`\mathrm{exp}:𝔰𝔲(2)SU(2)`$ is the exponential map and $`f`$ is any smooth function on $`M`$. We show that the covariant derivative $`_{\stackrel{~}{A}_n}:\mathrm{\Omega }^0(ad(P))\mathrm{\Omega }^1(ad(P))`$ on the associated bundle $`ad(P)`$ leaves the real line bundle $`ad(H_n)ad(P)`$ invariant. We thus calculate in our coordinate chart $`U`$:
$`_{\stackrel{~}{A}_n}s=_{\stackrel{~}{A}_n}(f𝐪_n)`$ $`=`$ $`\mathrm{d}(f𝐪_n)+[\stackrel{~}{A}_n,f𝐪_n,]`$
where by abuse of notation $`\stackrel{~}{A}_n`$ denotes the connection matrix of $`\stackrel{~}{A}_n`$ in the gauge (8). The first term equals:
$`\mathrm{d}(f𝐪_n)`$ $`=`$ $`\mathrm{d}f𝐪_n+f\mathrm{d}(\mathrm{sin}\mathrm{\Theta }\mathrm{cos}(n\varphi )𝐢+\mathrm{sin}\mathrm{\Theta }\mathrm{sin}(n\varphi )𝐣+\mathrm{cos}\mathrm{\Theta }𝐤)`$
$`=`$ $`\mathrm{d}f𝐪_n+f\mathrm{d}\mathrm{\Theta }\left(\mathrm{cos}\mathrm{\Theta }\mathrm{cos}(n\varphi )𝐢+\mathrm{cos}\mathrm{\Theta }\mathrm{sin}(n\varphi )𝐣\mathrm{sin}\mathrm{\Theta }𝐤\right)+`$
$`+fn\mathrm{d}\varphi \left(\mathrm{sin}\mathrm{\Theta }\mathrm{sin}(n\varphi )𝐢+\mathrm{sin}\mathrm{\Theta }\mathrm{cos}(n\varphi )𝐣\right),`$
and the second gives,
$$[\stackrel{~}{A}_n,f𝐪_n]=$$
$$=[\frac{n}{2}\left(\left(c\frac{1}{r}\right)\mathrm{d}\tau +\mathrm{cos}\mathrm{\Theta }\mathrm{d}\varphi \right)𝐪_n\frac{n}{2}\mathrm{d}\varphi 𝐤+\frac{1}{2}\mathrm{d}\mathrm{\Theta }(\mathrm{sin}(n\varphi )𝐢\mathrm{cos}(n\varphi )𝐣),f𝐪_n]=$$
$$=[\frac{n}{2}\mathrm{d}\varphi 𝐤+\frac{1}{2}\mathrm{d}\mathrm{\Theta }\left(\mathrm{sin}(n\varphi )𝐢\mathrm{cos}(n\varphi )𝐣\right),f𝐪_n]=$$
$$=fn\mathrm{d}\varphi \left(\mathrm{sin}\mathrm{\Theta }\mathrm{sin}(n\varphi )𝐢\mathrm{sin}\mathrm{\Theta }\mathrm{cos}(n\varphi )𝐣\right)+$$
$$+f\mathrm{d}\mathrm{\Theta }\left(\mathrm{cos}\mathrm{\Theta }\mathrm{cos}(n\varphi )𝐢\mathrm{cos}\mathrm{\Theta }\mathrm{sin}(n\varphi )𝐣+\mathrm{sin}\mathrm{\Theta }𝐤\right).$$
Adding the two above expressions up we see that
$$_{\stackrel{~}{A}_n}(f𝐪_n)=\mathrm{d}f𝐪_n,$$
showing that $`\stackrel{~}{A}_n`$ reduces to a $`U(1)`$-connection on $`H_nP`$. Now (9) shows that this $`U(1)`$-connection on $`H_n`$ has the same curvature than $`A_n`$ therefore they should coincide, in particular $`H_nH^n`$. Thus we proved that the Charap-Duff’s connection (8) is equivalent to our connection (6).
We finish this section by writing down the explicit gauge transformations on $`U^\pm `$ which transform our connection (6) to Charap-Duff’s (8). From (9) we can guess that the gauge transformations we are looking for should rotate the vector $`𝐪_n`$ into the unit vector $`𝐤`$. This transformation cannot be carried out continously over the whole $`S^2`$ by using only one transformation but there is no obstruction if we use two gauge transformations on the charts $`U^\pm `$ which are related along $`U^+U^{}`$ by an Abelian gauge transformation. Consider the gauge transformations $`g_n^\pm :U^\pm SU(2)S^3`$ given by<sup>4</sup><sup>4</sup>4By abuse of notation we will regard the unit quaternions $`𝐢,𝐣,𝐤`$ either elements of the Lie algebra $`𝔰𝔲(2)\mathrm{Im}`$ or of the group $`SU(2)S^3`$ depending on the context.
$$g_n^\pm (\tau ,r,\mathrm{\Theta },\varphi ):=\mathrm{exp}\left(\pm 𝐤\frac{n\varphi }{2}\right)\mathrm{exp}\left(𝐣\frac{\mathrm{\Theta }}{2}\right)\mathrm{exp}\left(𝐤\frac{n\varphi }{2}\right).$$
In this form we only see that $`g_n^\pm `$ are smooth gauge transformations on $`U`$. In order to be well defined as smooth maps $`g_n^\pm :U^\pm SU(2)`$ we have to show that they extend analytically over the appropriate poles. We show this for $`g^+`$ here, the case of $`g^{}`$ being similar. It is easily checked that the following gauge transformation in Descartes coordinates gives rise<sup>5</sup><sup>5</sup>5In other words the gauge transformation (10) pulls back to $`g^+`$ by the map given by (7). to $`g^+`$ after the coordinate transformation (7):
$`\left({\displaystyle \frac{x_3}{2r}}+{\displaystyle \frac{1}{2}}\right)^{1/2}\left({\displaystyle \frac{x_3}{2r}}+{\displaystyle \frac{1}{2}}𝐢{\displaystyle \frac{x_2}{2r}}𝐣{\displaystyle \frac{x_3}{2r}}\right).`$ (10)
In this form we see that the map $`g^+:USU(2)`$ extends analytically to $`U^+U`$, that is to points of $`M`$, where $`(\mathrm{\Theta }=0)`$ or equivalently $`x_3/r=1`$.
Let us prove that the above gauge transformations do indeed transform (8) into (6)! First we show that it rotates $`𝐪_n`$ into $`𝐤`$: Writing $`𝐪_n=\mathrm{sin}\mathrm{\Theta }\mathrm{cos}(n\varphi )𝐢+\mathrm{sin}\mathrm{\Theta }\mathrm{sin}(n\varphi )𝐣+\mathrm{cos}\mathrm{\Theta }𝐤=\mathrm{exp}(𝐤n\varphi )\mathrm{sin}\mathrm{\Theta }𝐢+\mathrm{cos}\mathrm{\Theta }𝐤`$, we can proceed as follows:
$$g_n^\pm (\mathrm{exp}(𝐤n\varphi )\mathrm{sin}\mathrm{\Theta }𝐢+\mathrm{cos}\mathrm{\Theta }𝐤)(g_n^\pm )^1=$$
$$=\mathrm{exp}\left(\pm 𝐤\frac{n\varphi }{2}\right)\mathrm{exp}\left(𝐣\frac{\mathrm{\Theta }}{2}\right)(\mathrm{sin}\mathrm{\Theta }𝐢+\mathrm{cos}\mathrm{\Theta }𝐤)\mathrm{exp}\left(𝐣\frac{\mathrm{\Theta }}{2}\right)\mathrm{exp}\left(𝐤\frac{n\varphi }{2}\right).$$
Since $`\mathrm{sin}\mathrm{\Theta }𝐢+\mathrm{cos}\mathrm{\Theta }𝐤=\mathrm{exp}(𝐣\mathrm{\Theta })𝐤`$, we can go further by writing
$$\mathrm{exp}\left(\pm 𝐤\frac{n\varphi }{2}\right)𝐤\mathrm{exp}\left(𝐤\frac{n\varphi }{2}\right)=𝐤$$
proving that the above gauge transformations $`g_n^\pm `$ send $`𝐪_n`$ into $`𝐤`$.
Finally we calculate that at one hand
$$g_n^\pm \mathrm{d}(g_n^\pm )^1=\frac{n}{2}\mathrm{d}\varphi 𝐤+\frac{n}{2}\mathrm{d}\varphi \mathrm{exp}\left(\pm 𝐤\frac{n\varphi }{2}\right)\mathrm{exp}(𝐣\mathrm{\Theta })\mathrm{exp}\left(𝐤\frac{n\varphi }{2}\right)𝐤+\frac{1}{2}\mathrm{d}\mathrm{\Theta }\mathrm{exp}(\pm 𝐤n\varphi )𝐣,$$
on the other hand
$$g_n^\pm \left(\frac{n}{2}\mathrm{d}\varphi 𝐤+\frac{1}{2}\mathrm{d}\mathrm{\Theta }(\mathrm{sin}(n\varphi )𝐢\mathrm{cos}(n\varphi )𝐣)\right)(g_n^\pm )^1=$$
$$=\frac{n}{2}\mathrm{d}\varphi \mathrm{exp}\left(\pm 𝐤\frac{n\varphi }{2}\right)\mathrm{exp}(𝐣\mathrm{\Theta })\mathrm{exp}\left(𝐤\frac{n\varphi }{2}\right)𝐤\frac{1}{2}\mathrm{d}\mathrm{\Theta }\mathrm{exp}(\pm 𝐤n\varphi )𝐣.$$
But these terms cancel each other except $`\frac{n}{2}\mathrm{d}\varphi 𝐤`$ demonstrating the desired result
$$g_n^\pm \stackrel{~}{A}_n(g_n^\pm )^1+g_n^\pm \mathrm{d}(g_n^\pm )^1=A_n^\pm $$
where $`A_n^\pm `$ are given by (6). Note that the two gauge transformations are related along $`U^+U^{}`$ by the Abelian gauge transformation
$$\mathrm{exp}(𝐤n\varphi )g_n^{}=g_n^+$$
yielding again $`A_n^{}𝐤n\mathrm{d}\varphi =A_n^+`$.
Thus we gave two proofs that the Charap–Duff instantons coincide with ours proving that these solutions are nothing but Abelian dyons carrying magnetic charge $`n`$ and electric charge $`n`$. Indeed, the electric charge is given by the integration of the electric field over an embedded two-sphere. By self-duality
$$\frac{1}{2\pi }\underset{S^2}{}\omega |_{S^2}=1,$$
hence it is clear that the general solution has electric charge $`n`$, too. In summary we see that the basic characteristic numbers of these solutions are their magnetic charge $`n`$ represented by the first Chern class of the $`U(1)`$-bundle $`H^n`$ instead of the first Pontryagin number $`2n^2`$.
## 4 $`L^2`$-cohomology
In this final section we show that we have found all the Abelian instantons on the Euclidean Schwarzschild manifold.
###### Theorem 4.1
Let $`\eta `$ be an $`L^2`$ harmonic form on $`M`$. Then it is a linear combination of $`\mathrm{d}\xi `$ and $`\mathrm{d}\xi `$. Consequently a self-dual $`L^2`$ harmonic $`2`$-form on $`M`$ is some constant multiple of $`\omega =\mathrm{d}\xi +\mathrm{d}\xi `$.
### Proof.
First of all the volume of $`(M,g)`$ is infinite. It can be seen by calculating:
$$\underset{M}{}\mathrm{𝟏}=\underset{0}{\overset{2\pi }{}}\underset{0}{\overset{\pi }{}}\underset{2m}{\overset{\mathrm{}}{}}\underset{0}{\overset{8\pi m}{}}r^2\mathrm{sin}\mathrm{\Theta }\mathrm{d}\tau \mathrm{d}r\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi =\mathrm{},$$
where we have used again the parameterization of the Euclidean Schwarzschild manifold given in the previous section. This implies that there are no $`L^2`$ harmonic $`0`$\- or equivalently $`4`$-forms. Now, as $`M`$ is Ricci-flat and complete, Corollary 1 of Dodziuk implies that there are no $`1`$\- and equivalently $`3`$-forms on $`M`$.
It remains to show that any $`L^2`$ harmonic $`2`$-form is a linear combination of $`\mathrm{d}\xi `$ and $`\mathrm{d}\xi `$. For this we use a recent result of Hitchin, namely Theorem 3 of which we cite in full:
###### Theorem 4.2 (Hitchin)
Let $`M`$ be a complete oriented Riemannian manifold and let $`G`$ be a connected Lie group of isometries such that the Killing vector fields $`X`$ it defines satisfy
$$|X|c^{}\rho (x_0,x)+c^{\prime \prime }.$$
Then each $`L^2`$ cohomology class is fixed by $`G`$.
(Here $`\rho `$ is the distance function of the Riemannian manifold.) We would like to apply this result to $`M`$ with $`GSO(3)`$ acting on $`M`$ by isometries of $`S^2`$. A glance at the metric (1) assures us that the Killing fields of this action have indeed linear growth. Thus it is sufficient to find all $`SO(3)`$-invariant harmonic $`2`$-forms on $`M`$. Let $`\eta `$ be such a form. In our coordinate chart $`U`$ it must have the shape:
$$\eta =f(\tau ,r)\mathrm{d}\tau \mathrm{d}r+\alpha _\tau (\tau ,r)\mathrm{d}\tau +\alpha _r(\tau ,r)\mathrm{d}r+\beta (\tau ,r),$$
where $`f(r,\tau )`$ is an $`SO(3)`$-invariant function on $`S^2`$, moreover $`\alpha _\tau (\tau ,r)`$ and $`\alpha _r(\tau ,r)`$ are $`SO(3)`$ invariant $`1`$-forms on $`S^2`$, and finally $`\beta (\tau ,r)`$ is an $`SO(3)`$-invariant $`2`$-form on $`S^2`$. However there are very few $`SO(3)`$-invariant forms on $`S^2`$. Namely only the constant functions and constant times the volume form of the round $`S^2`$ are $`SO(3)`$-invariant. It follows because $`SO(3)`$ acts transitively showing that only the constant functions and equivalently constant multiples of the volume form are the $`SO(3)`$-invariant $`0`$\- and $`2`$-forms respectively. Moreover there are no non-trivial $`SO(3)`$-invariant $`1`$-forms on $`S^2`$, which could be seen by looking at the dual vector field and seeing that the action of the $`U(1)`$ stabilizator of any point on the tangent space at that point has only the origin as its fixed point.
It follows that our $`SO(3)`$-invariant $`2`$-form must have the form:
$$\eta =f(\tau ,r)\mathrm{d}\tau \mathrm{d}r+h(\tau ,r)\mathrm{sin}\mathrm{\Theta }\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi $$
where $`\mathrm{sin}\mathrm{\Theta }\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi `$ is the volume form of the unit $`S^2`$ and $`f(\tau ,r)`$ and $`h(\tau ,r)`$ stand for a function on $`M`$ depending only on $`\tau `$ and $`r`$. Its Hodge-dual is given by
$$\eta =h(\tau ,r)\frac{1}{r^2}\mathrm{d}\tau \mathrm{d}r+r^2f(\tau ,r)\mathrm{sin}\mathrm{\Theta }\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi .$$
In order that both $`\eta `$ and $`\eta `$ be closed we need that neither $`h(r,\tau )`$ nor $`r^2f(r,\tau )`$ depend on $`\tau `$ or $`r`$ which means that $`\eta `$ must have the form:
$$\frac{c_1}{r^2}\mathrm{d}\tau \mathrm{d}r+c_2\mathrm{sin}\mathrm{\Theta }\mathrm{d}\mathrm{\Theta }\mathrm{d}\varphi ,$$
exactly as claimed. The result follows. $`\mathrm{}`$
## 5 Concluding Remarks
Previously we have proved that the self-dual solutions to the $`SU(2)`$ Yang–Mills equations over the Euclidean Schwarzschild manifold found by Charap and Duff correspond to Abelian dyons rather than non-Abelian ones. From the mathematical point of view we have seen that the curvatures of these solutions represent elements of the non-trivial second reduced $`L^2`$ cohomology group of the Euclidean Schwarzschild manifold. This identification enabled us to find all the Abelian instantons over this manifold.
The physical interpretation of these solutions is more subtle, however. In light of our results these solutions seem to describe a static electromagnetic dyon configuration surrounding the Schwarzschild black hole. Accepting this, we can interpret their Pontryagin numbers $`2n^2`$ as their three dimensional energy rather then their Euclidean action. Indeed, it is straightforward that the Euclidean Schwarzschild metric tends to the three dimensional flat metric of $`\times S^2`$ and can be extended as the flat metric to the whole $`^3`$ as $`m0`$ (i.e. as the Hawking temperature of the black hole tends to infinity) while neither solutions (6) nor their Euclidean action depends on $`m`$. Henceforth in the limit $`m0`$ we recover the static dyon of charge $`(n,n)`$ on flat space and such a configuration has energy $`2n^2`$ as it is well known.
The general (non self-dual) dyons of charge $`(k,n)`$ correspond to the general elements of the integer lattice $`\overline{H}_{L^2}^2(M,g)`$ in the reduced $`L^2`$-cohomology group of $`(M,g)`$. |
warning/0003/math0003078.html | ar5iv | text | # References
Representations of $`SU(1,1)`$ in Non-commutative Space Generated by the Heisenberg Algebra
H. Ahmedov<sup>1</sup> and I. H. Duru<sup>2,1</sup>
1. Feza Gürsey Institute, P.O. Box 6, 81220, Çengelköy, Istanbul, Turkey <sup>1</sup><sup>1</sup>1E–mail : hagi@gursey.gov.tr and duru@gursey.gov.tr.
2. Trakya University, Mathematics Department, P.O. Box 126, Edirne, Turkey.
Abstract: $`SU(1,1)`$ is considered as the automorphism group of the Heisenberg algebra $`H`$. The basis in the Hilbert space $`K`$ of functions on $`H`$ on which the irreducible representations of the group are realized is explicitly constructed. The addition theorems are derived.
March 2000
1. Introduction
Investigating the properties of manifolds by means of the symmetries they admit has a long history. Non-commutative geometries have become the subject of similar studies in recent decades. For example there exists an extensive literature on the $`q`$-deformed groups $`E_q(2)`$ and $`SU_q(2)`$ which are the automorphism groups of the quantum plane $`zz^{}=qz^{}z`$ and the quantum sphere respectively . Using group theoretical methods the invariant distance and the Green functions have also been written in these deformed spaces .
In the recent work we started to analyze yet another non-commutative space $`[z,z^{}]=1`$ ( i. e. the space generated by the Heisenberg algebra ) by means of its automorphism groups: We considered $`E(2)`$ group transformations in $`z,z^{}`$ space; and constructed the basis (which are written in terms of the Kummer functions) in this space where the unitary irreducible representations of $`E(2)`$ are realized . This analysis revealed a peculiar connection between the $`2`$-dimensional Euclidean group and the Kummer functions.
In the present work we continue to study the same non-commutative space $`[z,z^{}]=1`$, this time by means of the other admissible automorphism group $`SU(1,1)`$.
In Section 2 we define $`SU(1,1)`$ in the Heisenberg algebra $`H`$ and construct the unitary representations of the group in the Hilbert space $`X`$ where $`H`$ is realized.
In Section 3 we classify the invariant subspaces in the space of the bounded functions on $`H`$ where the irreducible representations of $`SU(1,1)`$ are realized.
In Section 4 we show that in the Hilbert space $`K`$ of the square integrable functions only principal series is unitary. We construct the orthonormal basis in $`K`$ which can be written in terms of the Jacobi functions.
Section 5 is devoted to the addition theorems. These theorems provide a group theoretical interpretation for the already existing identities involving the hypergeometric functions which all are actually the Jacobi functions. They may also lead to new identities.
2. Weyl representations of $`SU(1,1)`$
The one dimensional Heisenberg algebra $`H`$ is the 3-dimensional vector space with the basis elements $`\{z,z^{},1\}`$ and the bilinear antisymmetric product
$$[z,z^{}]=1.$$
(1)
The $``$-representation of $`H`$ in the suitable dense subspace of the Hilbert space $`X`$ with the complete orthonormal basis $`\{n\}`$, $`n=0,1,2,\mathrm{}`$ is given by
$$zn=\sqrt{n}n1,z^{}n=\sqrt{n+1}n+1.$$
(2)
Let us represent the pseudo-unitary group $`SU(1,1)`$ in the vector space $`H`$
$$g\left(\begin{array}{c}z\\ z^{}\end{array}\right)=\left(\begin{array}{cc}a& b\\ \overline{b}& \overline{a}\end{array}\right)\left(\begin{array}{c}z\\ z^{}\end{array}\right).$$
(3)
Due to
$$a\overline{a}b\overline{b}=1$$
(4)
the transformations (3) preserve the commutation relation
$$[gz,gz^{}]=[z,z^{}].$$
(5)
Therefore
$$gz=U(g)zU^1(g),gz^{}=U(g)z^{}U^1(g)$$
(6)
where $`U(g)`$ is the unitary representation of $`SU(1,1)`$ in $`X`$:
$$U(g_1)U(g_2)=U(g_1g_2),U^{}(g)=U^1(g)=U(g^1).$$
(7)
The Cartan decomposition for the group reads
$$g=k(\varphi )h(\alpha )k(\psi ),$$
(8)
where
$$k(\psi )=\left(\begin{array}{cc}e^{\frac{i\psi }{2}}& 0\\ 0& e^{i\frac{\psi }{2}}\end{array}\right),h(\alpha )=\left(\begin{array}{cc}\mathrm{cosh}\frac{\alpha }{2}& \mathrm{sinh}\frac{\alpha }{2}\\ \mathrm{sinh}\frac{\alpha }{2}& \mathrm{cosh}\frac{\alpha }{2}\end{array}\right).$$
(9)
For the subgroup $`k(\psi )`$ we have
$$U(k(\psi ))n=e^{i\frac{n\psi }{2}}n.$$
(10)
Let us choose the following realizations for $`z`$, $`z^{}`$ and $`X`$:
$$z=\frac{1}{\sqrt{2}}(x+\frac{d}{dx}),z^{}=\frac{1}{\sqrt{2}}(x\frac{d}{dx}),$$
(11)
$$xn=\mathrm{\Psi }_n(x),\mathrm{\Psi }_n(x)=\sqrt{\frac{e^{x^2}}{2^nn!\sqrt{\pi }}}H_n(x),$$
(12)
where $`H_n`$ is the Hermite polynomial. From
$$h(\alpha )z=\frac{1}{\sqrt{2}}(xe^{\frac{\alpha }{2}}+e^{\frac{\alpha }{2}}\frac{d}{dx})$$
(13)
and
$$_{\mathrm{}}^{\mathrm{}}𝑑x\overline{\mathrm{\Psi }_m(x)}\mathrm{\Psi }_n(x)=\delta _{nm}$$
(14)
we get
$$U(h(\alpha ))\mathrm{\Psi }_m(x)=e^{\frac{\alpha }{4}}\mathrm{\Psi }_m(e^{\frac{\alpha }{2}}x).$$
(15)
Matrix elements of $`U(h(\alpha ))`$ in the basis $`\{n\}`$ reads
$$U_{mn}(h)mU(h(\alpha ))n=e^{\frac{\alpha }{4}}_{\mathrm{}}^{\mathrm{}}𝑑x\overline{\mathrm{\Psi }_m(x)}\mathrm{\Psi }_n(e^{\frac{\alpha }{2}}x).$$
(16)
Evaluating this integral we get
$$U_{mn}(h)=\frac{2^{\frac{mn}{2}}}{(\frac{nm}{2})!}\sqrt{\frac{n!\mathrm{sinh}^{nm}\frac{\alpha }{2}}{m!\mathrm{cosh}^{n+m+1}\frac{\alpha }{2}}}F(\frac{m}{2},\frac{1m}{2};1+\frac{nm}{2};\mathrm{sinh}^2\frac{\alpha }{2})$$
(17)
if $`nm`$ and $`n+m`$ is even and
$$U_{mn}(h)=0$$
(18)
if $`n+m`$ is odd. For $`mn`$ one has to replace $`m`$, $`n`$ and $`\alpha `$ in the above formulas by $`n`$, $`m`$ and $`\alpha `$ respectively.
3. Irreducible representations of $`SU(1,1)`$ in $`H`$
The formula
$$T(g)F(z)=F(gz)$$
(19)
defines the representation of $`SU(1,1))`$ in the space $`K_0`$ of bounded operators in the Hilbert space $`X`$ representable as the finite sums
$$F=(f_n(\zeta )z^n+z^nf_n(\zeta )).$$
(20)
Here $`f_n(\zeta )`$ are functions of the self-adjoint operator $`\zeta =z^{}z`$. Using (6) we can rewrite (19) in the form
$$T(g)F(z)=U(g)F(z)U^{}(g)$$
(21)
With the one parameter subgroups $`g_1=h(ϵ)`$, $`g_2=k(\frac{\pi }{2})h(ϵ)k(\frac{\pi }{2})`$ and $`g_3=k(ϵ)`$ of $`SU(1,1)`$ we associate the linear operators $`E_k:K_0K_0`$
$$E_k(F)=\underset{ϵ0}{lim}\frac{1}{ϵ}(T(g_k)FF)$$
(22)
with the limit being taken in the strong operator topology. Inserting (21) into (22) we get ( with $`H_\pm =E_1iE_2,H=iE_3`$ )
$$H_{}(F)=\frac{1}{2}[F,z^2],H_+(F)=\frac{1}{2}[z^2,F],H(F)=\frac{1}{2}[\zeta ,F],$$
(23)
which implies the Lie algebra of $`SU(1,1)`$
$$[H_+,H_{}]=2H,[H,H_\pm ]=\pm H_\pm .$$
(24)
The irreducible representations labelled by pair $`(\tau ,ϵ)`$, $`\tau C`$ and $`ϵ=0,\frac{1}{2}`$ are given by the formulas
$$H_{}D_k^{(\tau ,ϵ)}=(k+\tau +ϵ)D_{k1}^{(\tau ,ϵ)},$$
(25)
$$H_+D_k^{(\tau ,ϵ)}=(k\tau +ϵ)D_{k+1}^{(\tau ,ϵ)},$$
(26)
$$HD_k^{(\tau ,ϵ)}=(k+ϵ)D_k^{(\tau ,ϵ)}.$$
(27)
(23) and (27) imply
$$D_k^{(\tau ,ϵ)}=z^{2(k+ϵ)}f_k^{(\tau ,ϵ)}(\zeta )$$
(28)
for $`k0`$ and
$$D_k^{(\tau ,ϵ)}=f_k^{(\tau ,ϵ)}(\zeta )z^{2(k+ϵ)}$$
(29)
for $`k<0`$. By substituting (28) in (25) and (26) with
$$f_k^{(\tau ,ϵ)}(\zeta )=\underset{n=0}{\overset{\mathrm{}}{}}\frac{()^n2^{n+k+ϵ}}{n!}C_{kn}z^nz^n$$
(30)
we get the recurrence relations
$$nC_{kn1}+\frac{k+ϵ+\tau }{2k+2ϵ+n1}C_{k1n}(2k+2ϵ+n)C_{kn}=0,$$
(31)
$$C_{kn+1}C_{kn+2}(k+ϵ\tau )C_{k+1n}=0,$$
(32)
which are solved by
$$C_{kn}=\frac{\mathrm{\Gamma }(1+\tau +ϵ+k+n)}{\mathrm{\Gamma }(1+2ϵ+2k+n)}.$$
(33)
Using
$$z^nz^n=\zeta (\zeta 1)\mathrm{}(\zeta n+1)$$
(34)
for $`k0`$ we get
$$f_k^{(\tau ,ϵ)}(\zeta )=(2)^k^{}\frac{\mathrm{\Gamma }(1+\tau +k^{})}{\mathrm{\Gamma }(1+2k^{})}F(\zeta ,1+\tau +k^{};1+2k^{};2),$$
(35)
where $`k^{}=k+ϵ`$. The functions $`f_k^{(\tau ,ϵ)}`$ for $`k<0`$ is shown to be defined from the expression
$$f_k^{(\tau ,ϵ)}(\zeta )=f_k^{(\tau ,ϵ)}(\zeta ).$$
(36)
From (25), (26) and (27) we conclude that $`SU(1,1)`$ admits the following irreducible representations:
i) $`T_{(\tau ,ϵ)}:(\tau +ϵ)\overline{}Z`$
ii) $`T_{(\tau ,ϵ)}^\pm :(\tau +ϵ)Z,\tau ϵ<0`$, that is $`\tau =\frac{1}{2},1,\frac{3}{2},\mathrm{}`$
iii) $`T_{(\tau ,ϵ)}^0:(\tau +ϵ)Z,\tau ϵ0`$, that is $`\tau =0,\frac{1}{2},1,\frac{3}{2}\mathrm{}`$
The corresponding invariant subspaces are:
i) $`V_{(\tau ,ϵ)}`$ generated by $`\{D^{(\tau ,ϵ})_k\}_{k=\mathrm{}}^{\mathrm{}}`$
ii) $`V_{(\tau ,ϵ)}^+`$ and $`V_{(\tau ,ϵ)}^{}`$ generated by $`\{D^{(\tau ,ϵ})_k\}_{k=\mathrm{}}^{\tau ϵ}`$ and $`\{D^{(\tau ,ϵ})_k\}_{k=\tau ϵ}^{\mathrm{}}`$
iii) $`V_{(\tau ,ϵ)}^0`$ generated by $`\{D^{(\tau ,ϵ})_k\}_{k=\tau ϵ}^{\tau ϵ}`$
4. Unitary irreducible representations of $`SU(1,1)`$ in $`H`$
We can define the norm in the subspace of $`K_0`$ with $`f_n(\zeta )`$ in (20) being the functions with finite support in $`Spect(\zeta )=\{0,1,2,\mathrm{}\}`$ as
$$F=\sqrt{tr(F^{}F)}.$$
(37)
Completion of this subspace leads to the Hilbert space $`K`$ of the square integrable functions in the linear space $`H`$ with the scalar product
$$(F,G)=tr(F^{}G).$$
(38)
Using (21), the unitarity of $`U(g)`$ and the property of the trace we conclude that the representation $`T(g)`$ in $`K`$ is unitary. (23) implies the real structure in the Lie algebra
$$H_\pm ^{}=H_{},H^{}=H.$$
(39)
To investigate the unitarity of the irreducible representations in the Hilbert space $`K`$ classified in the previous section we consider the orthogonality condition for the basis elements $`D_k^{(\tau ,ϵ)}`$. Using (2) and (28) we get
$$(D_k^{(\tau ,ϵ)},D_m^{(\tau ^{},ϵ^{})})=\delta _{mk}\delta _{ϵϵ^{}}\underset{n=0}{\overset{\mathrm{}}{}}\frac{(n+2k+2ϵ)!}{n!}\overline{f_k^{(\tau ,ϵ)}(n)}f_k^{(\tau ^{},ϵ)}(n).$$
(40)
Putting
$$s=1e^t,\lambda =1+2(k+ϵ)+\mu $$
(41)
in the formula
$`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\mathrm{\Gamma }(n+\lambda )}{n!\mathrm{\Gamma }(\lambda )}}s^nF(n,a;\lambda ;2)F(n,b;\lambda ;2)=`$
$`=(1s)^{a+b\lambda }(1+s)^{ab}F(a,b;\lambda ;{\displaystyle \frac{4s}{(1+s)^2}})`$ (42)
and taking first the limit $`\mu +0`$ and then $`t\mathrm{}`$ we obtain for $`\tau =\frac{1}{2}+i\rho `$, $`\rho R`$ the orthogonality relations
$$(D_k^{(\frac{1}{2}+i\rho ,ϵ)},D_m^{(\frac{1}{2}+i\rho ^{},ϵ^{})})=\delta _{mk}\delta _{ϵϵ^{}}\delta (\rho \rho ^{}).$$
(43)
In the deriving of the above relation we used
$$F(a,b;c;1)=\frac{\mathrm{\Gamma }(c)\mathrm{\Gamma }(cab)}{\mathrm{\Gamma }(ca)\mathrm{\Gamma }(cb)}$$
(44)
and the representation
$$\underset{t\mathrm{}}{lim}\frac{e^{izt}}{z+i0}=2\pi i\delta (z)$$
(45)
for the Dirac delta function. For other values of $`\tau `$ there is no orthogonality condition. Thus in $`K`$ only the representation $`T_{(\tau ,ϵ)}`$ with $`\tau =\frac{1}{2}+i\rho `$ of Section 3, which is the principal series is unitary.
5. The addition theorems
(i) Restriction of (19) on the subspace $`V_{(\tau ,ϵ)}`$ reads:
$$T(g)D_k^{(\tau ,ϵ)}=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}t_{nk}^{(\tau ,ϵ)}(g)D_n^{(\tau ,ϵ)}$$
(46)
or
$$U(g)D_k^{(\tau ,ϵ)}U^{}(g)=\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}t_{nk}^{(\tau ,ϵ)}(g)D_n^{(\tau ,ϵ)}$$
(47)
where
$`t_{kn}^{(\tau ,ϵ)}(g)={\displaystyle \frac{e^{i(k+ϵ)\varphi i(k+ϵ)\psi }}{(kn)!}}{\displaystyle \frac{\mathrm{\Gamma }(1+\tau ϵn)\mathrm{sinh}^{kn}\frac{\alpha }{2}}{\mathrm{\Gamma }(1+\tau ϵk)\mathrm{cosh}^{k+n+2ϵ}\frac{\alpha }{2}}}\times `$
$`\times F(\tau ϵn,1+\tau ϵn;1+kn;\mathrm{sinh}^2{\displaystyle \frac{\alpha }{2}})`$ (48)
are the matrix elements of the irreducible representations which are valid for $`kn`$. For $`k<n`$ one has to replace $`k`$ and $`n`$ on the right hand side by $`k`$ and $`n`$ respectively.
(ii) Restriction of (19) on the subspaces $`V_{(\tau ,ϵ)}^+`$ and $`V_{(\tau ,ϵ)}^{}`$ gives the following addition theorems:
$$U(g)D_k^{(\tau ,ϵ)}U^{}(g)=\underset{n=\mathrm{}}{\overset{\tau ϵ}{}}t_{nk}^{(\tau ,ϵ)}(g)D_n^{(\tau ,ϵ)}$$
(49)
and
$$U(g)D_k^{(\tau ,ϵ)}U^{}(g)=\underset{n=\tau ϵ}{\overset{\mathrm{}}{}}t_{nk}^{(\tau ,ϵ)}(g)D_n^{(\tau ,ϵ)}.$$
(50)
(iii) On the subspaces $`V_{(\tau ,ϵ)}^0`$ the addition theorem reads
$$U(g)D_k^{(\tau ,ϵ)}U^{}(g)=\underset{n=\tau ϵ}{\overset{\tau ϵ}{}}t_{nk}^{(\tau ,ϵ)}(g)D_n^{(\tau ,ϵ)}.$$
(51)
Sandwiching both sides of (47), (49), (50) and (51) between the states $`l`$ and $`s`$ we get
$$\underset{m,t=0}{\overset{\mathrm{}}{}}U_{lm}(g)\overline{U_{st}(g)}(D_k^{(\tau ,ϵ)})_{mt}=\underset{n}{}t_{nk}^{(\tau ,ϵ)}(g)(D_n^{(\tau ,ϵ)})_{ls}$$
(52)
Multiplying (47), (49), (50) and (51) by $`U(g)`$ from the right and sandwiching them between the states $`l`$ and $`s`$ we get
$$\underset{m=0}{\overset{\mathrm{}}{}}U_{lm}(g)(D_k^{(\tau ,ϵ)})_{ms}=\underset{m=0}{\overset{\mathrm{}}{}}\underset{n}{}t_{nk}^{(\tau ,ϵ)}(g)(D_n^{(\tau ,ϵ)})_{lm}U_{ms}(g)$$
(53)
Multiplying (47), (49), (50) and (51) by $`U^{}(g)`$ and $`U(g)`$ from the left and right respectively and sandwiching them between the states $`l`$ and $`s`$ we get
$$(D_k^{(\tau ,ϵ)})_{ls}=\underset{m,t=0}{\overset{\mathrm{}}{}}\underset{n}{}t_{kn}^{(\tau ,ϵ)}(g)U_{ts}(g)\overline{U_{ml}(g)}(D_n^{(\tau ,ϵ)})_{mt}$$
(54)
where
$$(D_k^{(\tau ,ϵ)})_{mt}=\sqrt{\frac{m!}{t!}}f_k^{(\tau ,ϵ}(t)\delta _{m,t+2k+2ϵ}$$
(55)
for $`k0`$ and
$$(D_k^{(\tau ,ϵ)})_{mt}=\sqrt{\frac{t!}{m!}}f_k^{(\tau ,ϵ}(m)\delta _{m,t+2k+2ϵ}$$
(56)
for $`k<0`$.
Finally we like to give two simple specific examples: Let $`g=h(\alpha )`$, $`ϵ,k=0`$ in (51) that is $`\tau `$ is positive integer. Taking $`s,l=0`$ in (52) and (53) we get
$$P_\tau (\mathrm{cosh}\alpha )=\frac{1}{\sqrt{\pi }\mathrm{cosh}\frac{\alpha }{2}}\underset{n=0}{\overset{\mathrm{}}{}}\frac{\mathrm{\Gamma }(n+\frac{1}{2})}{n!}\mathrm{tanh}^{2n}\frac{\alpha }{2}F(2n,1+\tau ;1+n;\mathrm{sinh}^2\frac{\alpha }{2})$$
(57)
and
$$1=\underset{n=0}{\overset{\tau }{}}\frac{()^n!}{(n!)^2}\frac{(\tau +n)!}{(\tau n)!}\mathrm{tanh}^{2n}\frac{\alpha }{2}F(\tau ,1+\tau ;1+n;\mathrm{sinh}^2\frac{\alpha }{2})$$
(58)
respectively. $`P_\tau `$ in (57) is the Legendre function. |
warning/0003/nlin0003031.html | ar5iv | text | # Direct-interaction electrodynamics of a two-electron atom
## I Action-at-a-distance electrodynamics
The Wheeler-Feynmanelectrodynamics developed from the Schwarzschild-Tetrode-Fokker direct-interaction functional. Equations of motion are derived from Hamilton’s principle for the action integral
$`S=_im_ic𝑑s_i+e^2_{ij}\delta \left(x_ix_j^2\right)x_ix_j𝑑s_i𝑑s_j,`$
where the four-vector $`x_i(s_i)`$ represents the four-position of particle $`i`$ parametrized by arc-length $`s_i`$ , double bars indicate quadri-vector modulus $`x_ix_j^2(x_ix_j)(x_ix_j)`$ and the dot indicates the usual Minkowski relativistic scalar product of four-vectors. (integration is to be carried over the hole particle trajectories, at least formally). The above action integral describes an interaction at the advanced and retarded light-cones with an electromagnetic potential given by half the sum of the advanced and retarded Liènard-Wierchert potentials . Wheeler and Feynman showed that electromagnetic phenomena can be described by this direct action-at-a-distance theory in complete agreement with Maxwell’s theory as far as the classical experimental consequences. This direct-interaction formulation of electrodynamics was developed to avoid the complications of divergent self-interaction, as there is no self-interaction in this theory, and also to eliminate the infinite number of field degrees of freedom of Maxwell’s theory . It was a great inspiration of Wheeler and Feynman in 1945, that followed a lead of Tetrode and showed that with the extra hypothesis that the electron interacts with a completely absorbing universe, the advanced response of this universe to the electron’s retarded field arrives at the present time of the electron and is equivalent to the local instantaneous self-interaction of the Lorentz-Dirac theory. The action-at-a-distance theory is also symmetric under time reversal, as the Fokker action includes both advanced and retarded interactions. Dissipation in this time-reversible theory becomes a matter of statistical mechanics of absorption. The area of Wheeler-Feynman electrodynamics has been progressing slowly but steadily since 1945: Quantization was achieved by use of the Feynman path integral technique and the effect of spontaneous emission was successfully described in terms of interaction with the future absorber, in agreement with quantum electrodynamics. It was also shown that it is possible to avoid the usual divergencies associated with quantum electrodynamics by use of proper cosmological boundary conditions. As far as understanding of the dynamics governed by the equations of motion, the state of the art is as follows: The exact circular orbit solution to the attractive two-body problem was proposed in 1946 and rediscovered by Schild in 1962. The 1-dimensional symmetric two-electron scattering is a special case where the equations of motion simplify a lot and it has been studied by many authors, both analytically and numerically . In this very special case the initial value functional problem surprisingly requires much less than an arbitrary initial function to determine a solution manifold with the extra condition of bounded manifold for all times. It was shown that the solution is uniquely determined by the interelectronic distance at the turning point if this distance is large enough (this minimum distance curiously evaluates to 0.49 Bohr radii by the action-at-a-distance theory, much larger than about one classical electronic radius that one would naively guess). As a result of this theorem, there is a single continuous parameter (the positive energy) describing the unique non-runaway symmetric orbit at that given positive energy.
The Noether’s four-constant of motion derived from the Fokker Lagrangian involves an integral over the past history. For example in the case of a hydrogen atom this four-momentum constant evaluates to
$`P^\lambda `$ $`=`$ $`m_p\dot{x}_p^\lambda +eA^\lambda (x_p)+m_e\dot{x}_e^\lambda eA^\lambda (x_e)`$
$`2e^2{\displaystyle _\tau ^{\mathrm{}}}𝑑\tau _p{\displaystyle _{\mathrm{}}^\tau }𝑑\tau _e\stackrel{´}{\delta }\left(x_ex_p^2\right)(x_px_e)\dot{x}_e\dot{x}_p`$
$`+2e^2{\displaystyle _{\mathrm{}}^\tau }𝑑\tau _p{\displaystyle _\tau ^{\mathrm{}}}𝑑\tau _e\stackrel{´}{\delta }\left(x_ex_p^2\right)(x_px_e)\dot{x}_e\dot{x}_p,`$
where $`\stackrel{´}{\delta }`$ represents the derivative of the delta function. Notice that because of this delta function, only finite portions of the trajectory are involved: actually an extent of length $`t2r_{12}/c`$ approximately. This non-local constant will behave very differently from the local Coulombian energy, that is known to confine orbits of a negative energy within a maximum separation distance. In the case where the particles acquire a large separation (unbound state), the hole past history is involved ($`t2r_{12}/c\mathrm{}`$) in the determination of the non-local energy constant.
As regards the mathematical structure of the equations of motion, for the case of a two-electron atom the acceleration of electron 1 is given by
$$a_1(t)=\frac{e}{m\gamma }\{E\frac{v_1(t)}{c^2}Ev_1(t)+\frac{v_1(t)}{c}\times B\},$$
(1)
where $`e`$ and $`m`$ are the electronic charge and mass, $`\gamma `$ $`1/\sqrt{(1\frac{v_1^2(t)}{c^2})^{1/2}}`$and $`E`$ and $`B`$ are the total electric and magnetic fields produced by electron 2 and the nucleus. In the action-at-a-distance theory these fields are given by the average of the retarded and advanced Liènard-Wiechert fields, calculated with the instantaneous position of the stationary nucleus and the retarded and advanced positions of electron 2 at the times $`t_2=t_{}`$ , which is defined by the implicit condition
$`R_{}|r_2(t_{})r_1(t_1)|=c(t_{}t_1),`$
where the minus and plus signs are the conditions for the retarded and advanced times respectively. The partial electric fields of electron 2 acting on electron 1 at time $`t_{1\text{ }}`$are
$`E_{}(x_1,x_2,v_2,a_2)`$ $`=`$ $`{\displaystyle \frac{e(n_{}\beta _2)}{\gamma _2^2(1n_{}\beta _2)^3R_{}^2}}`$
$`{\displaystyle \frac{e}{c}}\left[{\displaystyle \frac{n_{}\times \left\{(n_{}\beta _2)\times \dot{\beta }_2\right\}}{(1n_{}\beta _2)^3R_{}}}\right],`$
where $`R_{}n_{}r_2(t_{})r_1(t_1)`$ , $`\beta _2v_2/c,(e),\gamma _2(1\beta _2^2)^{1/2}`$ and $`c`$ is the speed of light. The advanced field $`E_+`$ is obtained from the above expression by replacing $`t_{}`$ by $`t_+`$ and $`c`$ by $`c.`$ The partial magnetic fields of electron 2 are
$`B_{}=\pm n_{}\times E_{},`$
where the $`\pm `$ is to ensure an outgoing Poynting vector $`(cE\times B)`$ for the retarded fields and an incoming Poynting vector for the advanced fields. The total electric field in equation (1) must include also the instantaneous Coulomb electric field of the stationary nucleus.
Equation (1) can suggest a paradox about causality, as the force depends on the future of particle 2. In the following, and to finish this introduction, we show that equation (1), when written properly, becomes a functional differential equation with delayed argument only, as first observed in. To outline the essentials of the explanation, let us first ignore the field of the nucleus and take the nonrelativistic limit of (1) ($`v_1=0`$). In this approximation the electric field $`E`$ entering in equation (1) evaluates to $`E=0.5E_+(x_1,x_{2+},v_{2+},a_{2+})+0.5E_{}(x_1,x_2,v_2,a_2)`$. Then we note that one can use equation (1) as an equation of motion for particle 2 , by solving the rearranged form of (1),
$`eE_+(x_1,x_{2+},v_{2+},a_{2+})`$ $`=`$ (3)
$`2ma_1(t)eE_{}(x_1,x_2,v_2,a_2),`$
for the most advanced acceleration of particle 2, $`a_{2+}a_2(t+\frac{d_+}{c}).`$ In the above form it is clear that the right hand side involves only functions evaluated at times prior to the most advanced time, defined by $`s=t+\frac{d_+}{c}`$, and no further advanced information is necessary, eliminating the ghost of dependence on the future. In the same way, the causal equation of motion of particle 1 is to be produced from the equation for particle 2 by solving for the most advanced acceleration of particle 1. For the special case of 1-dimensional motion of two electrons, $`E_+=E_+(x_1,x_{2+},v_{2+})`$ depends only on the advanced velocity, and (LABEL:causal) can easily be solved for this advanced velocity as a function of the past history. In the 3-dimensional case there is an extra complexity, as the acceleration appears in the Liènard-Wiechert partial field $`E_+`$ in the form $`n_{12}\times (n_{12}\times a_{2+})/r_{12}`$. The bad news is that the component of the acceleration along the advanced normal can not be solved for from the value of the double-vector-product only. Because of this degeneracy, equation (LABEL:causal) is an algebraic-differential equation, and the null direction of the left hand side of (LABEL:causal) is a constraint to be satisfied by the right hand side (the scalar product with $`n_{12}`$ must vanish). The numerically correct way to integrate this type of equation is by use of the modern integrators for algebraic-differential equations like DASSL adapted for retarded equations (which has never been done yet) or by dealing directly with the algebraic constraint . According to the standard classification of G. A. Kamenskii, equation (LABEL:causal) belongs to the class of differential-difference equations of neutral type. Even though more complex, the motion is still causally determined by the past trajectory, as we wanted to demonstrate, the price being an algebraic neutral delay equation.
As far as initial conditions go, the general theory on delay equations tells us that we need to provide an initial $`C^2`$ function describing the position of particle 2 from $`s\frac{(d_++d_\mathrm{\_})}{c}=t\frac{d_\mathrm{\_}}{c}`$ up to the initial instant $`s=0`$. The information on particle 1 needed is also to be provided over twice the retardation lag seen by particle 1. This is a short piece of trajectory for bound nonrelativistic atomic orbits, but for a ionized state or a runaway orbit this can be the whole past history! Unless further simplifications or conditions are added, this is the generic problem at hand. The 3-dimensional cases of atomic interest (e.g. helium) have never been studied, and they are more complex than the 1-d scattering because one can have negative energy bound states for example. Most relevant for physics is the question of the conditions for the existence of a bounded manifold solution, which still needs to be understood in the general case (it would be very curious if they turned out to be a discrete set of negative energies). The only existing analytical result in the 3-dimensional case is the linear stability of the Schonberg-Schild circular orbits, resulting in an infinite number of unstable solutions to the characteristic equation. The numerical treatment ofthe exact neutral equations displays instabilities and is generally difficult. In the following we resort to the Darwin approximation not as much as a mathematical approximation to the action-at-a-distance electrodynamics, but as a physical approximation of Lorentz-invariant dynamics in the atomic (shallow) energy range.
## II Numerical Calculations for the Coulomb Dynamics
To introduce our numerical calculations, we start from the scale-invariant Coulomb limit of the Tetrode-Fokker-Wheeler-Feynman interaction: Let $`e`$ and $`m`$ be the electronic charge and mass respectively and $`Ze`$ the nuclear charge of our two-electron atom, which in this work is assumed to have an infinite mass. All our numerical work uses a scaling which exploits the scale invariance of the Coulomb dynamics: Given a negative energy, there is a unique circular orbit at that energy with frequency $`\omega _o`$ and radius$`R`$ related by $`e^2/(m\omega _o^2R^3)=1/(Z\frac{1}{4})\zeta (Z)`$. We scale distance, momentum, time and energy as $`xRx`$, $`pm\omega _oRp`$ , $`\omega _odtd\tau `$ and $`Em\omega _o^2R^2\widehat{H}`$, respectively. In these scaled units, the Coulomb dynamics of the two-electron atom is described by the scaled Hamiltonian
$$\widehat{𝐇}=\frac{1}{2}(|\stackrel{}{p}_1|^2+|\stackrel{}{p}_2|^2)+\zeta (Z)\{\frac{1}{r_{12}}\frac{Z}{r_1}\frac{Z}{r_2}\},$$
(5)
where $`r_1|\stackrel{}{x}_1|,r_2|\stackrel{}{x}_2|,r_{12}|\stackrel{}{x}_1\stackrel{}{x}_2|`$(single bars represent euclidean modulus) and $`\beta \omega _oR/c`$. For a generic non-circular orbit, $`\beta `$ plays the role of a scale parameter, and we recover the value of the energy in ergs through $`E=mc^2\beta ^2\widehat{H}`$. Notice that $`\beta `$ does not appear in the scaled Hamiltonian, which is the scale invariance property. From the scaled frequency $`\widehat{w}`$ and scaled angular momentum $`\widehat{l}`$ we can recover the actual values in CGS units by the formulas
$$w=\frac{mc^2\zeta (Z)\beta ^3}{e^2/c}\widehat{w},l=\frac{e^2/c}{\zeta (Z)}\frac{\widehat{l}}{\beta }.$$
(6)
The only other analytic constant of the Coulomb dynamics, besides the energy (5) is the total angular momentum, and this dynamics in chaotic and displays Arnold diffusion, as proved in for a similar three-body system.
The numerical calculations were performed using a 9th-order Runge-Kutta embedded integrator pair. We chose the embedded error per step to be $`10^{14}`$, and after ten million time units of integration the percentage changes in energy and total angular momentum were less than $`10^6`$. As a numerical precaution we performed the numerical calculations using the double Kustanheimo coordinate transformation to regularize single collisions with the nucleus. As these alone are not enough for faithful integration, we checked that there was never a triple collision, as the minimum inter-electronic distance was about $`0.3`$ units while the minimum distance to the nucleus was $`0.01`$ units for all the orbits considered in this work. We also checked that along stable non-ionizing orbits we can integrate forward up to fifty thousand time units, reverse the integration, go backwards another fifty thousand units and recover the initial condition with a percentile error of $`10^5`$. For longer times this precision of back and forth integration degenerates rapidly, which is due to the combined effect of numerical truncation and stochasticity. The question of how far in time the numerical trajectories approximate shadowing trajectories in the present system is far from trivial , but we assume it to be a time at least of the order of these one hundred thousand units. (Energy conservation of one part in a million is achieved for much longer times, even one billion time units).
The study of orbits of a two-electron atom was greatly stimulated by the recent interest in semiclassical quantization, and these studies discovered two types of stable zero-angular-momentum periodic orbits for helium ($`Z=2`$): the Langmuir orbit and the frozen-planet orbit . A detailed study of the non-ionizing orbits of Coulombian helium was initiated in reference for plane orbits, and we describe some of their results below. There are basically two types of non-ionizing orbits: Symmetric if $`r_1=r_2`$ for all times and asymmetric if $`r_1r_2`$ generically. Symmetric orbits are produced by symmetric initial conditions like for example $`x_1(0)=x_2(0)`$ and $`v_1(0)=v_2(0)`$ or $`x_1(0)=x_2(0)`$ and $`v_1(0)=v_2(0)`$ with $`x_1(0)v_1(0)=0`$ Because (5) is symmetric under particle exchange, these orbits satisfy $`r_1=r_2`$ at all times, and therefore cannot ionize if $`H<0`$ (both electrons would have to ionize at the same time, which is impossible at negative energies). For example the double-elliptical orbits (two equal ellipses symmetrically displaced along the x-axis) discussed in are in this class. Double-elliptical orbits are known to be unstable and we find that they ionize in about one hundred turns because of the numerical truncation error. Most symmetric plane orbits are very unstable to asymmetric perturbations, with the exception of the Langmuir orbit for a small range of $`Z`$ values around $`Z=2`$ .
The simplest way to produce an asymmetric non-ionizing plane orbit is from the initial condition $`x_1=(r_1,0,0)`$ , $`\dot{x}_1=(0,v_1\sqrt{4/7},0)`$ , $`x_2=(1.0,0,0)`$ , $`\dot{x}_2=(0,\sqrt{4/7},0)`$, as suggested in . In Figure 1 we show the electronic trajectories for the first three hundred scaled time units along a two-dimensional non-ionizing orbit of $`Ca^{+18}(Z=20)`$ with $`r_1=1.4`$ and $`v_1=1.28442`$ in the above defined condition. We used a numerical refining procedure to finely adjust $`v_1`$as to maximize the non-ionizing time and this condition of Figure 1 does not ionize for one million time units. The orbit survives that far only for a very sharp band of values of $`v_1`$, other neighboring values producing quick ionization. This orbit was named double-ring torus in . The other possible type of non-ionizing orbit resulting from the above initial condition, depending on $`(r_1,v_1),`$ is what was named braiding torus in reference , with both electrons orbiting within the same region. A search over $`(r_1,v_1)`$ was conducted in , and it was found that most values of $`(r_1,v_1)`$ produce quick ionization except for a zero-measure set of $`(r_1,v_1)`$ values where braiding tori or double ring orbits are found. This suggests the general result that non-ionizing orbits are rare in phase space.
To search for general tridimensional non-ionizing orbits in phase space, it is convenient to introduce canonical coordinates $`\stackrel{}{x}_d`$ and $`\stackrel{}{x}_c`$
$`\stackrel{}{p}_d`$ $``$ $`(\stackrel{}{p}_1\stackrel{}{p}_2)/\sqrt{2},\stackrel{}{x}_d(\stackrel{}{x}_1\stackrel{}{x}_2)/\sqrt{2}`$ (7)
$`\stackrel{}{p}_c`$ $``$ $`(\stackrel{}{p}_1+\stackrel{}{p}_2)/\sqrt{2},\stackrel{}{x}_c(\stackrel{}{x}_1+\stackrel{}{x}_2)/\sqrt{2}.`$ (8)
Initial conditions with $`\stackrel{}{x}_c=`$ $`\stackrel{}{p}_c=0`$ describe double-elliptical orbits (and circular as a special case). To generate an elliptical initial condition, we exploit the scale invariance and set the energy to minus one. It is easy to check that elliptical orbits of the Hamiltonian (5) with an energy of minus one must have a total angular momentum of magnitude ranging from zero to two. To exploit the rotational invariance of (5), we can choose the plane defined at $`\stackrel{}{x}_c=`$ $`\stackrel{}{p}_c=0`$ by the angular momentum $`\stackrel{}{L}=`$ $`\stackrel{}{x}_d\times \stackrel{}{p}_d+\stackrel{}{x}_c\times \stackrel{}{p}_c=\stackrel{}{x}_d\times \stackrel{}{p}_d`$ to be the $`xy`$ plane. On this $`xy`$ plane a single number $`0<|\stackrel{}{x}_d\times \stackrel{}{p}_d|<2`$ (the angular momentum), determines completely the elliptical orbit. The next step in producing a generic orbit is to add all possible perturbations along $`\stackrel{}{x}_c`$ and $`\stackrel{}{p}_c`$ to the chosen elliptical orbit. These are six directions and once we are looking for bound oscillatory orbits, we can choose $`z_c=0`$, once $`z_c`$ has to cross the $`xy`$ plane at some point. These are five numbers to vary and plus the angular momentum of the elliptical orbit it totals six parameters. Our numerical search procedure consists in varying these six parameters over a fine grid, integrating every single initial condition until the distance from one electron to the nucleus is greater than twenty units, which is our ionization criterion. This criterion fails if the orbit has a very low angular momentum because these can go far away from the nucleus and come back, and therefore our search possibly misses low-angular-momentum non-ionizing orbits. As the majority of the initial conditions ionize very quickly, this search procedure is reasonably fast. We first perform a coarse search for ionization times above one thousand units and then refine in the neighborhood of each surviving condition to get conditions that do not ionize after one million time units.
Using the above numerical search procedure we found the tridimensional non-ionizing initial condition of Figure 2 for helium, a tridimensional double-ring orbit generated by the initial condition
$`x_1`$ $`=`$ $`(1.2812617,0.0147169,0.0)`$
$`x_2`$ $`=`$ $`(1.5511484,0.0147169,0.0)`$
$`p_1`$ $`=`$ $`(0.0194868,0.4398889,0.1094930)`$
$`p_2`$ $`=`$ $`(0.0194868,0.7972467,0.1094930),`$
which does not ionize before ten million turns. (After the search and refinement, we scaled this orbit’s energy to minus one, for later convenience). We also found the non-ionizing orbit orbit of Figure 3 for H-minus ($`Z=1`$), a tridimensional orbit generated by the condition
$`x_1`$ $`=`$ $`(1.9776507,0.3411364,0.0)`$
$`x_2`$ $`=`$ $`(1.2288121,0.3411364,0.0)`$
$`p_1`$ $`=`$ $`(0.0421302,0.5057782,0.2810539)`$
$`p_2`$ $`=`$ $`(0.0421302,0.4132970,0.2810539),`$
which does not ionize before one million turns (Coulombian energy of this condition is also minus one). This last orbit is fragile and numerically harder to find: as the first electron has an orbit very close to the positive $`Z=1`$ charge, there remains only a dipole field to bind the second electron. As the outer electron is much slower in the scaled units, we had to plot the first $`10000`$ time units of evolution to display the generic features of the trajectory. Non-ionizing orbits of $`H^{}`$ are very rare in phase space, which is reminiscent of the quantum counterpart, as the $`H^{}`$ ion is known to have only one quantum bound state at $`E0.55mc^2\alpha ^2`$, very close to the ionization threshold $`(0.5mc^2\alpha ^2)`$.
One remarkable fact about these non-ionizing orbits is that they all have a very sharp Fourier transform. This property makes them approximately quasi-periodic orbits. For example in Figure 4 we plot the fast Fourier transform of the orbit of Figure 2, performed using $`2^{16}`$ points. (It seems that there are at least two basic frequencies in the resonance structure of Figure 4). Even though these orbits look like quasi-periodic tori, there seems to be a thin stochastic tube surrounding each orbit, as evidenced by a small positive maximum Lyapunov exponent. We calculated numerically this maximum Lyapunov exponent by doubling the integration times up to $`T=10^7`$ and found that the exponent initially decreases but then saturates to a value of about $`0.001`$ for the orbits of Figures 1, 2 and 3. The gravitational three-body problem has recently been proved to display Arnold diffusion, and this numeriacally calculated positive Lyapunov exponent suggests that the same is true for the two-electron Coulombian atom.
## III Numerical calculations for the Darwin Dynamics
The numerical integrations in this section are performed using the Darwin approximation. The Darwin equations of motion are a $`\beta ^2`$ perturbation of the Coulomb dynamics, of size $`\beta ^210^4`$ for atomic energies. In the scaled units of section II the Darwin Hamiltonian is the following $`\beta ^2`$ perturbation of Hamiltonian (5)
$`\widehat{𝐇}_D`$ $`=`$ $`{\displaystyle \frac{1}{2}}(|\stackrel{}{p}_1|^2+|\stackrel{}{p}_2|^2)+\zeta (Z)\{{\displaystyle \frac{1}{r_{12}}}{\displaystyle \frac{Z}{r_1}}{\displaystyle \frac{Z}{r_2}}\}`$ (11)
$`{\displaystyle \frac{\zeta (Z)\beta ^2}{2r_{12}}}[\stackrel{}{p}_1\stackrel{}{p}_2+(\widehat{n}_{12}\stackrel{}{p}_1)(\widehat{n}_{12}\stackrel{}{p}_2)]`$
$`{\displaystyle \frac{\beta ^2}{8}}[|\stackrel{}{p}_1|^4+|\stackrel{}{p}_2|^4],`$
where $`\widehat{n}_{12}(\stackrel{}{x}_1\stackrel{}{x}_2)/r_{12}`$. The second line represents the Biot-Savart magnetic interaction plus the first relativistic correction to the static electric field and the last line describes the relativistic mass correction. Notice that these are both proportional to the small parameter $`\beta ^2`$, which makes them a small scale-dependent perturbation on the scale invariant Coulomb Hamiltonian (first line). It is possible to regularize the Darwin equations with the same double-Kustanheimo transformation, only that here one needs to define the regularized time using the higher powers $`dt=r_1^2r_2^2ds`$, instead of the lower powers $`dt=r_1r_2ds`$ used to regularize the Coulomb equations.
The main question we address numerically in this section is the dependence of the stability of a non-ionizing orbit with the energy scale of the orbit . Here we use the word stability to mean ionization-stability: We call an initial condition ionization-stable if any small perturbation of it produces another non-ionizing orbit. The scale-dependent Darwin terms (of size $`\beta ^2)`$ produce significant deviations from the Coulomb dynamics only in a time-scale of order $`1/\beta ^2,`$ which we find numerically to be the typical time for a non-ionizing Coulombian initial condition to ionize along the Darwin vector field. This poses a numerical difficulty if $`\beta `$ is too small because one has to integrate the orbit for very long times to investigate the stability. It turns out that ionization-stable orbits can be found at larger values of $`\beta `$ for larger values of $`Z`$ . Here the dynamical stability mechanism is reminiscent of quantum atomic physics, where the values of $`\beta `$ vary with the nuclear charge as $`\beta Z/137`$. Large values of $`Z`$ facilitate the numerical procedure and in the following we present the numerical investigation of the stability of non-ionizing orbits starting from the large $`Z`$ case.
Let us start with the $`Z=20`$ calcium ion two-electron system along the non-ionizing orbit of Figure 1 by fixing $`r_1=1.4`$ and $`v_1=1.28442.`$ in the condition defined in section II. To test the stability of the orbit at each value of $`\beta `$ we add a random perturbation of average size $`\beta ^2`$ to the initial condition and integrate the Darwin dynamics until either we find ionization or the time of integration is greater than $`10^7`$ time units We repeat this for at least twelve randomly chosen perturbations (because of the twelve degrees of freedom) and the minimum time to ionization is plotted in Figure 5 as a function of $`\beta `$. It can be seen that only for a narrow set of values around $`\beta 0.037`$ this minimum time to ionization was greater than $`10^6`$. or the other values it decreases rapidly to a value of about $`10^3`$. One could argue that for the other values of $`\beta `$ the non-ionizing initial condition has shifted away from the $`v_1=1.28442`$ initial condition and this being the reason that our orbit ionized. To test this, we fixed $`\beta `$ at a ”bad ” value for example $`\beta =0.02`$ and varied the plane initial condition in the neighborhood of this condition of Figure 1. We found that the minimum time to ionization was always about $`10^3`$ (also the maximum time before ionization was about $`10^3`$). We also searched in a bigger neighborhood, of size proportional to $`\beta `$. This suggests the interpretation that for the special resonant value of $`\beta =0.037`$ the net diffusive effect of the scale-dependent term vanishes, allowing a non-ionizing perturbed manifold. In order to have a direct interpretation (in atomic units) of the scale parameter $`\beta `$, it is convenient to scale to minus one the energy of the initial condition of Figure 1 (by exployting the Coulombian scale invariance). After this, the energy of the orbit in ergs evaluates to $`E=mc^2\beta ^2\widehat{H}=mc^2\beta ^2`$, and for $`\beta =0.037`$ this is approximately $`24.59`$ atomic units. The total angular momentum of this orbit is $`l_z=7.94\mathrm{}`$. This orbit’s energy is above the ionization continuum of the ion, $`E=mc^2\alpha ^2Z^2/2=200`$ atomic units, but it is still in the quantum range. It serves nevertheless to demonstrate that this dynamical system might exhibit non-ionizing stable orbits only at very sharply defined energy values.
For the orbits of Figures 2 and 3, the above procedure becomes prohibitively slow, as the value of $`\beta `$ are much smaller and one must integrate for very long times, much beyond the estimated shadowing time. To partially overcome this we used a larger amplitude random perturbation (of average size $`20\beta ^2`$), to produce faster ionization. The drawback with this is that the minimum ionization time does not show pronounced peaks, only the average ionization time still showing a signature of scale dependence. In Figure 6 we show this average time for the orbit of Figure 3. This property of sharply defined energies can possibly be found for the lower-lying energies below the ionization threshold as well. These orbits would involve configurations where the electrons come very close to the nucleus and acquire a large velocity. Even though our integrator is regularized, the correct physical electronic repulsion is greatly amplified when one electron has a relativistic velocity and the Darwin approximation can not describe the physics then. Actually, it is known that the Darwin interaction can produce unphysical effects when pushed to relativistic energies. We therefore do not expect to find these low-lying atomic energy scales with the present Darwin approximation and shall be contempt with these interesting result already.
For the same reason given above, we do not study here the frozen-planet periodic orbit (the two electrons performing one-dimensional periodic motion on the same side of the nucleus, with the inner electron rebounding from the origin, an artifact of regularization). The main problem being the failure of the Darwin approximation, as the inner particle goes to the speed of light. The correct relativistic dynamics can actually produce a new physical inner turning point very close to the origin but not at the origin as the regularized motion, and we discuss elsewhere.
Last, we consider the non-ionizing symmetric periodic orbit called the Langmuir orbit, where the two electrons perform symmetric bending motion shaped approximately like a semi-circle. For the Coulomb two-electron atom with $`Z=2`$ this orbit was found to have a zero maximum Lyapunov exponent. The orbit is therefore neutrally stable, which is the best one can expect from a periodic orbit of a Hamiltonian vector field. (Absolute stability violates the symplectic symmetry, which says that to every stability exponent $`\lambda `$ one should have a $`1/\lambda `$ exponent). It is a simple matter to obtain the Langmuir-like orbit for the Darwin Hamiltonian at any given value of $`\beta `$: all it takes is a little adjusting in the neighborhood of the Coulombian Langmuir condition. We attempted to investigate numerically any scale-dependent diffusion away from this Darwin-Langmuir condition for $`\beta `$ in the atomic range, but again the numerics is prohibitively slow at the time of writing this work.
## IV Conclusions and Discussion
The simplified dynamical mechanism behind resonant non-ionization seems to go intuitively as follows: The peculiar scale-invariant Coulomb dynamics determines the non-ionizing orbits within narrow ”stochastic tubes ”. The next step is the action of the small scale-dependent relativistic corrections that produce a slow diffusion of the orbit out of the thin tube in a time of the order of $`1/\beta ^2`$. After this, quick ionization follows. Only at very special resonant values of $`\beta `$ the relativistic terms leave the orbit within the tube, a resonant effect that depends on $`\beta `$, fixing the energy scale. In the literature, the escape to infinity from simpler to understand two-degree-of-freedom systems has been attributed to cantori, which, as is well known, can trap chaotic orbits near regular regions for extremely long times. In the present larger dimensional case it appears that resonances are also controlling the escape to infinity of one electron by the existence of extra resonant constants of motion. This seems to be in agreement with the numerical results of very sharp peaks for the minimum ionization time. We have tried to concentrate on the physics described by this combination of chaotic dynamics on a two-electron atom with inclusion of relativistic correction, while discussing this highly nontrivial result of nonlinear dynamics.
In references we noticed that a simple resonant normal form criterion gives a surprisingly good prediction for the discrete atomic energy levels of helium. The resonant structure was calculated using the Darwin interaction (11), which is the low-velocity approximation to both Maxwell’s and Wheeler-Feynman’s electrodynamics. As we saw in section II, the Coulombian non-ionizing orbits are far from circular, and these orbits would radiate even in dipole according to the time-irreversible Maxwell’s electrodynamics (circular orbits radiate only in quadrupole but are linearly unstable). It becomes then clear that the heuristic results of can only have a physical meaning in the context of a time-reversible theory (as the action-at-a-distance electrodynamics for example).
The combination of chaotic dynamics with relativistic invariance has never been explored numerically, and most known Lorentz-invariant dynamical systems are for one particle and possess trivially integrable dynamics. The situation gets unexpectedly much more complicated for more than one particle (apart from the trivial non-interacting many-particle system): Due to the famous no-interaction theorem, the relativistic description of two directly interacting particles is impossible within the Hamiltonian formalism and its set of ten canonical generators for the Poincare group . Description of interacting particles is possible only in the context of constraint dynamics, with eleven canonical generators and with the Dirac bracket replacing the Poisson bracket. For example the relativistic action-at-a-distance equations for two interacting electrons are non-local and possess only infinite-dimensional constrained Hamiltonian representations. The interested reader should consult some recently found two-body direct-interaction relativistic Lagrangian dynamical systems as well as the constraint-dynamics direct-interaction models recently used in chromodynamics and two-body Dirac equations. The nonlinear dynamics of these models could display interesting and so far inexplored dynamical behaviour.
It would be natural to wonder if one can find an analogous scale-dependent dynamics for a dynamical system describing the hydrogen atom, apparently the simplest example of Lorentz-invariant two-body relativistic dynamics of atomic interest. It turns out that hydrogen is not simpler than helium at all, but it appears to us that there is an essential difference which has actually made the interesting dynamics of a two-electron atom amenable to study already within the Darwin approximation: In a two-electron atom orbits with a negative energy can ionize, while in hydrogen this might be possible only if one includes all orders of the relativistic action-at-a-distance interaction. (As we saw in section I, the ”Noether’s energy constant” involves a segment of the past trajectory, and a negative value does not forbid ionization). Ionization with a negative energy would be impossible for hydrogen within the Darwin approximation (unless the electron goes to the speed of light). This is indication that in hydrogen the essential physics described by the action-at-a-distance electrodynamics is of non-perturbative character. The paradoxical result of the infinite linear instability of circular orbits in atomic hydrogen is another warning of this non-perturbative dynamics.
## V Acknowledgments
The author acknowledges the support of FAPESP, proc. 96/06479-9 and CNPQ, proc. 301243/94-8(NV). |
warning/0003/hep-th0003063.html | ar5iv | text | # Special Properties of Five Dimensional BPS Rotating Black Holes
## 1 Introduction
One of the most appealing features of String Theory is that it provides a finite quantum theory of gravity which reduces to the well tested General Relativity in the appropriate limit. Quantum gravitational effects become particularly important in at least two physical situations - the early universe and black holes physics - which are therefore laboratories to test the novel stringy physics. But whereas some significative progress has been done in understanding microscopics of black holes, stringy cosmology is still in a very early stage. One important reason for this distinction is supersymmetry.
Supersymmetry has allowed quantitative mapping between classical gravitational configurations and quantum states of strings. But one pays a price: preservation of supersymmetry requires the initial classical configuration to be dynamically very simple. In particular, it must contain a causal Killing vector field, in order to possess a Killing spinor. Whereas this is not an obstacle in understanding some features of black holes, since we may study extreme Reissner-Nordström (RN) type solutions, it becomes a major obstacle in studying cosmology, since there seems to be no easy way to reconcile supersymmetry with an expanding universe.
Even within the domain of black holes one has the natural desire of understanding more complex configurations than the aforementioned ones, in particular to include rotation, which we expect will be present in the black holes we find in Nature. Rotation makes the dynamical properties of a black hole spacetime richer, and therefore ‘less compatible’ with supersymmetry. In fact, regular (on and outside a horizon), rotating, supersymmetric black hole solutions are rare configurations in string theory. Let us consider the Heterotic string theory, compactified on a $`T^{10D}`$ torus to $`9D4`$ spacetime dimensions. For $`9D6`$, the low energy effective field theory will contain $`(362D)`$ one-form gauge potentials, giving rise to electric charges $`Q_a`$, $`a=1\mathrm{}362D`$. Of course, these charges have no absolute meaning in themselves; they can be mixed together by the $`O(10D,26D)`$ symmetry of the low energy effective action. In particular, the subgroup $`[SO(10D)\times SO(26D)]/[SO(9D)\times SO(25D)]`$ has dimension $`(342D)`$. Therefore, a $`9D6`$ dimensional black hole solution classified by mass, the maximal number of angular momentum parameters, $`l_i`$, $`i=1\mathrm{}[(D1)/2]`$=rank of $`SO(D1)`$, and two electric charges can be used to generate a solution with all $`(36D)`$ charges, by the use of the above subgroup. The solution with two charges is called the generating solution. Since the $`O(10D,26D)`$ symmetry does not act on the D dimensional metric, the geometry for the most general electrically charged rotating black hole solution can be seen at the level of the generating solution.
In $`D=5`$, the (reduced) Neveu-Schwarz field can be Hodge dualized to couple to ‘0-branes’. It carries a magnetic charge for a black hole. Hence, the generating solution will have 3 charges, and the most general electrically charged black hole 26 electric and 1 magnetic charge. In $`D=4`$, the one form fields also carry magnetic charges for black holes. Hence the generating solution has 4 charges and the most general black hole solution 56 charges, 28 electric and 28 magnetic<sup>2</sup><sup>2</sup>2The four dimensional case is actually more complex and the generating solution should have five charges; a case which has not been dealt with in the literature ..
The geometry of these generating solutions in the BPS limit has the following behaviour:
In $`D=4`$ ; the spacetime with non-vanishing angular momentum yields a naked (timelike) singularity. Putting all charges equal it reduces to the extreme (i.e. $`Q^2=M^2`$ and arbitrary $`J`$) Kerr-Newman (KN) spacetime of Einstein-Maxwell theory;
In $`D=5`$; a regular rotating black hole spacetime . Putting all charges equal it reduces to the Breckenridge-Myers-Peet-Vafa (BMPV) spacetime , of $`N=2`$, $`D=5`$ Supergravity;
In $`D6`$; a naked singularity. When only one $`l_i`$ is non-zero it reduces to a better behaved null singularity .
Regularity, therefore, distinguishes five dimensional, BPS rotating black holes. But why?
In section 2 we perform a comparative analysis between the extreme KN black hole and the BMPV black hole. In particular, we emphasize how a Hodge self duality condition, which can be implemented in a four dimensional transverse space to the world volume, allows for the harmonic function in the metric and the Sagnac connection to be independent. This is essential for the properties of the solution. We then give a simple example of using such duality condition. Specifically we find a ‘supersymmetric’ Brinkmann wave type solution to $`D=6`$ pure gravity and mention the corresponding five dimensional Kaluza-Klein rotating black hole.
In section 3 we exhibit the $`D=10`$ description of five dimensional, BPS, rotating black holes. We also analyse the interesting causal structure of the $`D=5`$ black holes from the ten dimensional perspective, finding that the unavoidable Closed Timelike Curves (CTC’s) of the five dimensional solution acquire a trivial character in ten dimensions (in the language of ). As far as we are aware this is the first example of resolving causal anomalies in higher dimensions. In order to give a complete description of the causal structure of this black holes, we review the string theory dual picture of the compactified spacetime in terms of an $`N=4`$ Super Conformal Field Theory, which has allowed the microscopic computation of the entropy . The point we wish to stress is the loss of unitarity in this theory when the spacetime undergoes a loss of causality. This unitarity bound can be seen by looking at general unitarity requirements for the conformal weights of SCFT operators. Alternatively one may examine the representations of an $`N=2`$ SCFT and the condition for these to be unitary, that arises from analysing the sign of the determinant of the Verma module.
Spacetimes where causality is violated may have odd effects related to geodesic motion. One such example is the “totally imprisoned incompleteness” found for the Lorentzian Taub-NUT metric . The phenomenon consists in the existence of a family of null geodesics which are imprisoned inside a compact region of spacetime whose boundary they reach within finite affine length. Such geodesics are therefore incomplete. However, they meet neither an s.p. (divergence of a curvature invariant) nor a p.p. singularity (divergence of the components of the Riemann tensor in a parallelly propagated frame).
The hypersurface from which they cannot be extended separates a causally well behaved spacetime region (the region initially found by Taub) from the outer reaches where closed timelike curves exist. We should remark, however, that there exists another family of null geodesics that passes through these surfaces.
In section 4.1 we explore an interesting effect arising for the BMPV spacetime whenever CTC’s arise outside the horizon - the over-rotating case - firstly pointed out in . The phenomenon -a ‘repulson’ effect- is shown still to be present when we consider test particles with charge, naturally following “charged geodesics”. For scalar waves we also try a non-minimal coupling to the geometry, in the Klein-Gordon equation, by including the Ricci scalar. Separation of variables can still be achieved, and a radial equation obtained. We conclude the universality of the effect for non accelerated observers. As for the Taub-NUT case we could find neither p.p. nor s.p. singularities, and nothing special seems to occur for an accelerated observer trying to overcome this natural obstacle. We also note that this repulson behaviour is distinct from singular repulsons recently studied and resolved in the context of AdS/CFT .
The uniqueness or ‘no-hair’ theorems for black holes tell us that a small set of independent quantities completely determines black hole spacetimes. Within Einstein-Maxwell theory such set includes solely ADM mass ($`M`$), angular momentum ($`J^i`$), electric ($`Q`$) and magnetic ($`P`$) charges. In spite of such restriction, black holes may have other multipole moments, albeit not independent. The magnetic dipole moment ($`\mu ^i`$) is the most natural one, since we expect any electrically charged rotating object to have it. In $`D=4`$ it is related to the previous quantities by $`\mu ^i`$ $`QJ^i/(2M)`$, therefore defining a constant of proportionality which is called the gyromagnetic ratio.
In ordinary electrodynamical systems, it follows from the definitions that $`g=1`$ for any rotating homogeneous distribution of charges. But for four dimensional Kerr-Newman black holes, $`g=2`$ in analogy with the quantum mechanical value for the electron (up to loop quantum corrections). The same value for $`g`$ is found for other heterotic black holes in four dimensions <sup>3</sup><sup>3</sup>3The charges for the Kerr-Newman-Sen black hole and for the ‘electron’ have different origins, however, in heterotic string theory. Therefore there is a distinction in their gyromagnetic ratios ., and some p-brane solutions in other dimensions, namely the membrane in $`D=11`$ . However, this is by no means universal. Kaluza-Klein black holes have a gyromagnetic ratio that depends on the several metric parameters . In the limit of large electric charge and vanishing magnetic charge, $`g`$ approaches one. This is the natural value for massive Kaluza-Klein modes in five dimensional Kaluza-Klein theory . The classical value arises because the charge is orbital motion in the compact direction. Another Kaluza-Klein black hole is the Dirichlet 0-brane supergravity solution in $`D=10`$. It is known to have $`g=1`$ . In fact, it is well known that the identification of the spectrum of IIA D0-branes with the Kaluza-Klein modes of the 11D graviton was a major guideline in the discovery of M-theory .
In finding the value of $`g`$ for the $`M2`$ and $`D0`$ branes, the superpartners technique was used. This method is usually applied to static BPS solutions, yielding a new solution with fermionic hair.<sup>4</sup><sup>4</sup>4The new configuration is only a solution to the full supergravity equations of motion if the method is carried out to all orders. Otherwise it solves the equations of motion only to some order in fermionic parameters. The interpretation of these “superpartners” is ambiguous due to the existence of bilinears of odd Grassmann numbers in the bosonic fields. Nevertheless, the method is useful in reading quantities like the gyromagnetic ratio. The technique was first used in and applied to the extreme RN background. The superpartners were found to have $`g=2`$.
In section 4.2, we compute the gyromagnetic ratio for both the BMPV background and its superpartners showing they are different. This is in contrast with the four dimensional case, where superpartners of the extreme RN background have the same $`g`$ as the Kerr-Newman. We then analyse the problem of a Dirac fermion in the BMPV spacetime and show it is isomorphic to the problem of a non-minimally coupled Dirac field in flat space. This coupling alters the usual value for $`g`$, creating a matching between the behaviour of the elementary particle and either the BMPV black hole or the superpartner of the static $`D=5`$ RN spacetime.
We close with a discussion.
## 2 BMPV versus Extreme KN and the self-duality condition
In $`D=4,5`$ the simplest supergravity theories containing a Maxwell field are the $`N=2`$ theories. The bosonic truncations of the general theories are, respectively, the Einstein-Maxwell theory and Einstein-Maxwell-Chern-Simons theory, with actions given by
$$𝒮^{(4)}=\frac{1}{16\pi G_4}d^4x\sqrt{g}\left[R\frac{1}{4}F^2\right],𝒮^{(5)}=\frac{1}{16\pi G_5}d^5x\left[\sqrt{g}(RF^2)\frac{2}{3\sqrt{3}}AFF\right].$$
(2.1)
Finding supersymmetric solutions to these theories involves solving the gravitino variation equations which are, respectively,
$$Dϵ\frac{1}{4}F_{ab}\mathrm{\Gamma }^{ab}\mathrm{\Gamma }ϵ=0,Dϵ+\frac{i}{4\sqrt{3}}\left(e^a\mathrm{\Gamma }_a^{bc}4e^b\mathrm{\Gamma }^c\right)F_{bc}ϵ=0,$$
(2.2)
where we have denoted the covariant derivative acting on spinors by $`Dϵ=dϵ+1/4w_{ab}\mathrm{\Gamma }^{ab}ϵ`$. For the four dimensional theory all solutions possessing Killing spinors were found by Tod . They split into two families, according to the existence of a null or a timelike Killing vector field. The former type corresponds to a family of gravitational waves, carrying electromagnetic fields, with the standard plane-fronted waves with parallel rays (pp-waves) arising as a special case. The latter case corresponds to the Israel-Wilson-Perjés (IWP) metrics . The IWP solutions can be written in the form:
$$\begin{array}{c}ds^2=|H|^2(dt+w_idx^i)^2+|H|^2\delta _{ij}dx^idx^j,\hfill \\ \\ F=_i(H^1)dtdx^i+\frac{1}{2}\left(|H|^2\delta ^{kl}ϵ_{ijk}_l(H^1)+2w_i_j(H^1)\right)dx^idx^j.\hfill \end{array}$$
(2.3)
The latin indices denote spatial coordinates and take values $`1,2,3`$. We are using cartesian coordinates for the spatial metric. The complex function $`H(x^i)`$ is required to solve Laplace’s equation in $`E^3`$, and $`|H|`$ represents its modulus. It completely determines the solution since the rotation vector, $`w^i`$, is determined by
$$\times 𝐰=i(HH^{}H^{}H),$$
(2.4)
where the curl is taken for $`E^3`$ and ‘\*’ denotes complex conjugation. The symbols $`,`$ denote, respectively, real and imaginary part of a complex quantity.
There are two distinct families within the IWP solutions, according to how one complexifies $`H(x^i)`$ . Complexifying the constants in the harmonic function leads to a multi-object generalisation of the charged Taub-NUT solution, whereas complexifying the coordinates in a specific way leads to the multi-object generalisation of the extreme (i.e. $`Q^2=M^2`$, where M,Q are the ADM mass and charge of the solution) Kerr-Newman (KN) spacetime. The one object limit of the latter - the usual KN solution - is obtained by choosing
$$H=1+\frac{M}{\sqrt{x^2+y^2+(zia)^2}},$$
(2.5)
and contact with the standard form in Boyer-Linquist coordinates is made by changing from cartesian to oblate spheroidal coordinates, $`(r,\theta ,\varphi )`$, with transformations
$$x+iy=\sqrt{(rM)^2+a^2}\mathrm{sin}\theta e^{i\varphi },z=(rM)\mathrm{cos}\theta ,$$
(2.6)
where the constant $`a`$ will be the angular momentum parameter.<sup>5</sup><sup>5</sup>5Aside: Within the Kerr-Newman family, only the non-rotating case or the extreme $`Q^2=M^2`$ case can be expressed in the IWP form (2.3). To see this notice that a necessary condition is that after completing the square, $`(dt+w)^2`$, the remaining (i.e. transverse) metric must be conformally flat. Since the Weyl tensor is identically zero in three dimensions, it cannot be used to check conformal flatness. But there is a conformal tensor in three dimensions, firstly defined by Eisenhart but often called the York tensor , which is defined as
$$Y_{abc}2_{[c}R_{b]a}+\frac{1}{2}g_{a[c}_{b]}R.$$
(2.7) The square parenthesis denote antisymmetrization with unit weight. For the Kerr-Newman spacetime the non-vanishing components for the York tensor applied to the transverse metric are proportional to $`a^2(Q^2M^2)`$ as claimed.
For the five dimensional case, not all supersymmetric solutions are known, but at least the equivalent to pp-waves and to the extreme Kerr-Newman are. The latter is known as the BMPV solution and given by the fields
$$\begin{array}{c}ds^2=H^2[dt+a_idx^i]^2+H\delta _{ij}dx^idx^j,\hfill \\ \\ F=\frac{\sqrt{3}}{2}[_i(H^1)dx^idt+_i(H^1a_j)dx^idx^j].\hfill \end{array}$$
(2.8)
The latin indices run now from 1 to 4. The solution is determined by two independent quantities. The real function $`H(x^i)`$ is required to be harmonic on $`E^4`$, and the $`E^4`$ one form $`a`$ is required to have a Hodge self-dual field strength. In hyperspherical coordinates $`(\rho ,\theta ,\varphi _1,\varphi _2)`$, obtained from cartesian coordinates as
$$x^1+ix^2=\rho \mathrm{cos}\theta e^{i\varphi _1},x^3+ix^4=\rho \mathrm{sin}\theta e^{i\varphi _2},$$
(2.9)
the metric on $`E^4`$ can be written as
$$ds_{E^4}=d\rho ^2+\rho ^2(d\theta ^2+\mathrm{sin}^2\theta d\varphi _1^2+\mathrm{cos}^2\theta d\varphi _2^2),$$
(2.10)
and the functions in (2.8) expressed as
$$a=\frac{J}{2\rho ^2}(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2),H=1+\frac{\mu }{\rho },$$
(2.11)
where $`J`$, $`\mu `$ are constants.
Comparing the two solutions, the crucial difference is that the KN is determined by one complex function, whereas the BMPV is determined by one real function and a self dual form, independent of one another. For the latter spacetime, the gravitational potential, determined by the harmonic function, $`H`$, and the dragging effects, determined by the Sagnac connection, $`a_i`$, will be independent, while for the former spacetime they are interconnected. There are several consequences:
The location of the null hypersurface defining the future event horizon, $`^+`$ (which, for these coordinates, is determined by the poles of the harmonic function), will depend on the angular momentum $`J`$ for the KN case, but not for the BMPV case. In particular we know that for the KN case only for $`J=0`$ will the harmonic function have real poles. In contrast the BMPV case will have a well defined horizon for non-vanishing $`J`$. The horizon becomes, however, ill-defined for large $`J`$, since the locus of $`^+`$ becomes a timelike hypersurface ;
For the BMPV case, the norm of the timelike Killing vector field, does not depend on the angular momentum. Therefore, there cannot be an ergoregion and the angular velocity of the horizon must be zero. Since there is an overall angular momentum the spacetime cannot be rigidly rotating.
This two cases are the only possibilities when trying to include rotation in a supersymmetric black hole spacetime: either we destabilise the horizon and create a naked singularity, or we keep a non-rotating horizon. A rotating horizon is incompatible with supersymmetry since it will create an ergoregion .
Comparing the two theories, two major differences are apparent: the dimension and the Chern-Simons term. The role of the Chern-Simons term was discussed in where it was conjectured that the stability of the solution (and the regularity) of the horizon was associated with the presence of such term with the particular coefficient required by supersymmetry. But the particular spacetime dimension, $`D=5`$, was not sufficiently appreciated. A simple comparison might help. The similarity between the bosonic sector of $`N=2,D=5`$ and $`N=8,D=11`$ Supergravity has been long known, particularly the identical Chern-Simons terms . The major difference is in the rank of the gauge field; a 1-form potential for the former versus a 3-form for the latter. They naturally couple to a black hole or a membrane. However, no regular, supersymmetric, rotating M-brane solution exists in the literature. In , the BPS limit of the non-extreme rotating M2-brane yields a spacetime with vanishing angular momentum (and an irregular horizon). Attempting to find such solution immediately reveals a major difference with the five dimensional case. In the derivation of it was crucial to use the fact that the one form in the metric describing the rotation has a field strength with a self-duality property in the transverse space to the world volume (as will be review in section 4.2.1). As it stands, this property cannot be used in $`D=11`$, and singles out $`D=5`$ as a special dimension.
We now give a simple example of how a self-duality condition can be useful in finding a solution with Killing spinors in a particular dimension. One can define a Brinkmann wave as a geometry that admits a covariantly constant null vector field: $`N=/v`$ obeying $`_\mu N=0`$. In general these waves can be described by a metric of the type
$$ds^2=du(dv+fdu+A_idx^i)+\widehat{\gamma }_{ij}dx^idx^j.$$
(2.12)
We are using light cone coordinates $`u=t+y`$, $`v=t+y`$. Both the scalar function $`f`$, the spatial vector $`A_i`$ and the spatial metric $`\widehat{\gamma }_{ij}`$ may admit $`u`$ and $`x^i`$ dependence. Pp-waves are the special case $`A_i=0`$ (or a pure gauge $`A_i`$, i.e., $`A_i=_iK`$ for some function $`K(x^i)`$) and $`\widehat{\gamma }_{ij}=\delta _{ij}`$. We will be interested in (2.12) with a flat spatial metric and no $`u`$ dependence. Then, demanding this geometry to be a solution to pure gravity (in arbitrary dimension) yields the constraints
$$^kF_{ki}=0,\mathrm{\Delta }f=\frac{1}{8}F^{ij}F_{ij},$$
(2.13)
where $`F_{ij}=_iA_j_jA_i`$. If in addition one is looking for a supersymmetric configuration, the spacetime must possess a supercovariantly constant spinor, or equivalently for pure gravity, a covariantly constant spinor. Such metrics have a restricted holonomy and obey
$$_vϵ=0,_uϵ+\frac{1}{16}F_{ij}\mathrm{\Gamma }^{ij}ϵ+\frac{1}{4}_if\mathrm{\Gamma }^{ui}ϵ=0,_iϵ\frac{1}{8}F_{ji}\mathrm{\Gamma }^{uj}ϵ=0.$$
(2.14)
For the pp-wave case, constant spinors with $`\mathrm{\Gamma }^uϵ=0`$ (or, in $`t,y`$ coordinates, $`(1+\mathrm{\Gamma }^{ty})ϵ=0`$) are solutions to (2.14) showing that these waves are one half supersymmetric. But for non trivial $`A_i`$ we have to deal also with the term $`F_{ij}\mathrm{\Gamma }^{ij}ϵ`$. A way to make this term vanish is to require the $`x^i`$ space to be four dimensional ($`E^4`$) and $`F_{ij}`$ to be the components of a Hodge self-dual two form in this space. We then find
$$F_{ij}\mathrm{\Gamma }^{ij}ϵ=\frac{1}{2}F_{ij}\mathrm{\Gamma }^{ij}(1+\mathrm{\Gamma }^{ty}\mathrm{\Gamma }^7)ϵ,$$
(2.15)
where $`\mathrm{\Gamma }^7\mathrm{\Gamma }^{ty1234}`$ is the chirality operator for the six dimensional spacetime. Thus, in $`D=6`$, (2.12) admits Killing spinors with one quarter of the degrees of freedom of a general constant spinor, due to the conditions
$$\mathrm{\Gamma }^7ϵ=ϵ,\mathrm{\Gamma }^uϵ=0.$$
(2.16)
The self-duality condition implies that the Maxwell equation in (2.13) is identically obeyed. To solve the Poisson type equation we parametrise $`E^4`$ with the hyperspherical coordinates (2.9). Then,
$$A=\frac{J}{\rho ^2}(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2),f=\frac{Q}{r^2}+\frac{J^2}{12r^6},$$
(2.17)
with $`Q`$ constant, is a solutions. Actually, the first term in $`f`$ could be any harmonic function on $`E^4`$. We therefore find the following geometry to be a Brinkmann wave solution to $`D=6`$ pure gravity and admitting Killing spinors:
$$ds^2=dt^2+dy^2+\left(\frac{Q}{r^2}+\frac{J^2}{12r^6}\right)(dtdy)^2+\frac{J}{r^2}(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2)(dydt)+ds_{E^4}.$$
(2.18)
Compactifying along the $`y`$ direction and performing Kaluza-Klein reduction, we find a 5D black hole type solution to Kaluza-Klein theory (in 5 dimensions). In the Einstein frame
$$\begin{array}{c}ds_{5,E}^2=\mathrm{\Delta }^{\frac{2}{3}}\left[dt+\frac{J}{2\rho ^2}(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2)\right]^2+\mathrm{\Delta }^{\frac{1}{3}}ds^2(E),\hfill \\ \\ A=\left(1\mathrm{\Delta }^1\right)dt+\frac{J}{2\rho ^2\mathrm{\Delta }}(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2),\hfill \\ \\ e^{2\varphi }=\mathrm{\Delta }1+\frac{Q}{\rho ^2}+\frac{J^2}{12\rho ^6}.\hfill \end{array}$$
(2.19)
The solution has some unusual features, but we will postpone a detailed examination to somewhere else.
In section 3 we will need a 10D Brinkmann wave compatible with a system involving D1 and D5 branes. Such geometry will be similar to (2.18) but without the $`J^2/12r^6`$ term. Some components of the Ricci tensor will then be non-zero, namely
$$R_{tt}=R_{yy}=R_{ty}=R_{yt}=\frac{J^2}{r^8}.$$
(2.20)
Therefore it is not a solution to pure gravity. It cannot be a solution to $`D=6`$, minimal supergravity either. In fact, the graviton multiplet is then $`(g_{MN},\mathrm{\Psi }_M^{Weyl},B_{MN}^{sd})`$. The gravitino is a Weyl vector-spinor and the field strength $`H_{MNP}^{sd}`$ derived from the potential two form $`B_{MN}^{sd}`$ is required to be self-dual.<sup>6</sup><sup>6</sup>6The self duality condition is crucial for the matching of on-shell degrees of freedom: $`g_{MN}`$ has $`(D2)(D1)/21=9`$; $`\mathrm{\Psi }_M^{Weyl}`$ has $`2^{\left[\frac{D}{2}\right]1}(D3)=12`$; an usual two form potential $`B_{MN}`$ has $`(D2)(D3)/2=6`$, which can be halved by the self duality condition, obtaining the necessary matching for supersymmetry. Hence, due to the Lorentzian $`(+++\mathrm{})`$ signature, $`H_{MNP}^{sd}`$ will have a vanishing energy momentum tensor, and any purely bosonic solution to simple $`D=6`$ supergravity must be Ricci flat.
In order to solve non trivially the Killing spinor conditions we required the space transverse to the Brinkmann wave propagation to be $`E^4`$. However, we may add more flat directions, $`z^\alpha `$, without spoiling the supersymmetry, as long as the metric functions $`A_i,f`$ do not depend on $`z^\alpha `$. In this way we build the $`D=10`$ configuration:
$$\begin{array}{c}ds^2=du(dv+fdu+A_idx^i)+\delta _{ij}dx^idx^j+\delta _{\alpha \beta }dz^\alpha dz^\beta ,\\ \\ B=\frac{A_i}{2}dudx^i,\end{array}$$
(2.21)
with $`\alpha =5\mathrm{}8`$. Assuming self-duality of $`F_{ij}`$ in the aforementioned sense, (2.21) is a supersymmetric configuration of $`D=10`$, type II supergravities.
Notice that the essential piece in this construction was to maintain an $`SO(4)`$ symmetry in the transverse space to the direction of propagation of the wave. This is an isometry of the space where the metric functions vary. The same isometry is present for a system of D1 branes inside D5-branes. Therefore we will be able to superimpose the systems and still have a supersymmetric solution. But the same is not true for the system of D2-D6-NS5 branes in type IIA that describe 4 dimensional black holes . From the higher dimensional viewpoint we can still see why we should expect different properties of 4 or 5 dimensional black holes.
For completeness we note that (2.21) with $`A`$ given by (2.17) and $`f=C/r^2`$ indeed solves the field equations for a graviton-axion configuration of $`D=10`$ type II supergravity, which read
$$R_{\mu \nu }\frac{1}{2}g_{\mu \nu }R=\frac{1}{4}\left(H_{\mu \sigma \tau }H_\nu ^{\sigma \tau }\frac{1}{6}g_{\mu \nu }H_{\sigma \tau \rho }H^{\sigma \tau \rho }\right),D_\mu H^{\mu \sigma \tau }=0.$$
(2.22)
Let us remark that the ten dimensional Brinkmann wave we have just described is not a particular case of the solutions derived in , due to the special dependence of the ‘rotation vector’ on some of the coordinates of the transverse space rather than only one of the light cone directions.
## 3 Causality in Supergravity and String Theory
In spite of fairly recent proposals for ‘infinitely large extra dimensions’ , the most accepted way to match the observed number of dimensions with the theoretical requirements of string theory is still compactification. One assumes, therefore, a special topology for the universe, namely that it is a $`K^6`$ bundle over $`M^4`$, where $`M^4`$ is a four dimensional manifold and $`K^6`$ a compact six dimensional space. Compactification induces a lower dimensional spectrum consisting of an infinite tower of states. Most often only the massless ground states are of any relevance at low energies. These are related to the higher dimensional fields through the Kaluza-Klein procedure. When spacetime is a twisted $`K^6`$ bundle over $`M^4`$, the four dimensional metric obtained by the procedure will not be conformal to the 10 dimensional one, implying a modification in the causal structure: light cones for 10 and 4 dimensional gravity do not coincide anymore.<sup>7</sup><sup>7</sup>7As an aside, let us remark on theories with variable speed of light. In fact, some recent attempts to deal with the cosmological problem of a (possibly) negative deacceleration parameter via gravity-scalar theories show a close technical similarity. In such theories, matter couples to gravity via a combination of the metric plus the derivatives of the scalar field. In our language it would be as if gravity was propagating in the whole ten dimensions, whereas matter would couple to a four dimensional Kaluza-Klein metric.
In this section we analyse the distinctions between higher and lower dimensional viewpoints for the particular example of the BMPV black hole. To complete the picture we then review the string theory dual description of this spacetime , and stress the relation between microscopic unitarity and macroscopic causality (which was implicit in ).
### 3.1 Causality and dimensional reduction
#### 3.1.1 The D1-D5-Brinkmann wave system in D=10
In 11 dimensions, a supersymmetric configuration was found describing the intersection of a five brane and a two brane on a string, with momentum along the string direction and with non-trivial angular momentum <sup>8</sup><sup>8</sup>8The reader should notice that in formula (82) of this reference the expressions for $`B_{t16}^{(11)}`$ and $`B_{t\varphi _26}^{11}`$ should read, respectively $`T`$ and $`J\mathrm{cos}^2\theta T/2r^2`$.. Using the following set of 11 dimensional coordinates we represent the configuration as:
| t | $`y_1`$ | $`y_2`$ | $`y_3`$ | $`y_4`$ | $`y_5`$ | $`y_6`$ | r | $`\theta `$ | $`\varphi _1`$ | $`\varphi _2`$ |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 5 | 5 | 5 | 5 | 5 | 5 | | | | | |
| 2 | 2 | | | | | 2 | | | | |
| | w | | | | | | | | | |
The last row refers to the spatial direction in which the wave carrying the momentum is propagating. Compactifying the $`y^6`$ direction to a circle and performing Kaluza-Klein reduction we obtain a solution of type II supergravity (IIA or IIB since no RR fields are excited). Transforming to the Einstein frame and S-dualizing we then get a solution to type IIB supergravity which can be interpreted as a $`D1`$-brane inside a $`D5`$ with a Brinkmann wave propagating along the string, as described in the last section:
$$\begin{array}{c}ds_E^2=f_5^{\frac{1}{4}}f_1^{\frac{3}{4}}\left[dt^2+dy_1^2+f_K(dtdy_1)^2+\frac{J}{\rho ^2}(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2)(dy_1dt)\right]+\hfill \\ +\left(\frac{f_1}{f_5}\right)^{\frac{1}{4}}ds^2(E_I^4)+f_5^{\frac{3}{4}}f_1^{\frac{1}{4}}ds^2(E_E^4),\hfill \\ \\ B_{RR}=f_1^1dtdy_1P\mathrm{cos}^2\theta d\varphi _1d\varphi _2+\frac{J}{2\rho ^2}f_1^1\left(dy_1dt\right)(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2),\hfill \\ \\ e^{2(\varphi \varphi _{\mathrm{}})}=\frac{f_5}{f_1}.\hfill \end{array}$$
(3.1)
Solutions with some similarities have been studied in . Setting $`J=0`$ we recover the standard D1-D5-pp wave system . The Euclidean space $`E_I^4`$ is parametrised by $`y_2,y_3,y_4,y_5`$, while the Euclidean space $`E_E^4`$ is parametrised by $`\rho ,\theta ,\varphi _1,\varphi _2`$. The three functions $`f_1`$, $`f_5`$, $`f_K`$ are given by:
$$f_5=1+\frac{P}{\rho ^2},f_1=1+\frac{Q}{\rho ^2},f_K=\frac{Q_{KK}}{\rho ^2}.$$
(3.2)
This solution is invariant under the transformations generated by four supercharges of the type IIB supersymmetry algebra. This corresponds to $`1/8`$ of the vacuum supersymmetry. Recall that the supersymmetry invariance of this theory is generated by two Weyl-Majorana spinors, $`ϵ_L`$,$`ϵ_R`$. The chirality conditions are $`\mathrm{\Gamma }^{11}ϵ_{L,R}=ϵ_{L,R}`$. The Killing spinors obey three sets of conditions. The presence of the D-string and the D5-brane give rise to, respectively:
$$\mathrm{\Gamma }^{ty_1}ϵ_L=ϵ_R,\mathrm{\Gamma }^{ty_1y_2y_3y_4y_5}ϵ_L=ϵ_R,$$
(3.3)
whereas the Brinkmann wave requires
$$\mathrm{\Gamma }^{ty_1}ϵ_L=ϵ_L,\mathrm{\Gamma }^{ty_1}ϵ_R=ϵ_R.$$
(3.4)
Of course, these are exactly the same conditions as in the $`J=0`$ case . The presence of angular momentum does not break any further supersymmetry, in the same way as it does not cost any more energy. However, as mentioned before, in solving the equation for the gravitino variation it becomes essential to use the fact that the one form representing the rotation, i.e. $`A_i`$ in $`A_idx^idt`$, is self-dual in the transverse space $`E_E^4`$.
Compactifying further on $`T^4\times S^1`$, we obtain a five dimensional solution describing a rotating black hole with three different charges. Using U-duality arguments, one can quantise the charges $`P,Q,Q_{KK}`$ in terms of stringy quantities . Denoting the radius of the circle by $`R`$ and the volume of $`T^4`$ by $`(2\pi )^4V`$ the result is:
$$P=Q_5g\alpha ^{},Q=\frac{Q_1\alpha ^3g}{V},Q_{KK}=\frac{g^2\alpha ^4N_R}{R^2V}.$$
(3.5)
The quantities $`Q_1,Q_5`$ and $`N_R`$ are integers and are counting, respectively, the number of D1-branes, D5-branes and units of right moving momentum, $`g`$ is the string coupling and $`\alpha ^{}`$ the Regge slope. Unlike these quantities, the angular momentum parameter $`J`$ is still continuous. We will address its quantisation below.
#### 3.1.2 The BMPV black hole in D=5
Let us describe the five dimensional configuration. The metric reads
$$ds_E^2=\left[f_1f_5(1+f_K)\right]^{\frac{2}{3}}\left[dt+\frac{J}{2\rho ^2}(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2)\right]^2+\left[f_1f_5(1+f_K)\right]^{\frac{1}{3}}ds^2(E_E^4),$$
(3.6)
while for the gauge fields we get
$$\begin{array}{c}A_i=\frac{Q_i}{\rho ^2+Q_i}dt+\frac{J}{2(Q_i+\rho ^2)}(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2),\\ \\ B=P\mathrm{cos}^2\theta d\varphi _1d\varphi _2\frac{J}{2(Q+\rho ^2)}dt(\mathrm{sin}^2\theta d\varphi _1\mathrm{cos}^2\theta d\varphi _2),\end{array}$$
(3.7)
where $`i=1,2`$, $`Q_1Q_{KK}`$ and $`Q_2Q`$. Hence $`A_1`$ is the Kaluza-Klein gauge field arising from compactifying $`y_1`$ and $`A_2`$ is the winding gauge field arising from the same compactification. The moduli are
$$e^{2(\varphi \varphi _{\mathrm{}})}=\frac{f_5}{f_1},e^{2\sigma _1}=\frac{1+f_K}{(f_5f_1^3)^{\frac{1}{4}}},e^{2\sigma _s}=\left(\frac{f_1}{f_5}\right)^{\frac{1}{4}},$$
(3.8)
where $`s=2,3,4,5`$. An equivalent solution was firstly obtained in . The ADM mass and the entropy of this black hole are
$$\begin{array}{c}M_{ADM}=\frac{RV}{\alpha ^4g^2}(Q+P+Q_{KK})=\frac{1}{g^2}\left(\frac{RgQ_1}{\alpha ^{}}+\frac{RVgQ_5}{\alpha ^3}+\frac{g^2N_R}{R}\right),\\ \\ S_{Sugra}=\frac{\pi ^2}{2G_5}\sqrt{PQQ_{KK}\frac{J^2}{4}}.\end{array}$$
(3.9)
$`G_5=\pi \alpha ^4g^2/(4VR)`$ is the five dimensional Newton’s constant which relates to the ten dimensional one ($`G_{10}=8\pi \alpha ^4g^2`$) by the moduli of the compact manifold. Notice that the ADM mass is unchanged from the static case as expected from supersymmetry. For the special case when all the three charges coincide, i.e. $`Q_{KK}=P=Q\mu `$, the configuration is equivalent to the one in . The RR two form potential is then encoded in the remaining gauge fields through
$$dB=d\overline{A}\overline{A}d\overline{A},$$
(3.10)
where $`\overline{A}A_1=A_2`$ and $``$ denotes Hodge duality with respect to the metric (3.6). All the scalars are constants. Therefore, in this special case, the solution is specified by the metric and the gauge field $`\overline{A}`$. In this form, the configuration can also be obtained as a solution to five dimensional $`N=2`$ supergravity . Using a Schwarzchild type radial coordinate $`r^2=\rho ^2+\mu `$, $`g_E`$ and $`\overline{A}`$ can be expressed as
$$\begin{array}{c}ds^2=\left(\mathrm{\Delta }_{10}\right)^2\left[dt+\frac{\mu \omega }{2(r^2\mu )}(d\gamma +\mathrm{cos}\beta d\alpha )\right]^2+\frac{dr^2}{\left(\mathrm{\Delta }_{10}\right)^2}+\frac{r^2}{4}[d\alpha ^2+d\beta ^2+d\gamma ^2+2\mathrm{cos}\beta d\alpha d\gamma ],\hfill \\ \\ A\frac{\sqrt{3}}{2}\overline{A}=\frac{\sqrt{3}\mu }{2r^2}\left[dt\frac{w}{2}\left(d\gamma +\mathrm{cos}\beta d\alpha \right)\right].\hfill \end{array}$$
(3.11)
The angles $`(\alpha ,\beta ,\gamma )`$ are Euler angles on $`SU(2)S^3`$, we defined $`w=J/2\mu `$ to make contact with and for future convenience we have introduced the notation
$$\mathrm{\Delta }_{ij}=1\left(\frac{\mu }{r^2}\right)^i\left(\frac{\omega }{r}\right)^{2j}.$$
(3.12)
The configuration (3.11) is the BMPV black hole, which in isotropic coordinates becomes (2.8). We define the quantities $`r_H`$, $`r_L`$, $`r_Q`$ and $`r_A`$ as, respectively the zeros of $`\mathrm{\Delta }_{10}`$, $`\mathrm{\Delta }_{21}`$, $`\mathrm{\Delta }_{11}`$, $`\mathrm{\Delta }_{01}`$. The first quantity defines the black hole horizon ($`r=r_H`$). The second quantity defines the surface $`r=r_L`$, which we call the Velocity of Light Surface (VLS). Inside the VLS there are closed timelike curves (CTC’s). We distinguish two cases. For $`r_L>r_H`$ there are naked CTC’s. This is the over-rotating case. For $`r_L<r_H`$ the CTC’s are hidden behind the horizon, which we refer to as the under-rotating case. Notice that
$$r_Lr_Hw^2\mu \frac{J^2}{4}Q_{KK}QP,$$
(3.13)
which is therefore the causality bound. We remark that the interpretation of $`r=r_H`$ as the black hole horizon only makes sense in the under-rotating case, since the horizon should be a null hypersurface which is no longer true in the over-rotating case. The interpretation of $`r_Q`$ and $`r_A`$ will be given in section 4.1.
At this point we should comment on some conventions. The standard treatment of $`N=2`$, $`D=5`$ Supergravity leads to a Bogomol’nyi bound of $`M\sqrt{3}|Q|/2`$ . This leads to a factor of $`\sqrt{3}/2`$ in the gauge potential, so that the field $`A`$ introduced in (3.11) is consistent with this treatment. This factor is then essential for supersymmetry. The potential $`\overline{A}`$ coming from dimensional reduction is consistent with a bound without the $`\sqrt{3}/2`$ factor. Therefore, one must perform a rescaling of the field to make contact with the five dimensional theory. Although using $`A`$ leads to a more awkward set of conventions, we will use it when dealing with the five dimensional viewpoint.
#### 3.1.3 Causality
The geometry (3.11) contains strong causality violations. Even a freely falling particle can move backwards in time (as seen from the observer at infinity) . What happens in $`D=10`$?
The geometry (3.1) does not have any ‘obvious’ closed timelike curves, in the sense that there is no periodic direction whose metric coefficient changes sign in some spacetime region. However if we choose to compactify the $`y_1`$ direction, we create CTC’s. The point is that there are linear combinations like $`t^\mu _\mu =2B/r(/\gamma )+A(/y_1)`$ that become timelike (we have chosen to parametrise the ten dimensional solution with Euler angles for these remarks). But only when the $`y_1`$ direction is made periodic, may the curves with such tangents become closed. Rewriting the $`10D`$ configuration with all the charges equal and changing to Schwarzchild coordinates as before, the condition for such curves to become null is:
$$\left|\frac{B}{A}\right|=\left(\frac{r_L}{r}\right)^3\pm \sqrt{\mathrm{\Delta }_{21}}.$$
(3.14)
Only for $`r<r_L`$ two distinct roots arise. Values of $`|B/A|`$ in between them correspond to timelike curves. This happens only inside the VLS, which can therefore be seen from the higher dimensional perspective as well. The condition for the curves to be closed is
$$\frac{B}{A}=\frac{r}{2R}q,$$
(3.15)
where $`qQ`$.
The universal covering space of the manifold with geometry (3.1) is therefore causally well behaved, whereas its non-simply connected compactification has CTC’s. Notice that this is quite different from creating causal anomalies in flat space by identifying the time coordinate or from the $`AdS`$ causal problems. In both these cases, it is a timelike direction that becomes compact, whereas in our case CTC’s become possible through the compactification of a spacelike direction. The similarity is, of course, they both can be resolved by going to the universal covering space.
The Kaluza-Klein reduction to five dimensions gives rise to the identifications
$$g_{MN}^{(10,E)}=\left(\begin{array}{c}g_{\mu \nu }^{(5,KK)}+e^{2\sigma _1}A_\mu A_\nu e^{2\sigma _1}A_\mu 0\\ e^{2\sigma _1}A_\nu e^{2\sigma _1}0\\ \mathrm{0\; 0}e^{2\sigma _s}\end{array}\right).$$
(3.16)
The superscripts on the metric refer to the Einstein or ‘Kaluza-Klein’ frame<sup>9</sup><sup>9</sup>9By Kaluza-Klein frame we mean the frame obtained from a Kaluza-Klein compactification starting from the higher dimensional Einstein frame and without rescaling the lower dimensional metric by using the moduli, as in using ansatz (3.16). The first row (and column) refers to the directions $`t`$ and $`E_E^4`$, the second to the direction $`y_1`$ and the last to the directions in $`E_I^4`$. The five dimensional Einstein frame is then obtained as
$$g^{(5,E)}=e^{\frac{2}{3}_{i=1}^5\sigma _i}g^{(5,KK)}.$$
(3.17)
The procedure eliminates the $`y_1`$ direction and effectively projects down the CTC’s into the $`\gamma `$ direction. So, the metric $`g_{\mu \nu }^{5,KK}`$ (or $`g_{\mu \nu }^{5,E}`$) has unavoidable closed timelike curves, whereas the combination
$$g_{\mu \nu }^{(5,KK)}+e^{2\sigma _1}A_\mu A_\nu $$
(3.18)
does not. The latter describes the local geometry of a manifold were any existing CTC’s will not be homotopic to a point.
In Figure 1 we illustrate the above procedure. The picture on the left describes the $`D=10`$ universal covering manifold, i.e., $`y_1R`$, and we illustrate the curve with tangent $`t^\mu _\mu `$. Step $`1`$ is the compactification of $`y_1`$, i.e., $`y_1S^1`$, so that the curve with tangent $`t^\mu _\mu `$ becomes closed. Step $`2`$ is the Kaluza-Klein reduction, upon which the curve is projected onto the $`\gamma `$ direction, corresponding to the causal anomalies seen in the BMPV black hole. We should remark that the pictures are misleading in two senses. Firstly, the manifold parametrised by $`y_1`$ and $`E_E^4`$ is a non trivial $`R`$ (or $`S^1`$) bundle over $`E_E^4`$, which is precisely what allows the curve described by $`t^\mu _\mu `$ to become timelike. Secondly, the $`\gamma `$ direction does not go around a non-trivial cycle of the manifold.
Can we associate other acausal spacetimes with better behaved higher dimensional configurations in a similar fashion? This would rely on performing a transformation of the type (3.18). Thus, one cannot apply the method to any acausal spacetime configuration, since not all examples have a gauge field. The Gödel manifold is the simplest example . But it turns out that this procedure does not work for the general five dimensional family of black holes . Moreover it is not something associated to supersymmetry either, since the four dimensional Kerr-Newman spacetime with $`Q^2=M^2`$ still exhibits the traditional closed timelike curves in the negative $`r`$ region after applying an analogous transformation to (3.18). Therefore, we have to conclude that this spacetime has a peculiar property, which, using Carter’s terminology , one can put as: the non-trivial CTC’s of the five dimensional spacetime arise as trivial CTC’s from the ten dimensional viewpoint.
### 3.2 Causality and Unitarity
Consider the ten dimensional supergravity D1-D5-brane solution, i.e., (3.1) with $`Q_{KK}=J=0`$, and $`y_1`$, $`E_I^4`$ compact. The D-brane tensions depend on the string coupling constant $`g`$ as $`1/g`$ (as can be seen from (3.9)) and the ten dimensional Newton’s as $`G_{10}g^2`$. Hence, the gravitational potential for D-branes goes as $`Vg`$. Decreasing the string coupling we turn off gravity and find a flat space configuration: a gas of open strings with boundary conditions determined by the presence of the D-branes.
The theory describing these open strings and effectively describing the D-branes is a 1+1 dimensional SCFT (Super Conformal Field Theory), which can be thought of as living on the worldvolume of the D1-branes inside the D5-branes . Within the AdS/CFT correspondence this is because the near horizon geometry of the $`D=10`$ configuration is $`AdS_3\times S^3\times T^4`$, and the 1+1 SCFT lives on the conformal boundary of the AdS piece. Since we compactified the supergravity configuration to 5D, our SCFT will be on the cylinder $`R\times S^1`$, were $`R`$ is the time direction. The amount of supersymmetry of this SCFT must be the same as in the classical configuration, i.e., $`1/4`$ of the vacuum supersymmetry. Thus, we have an $`N=4`$ SCFT on $`R\times S^1`$ describing the compactified supergravity D-brane configuration.
Supersymmetric states belong to short multiplets. These should be stable at any value of the string coupling. Therefore we expect our continuous variation of $`g`$ to leave the degeneracy of states unchanged, i.e., to be adiabatic. The SCFT has a central charge determined by the massless degrees of freedom of the open strings, $`\stackrel{~}{c}=N_B+N_F/2`$.<sup>10</sup><sup>10</sup>10In our setup the states describing the entropy are right moving. That is why the central charge has a tilde. But this is the usual Virasoro central charge, related to the complex dimension $`d_c`$ of the target space for the sigma model by $`\stackrel{~}{c}=3d_c`$. These are fairly easy to count: the massless excitations have the same number of bosonic ($`N_B`$) and fermionic ($`N_F`$) physical degrees of freedom, $`4Q_1Q_5`$, effectively carried by the (1,5) and (5,1) strings in the flat space D-brane configuration . Hence $`\stackrel{~}{c}=6Q_1Q_5`$.
Now we want to turn on momentum ($`Q_{KK}`$) and angular momentum ($`J`$) in the supergravity side and see the correspondence in the SCFT side.
The supergravity angular momentum is classified by the $`SO(4)`$ rotation group which is an isometry of the geometry (3.1) or (3.6) acting on the space transverse to the branes. We have to identify this symmetry on the SCFT side. It is known that an $`SO(4)`$ symmetry arises in the $`N=4`$ superconformal algebra, corresponding to endomorphisms in the graded algebra that rotate the fermionic generators $`G_m^i`$ (i=1..4) amongst themselves, i.e. R-symmetry. Such symmetry is gauged in the sense that linear combinations of the $`G_m^i`$ (for negative m) create states that carry $`(F_L,F_R)`$ charges of the $`U(1)_L\times U(1)_R`$ Cartan subalgebra of $`SO(4)`$. These two quantities are therefore to be identified with the two linearly independent spacetime angular momentum parameters
$$(J_L,J_R)(J^{21}J^{43},J^{21}+J^{43})=(0,\frac{\pi }{4G_5}J)=(0,\frac{VR}{\alpha ^4g^2}J).$$
(3.19)
The $`J^{ik}`$ is the angular momentum tensor that can be read off from the geometry (3.6) in the usual way. In this way, the $`J_L`$, $`J_R`$ become quantised in terms of $`(F_L,F_R)`$. Using the above results for the quantisation of $`P,Q,Q_{KK}`$, we may now express the entropy of the classical solution in terms of quantised charges:
$$S_{Sugra}=\frac{\pi ^2}{2G_5}\sqrt{PQQ_{KK}\frac{J^2}{4}}=2\pi \sqrt{Q_1Q_5N_R\frac{F_R^2}{4}}.$$
(3.20)
We have not identified $`N_R`$ in the SCFT side yet, i.e., established the correspondence between SUGRA momentum and a quantity in the SCFT. In supergravity the integer $`N_R`$ arises due to the compactification of the $`y_1`$ direction into $`S^1`$. It is therefore a $`U(1)`$ charge. The conformal algebra of the cylinder - the Virasoro algebra - has certainly a $`U(1)`$ subalgebra. States carry a $`U(1)`$ charge given by the eigenvalue of the zero mode in the theory, i.e., the Virasoro generator $`\stackrel{~}{L}_0`$. Hence, $`N_R`$ must correspond in the SCFT to the right moving level $`n_R`$ of the states.
What is the degeneracy of states of the form
$$|n_L,F_L;n_R,F_R>=|0,0;N_R,F_R>$$
(3.21)
in an $`N=4`$ SCFT of central charge $`\stackrel{~}{c}=6Q_1Q_5`$? For a 2D CFT on a cylinder, the degeneracy of states at highest available level $`M>>1`$ can be computed by using the modular invariance of $`T^2`$ in the Euclidean section , yielding, up to power corrections
$$d(M,\stackrel{~}{c})e^{2\pi \sqrt{\frac{M\stackrel{~}{c}}{6}}}.$$
(3.22)
The highest level available for the states (3.21) is not $`N_R`$, after we have fixed the other quantum numbers, namely $`F_R`$. In a CFT, states associated with a primary operator give a contribution to the total $`\stackrel{~}{L}_0`$ eigenvalue equal to their conformal weight. In particular, an operator crating states with charge $`F_R`$ has a conformal weight not smaller than $`3F_R^2/(2\stackrel{~}{c})`$ if we assume that the CFT is unitary, which demands all conformal weights to be non-negative . The total $`\stackrel{~}{L}_0`$ eigenvalue, $`N_R`$, is therefore the sum of $`3F_R^2/(2\stackrel{~}{c})`$ plus the conformal weight of other operators, $`N_R^{remain}`$, which is non-negative by unitarity:
$$N_R=\frac{3F_R^2}{2\stackrel{~}{c}}+N_R^{remain}Q_1Q_5N_R\frac{F_R^2}{4}.$$
(3.23)
This is the unitarity bound, which coincides with the causality bound (3.13). Moreover, the highest available level is now $`M=N_R^{remain}`$. This is the total level available to find as different combinations of operators. It therefore determines the degeneracy of states. It follows that
$$S_{SCFT}=\mathrm{ln}d(N_R^{remain},\stackrel{~}{c})=S_{Sugra}$$
(3.24)
The above bound can also be seen in the explicit constructions of unitary representations of the SCFT. First notice that the states we are interested in are in the left-moving ground state. So, we need to look only at the representation theory of an $`N=2`$ SCFT, which was worked out in . The generators are $`L_m`$, the Virasoro generators, $`T_m`$, the modes of a $`U(1)`$ current, and $`G_m^i`$, $`i=1,2`$, the modes of the supercurrent. The $`U(1)`$ symmetry is the only surviving piece of the $`SO(4)`$ symmetry of the $`N=4`$ algebra, which we want to think of as $`U(1)_R`$. There are 3 possible N=2 algebras, according to the moddings one can choose for the generators, via the boundary conditions. We are interested in the P algebra, where the fermions have periodic boundary conditions and therefore generalises the usual Ramond sector of the open string theory. The reason is that the spacetime angular momentum is carried by the fermionic modes of the (1,5) and (5,1) strings .
There are 3 possible classes of unitary representations for the P algebra. The ones with a two dimensional moduli space obey <sup>11</sup><sup>11</sup>11Note: $`\stackrel{~}{c}`$ in is the complex dimension of the target space, hence 3 times our usual Virasoro central charge.
$$2\left(\frac{\stackrel{~}{c}}{3}1\right)\left(h\frac{\stackrel{~}{c}}{24}\right)q^2+\frac{1}{4}\left(\frac{\stackrel{~}{c}}{3}+1\right)^20,$$
(3.25)
which, for $`\stackrel{~}{c}=6Q_1Q_5>>1`$, level $`h=N_R`$ and charge $`q=F_R`$ gives (3.23).
The upshot is that the charge (in CFT language) or angular momentum (in spacetime language) we create must be bounded by the energy we create; otherwise we violate unitarity or causality.
## 4 Other Properties of the BMPV spacetime
### 4.1 The ‘repulson’ behaviour
The first property we would like to address is related with the over-rotating case. Let us start by computing the charged geodesics. We follow the Hamilton-Jacobi method with the usual minimally coupled Hamiltonian
$$H=\frac{g^{\mu \nu }}{2}\left(p_\mu +qA_\mu \right)\left(p_\nu +qA_\nu \right)$$
(4.1)
where $`g_{\mu \nu }`$ and $`A_\mu `$ are given by (3.11). The ansatz for the action function is
$$S=Et+H(\alpha ,\beta ,\gamma )+W(r),H(\alpha ,\beta ,\gamma )=j_L\alpha +j_R\gamma +\chi (\beta ).$$
(4.2)
This generalises the construction in . The Killing tensor found there is reducible. Since each of the Killing vector fields into which it decomposes is a symmetry of the Maxwell field (in the sense that $`\mathrm{\pounds }_{L_3}A=\mathrm{\pounds }_{R_i}A=0`$, where $`\mathrm{\pounds }`$ denotes the Lie derivative and $`L_i(R_i)`$ the left (right) invariant vector fields on $`SU(2)`$), we conclude that the Killing tensor still commutes with the minimally coupled Hamiltonian. Therefore the quantity
$$j^2\left(L_1H\right)^2+\left(L_2H\right)^2+\left(L_3H\right)^2,$$
(4.3)
is still a constant of motion, as in the case of purely gravitational interactions. The following set of equations of motion is then obtained (for the BMPV black hole, the mass to charge ratio is $`M/Q=\sqrt{3}/2`$, but we keep $`Q`$ and $`M`$ in order to make the interpretation of the several terms clear):
$$\begin{array}{c}\left(\frac{dr}{d\lambda }\right)^2=E^2\mathrm{\Delta }_{21}\left(m^2+\frac{4j^2}{r^2}\right)\left(\mathrm{\Delta }_{10}\right)^2\frac{qQE}{r^2}\mathrm{\Delta }_{11}+\left(\frac{qQ}{2r^2}\right)^2\mathrm{\Delta }_{01}+\frac{2\omega j_R}{r^4}\left(qQ\frac{4}{3}ME\right)\mathrm{\Delta }_{10},\\ \\ \frac{dt}{d\lambda }=\frac{1}{\left(\mathrm{\Delta }_{10}\right)^2}\left[E\mathrm{\Delta }_{21}\frac{qQ}{2r^2}\mathrm{\Delta }_{11}\frac{4\omega Mj_R}{3r^4}\mathrm{\Delta }_{10}\right],\\ \\ \frac{d\gamma }{d\lambda }=\frac{4}{r^2}\left[\frac{j_Rj_L\mathrm{cos}\beta }{\mathrm{sin}^2\beta }+\frac{\omega }{4r^2\mathrm{\Delta }_{10}}\left(\frac{4}{3}MEqQ\right)\right].\end{array}$$
(4.4)
The $`\beta `$ and $`\alpha `$ equations of motion are the same as in the purely gravitational case . The quantity $`m^2`$ is the mass squared of the test particle, which arises in the usual fashion as the integral of motion associated with the metric Killing tensor. Notice that the RHS of the $`r`$ equation is even under CPT, whereas the RHS of the $`t`$ and $`\gamma `$ are odd, as expected from the form of the LHS. Also notice that the affine parameter $`\lambda `$ is not proper time $`\tau `$ for the massive case, but $`\tau =\lambda m`$, and $`m`$ can be made to vanish in the equations by defining energy, charge and angular momentum per unit mass.
We now discuss some features of these equations. There are four interesting surfaces: the horizon (at $`r=r_H`$), the VLS (at $`r=r_L`$) and the timelike surfaces $`r_Q`$, $`r_A`$, defined as the zeros of $`\mathrm{\Delta }_{11}`$ and $`\mathrm{\Delta }_{01}`$ respectively. The relative location of these surfaces is $`r_A<r_Q<r_L<r_H`$ ($`r_A>r_Q>r_L>r_H`$) for the under-rotating (over-rotating) case. The VLS bounds the region where light cones allow causal travelling into the past of the observer at infinity. The energy term in the $`t`$ equation changes sign corresponding to the possibility of geodesic time travelling, or equivalently, to a change of character from particle to antiparticle. The energy term in the radial equation also changes sign and becomes repulsive. The surface $`r=r_Q`$ couples to the Coulomb term both in radial and time equations. Both of them change signs when crossing this surface, i.e., the Coulomb interaction changes from attractive to repulsive or vice-versa. We call this surface ‘Coulomb Conjugation Surface (CCS)’. The $`r=r_A`$ surface corresponds to a change in sign for the general relativistic correction to the electromagnetic interaction (which is asymptotically subleading). Effectively $`q^2`$ changes sign there.
On the surface $`r=r_H`$, the right hand side of the $`r`$ equation of motion becomes
$$\left(E\frac{3qQ}{4M}\right)^2\left(1\left(\frac{r_L}{r_H}\right)^6\right).$$
(4.5)
This is always negative in the over-rotating case. We conclude, therefore, that the repulson like behaviour found in for the gravitational interactions is still true in general when charge interactions are taken into account: for the over-rotating solution there are no freely falling orbits, with or without charge, entering the $`r=r_H`$ surface. The case singled out by (4.5) corresponds to an energy to charge ratio for the test object of $`\sqrt{3}/2`$. With our conventions, within a supersymmetric theory $`Em\sqrt{3}/2|q|`$. Thus a particle for which (4.5) vanishes everywhere should have $`E=m`$, i.e. be at rest, so the mass to charge ratio is also $`\sqrt{3}/2`$\- a BPS particle. The behaviour is then better seen by rewriting the $`r`$ equation of motion:
$$\begin{array}{c}\left(\frac{dr}{d\lambda }\right)^2=\left(qQ\frac{4}{3}ME\right)\left[\frac{2\omega j_R}{r^4}\mathrm{\Delta }_{10}\frac{\omega ^2}{4r^6}\left(qQ\frac{4}{3}ME\right)\right]+\hfill \\ \\ +\left(E^2m^2\right)\frac{1}{r^2}\left(qQE\frac{4}{3}Mm^2\right)+\frac{1}{4r^4}\left((qQ)^2\frac{16}{9}(mM)^2\right)\frac{4j^2}{r^2}\left(\mathrm{\Delta }_{10}\right)^2.\hfill \end{array}$$
(4.6)
Since the BPS test particle must have for initial conditions the lowest possible quantum numbers, $`|E|=m,j=0`$, it will stay at rest in the spatial coordinates, i.e. a no-force configuration.
The repulson result is readily confirmed at the semi-classical level for charged scalar waves minimally coupled to the background electromagnetic field. We also introduce a non-minimal coupling to the geometry. We are still able to separate variables in the latter model, which follows from the Ricci scalar being a function of $`r`$ only:
$$R=\frac{2}{r^8}(2\mu ^2\omega ^2r^2\mu ^2).$$
(4.7)
The wave equation then takes the form
$$g^{\mu \nu }(D_\mu +iqA_\mu )(D_\nu +iqA_\nu )\mathrm{\Phi }=(m^2+\lambda R)\mathrm{\Phi }.$$
(4.8)
The non-minimal coupling might be thought of as a renormalisation of the mass. Following we use the ansatz $`\mathrm{\Phi }(x^\mu )=e^{itE}D_{j_L,j_R}^jF(r)`$ and change variables as
$$x=\frac{\mu }{r^2\mu },$$
(4.9)
in which case (4.8) can be written as
$$\frac{d^2F}{dx}=\left[A+\frac{B}{x}+\frac{C}{x^2}+\frac{D}{x^3}\frac{\lambda }{x^2}\left(\frac{\omega ^2x^3}{\mu (1+x)^3}\frac{x^2}{(1+x)^2}\right)\right]F.$$
(4.10)
A,B,C,D are given by
$$\begin{array}{c}A=\frac{1}{4\mu }\left(\left(\frac{r_L}{r_H}\right)^61\right)\left(E\mu \frac{qQ}{2}\right)^2,C=D+j(j+1)+\frac{E}{2}\left(\frac{qQ}{2}\mu E\right),\\ B=\frac{1}{4\mu }\left(\left(\frac{qQ}{2}\right)^2\mu ^2E^2\right)\frac{1}{2\mu }\left(\mu E\frac{Qq}{2}\right)^2\frac{\omega j_R}{2\mu }(qQ2\mu E),D=\frac{\mu }{4}(m^2E^2).\end{array}$$
(4.11)
Near the horizon the scalar equation (4.10) is approximated by keeping only the $`A`$ term on the right hand side. Oscillating solutions arise only if $`r_L<r_H`$, i.e. in the under-rotating case. Therefore the absorption cross section for scalar charged waves is zero in the over-rotating case, confirming the ‘repulson’ behaviour.
Now we ask if an accelerated observer can enter the ‘horizon’ in the over-rotating case. Obviously, there are timelike curves that can achieve that. The simplest example is a radial orbit with tangent vector
$$t^\mu _\mu =\frac{\sqrt{2}}{\mathrm{\Delta }_{10}}_t\mathrm{\Delta }_{10}_r,$$
(4.12)
which has a unit norm. But it is well known in general relativity that a generic timelike trajectory might not be ‘realistic’. The simplest example is given by the Reissner-Nördstrom spacetime. The repulsive character of the timelike singularity precludes any timelike trajectory with bounded proper acceleration to reach the physical singularity at $`r=0`$ (in Schwarzchild type coordinates) . The non real character of such curves is manifest in the fact one would need an infinite acceleration (and so an infinite payload for the rockets) to perform such a deed (even assuming one could survive the tidal forces). In the Reissner-Nördstrom case, the divergence in the proper acceleration is a consequence of the existence of an s.p. singularity. In our case, however, the most obvious curvature invariants ($`R`$,$`R_{\mu \nu }R^{\mu \nu }`$, $`R^{\mu \nu \sigma \tau }R_{\mu \nu \sigma \tau }`$) show no such singular behaviour on or outside the $`r=r_H`$ surface. Furthermore, we could not find any pp-singularity.<sup>12</sup><sup>12</sup>12Black holes with this kind of singular behaviour outside the horizon -‘naked black holes’- were studied in .
What happens to an accelerated observer? Along the trajectory described by (4.12), the proper acceleration gives
$$a^\mu a_\mu =t^\nu D_\nu t^\mu t^\tau D_\tau t_\mu =\frac{8\mu ^2}{r^6}\left(1+\frac{w^2}{r^2}\right).$$
(4.13)
There is no divergence here. Therefore it seems one would be able to travel into the $`r<r_H`$ region if (and only if) one would have a rocket and a (very) robust spaceship. This is clearly an odd behaviour.
Let us conclude this section by commenting on some recent work on repulsons. As mentioned above, the RN curvature singularity is repulsive. In particular, the extreme RN spacetime with negative mass has a repulsive naked singularity. This is the type of repulsons dealt with in . In order to make contact with the AdS/CFT conjecture, the repulsons therein are brane configurations with near horizon geometries containing an AdS piece. One example is (3.1) with $`J=Q_{KK}=0`$ and $`Q<0`$. Others may be obtained by T-duality. These repulsons are therefore quite different from our case, since the repulsive hypersurface is non-singular for the over-rotating BMPV spacetime.
### 4.2 The gyromagnetic ratio for the BMPV black hole
We also want to address the gyromagnetic ratio of these black holes. As seen before, from the asymptotic behaviour of the metric (3.11), we can read off how the two parameters $`\mu `$ and $`\omega `$ are related to the ADM mass and the two angular momenta parameters of the $`SO(4)`$ rotation group:
$$M_{ADM}=\frac{3\pi \mu }{4G_5},(J_L,J_R)=(0,\frac{\pi \mu \omega }{2G_5}).$$
(4.14)
To get the latter expression we can use the relation between Euler angles and the cartesian coordinates on $`E^4`$, $`X^i`$. Such relation is obtained by solving the embedding constraints for the $`S^3`$ embedding in $`E^4`$. We can then write
$$d\gamma +\mathrm{cos}\beta d\alpha =\frac{2(L^3)_{ki}X^kdX^i}{\rho ^2},$$
(4.15)
where $`L^3`$ is a Hodge self dual two form on $`E^4`$, and $`\rho =\sqrt{X^iX_i}`$. The spacetime angular momentum can now be expressed as
$$J^{ki}=\frac{\pi \mu \omega }{4G_5}(L^3)^{ki},$$
(4.16)
and our definitions of left and right angular momenta are $`J^{21}\pm J^{43}`$, with the ‘+’ sign for $`J_R`$.
The charge associated with the Maxwell field is given by<sup>13</sup><sup>13</sup>13An alternative quite frequent charge normalization is $`Q=(2A_{D2}G_D)^1_{S^{D2}}F^{\mu \nu }𝑑S_{\mu \nu }`$. Then, the charge, magnetic dipole moment and coefficient in Bogomol’nyi bound will differ from ours by a factor of $`\pi /2`$. In this section we set $`G_5=\pi /2`$. So our formulae are the same as the ones we would obtain with the alternative charge definition and setting $`G_5=1`$.
$$Q=\frac{1}{8\pi G_5}_{S^3}F^{\mu \nu }𝑑S_{\mu \nu }=\frac{\sqrt{3}\mu \pi }{2G_5},$$
(4.17)
and so the mass to charge ratio is $`\sqrt{3}/2`$, i.e. it saturates the Bogomol’nyi bound for the $`D=5`$, $`N=2`$ supergravity theory . We should remark that the normalization in (4.17) and for the ADM mass are consistent with Newtonian and Coulomb force laws of the form
$$𝐅_𝐍=\frac{8G_5}{3\pi }\frac{M_1M_2}{r^3},𝐅_𝐂=\frac{2G_5}{\pi }\frac{Q_1Q_2}{r^3},$$
(4.18)
which means that a gravitational-electrostatic force balance is achieved when $`4M_1M_2=3Q_1Q_2`$. An object experimenting a no-force condition on a BPS background is then expected to possess the same mass to charge ratio of $`\sqrt{3}/2`$, as seen in the last section.
Rotation endows a charged black hole with a magnetic dipole moment, which can be read off from the spatial components of the vector potential. Using (4.15) we get
$$\mu ^{ij}=\frac{\sqrt{3}\pi \mu \omega }{4G_5}(L^3)^{ij},$$
(4.19)
and from the usual relation we can read off the gyromagnetic ratio:
$$\mu ^{ij}=g\frac{Q}{2M}J^{ij}g=3.$$
(4.20)
Notice we would obtain the same value if we had used $`\overline{A}`$ in (3.11) instead, since an overall factor gives the same contribution to charge and magnetic moment.
#### 4.2.1 Black Hole superpartners
The field configuration (3.11) can be expressed very simply using an isotropic type radial coordinate, $`\rho `$ defined as $`\rho ^2+\mu =r^2`$. Then we have (2.8). The harmonic form $`H`$ and the one-form $`a`$ are given by (2.11) for a spacetime with a connected even horizon. The multi black hole configuration is obtained by using a harmonic function with $`N`$ poles:
$$H(x^i)=1+\underset{\alpha =1}{\overset{N}{}}\frac{\mu _\alpha }{|x^ix_\alpha ^i|^2},a=\frac{J}{4\mu }(L^3)_i^k_kHdx^i,$$
(4.21)
where $`x_\alpha ^i`$ are constants. A quite natural set of frames is
$$e^0=H^1(dt+a),e^i=H^{\frac{1}{2}}dx^i,$$
(4.22)
for which the field strength $`F`$ takes the form $`F=\sqrt{3}d(e^0)/2`$. Using the Cartan structure equation for the spin connection and expressing $`F`$ and the exterior derivative in terms of the frames we get :
$$\begin{array}{c}\omega _i^0=\frac{_iH}{H^{\frac{3}{2}}}e^0+\frac{f_{ij}}{2H^2}e^j,\omega _{ij}=\frac{f_{ij}}{2H^2}e^0+H^{\frac{3}{2}}\delta _{k[i}_{j]}He^k,\\ \\ F=\frac{\sqrt{3}}{2H^2}\left[H^{\frac{1}{2}}_iHe^ie^0+\frac{f_{ij}}{2}e^ie^j\right],d=e^0H_t+H^{\frac{1}{2}}e^i\left(_ia_i_t\right).\end{array}$$
(4.23)
The non-trivial supersymmetry variation of the $`D=5`$ simple supergravity theory for our bosonic background is the gravitino variation<sup>14</sup><sup>14</sup>14We follow the conventions of . However, the signature choice therein is $`(+)`$, so that our equations differ by a few factors of i. We choose for the flat Gamma matrices $`\mathrm{\Gamma }^{01234}=i`$, and make use of the four dimensional Majorana representation plus $`\mathrm{\Gamma }^4=i\mathrm{\Gamma }^0\mathrm{\Gamma }^1\mathrm{\Gamma }^2\mathrm{\Gamma }^3`$. Therefore, $`\mathrm{\Gamma }^0,\mathrm{\Gamma }^1,\mathrm{\Gamma }^2,\mathrm{\Gamma }^3`$ are real and $`\mathrm{\Gamma }^4`$ purely imaginary. Furthermore, $`\mathrm{\Gamma }^0,\mathrm{\Gamma }^4`$ are antisymmetric and the remaining gamma matrices symmetric.:
$$\begin{array}{c}\delta \mathrm{\Psi }=dϵ+\frac{1}{4}\omega _{ab}\mathrm{\Gamma }^{ab}ϵ+\frac{i}{4\sqrt{3}}\left(e^a\mathrm{\Gamma }_a^{bc}4e^b\mathrm{\Gamma }^c\right)F_{bc}ϵ\hfill \\ \\ =e^0(H_tϵ+[\frac{f_{ij}\mathrm{\Gamma }^{ij}}{8H^2}\frac{i_iH\mathrm{\Gamma }^i}{2H^{\frac{3}{2}}}](1i\mathrm{\Gamma }^0)ϵ)+e^k(H^{\frac{1}{2}}(_ka_k_t)ϵ+\frac{i\mathrm{\Gamma }^0_kH}{2H^{\frac{3}{2}}}ϵ\hfill \\ \\ [\frac{_iH\mathrm{\Gamma }_k^i}{4H^{\frac{3}{2}}}+\frac{if_{ki}\mathrm{\Gamma }^i}{2H^2}](1i\mathrm{\Gamma }^0)ϵ\frac{\mathrm{\Gamma }^{i0}}{4H^2}(f_{ik}f_{ik})ϵ).\hfill \end{array}$$
(4.24)
We have introduced the notation $`f_{ij}`$ for the components of the two form $`f=da`$, and $`f`$ is the Hodge dual of $`f`$ on $`E^4`$. One can check that the $`f`$ that follows from (2.11) (or its multi black hole generalisation) is a self dual form on $`E^4`$. Then, for
$$ϵ=H^{\frac{1}{2}}ϵ_0^K,ϵ_0^K=i\mathrm{\Gamma }^0ϵ_0^K,$$
(4.25)
we get $`\delta \mathrm{\Psi }=0`$, i.e. such $`ϵ`$ is a Killing spinor. However, if one chooses the constant spinor $`ϵ_0`$ obeying
$$ϵ_0^{AK}=i\mathrm{\Gamma }^0ϵ_0^{AK},$$
(4.26)
we get a non-trivial gravitino variation, which can be fed back into the vielbein and gauge field variations to yield the first order superpartners of the BMPV spacetime. These spinors are often called ‘anti-Killing’ spinors. We will comment more on the choice of this specific form for the ‘anti-Killing’ spinors below.
We now generate the superpartners for the BMPV (multi)-black hole spacetimes. To first non-trivial order the variations of the gravitino and gauge field are:
$$\begin{array}{c}\delta \mathrm{\Psi }=\frac{e^0}{H^2}\left[iH^{\frac{1}{2}}_iH\mathrm{\Gamma }^i\frac{1}{4}f_{ij}\mathrm{\Gamma }^{ij}\right]ϵ\frac{e^k}{H^{\frac{3}{2}}}\left[_kH+\frac{1}{2}_iH\mathrm{\Gamma }_k^i+\frac{if_{ki}\mathrm{\Gamma }^i}{H^{\frac{1}{2}}}\right]ϵ,\hfill \\ \\ \delta A=\frac{\sqrt{3}i}{2}\left(\overline{ϵ}\mathrm{\Psi }\overline{\mathrm{\Psi }}ϵ\right)=\frac{\sqrt{3}i}{2H^2}\left[\frac{e^0f_i^k}{2}H^{\frac{1}{2}}e^k_iH\right]\left(\overline{ϵ}\mathrm{\Gamma }_k^iϵ\right),\hfill \end{array}$$
(4.27)
while for the vielbein they follow from $`\delta e^a=\overline{ϵ}\mathrm{\Gamma }^a\mathrm{\Psi }\overline{\mathrm{\Psi }}\mathrm{\Gamma }^aϵ`$ yielding
$$\begin{array}{c}\delta e^0=\frac{i}{H^2}\left[\frac{e^0}{2}f_i^kH^{\frac{1}{2}}e^k_iH\right]\left(\overline{ϵ}\mathrm{\Gamma }_k^iϵ\right),\delta e^j=2iH^2\left[f_{ik}e^ke^0H^{\frac{1}{2}}_iH\right]\left(\overline{ϵ}\mathrm{\Gamma }^{ij}ϵ\right).\hfill \end{array}$$
(4.28)
In the holonomic basis the first order superpartner takes the following form:
$$\begin{array}{c}ds^2=H^2\left[1+iH^2f_i^j\left(\overline{ϵ}\mathrm{\Gamma }_j^iϵ\right)\right]dt^22H^2\left[a_k+i\left(H^2a_kf_i^j3_iH\delta _k^j\right)\left(\overline{ϵ}\mathrm{\Gamma }_j^iϵ\right)\right]dtdx^k\hfill \\ \\ +H\left[dx^kdx_k+\frac{4i}{H}\left(H^1_iHa_k+f_{ki}\right)\left(\overline{ϵ}\mathrm{\Gamma }_j^iϵ\right)dx^kdx^j\right],\hfill \end{array}$$
(4.29)
$$A=\frac{\sqrt{3}}{2H}\left[1+\frac{i}{2H^2}f_i^j\left(\overline{ϵ}\mathrm{\Gamma }_j^iϵ\right)\right]dt+\frac{\sqrt{3}}{2H}\left[a_k+i\left(\frac{1}{2H^2}a_kf_i^j_iH\delta _k^j\right)\left(\overline{ϵ}\mathrm{\Gamma }_j^iϵ\right)\right]dx^k.$$
(4.30)
Comparing the three terms of $`g_{0i}`$ with the three terms of $`A_i`$, we see that they have similar form, but with different coefficients:
The term with purely bosonic angular momentum, $`a_k`$, implies a relation $`J^{ki}=\mu ^{ki}/\sqrt{3}`$, leading to a gyromagnetic ratio $`g=3`$ as seen before;
The term with purely fermionic angular momentum, $`Hϵ^2`$, gives $`J^{ki}=\sqrt{3}\mu ^{ki}`$, implying $`g=1`$. This should be regarded as the gyromagnetic ratio for the fermionic superpartners of a static background;
The ‘mixed’ term, $`afϵ^2`$, yields $`J^{ki}=2/\sqrt{3}\mu ^{ki}`$ and gives rise to $`g=3/2`$.
As we move along the supermultiplet, the gyromagnetic ratio varies. By this we mean that $`g`$ for the BMPV black hole differs from the one for its superpartners. But notice that the supermultiplets are labelled by some value of the bosonic angular momentum, so that the superpartners of the $`D=5`$ RN black hole are not in the same supermultiplet as the BMPV black hole with $`a0`$.
The result to keep in mind is that the gyromagnetic ratio for the $`5D`$ black holes with some kind of angular momentum seems to cover the range $`1g3`$, with the extreme values attained for purely fermionic and bosonic angular momentum respectively. A similar behaviour will be seen when we study the Dirac equation.
Let us now justify the use of the form (4.26) for the anti-Killing spinors. It is quite clear that any choice of $`ϵ=f(x)ϵ_0`$, where $`fH^{\frac{1}{2}}`$ or $`ϵϵ^K`$, will lead to a non trivial gravitino variation and hence some kind of ‘superpartners’. The new set of fields will represent the same physical spacetime configuration as the initial one, just in a different ‘superframe’. This is in a very direct analogy with the excitation of the magnetic field by going to a frame moving with respect to a purely electric source. Then, we still have the same physical setup, i.e. some configuration of electric sources; the fact that we see a different set of fields (namely electric and magnetic) results from the fact that only the components of the electric plus the components of the magnetic field fit into a complete ‘relativistic’ multiplet, while each of these fields would fill a ‘Newtonian’ multiplet. Of course, the fundamental point is that the Lorentz group is a more fundamental symmetry than the Galilean group, which is a low energy approximation. Similarly, the superdiffeomorphisms group of a supergravity theory allows the same physical setup to be described by many different sets of fields. In many physical situations some of these sets of fields contain fermionic excitations while others will be purely bosonic. We can therefore gauge away the fermions, which can then be labelled as pure gauge. In many situations, however, the fermions cannot be gauged away (In the context of $`N=1,D=4`$ Supergravity, necessary and sufficient conditions for an excited gravitino to be pure gauge were given in ).
One way to deal with such gauge arbitrariness is to impose the tracelessness condition to the first order gravitino, i.e., $`\mathrm{\Gamma }^\mu \mathrm{\Psi }_\mu =0`$ . Physically this means that the first order gravitino is a pure spin $`3/2`$ excitation, since the $`\mathrm{\Gamma }^\mu \mathrm{\Psi }_\mu `$ projection of the gravitino transforms as a spin $`1/2`$ representation of the Lorentz group universal covering. Moreover, since $`\mathrm{\Psi }_\mu =\widehat{D}_\mu ϵ`$, where $`\widehat{D}_\mu `$ is the supercovariant derivative, we impose a condition on the existence of superpartners, namely that the candidate background has regular solutions to a modified Dirac equation $`\mathrm{\Gamma }^\mu \widehat{D}_\mu ϵ=0`$. Of course, we are interested in $`ϵ`$ obeying this equation but with $`\widehat{D}_\mu ϵ0`$, in order to get non-trivial superpartners. Our choice for anti-Killing spinors satisfies this criterion.
It has also been argued that a more fundamental criterion for suitable superpartners is to require the first order gravitino to be normalisable . For the superpartners of the static background it is very easy to show this is the case, just as for the $`D=4`$ RN spacetime :
$$|\mathrm{\Psi }|^2=_\mathrm{\Sigma }d^4x\sqrt{g_{(4)}}\mathrm{\Psi }_\mu ^{}\mathrm{\Psi }_\nu g^{\mu \nu }=2\pi ^2M_{ADM}((ϵ_0^{AK})^{}ϵ_0^{AK}),$$
(4.31)
where $`\mathrm{\Sigma }`$ is a spacelike surface.
#### 4.2.2 The gyromagnetic ratio for an electron
A test Dirac fermion with charge $`q`$ and mass $`m`$ interacting with gravitational and electromagnetic fields is described by a wave function obeying the minimally coupled Dirac equation, which takes the standard form
$$\left[\mathrm{\Gamma }^\mu \left(_\mu +\frac{1}{4}\omega _{ab\mu }\mathrm{\Gamma }^{ab}+iqA_\mu \right)m\right]\mathrm{\Psi }=0.$$
(4.32)
Using (4.23), the spin connection term can be rewritten in terms of the Maxwell tensor:
$$\left[\mathrm{\Gamma }^\mu \left(_\mu +iqA_\mu \right)\frac{1}{4\sqrt{3}}\mathrm{\Gamma }^0F_{ab}\mathrm{\Gamma }^{ab}m\right]\mathrm{\Psi }=0.$$
(4.33)
where $`\mathrm{\Gamma }^0`$ is the flat gamma matrix. Hence we see that the problem of studying a Dirac fermion in the background (3.11) can be restated as the study of a non-minimally coupled Dirac fermion interacting solely with an electromagnetic field. Of course this should be expected for a supersymmetric background, since the vanishing of the supercovariant derivative acting on some spinor means that there will be a cancellation of terms between the Maxwell field and the spin connection.
In four dimensional flat space, adding a non-minimal electromagnetic interaction of the type
$$\frac{iq}{8m}\mathrm{\Delta }g\mathrm{\Gamma }^{\mu \nu }F_{\mu \nu },$$
(4.34)
to the minimally coupled Dirac equation yields in the non relativistic limit, a magnetic dipole interaction in the Hamiltonian of the form
$$H_{dipole}=\frac{q}{2m}𝐬𝐁\left(2+\mathrm{\Delta }g\right)$$
(4.35)
The natural higher dimensional generalisation of this result yields for the gyromagnetic ratio associated with a Dirac fermion obeying (4.33) the non-standard value $`g=2+\mathrm{\Delta }g`$ and
$$\mathrm{\Delta }g\mathrm{\Psi }=\frac{2m}{\sqrt{3}q}i\mathrm{\Gamma }^0\mathrm{\Psi }$$
(4.36)
Therefore we arrive at the interesting conclusion that the gyromagnetic ratio of a spin $`1/2`$ particle depends on its spinor direction being parallel to the anti-Killing or Killing spinors. In particular we should get $`g=3`$ for the latter case and $`g=1`$ for the former, when the Dirac fermion is BPS. It is clear that there is a ‘conspiracy’ between the behaviour of the elementary particle and the behaviour of the black holes seen in the last section. Its meaning, however, is not quite clear. Moreover, this seems to be a particular property of the five dimensional family of black holes. For the $`D=4`$ extreme KN, such conspiracy does not arise: $`g=2`$ both for the bosonic background and for the superpartners of the extreme RN background . The Dirac equation in the extreme KN background cannot be easily cast into the non-minimally coupled flat space form. But a result similar to the five dimensional one would require $`\mathrm{\Delta }g=0`$, i.e., the vanishing of the non-minimal coupling. This is not expected to occur. Indeed, even for the $`D=4`$ extreme RN, the Dirac equation can easily be put in a form similar to (4.33) with $`\mathrm{\Delta }g=1/2`$.
Let us close this section by discussing the possible String Theory counterpart of these results. Massive string states have a Schwarzchild radius greater than their Compton wave length; in other words their mass is greater than the Planck mass (in ten dimensions). This led to the suggestion that such states should be identified with some extremal black holes in supergravity . One possible check on this conjecture was made for gyromagnetic ratios. On the String Theory side, the gyromagnetic ratio for heterotic states in the presence of a background gauge field were computed in . A matching was verified with the $`g`$ value for some black holes of $`D=4`$, $`N=4`$ supergravity coupled to 22 vector multiplets (i.e. the low energy field theory for heterotic on $`T^6`$) . On the string theory side, the computation follows from the knowledge of the action for the heterotic string in the presence of the background field. In our case, the black hole does not correspond to string states but rather to D-brane states. Since the open strings describing the D-branes do not couple to the Ramond-Ramond charge it is not clear how one could compute $`g`$ for the microscopic configuration.
## 5 Conclusions
In this paper we argued that within all known BPS, rotating, asymptotically flat stringy black holes, the five dimensional case is rather special. And that one may use these special spacetimes to learn more about the connections between microscopic and macroscopic gravity. Our framework was toroidally compactified string theory, but one may embed the BMPV geometry in M-theory compactified on more general Calabi-Yau spaces . In section 2 we performed a comparison with the typical irregular case: the four dimensional extreme Kerr-Newman. The special properties arise not only from the Chern-Simons term but also from the possibility of having a Hodge self dual rotation two-form. This is illustrated by studying a special class of gravitational waves.
The rich causal structure of these spacetimes also presents a novel feature: CTC’s homotopic to a point in the five dimensional spacetime are resolved in the universal covering of the ten dimensional uplifted geometry. Although CTC’s are clearly a non-perturbative effect in string theory (as the repulson effect of section 4.1), they manifest themselves at weak coupling by the loss of unitarity. The point here is that states violating (3.23) would imply the existence of states with negative $`\stackrel{~}{L}_0`$ eigenvalue. Consequently there would be negative norm states, i.e., non-unitary or ghost states.
Much controversy has surrounded causality as a fundamental principle in theories of gravity. Despite the doubts cast by the information paradox, unitarity is generally a more solid principle in quantum theories. In the same way we still have to understand quantitatively the microscopic description of the most fundamental black hole - Schwarzchild - it will be an interesting problem to understand the microscopic states associated with more fundamental acausal spacetimes, as the Gödel manifold, and scrutinise the role of unitarity.
Another interesting property of these black holes concerns the gyromagnetic ratio. In quantum field theory, requiring a ‘good’ behaviour for the tree level scattering amplitudes singles out $`g=2`$ as the most natural value for elementary particles . This is implemented by a non-minimal electromagnetic coupling for fields with spin higher than $`1/2`$. For spin 1 charged matter, another argument in favour of $`g=2`$ is that a coupling consistent with such gyromagnetic ratio gives rise to a non electromagnetic gauge symmetry . One could argue that a similar statement applies to black holes, if one is trying to interpret them as field theory realizations of fundamental states. It would therefore be interesting to understand in the microscopic context the results of section 4.2.
## Acknowledgments
I would like to thank Miguel Costa, Malcolm Perry, Harvey Reall and Paul Townsend for discussions. I am particularly grateful to Gary Gibbons for many discussions and suggestions. The author is supported by FCT (Portugal) through grant no. PRAXIS XXI/BD/13384/97. This work is also supported by the PPARC grant PPA/G/S/1998/00613. |
warning/0003/nucl-th0003022.html | ar5iv | text | # Tilted Pion Sources from Azimuthally Sensitive HBT Interferometry
## Abstract
Intensity interferometry in noncentral heavy ion collisions provides access to novel information on the geometry of the effective pion-emitting source. We demonstrate analytically that, even for vanishing pair momentum, the cross terms $`R_{ol}^2`$ and $`R_{sl}^2`$ of the HBT correlation function in general show a strong first harmonic in their azimuthal dependence. The strength of this oscillation characterizes the tilt of the major axis of the spatial emission ellipsoid away from the direction of the beam. Event generator studies indicate that this tilt can be large ($`>20^{}`$) at AGS energies which makes it by far the most significant azimuthally sensitive HBT signal at these energies. Moreover, transport models suggest that for pions this spatial tilt is directed opposite to the tilt of the directed flow ellipsoid in momentum space. A measurement of the azimuthal dependence of the HBT cross terms $`R_{ol}^2`$ and $`R_{sl}^2`$ thus probes directly the physical origin of directed pion flow.
PACS numbers: 25.75.+r, 07.60.ly, 52.60.+h
preprint: CERN-TH/2000-076,nucl-th/0003022
Two-particle momentum correlations between identical particles are commonly used to extract space-time and dynamical information about the particle emitting source in heavy ion collisions. The basis for this intensity interferometric method is the equation
$`C(𝐪,𝐊)`$ $`=`$ $`1+{\displaystyle \frac{\left|d^4xS(x,K)e^{iqx}\right|^2}{\left|d^4xS(x,K)\right|^2}},`$ (1)
$`q=p_1p_2`$, $`K=\frac{1}{2}(p_1+p_2)`$, which relates the phase-space density $`S(x,K)`$ of the source to the measured 2-particle correlation function $`C(𝐪,𝐊)`$. Experimental measurements of $`C`$ are usually parametrized in terms of the intercept $`\lambda (𝐊)`$ and the HBT radii $`R_{ij}^2(𝐊)`$ by
$$C(𝐪,𝐊)=1+\lambda (𝐊)\mathrm{exp}\left[\underset{i,j=o,s,l}{}q_iq_jR_{ij}^2(𝐊)\right].$$
(2)
In this Cartesian osl-system the relative momentum is decomposed into components parallel to the beam ($`l`$ = longitudinal), parallel to the transverse component of $`𝐊`$ ($`o`$ = out), and in the remaining third direction ($`s`$ = side).
Over the last decade the experimental frontier in studying identical two-particle correlations was defined by more and more differential measurements of the HBT radii $`R_{ij}^2(𝐊)`$. At both AGS and SPS energies, the longitudinal ($`K_L`$) and transverse ($`K_{}`$) pair momentum dependence is now well-studied for central and reaction-plane averaged non-central collisions. Most importantly, these studies have led to a detailed characterization of the longitudinal expansion and the transverse radial flow of the reaction zone. The next challenge is a similarly detailed study of the $`R_{ij}^2(𝐊)`$ as a function of the azimuthal orientation $`\mathrm{\Phi }`$ of the transverse pair momentum $`𝐊_{}`$ with respect to the impact parameter $`𝐛`$ in non-central collisions. This $`\mathrm{\Phi }`$-dependence reveals qualitatively new information about the space-time structure of the source and yields new insights on the underlying nature of flow. In fact, under reasonable assumptions all 10 components of the source’s spatial correlation tensor can be recovered.
An azimuthally sensitive HBT analysis involves all six parameters $`R_{ij}^2`$, all of which are functions of all 3 components of the pair momentum $`K_{}`$, $`Y`$ and $`\mathrm{\Phi }`$ . They provide information about the source in space-time according to the following relations :
$`R_s^2(K_{},\mathrm{\Phi },Y)=S_{11}\mathrm{sin}^2\mathrm{\Phi }+S_{22}\mathrm{cos}^2\mathrm{\Phi }S_{12}\mathrm{sin}2\mathrm{\Phi },`$ (3)
$`R_o^2(K_{},\mathrm{\Phi },Y)=S_{11}\mathrm{cos}^2\mathrm{\Phi }+S_{22}\mathrm{sin}^2\mathrm{\Phi }+S_{12}\mathrm{sin}2\mathrm{\Phi }`$ (4)
$`2\beta _{}S_{01}\mathrm{cos}\mathrm{\Phi }2\beta _{}S_{02}\mathrm{sin}\mathrm{\Phi }+\beta _{}^2S_{00},`$ (5)
$`R_{os}^2(K_{},\mathrm{\Phi },Y)=S_{12}\mathrm{cos}2\mathrm{\Phi }+\frac{1}{2}\left(S_{22}S_{11}\right)\mathrm{sin}2\mathrm{\Phi }`$ (6)
$`+\beta _{}S_{01}\mathrm{sin}\mathrm{\Phi }\beta _{}S_{02}\mathrm{cos}\mathrm{\Phi },`$ (7)
$`R_l^2(K_{},\mathrm{\Phi },Y)=S_{33}2\beta _lS_{03}+\beta _l^2S_{00},`$ (8)
$`R_{ol}^2(K_{},\mathrm{\Phi },Y)=\left(S_{13}\beta _lS_{01}\right)\mathrm{cos}\mathrm{\Phi }\beta _{}S_{03}`$ (9)
$`+\left(S_{23}\beta _lS_{02}\right)\mathrm{sin}\mathrm{\Phi }+\beta _l\beta _{}S_{00},`$ (10)
$`R_{sl}^2(K_{},\mathrm{\Phi },Y)=\left(S_{23}\beta _lS_{02}\right)\mathrm{cos}\mathrm{\Phi }`$ (11)
$`\left(S_{13}\beta _lS_{01}\right)\mathrm{sin}\mathrm{\Phi }.`$ (12)
The pair velocity $`𝜷=𝐊/K^0`$ arises from the on-shell constraint $`q^0=𝐪𝜷`$; it mixes spatial and temporal information. $`S_{\mu \nu }`$ denotes the spatial correlation tensor
$$S_{\mu \nu }=\stackrel{~}{x}_\mu \stackrel{~}{x}_\nu ,\stackrel{~}{x}_\mu =x_\mu \overline{x}_\mu ,(\mu ,\nu =0,1,2,3)$$
(13)
which measures the Gaussian width in space-time of the emission function $`S(x,K)`$ around the point of highest emissivity $`\overline{x}_\mu =\stackrel{~}{x}_\mu `$ :
$`\stackrel{~}{x}_\mu \stackrel{~}{x}_\nu (K)`$ $`=`$ $`{\displaystyle \frac{d^4x\stackrel{~}{x}_\mu \stackrel{~}{x}_\nu S(x,K)}{d^4xS(x,K)}}.`$ (14)
It is the inverse of the curvature tensor $`B_{\mu \nu }`$ :
$$S(x,K)N(K)S(\overline{x},K)\mathrm{exp}\left[\frac{1}{2}\stackrel{~}{x}^\mu B_{\mu \nu }\stackrel{~}{x}^\nu \right].$$
(15)
This approximation neglects non-Gaussian components of the emission function whose influence on the HBT radii can in most practical cases be neglected . We emphasize that in (13) $`S_{\mu \nu }`$ is defined in terms of Cartesian coordinates in an impact parameter fixed system, in which $`x_1=x`$ is parallel to the impact parameter $`𝐛`$ and $`x_3=z`$ lies in the beam direction.
The general relations (12) separate the explicit $`\mathrm{\Phi }`$-dependence of the HBT-radii (which is a consequence of the azimuthal rotation of the $`osl`$-system relative to $`(x_1,x_2,x_3)`$) from the implicit $`\mathrm{\Phi }`$-dependence of the space-time widths $`\stackrel{~}{x}_\mu \stackrel{~}{x}_\nu (K_{},Y,\mathrm{\Phi })`$ (which reflects a $`\mathrm{\Phi }`$-dependent change of the shape of the effective emission region) . Existing studies of (12) focussed on the detailed interplay between explicit and implicit $`\mathrm{\Phi }`$-dependences in the HBT radii $`R_s^2`$, $`R_o^2`$ and $`R_{os}^2`$ . Here we show, however, that some of the most striking features are found in analyzing the $`\mathrm{\Phi }`$-dependences of $`R_{ol}^2`$ and $`R_{sl}^2`$ which so far received less attention .
The following discussion is simplified significantly by the important observation that the implicit $`\mathrm{\Phi }`$-dependence of $`S_{\mu \nu }`$ is weak. It can be neglected relative to the explicit one given in (12) as long as the $`\mathrm{\Phi }`$-dependence of space-momentum correlations in the source is small compared to the thermal smearing, and for $`K_{}0`$ it vanishes completely. Studies with the RQMD model indicate that the first condition works well at least up to $`p_T=300`$ MeV/c for Au+Au collisions at 2 $`A`$ GeV . Beyond such model studies, a simple scale argument illustrates why neglecting the implicit $`\mathrm{\Phi }`$-dependence relative to the explicit one has a much wider kinematical region of validity than neglecting the implicit $`K_{}`$-dependence relative to the explicit one in (12): the latter is suppressed near $`K_{}=\mathrm{\hspace{0.17em}0}`$ ($`\beta _{}=\mathrm{\hspace{0.17em}0}`$), and it multiplies only space-time variances involving $`\stackrel{~}{t}`$ which are numerically small in practice. In contrast, the explicit $`\mathrm{\Phi }`$-dependence in (12) leads to prefactors $`\mathrm{cos}(n\mathrm{\Phi })`$, $`\mathrm{sin}(n\mathrm{\Phi })`$ oscillating between $`1`$ and $`1`$ even for $`K_{}=0`$ , and it multiplies the numerically large components of $`S_{\mu \nu }`$. The assumption of vanishing implicit $`\mathrm{\Phi }`$-dependence can be checked experimentally , and deviations can be quantified in a full harmonic analysis given elsewhere . Also, while it requires weak transverse flow, there is no such restriction on the longitudinal flow. Qualitatively, the main findings presented here do not depend on this assumption, but it simplifies our presentation and allows for a particularly intuitive geometric picture of the new effect discussed here.
With this proviso, the components $`S_{\mu \nu }`$ in (12) become $`\mathrm{\Phi }`$-independent constants which describe the same source being viewed from all angles $`\mathrm{\Phi }`$. We turn briefly to symmetry considerations at midrapidity. Considering collisions between equal mass nuclei, it can be rigorously shown that, as a consequence of point reflection symmetry around the spatial origin and mirror symmetry with respect to the reaction plane, five of the off-diagonal components $`S_{\mu \nu }`$ (all except $`S_{13}`$) oscillate symmetrically around zero. Coupled with the condition of vanishing implicit $`\mathrm{\Phi }`$-dependence this implies
$$S_{01}=0,S_{02}=0,S_{03}=0,S_{12}=0,S_{23}=0.$$
(16)
These equations and the fact that around midrapidity the average $`\beta _l`$ is zero (although average $`\beta _l^20`$) allow us to write the HBT radius parameters (12) in terms of 5 non-vanishing components only:
$`R_s^2=\frac{1}{2}\left(S_{11}+S_{22}\right)+\frac{1}{2}\left(S_{22}S_{11}\right)\mathrm{cos}2\mathrm{\Phi },`$ (17)
$`R_o^2=\frac{1}{2}\left(S_{11}+S_{22}\right)\frac{1}{2}\left(S_{22}S_{11}\right)\mathrm{cos}2\mathrm{\Phi }+\beta _{}^2S_{00},`$ (18)
$`R_{os}^2=\frac{1}{2}\left(S_{22}S_{11}\right)\mathrm{sin}2\mathrm{\Phi }`$ (19)
$`R_l^2=S_{33}+\beta _l^2S_{00},`$ (20)
$`R_{ol}^2=S_{13}\mathrm{cos}\mathrm{\Phi },`$ (21)
$`R_{sl}^2=S_{13}\mathrm{sin}\mathrm{\Phi }.`$ (22)
Since the $`\mathrm{\Phi }`$-dependences of $`R_s^2`$ and $`R_o^2`$ explicitly separate $`S_{22}`$ from $`S_{11}`$, the emission duration $`S_{00}`$ can now be determined without additional model assumptions , contrary to the case of collisions at $`b=\mathrm{\hspace{0.17em}0}`$.
Given the measured weak $`Y`$-dependence of the HBT-radii , Eqs. (22) can be used in practice also for event samples which are averaged over large $`Y`$-windows symmetric around $`Y=0`$. According to (22), the HBT radius parameters $`R_o^2`$, $`R_s^2`$ and $`R_{os}^2`$ all show second harmonic oscillations of the same strength $`\frac{1}{2}\left(S_{11}S_{22}\right)`$. This is the $`R_{o,2}^{c}{}_{}{}^{2}=R_{s,2}^{c}{}_{}{}^{2}=R_{os,2}^{s}{}_{}{}^{2}`$ rule for second harmonic coefficients ; leading deviations from this rule have been quantified and provide a consistency check on the assumption of negligible implicit $`\mathrm{\Phi }`$-dependence. More strikingly, $`R_{ol}^2`$ and $`R_{sl}^2`$ display purely first harmonic oscillation at midrapidity which are easier to measure. The expected identical amplitudes for these oscillations provide a further consistency check on our assumptions.
The required $`\mathrm{\Phi }`$-binning puts severe demands on the pair statistics. A first observation of azimuthally oscillating HBT radii was reported in . Such measurements require a reasonably accurate determination of the reaction plane; a typical uncertainty of 30 reduces the first and second harmonics in (12) by $`15`$% and $`45`$%, respectively, but these losses can be corrected for .
While the amplitude of the oscillations of $`R_o^2`$, $`R_s^2`$, and $`R_{os}^2`$ are given by the difference between the transverse source sizes in and perpendicular to the reaction plane, that of the oscillations of $`R_{sl}^2`$ and $`R_{ol}^2`$ is given by $`S_{13}\stackrel{~}{x}\stackrel{~}{z}`$. Parameterizing the source by an ellipsoid, a nonzero $`S_{13}`$ corresponds to a tilt in the reaction plane of the longitudinal major axis of the ellipsoid away from the beam direction. It can be characterized by a tilt angle
$$\theta _s=\frac{1}{2}\mathrm{tan}^1\left(\frac{2S_{13}}{S_{33}S_{11}}\right).$$
(23)
Rotating the spatial correlation tensor $`S_{\mu \nu }`$ by $`\theta _s`$ yields a purely diagonal tensor $`S^{}=R_y^{}(\theta _s)SR_y(\theta _s)`$ whose eigenvalues are the squared lengths of the 3 major axes.
We illustrate the role of the tilt angle (23) with a tilted Gaussian toy distribution with no space-momentum correlations:
$`S(x,K)=e^{E/T}\mathrm{exp}\left({\displaystyle \frac{x^2}{2\sigma _x^2}}{\displaystyle \frac{y^2}{2\sigma _y^2}}{\displaystyle \frac{z^2}{2\sigma _z^2}}{\displaystyle \frac{t^2}{2\sigma _t^2}}\right),`$ (24)
$`x^{}=x\mathrm{cos}\mathrm{\Theta }z\mathrm{sin}\mathrm{\Theta },z^{}=x\mathrm{sin}\mathrm{\Theta }+z\mathrm{cos}\mathrm{\Theta }.`$ (25)
To avoid relativistic complications, the “temperature” $`T`$ is kept small (20 MeV) in the following. Fig. 1 shows the projection of this source onto the reaction ($`xz`$) plane.
Using the model (25) with the parameters of Fig. 1 to randomly generate a set of phase-space points, we constructed a three-dimensional correlation function (with $`4\times 10^5`$ pairs with $`q<100`$ MeV/c) for each of eight $`45^{}`$-wide $`\mathrm{\Phi }`$ bins, according to the prescription and code of Pratt . Fitting each with the Gaussian parametrization (2) yields the HBT radii presented in Fig. 2. Treating the ten components of $`S_{\mu \nu }`$ as parameters, we perform a global fit with Eqs. (12) on these $`\mathrm{\Phi }`$-dependent radii. The fit results are indicated by solid lines in Fig. 2. Application of (23) to the fit results yields $`\theta _s=24.6^{}\pm 0.6^{}`$, in good agreement with the input value $`\mathrm{\Theta }=\mathrm{\hspace{0.17em}25}^{}`$. The diagonal elements $`S_{\mu \mu }^{}`$ reproduce, within statistical errors, the input values for the squares of the homogeneity lengths: $`\sigma _x=4.05\pm 0.05`$ fm, $`\sigma _y=4.97\pm 0.04`$ fm, $`\sigma _z=6.97\pm 0.04`$ fm, and $`\sigma _t=4.78\pm 0.45`$ fm. (The larger uncertainty in $`\sigma _t`$ arises from the low $`\beta _{}`$ of the pions in our example.)
While one may escape the effects of transverse flow (which may generate a $`\mathrm{\Phi }`$-dependent effective source) by selecting pion pairs at low $`K_{}`$, longitudinal flow, which generates $`zp_z`$ correlations, is generally stronger and cannot be cut away. Fortunately, since they are essentially orthogonal to the azimuthal dependences we are discussing, such correlations do not drastically alter the intuitive geometric picture we have discussed – the same source is still viewed from all angles $`\mathrm{\Phi }`$.
As an example we added a boost-invariant longitudinal flow component in $`z`$-direction to our toy source (scaled so that the collective flow velocity at $`z=\pm \sigma _z`$ is equal to the thermal velocity), leaving the geometry unchanged. This results in (i) an increase in the tilt angle $`\theta _s`$ from 25 to 33 and (ii) a reduction in $`S_{33}^{}`$ from 49 fm<sup>2</sup> to 31 fm<sup>2</sup>. The other components $`S_{\mu \mu }^{}`$ vary negligibly from the scenario without flow. Familiar from the case of azimuthally symmetric HBT, effect (ii) is understood in terms of a reduction in the length of homogeneity due to the flow : HBT correlations arise from particle pairs with close-by momenta; the space-momentum correlations induced by longitudinal flow then imply that they will be close-by in coordinate space as well. The increased tilt is similarly understood, by examining the spatial distribution of emission points for pions with low $`p_z`$. The shaded region in Fig. 1 shows the effective source for pions with $`|p_z|<40`$ MeV/$`c`$; it is clearly less prolate and more tilted than in the case of no flow (contour lines).
n
$`\mathrm{\Phi }`$-sensitive HBT studies and the measurement of spatial tilt connect for the first time the physics of directed flow with the space-time structure of the source . We consider this an essential step towards a full understanding of the phase-space dynamics of heavy-ion collisions. Although a detailed discussion must await a longer paper, we here shortly touch on the main physics points, using results from a realistic transport model.
We performed simulations of semiperipheral Au+Au collisions at 2 $`A`$ GeV with the RQMD (v2.3) model . The top panel of Fig. 3 shows $`p_x`$ – the average pion momentum in the reaction plane – as a function of momentum $`p_z`$ along the beam axis. Qualitatively consistent with experimental observations , a very weak negative directed flow (“anti-flow”) signal is observed – the average emission ellipsoid in momentum space is tilted to a negative angle with respect to the beam (the direction of directed proton flow defines the positive direction). The magnitude of the collective motion ($`10`$ MeV/$`c`$) is small compared to the typical $`p_T`$ scale ($`200`$ MeV/$`c`$); hence thermal smearing dominates. The bottom panel shows that, while the spatial distribution displays a richer structure than our toy model, it is nevertheless always characterized by a significant positive tilt – opposite the average tilt in momentum space. A full correlation function analysis of the RQMD events yields qualitatively similar results as those shown in Fig. 2.
This bears directly on the physical causes of directed pion flow at these energies. Detailed transport model studies have shown that pion reflection from (not absorption by) the nucleonic matter is at the root of directed pion flow at these energies. Focussing on the forward hemisphere, if absorption processes ($`\pi NN\mathrm{\Delta }NNN`$) were dominant in producing pion flow, we would expect an absence of $`\pi `$ emission points in the $`+x`$ quadrant, i.e. a negative tilt in coordinate space and in momentum space. Since it is the point of last scattering (as opposed to the original point of creation) which is relevant for HBT correlations , it is clear that reflection ($`\pi N\mathrm{\Delta }\pi N`$) from flowing participant or spectator baryons leads to a positive tilt in coordinate space as seen in Fig. 3: the reflected pions “illuminate” the coordinate-space anisotropies of the nucleonic matter. In this simple picture, then, the sign of $`\theta _s`$ immediately distinguishes between these two possibilities.
The arrows in Fig. 3 represent the average momenta of pions for different values of $`z`$. The resulting structure further underscores the importance of pion rescattering: Clearly, the more numerous pions from the high-density region around $`z=0`$ dominate, generating the anti-flow signal seen in experiment. However, pions from the more dilute large-$`|z|`$ region have less opportunity for rescattering and so retain the positive $`p_xp_z`$ correlation of their (flowing) parent $`\mathrm{\Delta }`$’s. Similar considerations generate a sign change in the pion flow as the impact parameter is varied in transport models .
In summary, for non-central collisions all ten components of the spatial correlation tensor $`S_{\mu \nu }`$ are accessible by $`\mathrm{\Phi }`$-dependent HBT measurements. Based on symmetry and scale considerations we argue that for low $`K_{}`$ the explicit $`\mathrm{\Phi }`$-dependence of Eqs. (12) dominates. Consistency relations allow to check whether this is true in practice. The spatial correlation tensor $`S_{\mu \nu }`$ can then be extracted completely from a global fit to the six $`\mathrm{\Phi }`$-dependent HBT radii. At midrapidity, the five nonvanishing components of $`S_{\mu \nu }`$ correspond to the four spacetime lengths of homogeneity and a tilt of the source in the reaction plane, away from the beam direction. This tilt, which may be quite large at AGS energies, causes striking and relatively easily measurable first-order harmonic oscillations in $`R_{ol}`$ and $`R_{sl}`$ and can give a direct experimental handle on the origin of pion flow at these energies.
The work of M.A.L. is supported by NSF Grant PHY-9722653 and that of U.H. by DFG, GSI and BMBF. |
warning/0003/astro-ph0003399.html | ar5iv | text | # Comments on the paper “Polarimetric Constraints on the Optical Afterglow Emission from GRB 990123” by Hjorth et al. (Science, 26 1999)
<sup>1</sup><sup>1</sup>institutetext: Theoretical Physics Division, Bhabha Atomic Research Centre
Mumbai 400085, India, amitra@apsara.barc.ernet.in (Received ; Accepted )
## Abstract
GRB 990123 is the most luminous event detected so far, and in an important paper, Hjorth et al. (HJ (1999)) reported an upper limit on the degree of linear polarization of the optical afterglow for this burst ($`P<2.3\%`$). One of the interprtations for this small value of $`P`$ was that the emission was probably due a relativistic jet with ordered magnetic field, and the viewing angle in the lab frame $`\theta ^{}\mathrm{\Gamma }^1`$, where $`\mathrm{\Gamma }`$ is the bulk Lorentz factor of the jet at the time of the optical emission. We point out that this conclusion resulted from a confusion between the angles measured in the lab frame and in the plasma rest frame. For ordered magnetic field, one would actually obtain a large value of $`P`$ because the above mentioned angle would correspond to a very large angle, $`\theta \pi /2`$, in the plasma rest frame. And this is probably the case with the blazars. On the other hand, it is indeed possible to have $`P0`$ if the magnetic field of GRB 990123 was completely chaotic and the viewing angle was considerably smaller than the semi-angle of the jet.
###### Key Words.:
Gamma rays: bursts – Gamma rays: theory
The above mentioned paper by Hjorth et al. (HJ (1999)) is very important in that it imposes an upper limit on the degree of linear polarization ($`P<2.3\%`$) for the optical afterglow of most luminous Gamma Ray Burst 990123. When corrected for various effects which might affect the true intrinsic polarization, this paper gave a value of $`P=0.0\pm 1.4\%`$, and concluded that “this is consistent with no linear polarization”, and an upper limit of $`P<2.3\%`$ was set at 95% confidence limit.
Hjorth et al. then attempted to interpret this result. Since a spherical fireball too can develop a somewhat clumpy magnetic field structure and may result in a significant value of $`P<10\%`$ (Gruzinov & Waxman GW (1999)), Hjorth noted that the observed low value of $`P`$ is consistent with emission from a spherical fireball. Then they also explored whether such a low value of $`P`$ is consistent with emission from a narrow ultra relativistic jet. Specifically, they considered the possibility whether the observed value of $`P`$ could be very low if the jet is observed at an angle $`\theta ^{}\mathrm{\Gamma }^1`$ where $`\mathrm{\Gamma }`$ is the bulk Lorentz factor of the jet at the time of optical emission. The reason that they considered this particular value of $`\theta ^{}`$ is the following: For a random GRB event, “for smaller viewing angles, the solid angle decreases and for larger angles the flux drops”. At the time of the optical emission, the estimated value of $`\mathrm{\Gamma }1020`$ and the corresponding $`\theta ^{}3^{}6^{}`$. To seek an answer to the question mentioned above, Hjorth et al. banked on a work by Celloti & Matt (CM (1994), CM) CM (1994). From Fig. 2 & 3 of CM, Hjorth et al. concluded that the value of $`P0`$ for $`\theta ^{}3^{}6^{}`$, :
“The measured value of polarization is therefore consistent with a small angle between the jet axis and our line of sight”.
We would like to point out here that while arriving at this conclusion, Hjorth et al. confused between the angle measured in the observer’s frame ($`\theta ^{}`$) and the angle measured in the rest frame of the plasma emitting the radiation ($`\theta `$). The absissa of Figs. 2 and 3 in CM is $`\theta `$ and not $`\theta ^{}`$ contrary to what has been considered by Hjorth et al. For small $`\theta ^{}`$, these two angles are connected by the well known special relativistic formula
$$\mathrm{sin}\theta =\frac{2\mathrm{\Gamma }\theta ^{}}{1+\mathrm{\Gamma }^2\theta _{}^{}{}_{}{}^{2}}$$
(1)
Thus for $`\theta ^{}\mathrm{\Gamma }^1`$, $`\theta \pi /2`$! In other words, in plasma rest frame, the small lab frame angle translates into a very large value. Then the Figs. 2 and 3 of CM would suggest a large value of $`P22\%`$ rather than $`P0\%`$. In the most popular classification scheme of the Active Galactic Nuclei, the viewing angle is smallest for the radio selected blazars and they are indeed found to have very high values of $`P1040\%`$. In fact this point was clearly mentioned by CM:
“As already mentioned, relativistic beaming effects, which are believed to affect the emission of blazars, imply that the observed radiation can be emitted at large angles in the plasma frame, even if the line of sight is close to the jet axis. Therefore, these sources could also show high X-ray polarization.”
Thus, the interpretation by Hjorth et al., in the framework of the work by CM, that the absence of polarization is consistent with a jet interpretation is incorrect and resulted from a confusion between angles measured in the lab frame and plasma rest frame. For the so-called galactic micro-quasars, it is believed that the viewing angle of the jet is larger as compared to the radio selected blazars The micro-quasars too display a fairly high value of $`P1015\%`$ which is however, smaller than the average linear polarization observed by the radio-selected blazars (Mirabel & Rodriguez Mira (1999)). And thus these two objects broadly support a scheme in which the degree of linear polarization increases with decreasing viewing angle of the relativistic jet. Physically this implies that a relativistic turbulent astrophysical jets may be endowed with some ordered magnetic field.
Having said this, we emphasize that, if one moves away from the uniform magnetic field configuration of CM, it might indeed be possible that $`P0`$ for a very small $`\theta ^{}`$ for the highly unusual case where there is no large scale ordered magnetic either along the axis of the jet or in a direction perpendicular to it (in such a case, however, it would be difficult to undrstand the phenomenon of jet confinement and acceleration). In particular, if the magnetic field lies in a plane perpendicular to the jet axis and is disordered on large scale, one would obtain zero linear polarization if the jet is viewd along the jet axis. This would be so because there would be total (axial) symmetry around the viewing direction. However, even in this case, if the jet is viewed off-axis, there would be a temporally variable finite $`P`$ which can be parameterized as (Ghisellini & Lazzati GL (1999))
$$P_{max}0.19P_0\left(\frac{\theta }{\theta _c}\right)(0.05\frac{\theta }{\theta _c}1;1^{}\theta _c15^{})$$
(2)
Here $`P_06070\%`$ is the intrinsic synchrotron polarization, $`\theta _c`$ is the semi-angle of the jet, and a radiation spectral index of 0.6 is assumed. Also, now, we have dropped the prime from the lab frame angle. Thus the value can reach $`10\%`$ if the jet is viewed along its edge. This conclusion is reinforced by model of relativistic jet acceletation by supposed completely tangled magnetic field (Heinz & Begelman HB (1999)).
However if supposed GRB jets are to be modelled on their AGN counterparts, it may be necessary to consider the existence of some ordered magnetic field in order to understand jet collimation away from the central engine, and one may look forward to detect even much higher value of $`P`$.
Coming back to the work of CM, it dealt with self Snchrotron Compton model of radiation emission, whereas GRBs afterglows are generally explained as direct Snychrotron emission. For a given magnetic field configuration, the polarizationn is in general higher in the latter case. To conclude, although the result obtained by Hjorth et al. that $`P<2.3\%`$ for the optical afterglow of GRB 990123 is important, the interpretation that this low value of $`P`$ could be understood if the viewing angle is $`\mathrm{\Gamma }^1`$ is incorrect if there is an ordered magnetic field in the jet. On the other hand, such a low value of $`P`$ may occur if the jet magnetic field is completely tangled with no component of field along the axis, and if the viewing angle is much smaller than the semi-angle of the jet. If we accept this latter interpretation, we need to understand the following: If there is usually an ordered magnetic field, over and above small scale chaotic magnetic field, in relativistic turbulent plasma associated with blazars on scales of $`100`$pc or more and for the so called micro-quasars, why the relativistic plamsa responsible for the emission of GRB 990123 was completely turbulent on all scales. Considering all such aspects, it seems that the afterglow of GRB 990123 was more likely to be quasispherical rather than jet type. This is so because even for a quasispherical afterglow, the observed spectral steepening can be explained if it were propagating within a thick presupernova wind (Dai & Lu DL1 (1999)). Yet, we caution that no definitive stand should be taken at this juncture. A broader discussion on the topic of jet collimation, acceleration, magnetic field generation and expected degree of linear polarization is beyond the scope of this research note. |
warning/0003/cond-mat0003172.html | ar5iv | text | # A minimal model for slow dynamics: Compaction of granular media under vibration or shear
## 1 Introduction
The issue of slow dynamics, rare events and anomalous diffusion is object to ongoing research in statistical physics . One example for an experimental realization of slow dynamics is the compaction of granular materials, being of interest for both industrial applications and research. Many compaction experiments have been carried out in the last decades, see e.g. , but the evolution of density with time is far from being well understood. Recent laboratory experiments concern pipes filled with granular media and periodically accelerated in order to allow for some reorganization . The compaction dynamics was obtained to be logarithmically slow and could be reproduced with a simple parking model . Alternative experiments concern a sheared block of a granular model material (monodisperse glass spheres) and display a similar dynamics . Numerical model approaches, like a frustrated lattice gas (the so-called “Tetris” model) also lead to this slow dynamic behavior, as well as some theory based on stochastic dynamics . The fact that a peculiar dynamics is reproduced by so many models indicates that it is a basic and essential phenomenon.
Rather than modeling granular systems in all details, e.g. in the framework of molecular dynamics or lattice gas simulations , we will propose a very simple model based on the picture of a random walk in a random energy landscape, for a review see , and even simpler than recent, very detailed considerations in the same spirit . Random walks have been examined, for example, on fractals and ultrametric spaces, where the continuous time random walk was introduced in order to allow a mathematical treatment . If a random walker is situated in a uncorrelated, random, fractal energy landscape, the process is called Sinai-diffusion . The aim of this study is to show that the Sinai model is in qualitative agreement with the compaction dynamics of granular media. An issue not addressed here is a quantitative adjustment which can be reached, for example, by introducing correlations in the EL.
## 2 Summary of the experimental results
The subject of compaction of granular material has been recently revisited through careful experiments carried out by Knight et al. and Nowak et al. . The experiment consists of a vertical cylinder, full of monodisperse beads, which is submitted to successive distinct taps of controlled acceleration (vertical vibration). The measurement of the mean volume fraction after each tap gives a precise information about the evolution of the compaction. From the first experiments performed with taps of constant amplitude, the increase in volume fraction was found to be a very slow process well fitted by the inverse of a logarithm of the number of taps. Nowak et al. have then studied the compaction under taps of variable amplitude, and showed that irreversible processes occur during the compaction. Starting with a loose packing, the evolution of the volume fraction is not the same when increasing the amplitude of vibration as when decreasing. The first branch appears to be irreversible, whereas the second is reversible.
We have recently performed a compaction experiment based on cyclic shear applied to an initially loose granular packing . A parallelepipedic box full of beads is submitted to a horizontal shear through the periodic motion of two parallel walls at amplitude $`\theta _{\mathrm{max}}`$, see Fig. 1(a). Compaction occurs during this process, leading to crystallisation of the beads in the case of a monodisperse material. The control parameter in this configuration is the maximum amplitude of shear $`\theta _{\mathrm{max}}`$ (inclination angle of the walls). The measurement of the mean volume fraction $`\varphi `$ shows that compaction under cyclic shear is a very slow process as in the vertical vibration experiments (typically $`5\times 10^4`$ shear cycles or taps). The higher the shear amplitude the more efficient is the compaction (shear amplitudes up to $`\theta _{\mathrm{max}}<12.5^{}`$ were examined), when starting with a loose packing of the same initial volume fraction. More surprising results arise when the packing is submitted to a sudden change in shear amplitude. We have observed that a “jump” in volume fraction occurs which is opposite and proportional to the change in $`\theta _{\mathrm{max}}`$. Sudden increase (resp. decrease) in the shear amplitude decreases (resp. increases) the volume fraction (Fig. 1). The response is very rapid (less than 20 cycles) and quasi-independent of the state before the angle change. For more detailed experimental results see .
## 3 The Model
A naive picture that evolves out from the experimental observations is the analogy between the packing of beads and a thermal system seeking for a minimum of energy in a very complex potential-energy landscape . Due to some agitation (shear) compaction occurs and the total potential energy of the packing decreases. Starting from a loose packing at $`n=0`$, the vibrational or shear excitation can be seen as the analog of the temperature in the sense that the excitation allows for an exploration of the phase space. In a granular packing of monodisperse spheres, the absolute minimum of the energy is obtained for a perfect (fcc) or (hc) crystal with volume fraction $`\varphi =0.74`$. If the energy landscape is complex with a lot of different scale valleys or hills, one can understand that an efficient, fast compaction will be obtained with high temperatures: the system is then able to escape the deep local valleys and find a valley with lower potential energy. A decrease of the temperature is then needed to explore local fine-scale minima. The goal of this paper is simply to explore this idea by studying the dynamics of random walkers on a random landscape (the Sinai model), and to show that this very naive picture gives results in qualitative agreement with the experimental observations.
Our model is based on the assumption that all possible configurations of a granulate in a given geometry can be mapped onto an “energy landscape” (EL). Since a simple two-level system (as used to model the dynamics in simple glasses) does not lead to the experimentally observed phenomenology, we assume a fractal energy landscape created by a random walk in energy with phase space coordinate $`x`$. A typical EL with energy $`V(x)`$ is schematically shown in Fig. 2. The stepsize in energy is $`\mathrm{\Delta }V`$, the mean of the energy landscape is $`V_{\mathrm{mean}}`$ and its absolute minimum is $`V_{\mathrm{min}}`$. Here, the EL is symmetric to its center, in order to allow for periodic boundary conditions in $`x`$. Given some EL, the granulate is now modeled as an ensemble of random walkers diffusing on the EL, with a temperature $`T_{RW}`$. The analog to the density of a granular packing is the rescaled energy of an ensemble of random walkers in the energy landscape
$$\nu =1\frac{EV_{\mathrm{min}}}{V_{\mathrm{mean}}V_{\mathrm{min}}},$$
(1)
which we denote as density in the following. For the (random) initial configuration, the energy will be $`EV_{\mathrm{mean}}`$ so that $`\nu =0`$; for the close-packing configuration, i.e. all RWs are in the absolute minimum, one has $`EV_{\mathrm{min}}`$ and thus $`\nu =1`$. The energy landscape has the size $`L`$, the ensemble of RWs consists of $`R`$ random walkers and $`S`$ is the number of steps performed.
Since the energy of the RWs in the EL corresponds to the potential energy of the packing, we interpret the (constant) stepsize $`\mathrm{\Delta }V`$ (used here as one possibility for the construction of the EL) as a typical activation energy barrier. The maximum of $`V`$ corresponds to a random loose, local packing density, the mean to the (initial) random close packing, and the absolute minimum to the hexagonal close packing. For one RW, the probability to jump within time $`\mathrm{\Delta }t`$ from one site $`x_i`$ to its neighbor-site $`x_{i\pm 1}`$ is
$$p_\pm (x_i)=\mathrm{min}[1,\mathrm{exp}(\mathrm{\Delta }_\pm (x_i)/T_{RW})],$$
(2)
with $`\mathrm{\Delta }_\pm (x_i)=V(x_{i\pm 1})V(x_i)`$. Since $`\mathrm{\Delta }V`$ is the only energy scale of the system, we define the dimensionless energy steps $`\delta _\pm ^i=\mathrm{\Delta }_\pm (x_i)/\mathrm{\Delta }V`$ and the dimensionless temperature $`T=T_{RW}/\mathrm{\Delta }V`$. Written in dimensionless parameters, the jump probabilities are thus $`p_\pm ^i=p_\pm (x_i)=\mathrm{min}[1,\mathrm{exp}(\delta _\pm ^i/T)]`$, so that a particle always jumps downhill ($`\delta _\pm ^i0`$), but jumps uphill only with a probability $`e_0=\mathrm{exp}(1/T)`$ (for $`\delta _\pm ^i>0`$), at finite temperature. The limits $`T0`$ and $`T\mathrm{}`$ correspond thus to immobile particles or to a homogeneous RW, respectively.
The discrete master equation for the probability density $`n^i(t)`$, to find a particle at time $`t`$ at site $`i`$ of the EL, is
$$n^i(t+\mathrm{\Delta }t)n^i(t)=\left[p_+^i+p_{}^i\right]n^i(t)+p_{}^{i+1}n^{i+1}(t)+p_+^{i1}n^{i1}(t)$$
(3)
and is (straightforwardly) simulated with $`R=200`$ random walkers in an energy landscape with $`L=5000`$ sites, if not explicitly mentioned. The time interval $`\mathrm{\Delta }t`$ corresponds to one Monte-Carlo step but not to one shear-cycle $`n`$. For the sake of simplicity, we measure $`x`$ in units of the distance between neighboring sites $`\mathrm{\Delta }x=x_{i+1}x_i=1`$, and time in units of $`\mathrm{\Delta }t`$. The diffusion constant of a homogeneous random walk ($`p_\pm =1/2`$ or $`T\mathrm{}`$) is thus $`D=\mathrm{\Delta }x^2/\mathrm{\Delta }t=1`$. For a constant occupation probability (initial density $`n_c=n^i(t=0)=R/L=\mathrm{const}.`$), one can extract the diffusion constant $`D_c`$ as function of the temperature, since $`p_\pm =1/2`$ and $`p_\pm =e_0/2`$ occur with equal probability, so that
$$D_c(T)=\frac{1+e_0}{2}=\frac{1+\mathrm{exp}(1/T)}{2}.$$
(4)
In Fig. 3, $`D_c=R_2(t)/\sqrt{t}`$ is plotted against $`T`$ after different times $`t`$, with $`R_2(t)=(x(t)x_0)^2`$. For large $`T`$, the system does not feel the EL and behaves like an ensemble of homogeneous RWs, whereas its behavior becomes subdiffusive after several steps for small $`T`$ when the EL is explored.
In a situation with $`T0`$, after a rather short transient, all RW will occupy a local minimum, a valley ($``$), wheras the unstable local configurations hill ($``$), or left- ($`/`$) and right-slope ($`\backslash `$) cannot be occupied. if $`T`$ is small enough, the random walkers stay trapped. Note that this statement is strictly true for $`T=0`$ or when a fixed number of shear-cycles or taps is implied. In the Sinai diffusion model, for a finite system, the RW will always find the global minimum – the temperature only determines the time-scale of this process . Since the duration of an experiment is limited, the global minimum cannot be found by all particles if the phase space volume $`L`$ is large enough.
## 4 Results and Discussion
In parallel to the tapping experiments by Nowak et al. , we present simulations of our model using a periodic time series for the temperature. The temperature is kept constant for $`S`$ steps and then increased by $`\mathrm{\Delta }T=0.1`$, where it is kept again constant for $`S`$ steps. $`T`$ is initially zero and then raised up to $`T=2`$. From this state, $`T`$ is decreased to zero and the loop is repeated seven times. In Fig. 4, the simulation results are displayed for different $`S`$ as given in the inset. In the initial branch with increasing $`T`$, the density increases and slowly decreases for large $`T`$. This branch is irreversible, but the periodic loops show almost reversible behavior. For very short loops (small $`S`$), the density continuously increases, for longer loops the behavior of the system is reversible. Note that the system also shows hysteretic behavior, the density at decreasing $`T`$ is below the density at increasing $`T`$.
In summary, we presented a very simple model for the dynamics of the compaction of granular media due to an external agitation. Our model is not as detailed as others (parking lot or frustrated lattice-gas models), but it is extremely simple and still shows qualitative agreement with two different types of experiment. Its simplicity allows for future analytical treatments. However, such a simple model arises, besides many others, two major questions: (i) Is it possible to link the real configuration phase space with the an energy landscape? (ii) Is it correct to do the analogy between the experimental external excitations like shear or tapping with a temperature?
## Acknowledgements
S. L. acknowledges the support of the I.U.S.T.I., Marseille and thanks for the hospitality. Furthermore, helpful discussions with A. Blumen and I. Sokolov are appreciated. |
warning/0003/hep-th0003244.html | ar5iv | text | # Gravitational instantons and internal dimensions.
## I Introduction
Consistent unified field theories which include gravity appear to indicate that the Universe has more than four spacetime dimensions. An obvious problem which follows is how to interpret these unseen extra dimensions? One approach that has been followed is to postulate that only four of these are observable, the extra dimensions have managed to become compact and are unobservably small. Recently however there has been a tremendous amount of interest in the effective five dimensional cosmologies associated with Branes, in which the fifth dimension can be macroscopic in size, yet remain unobservable at low energies. In general, these compactified spaces are assumed as part of the initial metric ansatz, and the cosmology of such metrics is then determined. Although this is a natural approach to take, it does not address the issue of whether such an initial condition is to be expected in string or M -theory, for example. Is there any way in which we can calculate the probability of the Universe possessing such compact internal dimensions as an initial condition? It would be of great interest if it could be shown that quantum cosmology predicts a manifold with compact extra dimensions as the most likely initial configuration.
Symmetry arguments usually provide a very powerful tool for determining which instanton solutions should provide the dominant contribution (i.e. those with lowest Euclidean action) to the Hartle Hawking path integral , hence providing the most likely background spacetime. An example is the Hawking-Moss instanton, involving a scalar field $`\varphi `$ with potential $`V(\varphi `$). Assuming the potential had a stationary point at some non-zero value they obtained in four spacetime dimensions an O(5) symmetric instanton solution where $`\varphi `$ is constant and the Euclidean manifold is a four sphere.
However, Coleman and De Lucia obtained an instanton solution of lower action with O(4) symmetry which was non-singular and corresponded to the nucleation of a bubble of true vacuum in a sea of false vacuum deSitter space. It was used in the earliest versions of open inflation , because the interior of such a bubble is in fact an open universe. Hawking and Turok took these solutions one step further, dropping the requirement for non-singular instanton configurations; they obtained solutions where the scalar field potential increased monotonically from a single minimum. These solutions also allowed for a natural continuation to an open universe which was inflating. Moreover, although the instanton solutions themselves were singular their action was finite. Indeed they demonstrated a family of solutions which had lower action than the more symmetric O(5) solution! The notion that the O(4) symmetry of the Hawking-Turok instanton was responsible for the low action was tested in . Treating the instanton as a foliation of squashed rather than round three spheres, it was found that the O(4) instanton was the lowest action solution within this family. In an interesting paper Garriga proposed a resolution to the problem of having a singularity in the solution; singular instantons can arise from compactifications of regular higher dimensional instantons when viewed as lower dimensional objects.
In this paper we investigate the nature of instanton solutions for the largest range of cosmologically relevant higher dimensional metrics that have been studied to date. Our results will be of relevance for the study of any higher dimensional model which involves compactifications on Einstein metrics, i.e. models of string cosmology involving compactifications on tori, supergravity compactifications on spheres and string theories where the compactified dimensions are Calabi-Yau manifolds. In particular we will be investigating instanton solutions arising from the metric ansatz,
$`\mathrm{ds}^2`$ $`=`$ $`\mathrm{d}\xi ^2+\mathrm{a}_{(1)}^2(\xi )\mathrm{ds}_{(1)}^2+\mathrm{a}_{(2)}^2(\xi )\mathrm{ds}_{(2)}^2+\mathrm{}+\mathrm{a}_{(\mathrm{T})}^2(\xi )\mathrm{ds}_{(\mathrm{T})}^2.`$ (1)
The only restriction on the $`ds_{(i)}^2`$’s is that they are Einstein metrics on compact manifolds; the Ricci tensor is proportional to the metric. Of the many solutions that exist, we will see how a class of these instantons may be continued to a four dimensional inflating universe, with a number of static extra dimensions.
In general, because of the non-linear nature of the equations, the solutions for the scale factors $`a_i`$ are obtained numerically, and from these we can study the action of the (generically singular) instantons. The most important result we obtain is that the family of singular instantons of this type can provide a local minima of the action for non trivial extra dimensions. However, it turns out that in all the cases we examined the action of these local minima remains higher than that of the corresponding higher dimensional Hawking-Turok instanton. The implication of such a result is important. The symmetry properties associated with the Hawking-Turok instanton appear to determine the most likely instanton configuration, at least for the cases involving Einstein metrics.
The layout of the rest of the paper is as follows: In section II we derive the field equations and action associated with our metric. Section III contains the results of our numerical and analytical analysis and presents the nature of the local minima of the action. It also contains the comparison of these instantons with the equivalent higher dimensional Hawking-Turok case and shows how the latter always lead to a lower Euclidean action. Section IV presents exact solutions for the case of a cosmological constant replacing the scalar field potential. We also mention the analytical continuation of our solutions to a space-time with a lorentzian signature and demonstrate the existence of solutions where the internal dimensions remain static while the four dimensional spacetime is inflating. We draw our conclusions in section V.
## II Derivation of Field Equations
Our starting point is a manifold $``$ which has a metric structure imposed on it, and a scalar field $`\varphi `$ living on it with potential $`𝒱(\varphi )`$. By using the usual torsion free, metric connection on $``$ we can describe the equations of motion for the metric and for $`\varphi `$ which follow from the Einstein-Hilbert action.
$`S_\mathrm{E}`$ $`=`$ $`{\displaystyle _{}}\eta \left[{\displaystyle \frac{1}{2\kappa ^2}}+{\displaystyle \frac{1}{2}}(\varphi )^2+𝒱(\varphi )\right]+\text{boundary term}.`$ (2)
Here, $`\kappa ^2=8\pi /m_{\mathrm{Pl}}^2`$ (scaled to unity for the rest of the paper), and the boundary terms are such that the action does not contain second derivatives of the metric. $`\eta `$ is the volume form and $``$ is the Ricci scalar of the connection.
As mentioned earlier we consider the manifold $``$ as a foliation in Euclidean time of a product of boundary-less manifolds. At any given time $`\xi `$ we can then write $`(\xi )`$ as a Cartesian product of $`_{(i)}`$, $`i=1\mathrm{}T`$ each with dimension $`n_{(i)}`$, where for convenience we define $`N=n_{(1)}+n_{(2)}+\mathrm{}+n_{(T)}`$. To endow $``$ with a metric structure we start by putting a metric on each of the $`_{(i)}`$, denoted $`\mathrm{ds}_{(\mathrm{i})}^2`$. The metric structure we impose on $``$ then follows by introducing a $`\xi `$ dependent scale factor, $`a_{(i)}`$, for each $`_{(i)}`$; providing information about the relative size of the $`_{(i)}`$ at any given $`\xi `$. The resulting metric is then given by Eq. (1).
This form for the metric is very general. It includes a wide class of metrics commonly considered in cosmology, such as those leading to the Coleman-De Lucia instanton , Kantowski-Sachs instantons , Hawking-Turok instanton , most of the models of string cosmology arising out of compactifications on tori (for a review see ), compactifications of string theory on Calabi-Yau spaces (for a review see ), and some compactifications of Supergravity theories on spheres (for a review see ). For example in $`T=1`$ and $`_{(1)}`$ is a three sphere with its standard round metric. A more exotic Kantowski-Sachs metric was considered in , there $`T=2`$ with $`_{(1)}=S^1`$, $`_{(2)}=S^2`$.
The equations of motion for the scale factors are derived using the Einstein-Hilbert action, for which we need to calculate the components of the Riemann tensor. This is made simpler by using methods from differential geometry , so we start by defining an orthonormal basis of one forms on $``$.
$`\omega ^0`$ $`=`$ $`\mathrm{d}\xi `$ (3)
$`\omega _{(i)}^{\overline{\mu }}`$ $`=`$ $`a_{(i)}\overline{\omega }_{(i)}^{\overline{\mu }}`$ (4)
The notation we are using is that barred quantities correspond to properties on the submanifolds $`_{(i)}`$. So, in (3) the $`\overline{\omega }_{(i)}^{\overline{\mu }}`$ are an orthonormal basis of one forms with respect to the metric $`\mathrm{ds}_{(\mathrm{i})}^2`$ and $`\overline{\mu }=1\mathrm{}n_{(i)}`$, whereas the $`\omega _{(i)}^{\overline{\mu }}`$ are in the orthonormal basis of $`\mathrm{ds}^2`$. The notation for the orthonormal basis of $`\mathrm{ds}^2`$ ($`\omega ^\mu `$) is such that $`\omega ^\mu =\omega _{(i)}^{\overline{\mu }}`$, $`\mu =1\mathrm{}N`$. So because $`\overline{\mu }=1\mathrm{}n_{(i)}`$ we find $`\overline{\mu }=\mu (n_{(1)}+n_{(2)}+\mathrm{}+n_{(i1)})`$. This should be unambiguous (although it might not seem so at first glance!) as a barred index always appears on a quantity with a subscript $`(i)`$ saying which $`_{(i)}`$ it lives on.
The connection one forms on the $`_{(i)}`$ ($`\overline{\mathrm{\Theta }}_{(i)\overline{\nu }}^{\overline{\mu }}`$) are taken to be torsion free metric connections,
$`\mathrm{d}\overline{\omega }_{(\mathrm{i})}^{\overline{\mu }}+\overline{\mathrm{\Theta }}_{(\mathrm{i})\overline{\nu }}^{\overline{\mu }}\overline{\omega }_{(\mathrm{i})}^{\overline{\nu }}`$ $`=`$ $`0`$ (5)
$`\overline{\mathrm{\Theta }}_{(i)\overline{\nu }\overline{\mu }}`$ $`=`$ $`\overline{\mathrm{\Theta }}_{(i)\overline{\mu }\overline{\nu }}\overline{\mu },\overline{\nu }=1\mathrm{}n_{(i)}.`$ (6)
The connection forms on $``$ satisfy similar relations
$`\mathrm{d}\omega ^\mu +\mathrm{\Theta }_\nu ^\mu \omega ^\nu `$ $`=`$ $`0`$ (7)
$`\mathrm{\Theta }_{\nu \mu }`$ $`=`$ $`\mathrm{\Theta }_{\mu \nu }\mu ,\nu =0\mathrm{}N.`$ (8)
To evaluate the connection one forms we use (3) to find
$`\mathrm{d}\omega ^0`$ $`=`$ $`0`$ (9)
$`\mathrm{d}\omega _{(\mathrm{i})}^{\overline{\mu }}`$ $`=`$ $`\alpha _{(i)}^{}\omega ^0\overline{\omega }_{(i)}^{\overline{\mu }}+a_{(i)}\mathrm{d}\overline{\omega }_{(\mathrm{i})}^{\overline{\mu }},`$ (10)
where we have introduced $`\alpha _{(i)}=\mathrm{ln}(a_{(i)})`$ and the prime denotes the derivative with respect to $`\xi `$. Taking the definition of $`\mathrm{\Theta }_\nu ^\mu `$ in (7) and using (5) for the individual $`_{(i)}`$ we find
$`\mathrm{\Theta }_{(i)\overline{\mu }}^0`$ $`=`$ $`\alpha _{(i)}^{}\omega _{(i)}^{\overline{\mu }}`$ (11)
$`\mathrm{\Theta }_{(i)\overline{\nu }}^{\overline{\mu }}`$ $`=`$ $`\overline{\mathrm{\Theta }}_{(i)\overline{\nu }}^{\overline{\mu }}\overline{\mu },\overline{\nu }=1\mathrm{}n_{(i)}.`$ (12)
So, if the indices on $`\mathrm{\Theta }_\nu ^\mu `$ correspond to different $`_{(i)}`$ then that connection form vanishes. We must also take care to note that $`\overline{\mathrm{\Theta }}_{(i)\overline{\nu }}^{\overline{\mu }}`$ is defined using the basis on $`_{(i)}`$ ($`\overline{\omega }_{(i)}^{\overline{\mu }}`$) whereas $`\mathrm{\Theta }_{(i)\overline{\nu }}^{\overline{\mu }}`$ uses that on $``$ ($`\omega _{(i)}^{\overline{\mu }}`$).
Now that the connection forms on $``$ are known, in terms of those on $`_{(i)}`$, we may calculate the curvature two forms ,
$`R_\nu ^\mu `$ $`=`$ $`\mathrm{d}\mathrm{\Theta }_\nu ^\mu +\mathrm{\Theta }_\rho ^\mu \mathrm{\Theta }_\nu ^\rho \mu ,\nu ,\rho =0\mathrm{}\mathrm{N}.`$ (13)
There is an analogous expression for the curvatures on $`_{(i)}`$, where the appropriate barred connections are used. Using (5) one finds for the curvature forms on $``$.
$`R_{(i)\overline{\mu }}^0`$ $`=`$ $`[\alpha _{(i)}^{\prime \prime }+(\alpha _{(i)}^{})^2]\omega ^0\omega _{(i)}^{\overline{\mu }}`$ (14)
$`R_{(i)\overline{\nu }}^{\overline{\mu }}`$ $`=`$ $`\overline{R}_{(i)\overline{\nu }}^{\overline{\mu }}(\alpha _{(i)}^{})^2\omega _{(i)}^{\overline{\mu }}\omega _{(i)}^{\overline{\nu }}`$ (15)
$`R_{(i,j)\overline{\overline{\nu }}}^{\overline{\mu }}`$ $`=`$ $`\alpha _{(i)}^{}\alpha _{(j)}^{}\omega _{(i)}^{\overline{\mu }}\omega _{(j)}^{\overline{\overline{\nu }}}`$ (16)
The notation for the last equation of (14) is that $`ij`$ and the single barred index corresponds to $`_{(i)}`$ with the double barred index living on $`_{(j)}`$. Again we stress that $`\overline{R}_{(i)\overline{\nu }}^{\overline{\mu }}`$ is defined with the $`\overline{\omega }_{(i)}^{\overline{\mu }}`$ basis, which differs from the basis on $``$ by a factor of $`a_{(i)}(\xi )`$.
Einstein’s equations relate the Ricci tensor to the stress-energy tensor. For the above curvature two forms we use $`R_\nu ^\mu =\frac{1}{2}_{\nu \rho \sigma }^\mu \omega ^\rho \omega ^\sigma `$, enabling us to calculate the Ricci tensor $`_{\mu \nu }=_{\mu \rho \nu }^\rho `$ and Ricci scalar $`=_\mu ^\mu `$:
$`_{00}`$ $`=`$ $`n_{(1)}[\alpha _{(1)}^{\prime \prime }+(\alpha _{(1)}^{})^2]n_{(2)}[\alpha _{(2)}^{\prime \prime }+(\alpha _{(2)}^{})^2]\mathrm{}`$ (17)
$`_{\overline{\mu }\overline{\nu }}^{(i)}`$ $`=`$ $`{\displaystyle \frac{1}{a_{(i)}^2}}\overline{}_{\overline{\mu }\overline{\nu }}^{(i)}(\mu \nu )`$ (18)
$`_{\overline{\mu }\overline{\mu }}^{(i)}`$ $`=`$ $`\alpha _{(i)}^{\prime \prime }+{\displaystyle \frac{1}{a_{(i)}^2}}\overline{}_{\overline{\mu }\overline{\mu }}^{(i)}\alpha _{(i)}^{}[n_{(1)}\alpha _{(1)}^{}+n_{(2)}\alpha _{(2)}^{}+\mathrm{}]`$ (19)
$``$ $`=`$ $`2\left(n_{(1)}\alpha _{(1)}^{\prime \prime }+n_{(2)}\alpha _{(2)}^{\prime \prime }+\mathrm{}\right)\left(n_{(1)}(\alpha _{(1)}^{})^2+n_{(2)}(\alpha _{(2)}^{})^2+\mathrm{}\right)`$ (20)
$`\left(n_{(1)}\alpha _{(1)}^{}+n_{(2)}\alpha _{(2)}^{}+\mathrm{}\right)^2+\left({\displaystyle \frac{1}{a_{(1)}^2}}\overline{}^{(1)}+{\displaystyle \frac{1}{a_{(2)}^2}}\overline{}^{(2)}+\mathrm{}\right),`$ (21)
where the repeated $`\overline{\mu }`$ index in (19) is not summed over. The Einstein tensor is defined by $`G_{\mu \nu }=_{\mu \nu }\frac{1}{2}g_{\mu \nu }`$, which with the Einstein-Hilbert action (2) leads to
$`G_{00}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\varphi ^2𝒱(\varphi )`$ (22)
$`G_{\overline{\mu }\overline{\nu }}^{(i)}`$ $`=`$ $`0`$ (23)
$`G_{\overline{\mu }\overline{\mu }}^{(i)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\varphi ^2𝒱(\varphi )\text{(no sum)}.`$ (24)
The key breakthrough now is to realise that for (23) and (24) to be consistent then $`\overline{}_{\overline{\mu }\overline{\nu }}^{(i)}`$ must be constants for $`\overline{\mu }=\overline{\nu }`$ and vanish otherwise. As we are using an orthonormal basis, this is precisely the statement that the metrics $`\mathrm{ds}_{(\mathrm{i})}^2`$ are Einstein metrics. In fact the equations are independent of what that metric is because the only effect a different Einstein metric has is to change the constant of proportionality between the Ricci and metric tensor, which may then be absorbed into the scale factors (if it is non-vanishing). This means we may replace $`\overline{}_{\overline{\mu }\overline{\mu }}^{(i)}`$ in (19) by $`\mathrm{\Lambda }_{(i)}=0,\pm 1`$ and $`\overline{}^{(i)}`$ in (20) by $`n_{(i)}\mathrm{\Lambda }_{(i)}`$ without loss of generality, as long as we remember to rescale the action appropriately.
We may also see what boundary terms are required in (2) by integrating the Ricci scalar by parts to find the boundary contribution. The volume form on $``$ is given by the wedge product of the volume forms on the $`_{(i)}`$,
$`\eta `$ $`=`$ $`a_{(1)}^{n_{(1)}}a_{(2)}^{n_{(2)}}\mathrm{}a_{(T)}^{n_{(T)}}\omega ^0\eta _{(1)}\eta _{(2)}\mathrm{}\eta _{(T)}.`$ (25)
By defining $`V_{(i)}`$ as the volume of $`_{(i)}`$ and $`\beta `$ to be the product of the scale factors we find that the action, including boundary terms, is given by
$`S_\mathrm{E}`$ $`=`$ $`V_{(1)}V_{(2)}\mathrm{}V_{(T)}\left\{\left[{\displaystyle \frac{\beta }{\xi }}\right]_{\xi _S}^{\xi _N}+{\displaystyle d\xi \beta \left[\frac{1}{2}+\frac{1}{2}\varphi ^2+𝒱(\varphi )\right]}\right\}`$ (26)
$`=`$ $`V_{(1)}V_{(2)}\mathrm{}V_{(T)}\left\{\left[{\displaystyle \frac{\beta }{\xi }}\right]_{\xi _S}^{\xi _N}{\displaystyle \frac{2}{n_{(1)}+n_{(2)}+\mathrm{}+n_{(T)}1}}{\displaystyle d\xi \beta (\xi )𝒱(\varphi )}\right\}`$ (27)
$`\beta (\xi )`$ $`=`$ $`a_{(1)}^{n_{(1)}}a_{(2)}^{n_{(2)}}\mathrm{}a_{(T)}^{n_{(T)}}.`$ (28)
In arriving at the second equation we have used the trace of Einstein’s equations (22)-(24) to eliminate the scalar curvature and scalar field kinetic terms. The quantities $`\xi _S`$ and $`\xi _N`$ refer to the range of the $`\xi `$ coordinate, with $`\xi _N`$ being the ‘north’ pole and $`\xi _S`$ referring to the ‘south’ pole of the instanton taken to be $`\xi =0`$. To save writing out the volumes of all the submanifolds we shall call the term in the curly braces of (27) the reduced action.
The preceding calculation shows that for metric (1) to be consistent then the metrics on the $`_{(i)}`$ must be Einstein. Given this, the evolution equations for the scale factors $`a_{(i)}(\xi )`$ depend only on the value of the ‘cosmological constants’ $`\mathrm{\Lambda }_{(i)}`$ and not on the detailed topology or geometry of the manifold. This is potentially very significant, for any statements we can make about the evolution of the scale factors cover a very large class of manifolds, all those admitting an Einstein metric. Whilst there are manifolds which do not admit an Einstein metric due to topological restriction, there is a large range which do. For example, all semi-simple Lie groups have a Killing metric which is Einstein, along with a large class of quotient spaces. It is noted that any given manifold may have more than one Einstein metric, .
## III Numerical solutions and Actions for Instanton configurations.
To make some specific predictions we shall in this section numerically investigate the case where there are just two submanifolds $`_a`$, $`_b`$ with scale factors $`a(\xi )`$ and $`b(\xi )`$ respectively. The dimension of and ‘cosmological constant’ associated with these manifolds are taken as $`n_a`$, $`n_b`$ and $`\mathrm{\Lambda }_a`$, $`\mathrm{\Lambda }_b`$ respectively. The equations of motion are then,
$`{\displaystyle \frac{1}{2}}n_a(n_a1)\alpha ^2+{\displaystyle \frac{1}{2}}n_b(n_b1)\beta ^2+n_an_b\alpha ^{}\beta ^{}{\displaystyle \frac{1}{2}}{\displaystyle \frac{n_a\mathrm{\Lambda }_a}{a^2}}{\displaystyle \frac{1}{2}}{\displaystyle \frac{n_b\mathrm{\Lambda }_b}{b^2}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\varphi ^2𝒱`$ (29)
$`(n_a1){\displaystyle \frac{a^{\prime \prime }}{a}}+n_b{\displaystyle \frac{b^{\prime \prime }}{b}}+{\displaystyle \frac{1}{2}}(n_a2)\left[(n_a1)\alpha ^2{\displaystyle \frac{\mathrm{\Lambda }_a}{a^2}}\right]+{\displaystyle \frac{1}{2}}n_b\left[(n_b1)\beta ^2{\displaystyle \frac{\mathrm{\Lambda }_b}{b^2}}\right]+n_b(n_a1)\alpha ^{}\beta ^{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\varphi ^2𝒱`$ (30)
$`n_a{\displaystyle \frac{a^{\prime \prime }}{a}}+(n_b1){\displaystyle \frac{b^{\prime \prime }}{b}}+{\displaystyle \frac{1}{2}}n_a\left[(n_a1)\alpha ^2{\displaystyle \frac{\mathrm{\Lambda }_a}{a^2}}\right]+{\displaystyle \frac{1}{2}}(n_b2)\left[(n_b1)\beta ^2{\displaystyle \frac{\mathrm{\Lambda }_b}{b^2}}\right]+n_a(n_b1)\alpha ^{}\beta ^{}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\varphi ^2𝒱`$ (31)
$`\varphi ^{\prime \prime }+(n_a\alpha ^{}+n_b\beta ^{})\varphi ^{}`$ $`=`$ $`{\displaystyle \frac{𝒱}{\varphi }}`$ (32)
In solving these equations, we used a simple potential, namely $`V(\varphi )=\frac{1}{2}\varphi ^2`$, although our results do not qualitatively depend on the exact shape of the potential. The main solutions of interest here are the cases where the south pole is regular and there is a curvature singularity at the north pole . Other cases where both the north and south poles are singular have been studied . We shall not concentrate on these cases here as the interesting features we wish to discuss are found in the case where only the north pole is singular.
As we want the south pole to be a smooth endpoint this places conditions on the scale factors. In order to avoid a conical curvature singularity then only one scale factor may vanish there, with all others approaching a constant. We order the $`_{(i)}`$ such that $`a_{(1)}`$ vanishes at $`\xi =0`$, according to
$`a_{(1)}(\xi 0)\sqrt{{\displaystyle \frac{\mathrm{\Lambda }_{(i)}}{n_i1}}}\xi ,`$ (33)
We see then that we must have $`\mathrm{\Lambda }_{(1)}>0`$ for everything to be well defined and for the solution to be non-trivial.
Before we proceed we need to know what effect the singularity is going to have on the action; in particular, does it remain finite? To decide this we make the assumption that at the singularity the potential is not important, although there are exceptions when exponential potentials are used . We may then integrate (32) to obtain
$`\varphi ^{}(\xi \xi _N)(a^{n_a}b^{n_b})^1.`$ (34)
Now assume a polynomial behaviour for the scale factors near the singularity of the form
$`a(\xi \xi _N)(\xi \xi _N)^pb(\xi \xi _N)(\xi \xi _N)^q.`$ (35)
This is consistent with (29-32) so long as $`(q,p1)`$. Then the dominant behaviour on the left hand side of (29) is $`(\xi \xi _N)^2`$, giving $`n_ap+n_bq=1`$. The volume factor, $`\beta =a^{n_a}b^{n_b}`$ (27), then goes linearly to zero at $`\xi _N`$. Moreover, as $`\varphi ^{}`$ is diverging as $`(\xi \xi _N)^1`$, $`\varphi `$ diverges logarithmically which is slow enough that the linear volume factor renders the singularity integrable for our potential.
We include some representative results below for the case of two Einstein metrics of dimensions $`n_{(1)}=3`$ and $`n_{(2)}=2`$. To be specific we have chosen to take the value of $`\mathrm{\Lambda }_{(i)}=n_{(i)}1`$, which is the appropriate value for the round metric on S$`^{n_{(i)}}`$. We shall explain the reason for this shortly.
The second of the plots above is a magnification of some structure on the first, showing a minima with negative reduced action. So, just as the Hawking-Turok instanton had a minima in the action so does this new solution which has non-trivial ‘spatial’ topology. The issue we need to address now is which has a lower action and is therefore more likely as initial conditions. For this we need to know the full Euclidean action rather than the reduced action, this means that the volume prefactors in (27) must be found. Naively it may seem that the total action may be made arbitrarily large and negative by choosing the Einstein metrics such that the volume of the manifolds is arbitrarily large. However, a result from Riemannian geometry, Bishop’s theorem implies that for positive $`\mathrm{\Lambda }`$ the metric which maximises the volume of the manifold is the round sphere. So by looking at the volumes appropriate for the round metric we are looking at the lowest possible action for all $`\mathrm{\Lambda }_{(i)}>0`$ instantons. The volume of S<sup>n</sup> with the round metric is
$`\text{V}ol(\mathrm{S}^\mathrm{n})`$ $`=`$ $`2\pi ^{\frac{n+1}{2}}/\mathrm{\Gamma }\left({\displaystyle \frac{n+1}{2}}\right),`$ (36)
so for our example of $`n_{(1)}=3`$, $`n_{(2)}=2`$ the total action is $`S_\mathrm{E}=(2\pi ^2)(4\pi )(0.048)=11.9`$, where the reduced action of -0.048 follows from the minima of Fig. 2. We now need to see how this compares to the Hawking-Turok solution in higher dimensions. The starting point is the metric ansatz. This corresponds to (1) with $`T=1`$ and $`\mathrm{ds}_{(1)}^2`$ the round metric on S<sup>n</sup>, which gives the equations
$`{\displaystyle \frac{a^{\prime \prime }}{a}}`$ $`=`$ $`\left(2{\displaystyle \frac{𝒱(\varphi )}{n(n1)}}+{\displaystyle \frac{\varphi ^2}{n}}\right)`$ (37)
$`\varphi ^{\prime \prime }+n\varphi ^{}{\displaystyle \frac{a^{}}{a}}`$ $`=`$ $`𝒱^{}(\varphi ).`$ (38)
The action is:
$$S_E=\text{V}ol(\mathrm{S}^\mathrm{n})\left((\mathrm{a}(\xi _\mathrm{N})^\mathrm{n})^{}+\frac{2}{\mathrm{n}1}d\xi (\mathrm{a}(\xi )^\mathrm{n}𝒱(\varphi ))\right).$$
(39)
We may get an approximate solution to these equations by following the process laid out in . To start we integrate (38) to find
$`\left(a(\xi _N)^n\varphi ^{}(\xi _N)\right)`$ $`=`$ $`{\displaystyle _0^{\xi _N}}d\xi a(\xi )^n{\displaystyle \frac{𝒱}{\varphi }},`$ (40)
then we make the approximation that at $`\xi _N`$ the constraint equation (29) yields,
$`a^{}(\xi _N)`$ $``$ $`{\displaystyle \frac{a(\xi _N)\varphi ^{}(\xi _N)}{\sqrt{n(n1)}}}`$ (41)
The action is then found from (39) by taking the scalar field to be the constant $`\varphi _0=\varphi (\xi =0)`$,
$$S_E\text{V}ol(\mathrm{S}^\mathrm{n})(\frac{2𝒱(\varphi _0)}{\mathrm{n}1}+\left(\frac{\mathrm{n}}{(\mathrm{n}1)}\right)^{1/2}𝒱,_{\varphi _0})\mathrm{d}\xi \mathrm{a}(\xi )^\mathrm{n},$$
(42)
The next approximation is that $`a(\xi )H^1\mathrm{sin}(H\xi )`$, where $`H^2=\frac{2𝒱(\varphi _0)}{n(n1)}`$. This then leaves us with,
$$S_E(\frac{2𝒱(\varphi _0)}{n1}+\left(\frac{n}{(n1)}\right)^{1/2}𝒱,_{\varphi _0})\frac{I_n}{H^{(n+1)}}Vol(\mathrm{S}^\mathrm{n}),$$
(43)
$`I_n=\{\begin{array}{cc}\frac{2^n[((n1)/2)!]^2}{n!}\hfill & \text{if n is odd}\hfill \\ \frac{(n1)!\pi }{2^{(n1)}(n/21)!(n/2)!}\hfill & \text{if n is even}.\hfill \end{array}`$ (46)
For example we find, $`I_2=\pi /2,I_3=4/3,I_4=3\pi /8,I_5=16/15`$. For the archetypal harmonic potential, $`𝒱=\frac{1}{2}\varphi ^2`$, we obtain an estimate for the location of the minima to be $`(\varphi _0)_{\mathrm{min}}n\sqrt{\frac{n}{(n1)}}`$, which is approximately linear in n. The corresponding value for the action is
$$S_{\mathrm{min}}(n)I_n\left[\frac{n1}{n}\right]^{(n1)}Vol(\mathrm{S}^\mathrm{n}).$$
(47)
The full numerical solutions to (37)-(39) are given in Fig. 3, where the reduced action is found for a range of dimensions. The behaviour is well explained by the approximation scheme, which describes the positions of the minima increasing as $`n`$ increases, along with the value of the reduced action which also increases with n.
We are now in a position to compare the total action we found for the product ansatz, -11.9, and this more symmetric case. The example we found before, Fig. 2, was for a six dimensional manifold. the total action of the Hawking-Turok solution in six dimensions is seen from Fig. 3 as $`\pi ^3(0.495)=15.3`$, which is lower. We have checked this for a number of dimensions, using $`\mathrm{\Lambda }_a`$, $`\mathrm{\Lambda }_b>0`$, and found that although a minima existed it had a higher action than the corresponding Hawking-Turok instanton. Although this is not a proof that the lowest action will not be of the product form, we see no reason to suspect otherwise.
Now let us proceed to investigate analytically the behaviour of the action for a more general class of our instantons to see if we can understand why a minima appears at all. Consider eqn (29). If we take p and q of (35) to both be less than one then, using our knowledge of the asymptotic behaviour of the field and scale factors, we find that near a singular pole this equation becomes,
$`{\displaystyle \frac{1}{2}}n_a(n_a1)\alpha ^2+{\displaystyle \frac{1}{2}}n_b(n_b1)\beta ^2+n_an_b\alpha ^{}\beta ^{}={\displaystyle \frac{1}{2}}\varphi ^2.`$ (48)
Let us consider the case $`n_a,n_b>>1`$. Then if we multiply eqn. (48) through by $`a^{2n_a}b^{2n_b}`$ we find that the left hand side becomes approximately equal to the boundary term in the action squared. Using this observation we can once more follow through the analysis of Hawking and Turok but this time the analysis will apply to the case of 2 submanifolds. The analysis here is the same as that which lead to (42), except that we must take $`n_a`$, $`n_b>>1`$ in the constraint equation to find
$`S_E=(𝒱,_{\varphi _0}{\displaystyle \frac{2𝒱(\varphi _0)}{(n_a+n_b1)}})V_{(1)}V_{(2)}{\displaystyle }\mathrm{d}\xi \mathrm{a}(\xi )^{\mathrm{n}_\mathrm{a}}\mathrm{b}(\xi )^{\mathrm{n}_\mathrm{b}}`$ (49)
The next step is to make the approximation that $`a(\xi )H^1\mathrm{sin}(H\xi )`$ and $`b(\xi )=b(\xi =0)=`$constant. The end result is that the action is approximated by the value obtained in (43) multiplied by $`b(\xi =0)^{n_b}`$ and an extra submanifold volume factor. This then explains why a valley is found in Fig. 2 which is parallel to the $`b(0)`$ axis, getting deeper as $`b(0)`$ increases. Clearly the valley cannot keep getting arbitrarily deep, so at some point these approximations break down. We find that the weak link in our chain of reasoning is the assumption that we make about the behaviour of the scale factors as they approach the north pole. For small $`b(0)`$ then we find that $`a(\xi )`$ decreases to zero at $`\xi _N`$, and our approximation works. When $`b(0)`$ is larger than some critical value then instead of decreasing to zero, $`a(\xi )`$ increases and diverges at $`\xi _N`$ and the approximation of treating it as a sine breaks down.
The important point to take away from all this is that some of our types of instanton with a ‘warp product space’ topology have local minima of action in parameter space, but the corresponding Hawking Turok instanton (with appropriate dimensionality) will still dominate over them in the semi-classical approximation to the Hartle Hawking wavefunction.
### A A conjecture.
We have seen that the metric associated with the scale factor which vanishes at $`\xi _S`$ must have $`\mathrm{\Lambda }_1>0`$, (33). The same constraint does not apply to the other manifolds. If we were to allow $`\mathrm{\Lambda }_{i>1}0`$ then Bishop’s theorem does not put a limit on the total volume of the $`_{(i>1)}`$; so if the reduced action had a minima with a negative value then (27) could be made arbitrarily negative by increasing $`V_{(i>1)}`$. We would therefore expect that negative values of the reduced action occur only if $`\mathrm{\Lambda }_{(i)}>0`$ for all $`i`$. We have checked this for the case of two submanifolds of various dimensions, always confirming this conjecture.
## IV Analytical solutions and analytical continuation.
The equations are simplified considerably if we actually drop the scalar field $`\varphi `$, and replace its potential with a cosmological constant $`\mathrm{\Lambda }`$. We then obtain the following analytical solutions . The first is given by,
$`a_{(1)}={\displaystyle \frac{1}{\sqrt{(n_{(1)}1)}\chi }}sin(\chi \xi )`$ (50)
$`n_{(1)}(n_{(1)}1)\chi ^2=2\mathrm{\Lambda }{\displaystyle \underset{i>1}{}}({\displaystyle \frac{n_{(i)}\mathrm{\Lambda }_{(i)}}{a_{(i)}^2}})`$ (51)
and for $`i>1`$,
$`a_{(i)}=\sqrt{{\displaystyle \frac{\mathrm{\Lambda }_{(i)}}{n_{(1)}\chi ^2}}},`$ (52)
where $`n_{(1)}>1`$. There is a similar solution when $`n_{(1)}=1`$. It is also possible that $`\chi `$ may be taken as imaginary, in which case the trigonometric function become hyperbolic, rendering the instanton non-compact. One can still create finite size instantons in this case by introducing domain walls at some value $`\xi _W`$. This creates a discontinuity in the gradient of $`a_{(1)}`$ causing the scale factor to decrease, if the wall tension is large enough. We can see that the limiting behaviour $`a_{(1)}(\xi 0)1/\sqrt{(n_{(1)}1)}`$ is consistent with (33) for $`\mathrm{\Lambda }_{(1)}=1`$, meaning that the metric is regular at the end points. Equation (52) also shows that if we have $`\chi `$ real then $`\mathrm{\Lambda }_{(i>1)}>0`$, and for imaginary $`\chi `$ $`\mathrm{\Lambda }_{(i>1)}<0`$. Ricci flat submanifolds would mean a vanishing scale factor for that manifold, so in effect they are not present.
The second analytical solution is,
$`a_{(1)}={\displaystyle \frac{1}{\sqrt{(n_{(1)}1)}\chi }}sin(\chi \xi )`$ (53)
$`a_{(2)}={\displaystyle \frac{1}{\sqrt{(n_{(2)}1)}\chi }}cos(\chi \xi )`$ (54)
$`(2n_{(1)}n_{(2)}+n_{(1)}(n_{(1)}1)+n_{(2)}(n_{(2)}1))\chi ^2=2\mathrm{\Lambda }{\displaystyle \underset{i>2}{}}({\displaystyle \frac{n_{(i)}\mathrm{\Lambda }_{(i)}}{a_{(i)}^2}})`$ (55)
and for $`i>2`$,
$`a_{(i)}=\sqrt{{\displaystyle \frac{\mathrm{\Lambda }_{(i)}}{(n_{(1)}+n_{(2)})\chi ^2}}},`$ (56)
where $`n_{(1)}>1`$ and $`n_{(2)}>1`$ although there is a similar solution when they both are equal to one. As before, if $`\chi `$ is imaginary we require a domain wall to make the instanton compact. This solution also requires $`\mathrm{\Lambda }_{(1)}=1`$ and $`\mathrm{\Lambda }_{(2)}=1`$ if the instanton is to close off in a regular manner at the ‘north’ and ‘south’ poles.
Although the majority of the instantons considered in this paper do not analytically continue to lorentzian space times where some of the dimensions are compactified we can use these analytical solutions to demonstrate that there are some that do. One subset of the solutions given above is the product of S<sup>4</sup> with S<sup>n</sup>. This can be analytically continued to a four dimensional deSitter space with a static S<sup>n</sup> as the internal manifold.
It should be noted that the static nature of this internal manifold is not stable to perturbations. This is of course the manifestation in this context of the problem of stabilising moduli fields in cosmology. We do not attempt to resolve this difficulty here. Some recent mechanisms for stabilising moduli fields in cosmology can be found in and .
## V Conclusions
In this paper, we have derived the equations of motion for a specific warp product of general Einstein metrics. The main conclusion we can draw is that instantons which continue to spaces with compact ‘extra’ dimensions of the form considered here do not have lower action than the corresponding higher-dimensional Hawking-Turok instanton. However, non trivial minima of the action do occur if the Einstein metrics all have positive $`\mathrm{\Lambda }_i`$. These results are significant: First, they seem to indicate that the symmetry arguments used by Hawking and Turok in their letter can be applied to higher dimensional cases. Secondly, our analysis applies to a wide range of metrics and cosmological scenarios. Our particular comparison of the instantons involved n-dimensional spheres as our internal compact dimensions. Bishop’s theorem then implies that these will provide the lowest possible action for such spacetimes with compact internal dimensions, hence our results apply to any Einstein metric – they will always be beaten by the corresponding Hawking-Turok instantons. This result strongly suggests to us that if the initial quantum state of the universe were to be described by the ‘Hartle Hawking proposal’ then it would be difficult to explain the presence of extra compact dimensions.
## Acknowledgements
We are grateful to J. Garriga, S. Gratton, N. Manton and T. Wiseman for useful conversations. EJC, JG and PS are all supported by PPARC. The numerical work was carried out in the UK-CCC COSMOS Origin 2000 supercomputer, supported by Silicon Graphics/Cray Research, HEFCE and PPARC. |
warning/0003/hep-ph0003123.html | ar5iv | text | # I Introduction
## I Introduction
In QCD, as a result of extensive perturbative calculations, some observables are now known to the next–to–next–to–leading order (NNLO, $``$$`a^3`$) in the power expansion in the strong coupling parameter $`a\alpha _s/\pi `$. Knowing such truncated perturbation series (TPS), the question of their resummation is gaining importance, especially if the typical process energies associated with the observable are low and thus the relevant coupling parameter is large. In such cases, it is to be expected that additional perturbative and nonperturbative effects, not explicitly contained in the TPS, will be numerically important. Many methods of resummation, based on the available TPS, try to incorporate such effects. Some of these methods eliminate the dependence on the renormalization scale (RScl) and scheme (RSch) by fixing them in the TPS itself in a judicious way – these methods could be regarded as renormalization–group–improved methods of resummation: BLM–fixing motivated by the large–$`n_f`$ considerations , Stevenson’s principle of minimal sensitivity (PMS) , Grunberg’s effective charge method (ECH) (cf. Ref. for a related method). Some of the recent works on resummations are the method of “commensurate scale relations” and related approaches , a method using an analytic form of the coupling parameter , ECH–related approaches , and an expansion in the two–loop coupling parameter . In the past few years, Padé approximants (PA’s) were also shown to be a rather successful method of resummation , especially since the resummed results show in general weakened RScl– and RSch–dependence. The diagonal Padé approximants (dPA’s) are particularly well motivated for observables since they are RScl–independent in the approximation of the one–loop evolution of the coupling $`\alpha _s(Q^2)`$ . In addition, PA’s go in their form beyond the polynomial form of the TPS on which they are based, thus contain a strong mechanism of quasianalytic continuation from the weak– into the strong–coupling regime, and can consequently incorporate some of the nonperturbative effects into the resummed result.
Recently, an extension of the method of dPA’s was presented which leads to the exact perturbative RScl–independence of the resummed result. We then extended the method so that it is applicable also to NNLO TPS’s , and suggested there a way to fix the RSch by applying the principle of minimal sensitivity (PMS). The way suggested in does not work properly in practice since no minimum of the PMS equation $`A/c_2=0`$ \[Eq. (40) there\] exists. The dependence of our approximants on the RSch–parameters $`c_2`$ and $`c_3`$ of the original TPS is a significant problem when the approximants are applied to the low–energy observables like the Bjorken polarized sum rule (BjPSR) at low $`Q_{\mathrm{photon}}\sqrt{3}`$ GeV .
This problem is addressed in the present paper. For the case of NNLO TPS, an extended version $`𝒜`$ of our approximant is constructed where the dependence on the leading RSch–parameter $`c_2`$ is eliminated by a variant of the PMS. Subsequently, the sub–leading RSch–parameter $`c_3`$ is adjusted so that the approximant reproduces the correct location of the leading infrared renormalon pole. The latter procedure is carried out in the concrete example of the BjPSR. The same method of $`c_3`$–fixing is then applied to Grunberg’s ECH and Stevenson’s TPS–PMS approximants. Hence, in the approximants we use $`\beta `$–functions which go beyond the last perturbatively calculated order in the observable (NNLO). Further, a PA–type of quasianalytic continuation for the $`\beta `$–functions is used in all these approximants. The resulting predictions for $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$ from the BjPSR are presented, along with those when PA’s are applied to the BjPSR, and compared with the world average. Differences between between our approximant and the other methods (ECH, TPS–PMS; PA’s) are pointed out.
## II Construction of $`c_2`$–independent approximants
Consider a QCD observable $`S`$ with negligible mass effects and known NNLO TPS
$`S_{[2]}`$ $`=`$ $`a_0(1+r_1a_0+r_2a_0^2),`$ (1)
$`\mathrm{with}:a_0`$ $``$ $`a(\mathrm{ln}Q_0^2;c_2^{(0)},c_3^{(0)},\mathrm{}),r_1=r_1(\mathrm{ln}Q_0^2),r_2=r_2(\mathrm{ln}Q_0^2;c_2^{(0)}).`$ (2)
We denoted $`a\alpha _s/\pi `$; $`Q_0`$ is the Euclidean RScl; $`c_j^{(0)}`$ ($`j2`$) are the RSch–parameters used in the TPS. The coupling parameter $`a(\mathrm{ln}Q^2;c_2^{(0)},\mathrm{})`$ in this RSch evolves according to the renormalization group equation (RGE) $`a/\mathrm{ln}Q^2=\beta (a;c_2^{(0)},c_3^{(0)},\mathrm{})`$. Here, the $`\beta `$–function has the power expansion $`\beta (a)=\beta _0a^2(1+c_1a+c_2^{(0)}a^2+c_3^{(0)}a^3+\mathrm{})`$, and $`\beta _0`$ and $`c_1`$ are RScl– and RSch–invariant. This RGE can be integrated (see Appendix of )
$$\beta _0\mathrm{ln}\left(\frac{Q_0^2}{\stackrel{~}{\mathrm{\Lambda }}^2}\right)=\frac{1}{a_0}+c_1\mathrm{ln}\left(\frac{c_1a_0}{1+c_1a_0}\right)+_0^{a_0}𝑑x\left[\frac{1}{x^2(1+c_1x)}+\frac{\beta _0}{\beta (x;c_2^{(0)},c_3^{(0)}\mathrm{})}\right],$$
(3)
where $`a_0a(\mathrm{ln}Q_0^2;c_2^{(0)},c_3^{(0)},\mathrm{})`$ and $`\stackrel{~}{\mathrm{\Lambda }}`$ is a universal scale ($``$$`0.1`$ GeV). When subtracting (3) from the analogous equation for $`aa(\mathrm{ln}Q^2;c_2,c_3,\mathrm{})`$, an equation is obtained which relates $`a`$ with $`a_0`$, i.e., determines $`a`$ in terms of $`a_0`$. This equation then determines also the expansion of $`a`$ in powers of $`a_0`$. From now on, we fix the “$`\mathrm{\Lambda }`$–convention” to $`\mathrm{\Lambda }=\stackrel{~}{\mathrm{\Lambda }}`$.
We make the following ansatz for our approximant, motivated by the RScl–invariant (but not RSch–invariant) approximant of Refs. :
$$\sqrt{𝒜_{S^2}}=\left\{\stackrel{~}{\alpha }\left[a(\mathrm{ln}Q_1^2;c_2^{(1)},c_3^{(1)},\mathrm{})a(\mathrm{ln}Q_2^2;c_2^{(2)},c_3^{(2)},\mathrm{})\right]\right\}^{1/2}\left(=S_{[2]}+𝒪(a_0^4)\right),$$
(4)
where we regard now the parameters $`c_2^{(j)},c_3^{(j)},\mathrm{}`$ ($`j=1,2`$) as fixed numbers, and $`c_2^{(1)}c_2^{(2)}`$. Five parameters in the approximant ($`\stackrel{~}{\alpha }`$, $`Q_1^2`$, $`Q_2^2`$, $`c_2^{(1)}`$, $`c_2^{(2)}`$) can be fixed by applying five conditions to the approximant. Three conditions are obtained from the so called minimal requirement: When we expand the approximant back in powers of $`a_0`$, the first three coefficients of the original TPS (1) have to be reproduced. The additional two conditions are obtained by a variant of the PMS
$$(𝒜_{S^2}/c_2^{(1)})|_{c_2^{(2)}}a_0^6(𝒜_{S^2}/c_2^{(2)})|_{c_2^{(1)}}.$$
(5)
This allows us to fix $`c_2^{(j)}`$’s. If we took in (4) $`c_2^{(1)}=c_2^{(2)}`$ ($`c_2`$), and $`c_k^{(1)}=c_k^{(2)}`$ ($`k3`$), i.e., the approximants of , we would obtain $`𝒜_{S^2}/c_2=10c_1a_0^5+𝒪(a_0^6)\sim ̸a_0^6`$, i.e., the PMS condition would not be satisfied. This is the main reason why we take two different (leading) parameters $`c_2^{(j)}`$ in the two $`a`$’s in (4). Further, since the two energy scales in (4), and in the approximants of , are $`Q_1^2Q_2^2`$, it does not appear unnatural to have $`c_2^{(1)}c_2^{(2)}`$. But the (subleading) parameters $`c_3^{(j)}`$ cannot be fixed by such an approach since
$$(𝒜_{S^2}/c_3^{(s)})|_{\delta c_3}=2a_0^5+𝒪(a_0^6)\mathrm{where}:c_3^{(s)}(c_3^{(1)}+c_3^{(2)})/2,\delta c_3(c_3^{(1)}c_3^{(2)}).$$
(6)
The same problem arises in the NNLO polynomial approximants ECH and TPS–PMS where $`𝒜_𝒮/c_3=(1/2)a_0^4+𝒪(a_0^5)`$ \[$`(𝒜_𝒮)^2/c_3=a_0^5+𝒪(a_0^6)`$\]. We will take, for simplicity, $`c_3^{(1)}=c_3^{(2)}c_3`$, and the value of $`c_3`$ will be fixed later.
Conditions (5) then depend also on $`\delta c_4(c_4^{(1)}c_4^{(2)})`$ which we set equal to zero to avoid further (presumably unnecessary) complications. Then the set of the five equations determining $`\stackrel{~}{\alpha }`$, $`Q_j^2`$ and $`c_2^{(j)}`$ ($`j=1,2`$) reads
$`y_{}^4y_{}^2z_0^2(c_2^{(s)})+y_{}{\displaystyle \frac{5}{4}}c_1\delta c_2{\displaystyle \frac{3}{16}}(\delta c_2)^2=0,`$ (7)
$`\{27(\delta c_2)^3157c_1(\delta c_2)^2y_{}8\delta c_2y_{}^2[27c_1^2+12c_2^{(s)}+34y_{}^28z_0^2(c_2^{(s)})]`$ (9)
$`+48c_1y_{}^3[13y_{}^23z_0^2(c_2^{(s)})]\}[5c_1\delta c_2+16y_{}^38z_0^2(c_2^{(s)})y_{})]^1=0,`$
$`\{27(\delta c_2)^4315c_1(\delta c_2)^3y_{}+64z_0^4(c_2^{(s)})y_{}^2[7c_1^22c_2^{(s)}+3z_0^2(c_2^{(s)})]4\delta c_2(20c_1y_{}3\delta c_2)`$ (12)
$`\times \left[2c_2^{(s)}y_{}^22c_2^{(s)}z_0^2(c_2^{(s)})+12z_0^2(c_2^{(s)})y_{}^2+3z_0^4(c_2^{(s)})+7c_1^2\left(y_{}^2+z_0^2(c_2^{(s)})\right)\right]`$
$`+36(\delta c_2)^2y_{}^2[z_0^2(c_2^{(s)})+25c_1^2]\left\}\right[5c_1\delta c_2+16y_{}^38z_0^2(c_2^{(s)})y_{}]^1=0,`$
$`r_1+{\displaystyle \frac{1}{2}}c_1{\displaystyle \frac{1}{4}}{\displaystyle \frac{\delta c_2}{y_{}}}=y_+,\stackrel{~}{\alpha }={\displaystyle \frac{1}{2y_{}}},`$ (13)
where we use notations
$`y_\pm `$ $``$ $`{\displaystyle \frac{1}{2}}\beta _0\left[\mathrm{ln}{\displaystyle \frac{Q_1^2}{Q_0^2}}\pm \mathrm{ln}{\displaystyle \frac{Q_2^2}{Q_0^2}}\right],\delta c_2c_2^{(1)}c_2^{(2)},c_2^{(s)}{\displaystyle \frac{1}{2}}(c_2^{(1)}+c_2^{(2)}),`$ (14)
$`z_0^2`$ $``$ $`\left(2\rho _2+{\displaystyle \frac{7}{4}}c_1^2\right)3c_2^{(s)}z_0^2(c_2^{(s)}),\rho _2=r_2r_1^2c_1r_1+c_2^{(0)}.`$ (15)
Here, $`\rho _2`$ is an RScl– and RSch–invariant, and therefore it is straightforward to see that the solutions of the system (7)–(13) for $`Q_j^2`$ and $`c_2^{(j)}`$ ($`j=1,2`$) and for $`\stackrel{~}{\alpha }`$ are independent of the original choice of the RScl ($`Q_0^2`$) and of the RSch ($`c_2^{(0)},c_3^{(0)},\mathrm{}`$). Eqs. (9)–(12) originate from PMS conditions (5), and the other three identities from the minimal condition. In particular, the latter three identities (7) and (13) show that $`\stackrel{~}{\alpha }`$ and the scales $`Q_1^2`$ and $`Q_2^2`$ are $`Q_0^2`$–independent irrespective of whether $`\delta c_20`$ or $`\delta c_2=0`$.
The coupled system of three equations (7)–(12) for the three unknowns $`c_2^{(j)}`$ ($`j=1,2`$) and $`y_{}\beta _0\mathrm{ln}(Q_1/Q_2)`$ can be solved numerically. The solutions which give $`|\stackrel{~}{\alpha }|1`$ or $`|\stackrel{~}{\alpha }|1`$ must be discarded because they would cause numerical instabilities in the approximant, and they would not make sense physically either – one of the scales $`Q_1`$, $`Q_2`$ would be orders of magnitude different from the other. There are apparently two possibilities: 1.) $`y_{}`$, $`c_2^{(s)}`$ and $`\delta c_2`$ are all real numbers; 2.) $`c_2^{(s)}`$ is real, $`y_{}`$ and $`\delta c_2`$ are imaginary numbers. In both cases, the approximant itself would be real, as it shoud be.
If there are several solutions which give different values for the approximant, we should choose (again within the PMS–logic) among them the solution with the smallest curvature with respect to $`c_2^{(1)}`$ and $`c_2^{(2)}`$.
## III Application to the Bjorken polarized sum rule; $`c_3`$-fixing
The Bjorken polarized sum rule (BjPSR) involves the isotriplet combination of the first moments over $`x_{\mathrm{Bj}}`$ of proton and neutron polarized structure functions
$$_0^1𝑑x_{\mathrm{Bj}}\left[g_1^{(p)}(x_{\mathrm{Bj}};Q_{\mathrm{ph}}^2)g_1^{(n)}(x_{\mathrm{Bj}};Q_{\mathrm{ph}}^2)\right]=\frac{1}{6}|g_A|\left[1S(Q_{\mathrm{ph}}^2)\right],$$
(16)
where $`p^2=Q_{\mathrm{ph}}^2`$$`<0`$ is $`\gamma ^{}`$ momentum transfer. At $`Q_{\mathrm{ph}}^2=3\mathrm{G}\mathrm{e}\mathrm{V}^2`$ where three quarks are assumed active ($`n_f=3`$), and if taking $`\overline{\mathrm{MS}}`$ RSch and RScl $`Q_0^2=Q_{\mathrm{ph}}^2`$, we have :
$`S_{[2]}(Q_{\mathrm{ph}}^2;Q_0^2=Q_{\mathrm{ph}}^2;c_2^{\overline{\mathrm{MS}}},c_3^{\overline{\mathrm{MS}}})`$ $`=`$ $`a_0(1+3.583a_0+20.215a_0^2),`$ (17)
$`\mathrm{with}:a_0=a(\mathrm{ln}Q_0^2;c_2^{\overline{\mathrm{MS}}},c_3^{\overline{\mathrm{MS}}},\mathrm{}),n_f`$ $`=`$ $`3,c_2^{\overline{\mathrm{MS}}}=4.471,c_3^{\overline{\mathrm{MS}}}=20.99.`$ (18)
Solving numerically the system of equations (7)–(13), we obtain one solution only
$$c_2^{(1)}=1.465,c_2^{(2)}=5.137,Q_1=0.594\mathrm{GeV},Q_2=1.164\mathrm{GeV}(\stackrel{~}{\alpha }=0.3301).$$
(19)
This solution is independent of the choice of RScl and RSch. For the time being, we will set the higher parameters $`c_k^{(j)}=0`$ ($`k4`$, $`j=1,2`$). Now our approximant depends only on the still free parameter $`c_3`$. This dependence is numerically significant. For a typical value $`a_0=0.09`$ \[$``$ $`\alpha _s^{\overline{\mathrm{MS}}}(3\mathrm{G}\mathrm{e}\mathrm{V}^2)0.283`$, $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)0.113`$\], the approximant (4) gives $`0.1523`$ and $`0.1632`$ when $`c_3=0,c_3^{\overline{\mathrm{MS}}}`$, respectively, i.e. a difference of $`7.2\%`$. In the case of the ECH and TPS–PMS approximants for the BjPSR, the respective differences are $`3.8\%`$ and $`4.0\%`$.
Parameter $`c_3`$ characterizes the $`\mathrm{N}^3\mathrm{LO}`$ term in the corresponding $`\beta `$–functions \[cf. Eq. (3)\], and information on its value cannot be obtained from the NNLO TPS \[cf. Eq. (6)\]. Therefore, to fix $`c_3`$, we should incorporate into the approximants a known piece of (nonperturbative) information beyond the NNLO TPS (17). Natural candidates for this are the known locations of the poles of the leading infrared renormalon ($`\mathrm{IR}_1`$: $`z_{\mathrm{pole}}=1/\beta _0`$) or ultraviolet renormalon ($`\mathrm{UV}_1`$: $`z_{\mathrm{pole}}=1/\beta _0`$), i.e., the poles of the Borel transform $`B_S(z)`$ of $`S`$ closest to the origin. Large–$`\beta _0`$ evaluations , based on the formulas of and using simple Borel transforms in a variant of the V–scheme (RScl $`Q_0=Q_{\mathrm{ph}}\mathrm{exp}(5/6)`$ and one–loop–evolved $`a`$), suggest that the $`\mathrm{UV}_1`$ contributions to the BjPSR at $`Q_{\mathrm{ph}}^2=2`$-$`3\mathrm{GeV}^2`$ are suppressed in comparison to the $`\mathrm{IR}_1`$ contributions by a factor 3–4 (cf. their Fig. 2).
Therefore, we will fix $`c_3`$ in the three approximants by incorporating in them the information on the location of the $`\mathrm{IR}_1`$ pole $`z_{\mathrm{pole}}=1/\beta _0`$ ($`=4/9`$). For that, we employ RScl– and RSch–invariant Borel transforms. Simple Borel transforms are not RScl/RSch–invariant, the use of their TPS’s leads to RScl/RSch–dependent $`c_3`$–fixing, which we want to avoid. We use a variant of the invariant Borel transform $`B(z)`$ introduced by Grunberg , who in turn introduced it on the basis of the modified Borel transform of Ref.
$$S(Q_{\mathrm{ph}}^2)=_0^{\mathrm{}}𝑑z\mathrm{exp}\left[\rho _1(Q_{\mathrm{ph}}^2)z\right]B_S(z),$$
(20)
where $`\rho _1`$ is the first RScl/RSch–invariant of $`S`$: $`\rho _1=r_1+\beta _0\mathrm{ln}(Q_0^2/\stackrel{~}{\mathrm{\Lambda }}^2)=\beta _0\mathrm{ln}(Q_{\mathrm{ph}}^2/\overline{\mathrm{\Lambda }}^2)`$. Here, $`\stackrel{~}{\mathrm{\Lambda }}`$ is the universal scale of Eq. (3), and $`\overline{\mathrm{\Lambda }}`$ a scale which depends on the choice of $`S`$ but is RScl/RSch–invariant and $`Q_{\mathrm{ph}}`$–independent. The $`\rho _1(Q_{\mathrm{ph}}^2)`$ is, up to an additive constant (the latter not affecting the poles of $`B_S`$), equal to $`1/a^{(1\mathrm{loop})}(Q_{\mathrm{ph}}^2)`$. Thus, $`B_S(z)`$ of (20) reduces to the simple Borel transform, up to a factor $`\mathrm{exp}(cz)`$, once higher than one–loop effects are ignored. The coefficients of the power expansion of $`B_S(z)`$ of (20) are RScl/RSch–invariant, in contrast to the case of the simple Borel transform. These invariant coefficients can be related with coefficients $`r_n`$ of $`S`$ most easily in a specific RSch $`c_k=c_1^k`$ ($`k2`$)
$$B_S(z)=(c_1z)^{c_1z}\mathrm{exp}(r_1z)\underset{0}{\overset{\mathrm{}}{}}\frac{(\stackrel{~}{r}_nc_1\stackrel{~}{r}_{n1})}{\mathrm{\Gamma }(n+1+c_1z)}z^n(c_1z)^{c_1z}\overline{B}_S(z).$$
(21)
Here, $`\stackrel{~}{r}_n`$ is the coefficient at $`\stackrel{~}{a}^{n+1}`$ in the expansion of $`S`$ in powers of $`\stackrel{~}{a}a(\mathrm{ln}Q_0^2;c_1^2,c_1^3,\mathrm{})`$; by definition $`\stackrel{~}{r}_1=0`$, $`\stackrel{~}{r}_0=1`$. Thus, the expansion of the approximant $`\sqrt{𝒜_{S^2}}(c_3)`$ in powers of $`\stackrel{~}{a}`$ leads to the expansion of the (reduced) Borel transform $`\overline{B}_\sqrt{𝒜}(z)`$ in powers of $`z`$. The coefficients starting at $`z^3`$ are predictions of the approximant and $`c_3`$–dependent: $`\overline{B}_\sqrt{𝒜}(z)=1+\overline{b}_1z+\overline{b}_2z^2+\overline{b}_3z^3+\mathrm{}`$, with $`\overline{b}_10.7516`$, $`\overline{b}_20.4209`$, $`\overline{b}_3(2.664+0.1667c_3)`$, etc. Terms with high powers of $`z`$ are not reliable, because the approximant is based on an NNLO TPS $`S_{[2]}`$ with only two terms beyond the leading order. We then employ Padé approximants (PA’s) of power expansion of $`\overline{B}_\sqrt{𝒜}`$, since they are efficient in determining the pole structure of $`\overline{B}_\sqrt{𝒜}`$. We performed the expansion of $`\sqrt{𝒜}(c_3)`$ up to $``$$`\stackrel{~}{a}^7`$, obtaining the expansion of $`\overline{B}_\sqrt{𝒜}(z)`$ up to $``$$`z^6`$. This allowed us to construct $`\mathrm{PA}_{\overline{B}}`$’s of as high order as $`[3/3]`$ or $`[4/2]`$. The value of $`c_3`$ in $`\mathrm{PA}_{\overline{B}}`$ was then adjusted to achieve $`z_{\mathrm{pole}}=1/\beta _0`$ ($`=4/9`$). The resulting values of $`c_3`$ are presented in the second column ($`\mathrm{TPS}_\beta `$) of Table I.
We carried out the analogous $`c_3`$–fixing for the polynomial approximants ECH and TPS–PMS<sup>1</sup><sup>1</sup>1 The ECH approximant is $`𝒜_S^{(\mathrm{ECH})}(c_3)=a(\mathrm{ln}Q_{\mathrm{ECH}}^2;\rho _2,c_3,\mathrm{})`$; the TPS–PMS approximant is $`𝒜_S^{(\mathrm{PMS})}(c_3)=a_{\mathrm{PMS}}\rho _2a_{\mathrm{PMS}}^3/2`$, with $`a_{\mathrm{PMS}}(c_3)=a(\mathrm{ln}Q_{\mathrm{ECH}}^2;3\rho _2/2,c_3,\mathrm{})`$, $`Q_{\mathrm{ECH}}^2=Q_0^2\mathrm{exp}(r_1/\beta _0)`$. to the BjPSR, and $`c_3`$ predictions for them are also included in Table I (columns with “$`\mathrm{TPS}_\beta `$”). These entries in the Table suggest the values $`c_312.5,17,16`$ for $`\sqrt{𝒜_{S^2}}`$, ECH, and TPS–PMS, respectively. The predictions of $`\mathrm{PA}_{\overline{B}}`$’s of intermediate order ($`[3/1]`$, $`[4/1]`$, $`[2/2]`$, $`[3/2]`$, $`[1/3]`$, $`[2/3]`$) appear to give the most stable results. Predictions of the higher order $`\mathrm{PA}_{\overline{B}}`$’s gradually lose predictability (predicted $`c_3`$’s can even become complex) because of the afore–mentioned overdetermination. The lowest order $`\mathrm{PA}_{\overline{B}}`$’s are unreliable due to their too simple structure.
The possibility to adjust the $`\mathrm{N}^3\mathrm{LO}`$ coefficient $`r_3`$ of (17) in a similar way, was apparently first mentioned by the authors of Ref. . They referred to PA’s ($`[2/1]`$) of the simple Borel transform, so their (PA–resummed) predictions would depend on the choice of the RScl and RSch. A systematic method to optimize the perturbative expansion by including the information on the location of the $`\mathrm{IR}_1`$ pole was suggested in Ref. .
Up until now we have taken the higher order parameters $`c_k^{(j)}`$ ($`k4`$, $`j=1,2`$) in our approximant (and in the ECH and TPS–PMS) to be zero, thus truncating the corresponding $`\beta `$–functions ($`\mathrm{TPS}_\beta `$). However, since the considered observable has low process energy $`Q_{\mathrm{ph}}1.73`$ GeV, we expect the higher order terms $`c_k^{(j)}x^{k+2}`$ ($`k4`$, $`x\alpha _s/\pi `$) of the $`\beta `$–function to contribute significantly to the determination (via evolution) of the relevant coupling parameters of the approximants. This leads us immediately to the question of quasianalytic continuation of the $`\beta (x)`$ functions from the small–$`x`$ into the large–$`x`$ regime. We can choose again Padé approximants (PA’s) as a tool of this quasianalytic continuation, keeping $`c_3`$ as the only free parameter, and subsequently determine $`c_3`$ in the afore–mentioned way.
In $`\sqrt{𝒜_{S^2}}(c_3)`$ there are $`\beta `$–functions characterized by the RSch–parameters $`(c_2^{(1)},c_3,\mathrm{})`$ (RSch1) and $`(c_2^{(2)},c_3,\mathrm{})`$ (RSch2) and determining the evolution and values of $`a_1a(\mathrm{ln}Q_1^2;c_2^{(1)},c_3,\mathrm{})`$ and $`a_2a(\mathrm{ln}Q_2^2;c_2^{(2)},c_3,\mathrm{})`$, respectively. In the (NNLO) ECH and the TPS–PMS approximants, the RSch–sets are $`(\rho _2,c_3,\mathrm{})`$ and $`(3\rho _2/2,c_3,\mathrm{})`$, respectively. For RSch1, ECH and TPS–PMS RSch, we have at first the freedom to construct $`[2/3]`$, $`[3/2]`$, or $`[4/1]`$ $`\mathrm{PA}_\beta `$’s. For RSch2, the additional condition $`c_4^{(2)}=c_4^{(1)}`$ ($`\delta c_4=0`$) has to be fulfilled. Since $`c_4^{(1)}`$ is a unique function of $`c_3`$ once a $`\mathrm{PA}_{\beta 1}`$ choice has been made for RSch1, we then have for $`\mathrm{PA}_{\beta 2}`$ of RSch2 the possibilities $`[2/4],[3/3],[4/2],[5/1]`$. For each choice of $`\mathrm{PA}_\beta `$’s, we essentially repeat the afore–mentioned procedure of determining the value of $`c_3`$. We consider the best choice of $`\mathrm{PA}_\beta `$’s the one giving the most stable prediction of $`c_3`$ over various PA’s $`[M/N]_{\overline{B}}`$ of the approximant’s invariant Borel transform. This turns out to be for $`\sqrt{𝒜_{S^2}}(c_3)`$ the choice ($`[2/3]_{\beta 1}`$, $`[2/4]_{\beta 2}`$), although ($`[2/3]_{\beta 1}`$, $`[5/1]_{\beta 2}`$) give virtually the same and almost as stable $`c_3`$–predictions. For the ECH and TPS–PMS the choice is $`[3/2]_\beta `$. The predictions for $`c_3`$ are given in Table I (columns with “$`\mathrm{PA}_\beta `$”). Those from PA’s $`[M/N]_{\overline{B}}`$ of intermediate order are significantly more stable than the corresponding ones with truncated $`\beta `$–functions (“$`\mathrm{TPS}_\beta `$”). This is a numerical indication that the PA–resummation of the $`\beta `$–functions improves the ability of the approximants to discern nonperturbative effects in the considered observable. The “$`\mathrm{PA}_\beta `$”–entries in Table I give us approximate values $`c_3=15.5,20,19`$ for our, the ECH, and the TPS–PMS approximant, respectively.
There is yet another argument in favor of the above $`\mathrm{PA}_\beta `$ choices. The chosen $`[2/3]_{\beta 1}`$ and $`[2/4]_{\beta 2}`$ (or: $`[5/1]_{\beta 2}`$) have positive poles with mutually similar values: $`x_{\mathrm{pole}}=0.334,0.325`$ (or: $`0.291`$), respectively. The value of $`x_{\mathrm{pole}}`$ ($`=\alpha _{\mathrm{pole}}/\pi `$) indicates a point where “a strong and an asymptotically–free phase share a common infrared attractor” . Thus, it is reasonable to expect that only those RSch’s whose $`\beta (x)`$–functions have about the same value of $`x_{\mathrm{pole}}`$ are suitable for the use in calculation of nonperturbative effects (on the other hand, in purely perturbative QCD, all RSch’s are formally equivalent). Hence, the mutual proximity of $`x_{\mathrm{pole}}`$’s of RSch1 and RSch2 $`\mathrm{PA}_\beta `$’s is now yet another indication that these $`\mathrm{PA}_\beta `$’s are the reasonable ones. What happens if we choose for RSch1 and RSch2 other $`\mathrm{PA}_\beta `$’s? In such cases, we always end up with one of the following situations: Either the two corresponding positive $`x_{\mathrm{pole}}`$ values are far apart, or both values are unphysically small, or one (positive) $`x_{\mathrm{pole}}`$ doesn’t exist, or there are no predictions for $`c_3`$ (not even unstable), or $`x_{\mathrm{pole}}`$ values are unstable under the change of $`c_3`$ in the interesting region $`c_312`$$`16`$. So, the choice $`[2/3]_{\beta 1}`$ and $`[2/4]_{\beta 2}`$ (or $`[5/1]_{\beta 2}`$) in our approximant is not just the choice giving the most stable $`c_3`$–predictions, it is also the only choice giving mutually similar (and reasonable) values of $`x_{\mathrm{pole}}`$ of RSch1 and RSch2. Further, the choice $`[3/2]_\beta `$ for the ECH and TPS–PMS RSch’s gives us $`x_{\mathrm{pole}}`$ values similar to the ones previously mentioned: $`x_{\mathrm{pole}}=0.263`$ for ECH with $`c_3=20`$; $`x_{\mathrm{pole}}=0.327`$ for TPS–PMS with $`c_3=19`$. Even other choices of $`\mathrm{PA}_\beta `$ for the ECH and TPS–PMS RSch’s ($`[2/3]_\beta `$, $`[4/1]_\beta `$), which also give rather stable and very similar $`c_3`$–predictions, give us $`x_{\mathrm{pole}}0.27`$$`0.41`$. Therefore, we see in all cases a clear correlation between the stability of the $`c_3`$–predictions on the one hand and $`x_{\mathrm{pole}}0.3`$$`0.4`$ on the other hand. Finally, $`[2/3]_\beta `$ is then the good choice for $`\overline{\mathrm{MS}}`$ RSch since it has $`x_{\mathrm{pole}}=0.311`$ ($`c_3^{\overline{\mathrm{MS}}}`$, cf. Eq. (18), has been determined in Ref. ). The choices $`[3/2]_\beta `$ and $`[4/1]_\beta `$ for $`\overline{\mathrm{MS}}`$ give $`x_{\mathrm{pole}}=0.119,0.213`$, respectively, which are further away from $`0.3`$$`0.4`$.
Now that all the hitherto unknown parameters in $`\sqrt{𝒜}`$ of (4) and in the ECH and TPS–PMS have been determined, we use the approximants to predict the values of $`\alpha _s^{\overline{\mathrm{MS}}}(3\mathrm{G}\mathrm{e}\mathrm{V}^2)`$ ($`=\pi a_0`$) from the measured values of the BjPSR $`S(Q_{\mathrm{ph}}^2=3\mathrm{G}\mathrm{e}\mathrm{V}^2)`$. Experimental values at $`Q_{\mathrm{ph}}=\sqrt{3}`$ GeV are given in (their Table 4) and are based on SLAC data
$$\frac{1}{6}|g_A|[1S(Q_{\mathrm{ph}}^2)]=0.177\pm 0.018S(Q_{\mathrm{ph}}^2)=0.155\pm 0.086.$$
(22)
where the constant $`|g_A|`$ is known from $`\beta `$–decay measurements: $`|g_A|=1.257(\pm 0.2\%)`$. The experimental uncertainties are high, mainly because of the effects of perturbative evolution on the small–$`x_{\mathrm{Bj}}`$ extrapolation of the polarized structure functions appearing in the sum rule (16), as explained in Ref. . We vary $`a_0`$ in our, and any other, approximant for the BjPSR $`S`$ in such a way that the values (22) are reproduced. We then obtain the predictions for $`\alpha _s^{\overline{\mathrm{MS}}}(3\mathrm{G}\mathrm{e}\mathrm{V}^2)`$ given in Table II. Given are always three values for $`\alpha _s`$, corresponding to the three values of $`S`$ (22). The results are given for our, the ECH and the TPS-PMS approximants, all with the described $`c_3`$–fixing and with the afore–mentioned PA–type resummation of the pertaining $`\beta `$–functions: $`[2/3]_\beta `$ (RSch1), $`[2/4]_\beta `$ (RSch2), $`[3/2]_\beta `$ (ECH), $`[3/2]_\beta `$ (TPS–PMS), $`[2/3]_\beta `$ ($`\overline{\mathrm{MS}}`$).
Given are also predictions of such approximants when the $`\beta `$–functions are TPS’s ($`c_k^{(j)}=0`$ for $`k4`$). To highlight the importance of $`c_3`$–fixing, we included predictions of these approximants (with $`\mathrm{TPS}_\beta `$) when we set $`c_3=0`$ in them. In addition, predictions of the following approximants are included in Table II: TPS $`S_{[2]}`$ (17) (NNLO TPS); TPS $`S_{[3]}`$ with $`r_3=128.05`$ ($`\mathrm{N}^3\mathrm{LO}`$ TPS); off–diagonal Padé approximants (PA’s) $`[1/2]_S`$ and $`[2/1]_S`$; square root of the diagonal PA (dPA) $`[2/2]_{S^2}`$, which is based solely on the TPS $`S_{[2]}`$ (17), as are the previous two off–diagonal PA’s; $`[2/2]_S`$ is the dPA constructed on the basis of the $`\mathrm{N}^3\mathrm{LO}`$ TPS $`S_{[3]}`$ with $`r_3=128.05`$. For $`[2/2]_S`$ and $`\mathrm{N}^3\mathrm{LO}`$ TPS we took the value $`r_3=128.05`$ (in $`\overline{\mathrm{MS}}`$, at RScl $`Q_0^2=3\mathrm{G}\mathrm{e}\mathrm{V}^2`$) because then the $`[1/2]`$ PA of the invariant Borel transform $`\overline{B}_S`$ (21) predicts the correct $`\mathrm{IR}_1`$ pole $`z_{\mathrm{pole}}=1/\beta _0`$. Numbers in Table II are with four digits so that predictions of various methods can be easily compared.
Table II includes predictions for $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$. They were obtained from $`\alpha _s^{\overline{\mathrm{MS}}}(3\mathrm{G}\mathrm{e}\mathrm{V}^2)`$ by evolution via four–loop RGE, using the values of the four–loop $`\overline{\mathrm{MS}}`$ coefficient $`c_3(n_f)`$ and the corresponding three–loop matching conditions for the flavor thresholds. In the matching, we used the scale $`\mu (n_f)=\kappa m_q(n_f)`$ above which $`n_f`$ flavors are active, with $`\kappa =2`$, and $`m_q(n_f)`$ being the running quark mass $`m_q(m_q)`$ of the $`n_f`$’th flavor. If increasing $`\kappa `$ from $`1.5`$ to $`3`$, the predictions for $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$ decrease by at most $`0.15\%`$. If we use $`[2/3]_\beta `$ instead of $`\mathrm{TPS}_\beta `$ in the evolution from $`3\mathrm{G}\mathrm{e}\mathrm{V}^2`$ to $`M_Z^2`$, $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$ decreases by less than $`0.04\%`$.
In Fig. 1 we present various approximants as functions of $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$. Our, ECH and TPS–PMS approximants have TPS $`\beta `$–functions and $`c_3=12.5,17,16`$, respectively (by the described $`\mathrm{IR}_1`$ pole requirement). These three approximants, when the $`\beta `$–functions are resummed by PA’s, and $`c_3=15.5,20,19`$, respectively ($`\mathrm{IR}_1`$ pole), are presented in Fig. 2. The three approximants with $`\mathrm{TPS}_\beta `$’s (from Fig. 1) are included in Fig. 2 for comparison.
If we reexpand the approximants in powers of $`a_0`$ (RScl $`Q_0^2=Q_{\mathrm{ph}}^2`$, in $`\overline{\mathrm{MS}}`$, $`n_f=3`$), predictions for coefficient $`r_3`$ at $`a_0^4`$ of expansion (17) are obtained. Our approximant $`\sqrt{𝒜_{S^2}}(c_3)`$, with $`c_3=15.5`$, predicts $`r_3=125.8c_3^{\overline{\mathrm{MS}}}/2+c_3130.8`$, and the ECH $`𝒜_S^{(\mathrm{ECH})}=a(\mathrm{ln}Q_{\mathrm{ECH}}^2;\rho _2,c_3,\mathrm{})`$, with $`c_3=20.`$, predicts $`r_3=129.9+(c_3^{\overline{\mathrm{MS}}}+c_3)/2129.4`$. Both predictions agree well with that of $`r_3129.9`$ ($`130.`$) which was obtained from the ECH under the assumption $`(c_3^{\overline{\mathrm{MS}}}+c_3)0`$ (note that $`c_3^{\overline{\mathrm{MS}}}21.0`$ was not known at the time was written).
## IV Discussion and conclusions
The results presented in Table II and in Figs. 12 show clearly that nonperturbative effects, as reflected in the mechanism of quasianalytic continuation from the small–$`a`$ into large–$`a`$ regime and in the presence of the leading infrared renormalon ($`\mathrm{IR}_1`$) pole, play an important role in the BjPSR at low photon transfer momenta $`Q_{\mathrm{ph}}1.73`$ GeV. These effects decrease the predicted value of $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$ by very substantial amounts. Our approximant gives the BjPSR–prediction $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)=0.1120_{0.0219}^{+0.0047}`$ (see Table II). Availability of additional data on polarized structure functions, especially in the low–$`x_{\mathrm{Bj}}`$ regime, may significantly reduce the uncertainties of the BjPSR–predictions for $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$.
The present world average is $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)=0.1173\pm 0.0020`$ by Ref. , and $`0.1184\pm 0.0031`$ by Ref. . The NNLO TPS predictions of the considered BjPSR ($`0.1171_{0.0260}^{+0.0102}`$, see Table II) cover the entire world average interval and more. However, when the afore–mentioned two classes of nonperturbative effects are taken into account, e.g. via the use of our or the ECH approximants and by the described $`c_3`$–fixing, we obtain an upper bound $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)_{\mathrm{max}}0.117`$ – see Table II. But this upper bound does not surpass the central values of the afore–mentioned world averages. The central value of $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$ extracted from the BjPSR ($`0.112`$) is significantly lower than the world average.
What could be the reason for this? One speculative possibility would be that some of the Feynman diagrams contributing to the $`\mathrm{N}^3\mathrm{LO}`$ term (yet unknown) of the BjPSR have a genuinely new topology not appearing in the lower diagrams, and that such new topology diagrams push the predicted values of $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)`$ significantly upwards. The resummation methods based on the NNLO TPS cannot “foresee” such contributions . In this context, we note that the afore–described $`c_3`$–fixing in our, ECH and TPS–PMS approximants enables these approximants to be based on more than just the information contained in the NNLO TPS and in the RGE. However, since the location of the $`\mathrm{IR}_1`$ pole can be determined by large–$`\beta _0`$ considerations, the described $`c_3`$–fixing apparently does not incorporate information from those possible higher–loop diagrams whose topologies are genuinely new.
Another possible reason for the difference between our $`\alpha _s^{\overline{\mathrm{MS}}}`$–predictions and those of the world average could for example lie in a hitherto underestimated relevance of nonperturbative contributions and of higher order perturbative terms in the numerical analyses of data for some QCD observables. This possibility should be seen also in view of the fact that (some) NNLO contributions ($``$$`a^3`$) are not yet theoretically known for several of the quantities whose data have been analyzed to predict the world average . However, lower values are allowed by some recent analyses beyond the NLO: $`\alpha _s^{\overline{\mathrm{MS}}}(M_Z^2)=0.118\pm 0.006`$ from the CCFR data for $`x_{\mathrm{Bj}}F_3`$ structure function from $`\nu N`$ DIS (NNLO); $`0.112_{0.012}^{+0.009}`$ from Gross–Llewellyn–Smith sum rule (NNLO); $`0.115\pm 0.004`$ from lattice computations.
From the theoretical point of view, we are dealing with three types of resummation approximants for NNLO TPS’s of QCD observables in the present paper:
1. Padé approximants (PA’s) provide an efficient mechanism of quasianalytic continuation. However, they do not possess RScl– and RSch–invariance, although their dependence on the RScl and on the leading RSch–parameter $`c_2`$ is in general weaker than that of the original TPS. In addition, they implicitly possess a $`c_3`$–dependence, but this dependence has no special role since there is also $`c_2`$– and RScl–dependence.
2. Grunberg’s ECH and Stevenson’s TPS–PMS methods do not possess a strong mechanism of quasianalytic continuation, except the one provided by the RGE–evolution of the coupling parameter $`a`$ itself.<sup>2</sup><sup>2</sup>2 In the one–loop limit, this amounts to the $`[1/1]`$ PA quasianalytic continuation for $`a`$ (ECH). This is so because these approximants do not go beyond the polynomial form in terms of the coupling parameter $`a`$. On the other hand, these approximants do achieve RScl– and $`c_2`$–independence, since they represent a judicious choice of the RScl and of $`c_2`$ in the TPS. They possess a $`c_3`$–dependence.
3. Our approximants provide an efficient mechanism of quasianalytic continuation, since they reduce to the diagonal PA expression $`[2/2]_{S^2}^{1/2}`$ in the one–loop limit (when all $`c_k^{(0)},c_k^{(j)},c_k0`$ for $`k1`$). At the same time, they possess invariance under the change of the RScl and of the leading RSch–parameter $`c_2`$. They possess a $`c_3`$–dependence.
4. The dependence on $`c_3`$ (and on $`c_k`$, $`k4`$) parameters in our, ECH and TPS–PMS approximants allows us to incorporate into them important nonperturbative information about the location of the leading IR renormalon pole. Further, it allows us to use in these approximants resummed $`\beta `$–functions (PA–type), thus presumably additionally strengthening the effects of quasianalytic continuation mechanism. These approximants are then fully independent of the RScl and RSch of the original TPS.
The leading higher–twist term contribution to the BjPSR ($``$$`1/Q_{\mathrm{ph}}^2`$) , or a part of it, is implicitly contained in our approximant, as well as in the ECH and the TPS–PMS, via the afore–mentioned $`c_3`$–fixing. The described approach implicitly gives an approximant–specific prescription for the elimination of the (leading IR) renormalon ambiguity. It is not clear which approximant accounts for the $``$$`1/Q_{\mathrm{ph}}^2`$ terms in the best way.
A more detailed and extensive presentation of the subject will appear shortly .
Acknowledgement: This work was supported by the Korean Science and Engineering Foundation (KOSEF). |
warning/0003/gr-qc0003084.html | ar5iv | text | # The Bianchi Type 𝐼 minisuperspace model
## I Introduction
Quantum cosmology was initiated by DeWitt in the late sixties. The canonical quantization procedure which in the twenties was the key to the theory of Quantum mechanics, was again used to derive a “Schrödinger equation” for gravity. The idea was that many of the cosmological mysteries which had their cause in the initial epoch of the universe, could be explained by Quantum cosmology (QC). Through the work of DeWitt, Misner, Ryan etc., QC had prosperous days in the next decade. Suddenly the interest came to a hold, and the number of articles was drasticly reduced. But the interest for QC was again renewed in the eighties pioneered through the work of Hartle and Hawking. They based their work on the boundary condition for a quantum state of the universe. Soon after, Vilenkin and Linde proposed their version of a suitable boundary condition.
Most of the previous work on QC is based on FRW and deSitter universe models although some authors have studied anisotropic models (for instance ). The spatial sections of the FRW and deSitter models correspond to globally homogeneous, isotropic, simply connected spaces. In recent years a paper by the mathematician Thurston has drawn many a physicist’s attention. His main conjecture, which still remains to be proven, claims that there are essentially 8 types of 3-manifolds. Concerning cosmology and especially QC the most interesting are their compact quotients which show interesting properties. In the nineties several authors investigated locally homogeneous spacetimes with non-trivial topologies. Hawking and Turok studied quantum creation of hyperbolic universes in the context of the no boundary proposal. Coule and Martin in addition to Costa and Fagundes studied cosmologies with compact hyperbolic spatial sections. We also want to mention the work of Fagundes already in the early eighties, in which he studies compact cosmologies which in the Thurston classification are quotients of $`\times ^2`$. In this paper we shall use a locally homogeneous anisotropic model that in the Thurston classification has the covering space $`E^3`$. We will not consider in any detail the covering space $`E^3`$ itself (which is simply connected) but rather one of its compact quotients, the three torus, $`T^3^3/^3`$ (which is a multiply connected space with a non-trivial fundamental group). The possibility of multiply connected spatial sections of the Bianchi metrics has again renewed the interest for QC. The space $`T^3`$ allows a Bianchi Type $`I`$ metric according to the Bianchi classification.
Using Misner’s notation the Bianchi metrics can be written as:
$`ds^2=N(t)^2dt^2+e^{2\alpha (t)}[e^{2\beta (t)}]_{ij}\chi ^i\chi ^j`$where $`\beta =\text{diag}(\beta _++\sqrt{3}\beta _{},\beta _+\sqrt{3}\beta _{},2\beta _+)`$ and $`\chi ^i=R_{ik}\omega ^k`$ where $`R_{ik}`$ is an $`SO(3)`$ matrix which can be parameterized by its *Euler angles* $`(\theta ,\varphi ,\psi )`$ i.e. $`𝐑=e^{\varphi \kappa _3}e^{\theta \kappa _1}e^{\psi \kappa _3}`$. The matrices $`(\kappa _1,\kappa _2,\kappa _3)`$ are the generators of the Lie group $`SO(3)`$. The invariant 1-forms $`\omega ^i`$ obey the corresponding Bianchi Type Lie algebra: $`d\omega ^i=\frac{1}{2}C_{jk}^i\omega ^j\omega ^k`$.
The Einstein-Hilbert action is given byIn this paper we use the following conventions: $`c=16\pi G=\mathrm{}=1`$. $`S=_Md^4x\sqrt{g}(R2\mathrm{\Lambda })+2_Md^3x\sqrt{h}K`$. In the case of a spatial homogeneous spacetime like the Bianchi Types, an invariant basis of forms on the spatial hypersurfaces can be found. Thus, three of the dimensions in the action may be integrated. We assume that we have a topology of a 3-torus, $`T^3`$, and for simplicity’s sake the volume will be set equal to 1. Choosing compact spatial sections are done for two reasons: Firstly because the spatial integration of the action is then finite, but also because the solution space of non-isomorphic metrics has a dimension equal to the dimension of the (true) phase space.
## II The moduli space of the toroidal Bianchi Type $`I`$ universe and the Principle of symmetric criticality
Let us briefly review how we construct the torus $`T^3`$ as a quotient space of $`E^3`$. It is important that we differenciate between the symmetry group that each hypersurface possesses and the symmetry group the hypersurfaces possess as spatial sections in the four dimensional manifold. For a more thorough investigation on this construction consult .
The full symmetry group of $`E^3`$ is $`IO(3)`$, translations and $`O(3)`$ rotations. The Bianchi metrics admit a 3 dimensional transitive group of isometries. The 3 dimensional group corresponding to the Bianchi Type $`I`$ Lie algebra is translations in three dimensions, thus isomorphic to $`^3`$. $`^3`$ is a normal Lie subgroup of $`IO(3)`$ acting simply transitive on the spaces $`^3`$. We will therefore consider the manifold $`^3`$ with a simply transitively symmetry group $`^3`$ defined in the obvious way. We will call this construction $`\widehat{}^3`$. These spaces have exactly the symmetry allowed by the Bianchi Type $`I`$ Lie algebra. To construct a three torus $`T^3`$ as a quotient of $`\widehat{}^3`$ we can do the following: We find a freely and properly discontinuous subgroup $`\mathrm{\Gamma }`$ ofThe symmetry group of $`\widehat{}^3`$ will actually be $`^3\stackrel{~}{\times }D_2`$ where $`D_2`$ is the dihedral group and $`\stackrel{~}{\times }`$ is the semi-direct product. $`D_2`$ has 4 elements: The unit element and rotations around the 3 axis by an angle of $`\pi `$. Sym$`(\widehat{}^3)`$, and identify points in $`^3`$ under the action of $`\mathrm{\Gamma }`$. This will ensure that the resulting quotient space is a smooth manifold. We are interested in subgroups $`\mathrm{\Gamma }`$ so that the resulting quotient has the topology of a three torus. These subgroups $`\mathrm{\Gamma }`$ can be characterized by three vectors $`𝐚,𝐛,𝐜^3`$ so that $`[𝐚,𝐛,𝐜]GL^+(3)`$. These vectors will be the generators of the fundamental group of the torus $`T^3`$ and points in $`\widehat{}^3`$ are identified under the action (which in this case is addition) of the vectors $`𝐚,𝐛,𝐜`$. $`\mathrm{\Gamma }`$ will then be $`\mathrm{\Gamma }=\{(n𝐚+m𝐛+k𝐜)^3|n,m,k\}`$. The moduli space can therefore be described by a matrix $`[𝐚,𝐛,𝐜]GL^+(3)`$ apart from the freedom of discrete modular transformations represented by conjugation with respect to matrices in $`GL^+(3,)`$. We will not fix the gauge with respect to these transformations in this paper. We will simply ignore them. The actual parameter space is therefore Teichmüller space. Through the homeomorphism $`GL^+(3)^6\times SO(3)`$ we will parametrize the SO(3) sector by the Euler angles $`(\theta ,phi,\psi )`$.
In this paper we will follow a slightly different but equivalent point of view. We consider a cube $`C`$ in $`^3`$. We parametrize the cube by $`C=\{(x,y,z)|0x,y,z1\}`$ and identify points on the boundary of the cube so that we have a toroidal topology. The moduli space will be determined by the linear mapping $`l_𝐋(𝐱)=\mathrm{𝐋𝐱}`$. In the above description the matrix $`𝐋`$ is simply $`𝐋=[𝐚,𝐛,𝐜]`$ , $`𝐋GL^+(3)`$. For the Bianchi Type $`I`$ universe the invariant 1-forms can be written locally as $`\omega ^i=K_j^idx^j`$ where $`K_j^i`$ is a constant matrix. After identification the coordinates $`x^i`$ become angular-like variables with periodicity 1.
For a minisuperspace model one always has to ensure that the equations of motion obtained from a locally variational principle are equivalent to the Einstein field equations. If this is true we say that the Principle of symmetric criticality holds. It has been known for a while that the Bianchi Class B models fail to satisfy this principle . To prove consistency of the minisuperspace model used in this paper we will use a theorem by Torre and collaborators . This theorem states
> The principle of symmetric criticality (PSC) is valid for for any metric field theory derivable from a local Lagrangian density if and only if the following 2 conditions are satisfied at each point $`x`$ in the region of spacetime under consideration.
>
> 1. $`H^q(G,I_x)0`$
> 2. $`V_x^I(V_x^I)_0=0`$
We will explain the symbols as we verify the principle of symmetric criticality in our case.
$`G`$ is the symmetry group which is used for reduction. In our case we have $`G=^3\stackrel{~}{\times }D_2`$ . The orbits of $`G`$ in the spacetime have dimension $`q`$, i.e. $`q=3`$. $`I_x`$ is the isotropy group for a point $`x`$. For a Bianchi Type $`I`$ universe with toroidal spatial sections each point $`x`$ has $`I_x^3\times D_2G`$. $`H^q(G,I_x)`$ is the Lie algebra cohomology of $`G`$ relative to $`I`$. In our case it is easy to see that since $`H^3(T^3)0`$ where $`H^3(T^3)`$ is the deRahm cohomology class of the $`T^3`$, we will have $`H^q(G,I_x)0`$. The first condition is therefore fulfilled. $`V_x`$ is the vector space of symmetric rank 2 tensors at a point $`x`$ in the spacetime, and $`V_x^I`$ is the vector space of $`I_x`$-invariant symmetric rank 2 tensors at $`x`$. $`(V_x^I)_0`$ is the annihilator of $`V_x^I`$ , i.e. linear functions on $`V`$ which vanish on $`V^I`$. By for instance direct calculation it is not difficult by ordinary linear algebra to show that also the second condition holds. Alternatively we can use the fact that the group action acts transversally on the bundle of metrics. Thus the principle of symmetric criticality holds for the Bianchi Type $`I`$ universe with toroidal spatial topology.
## III The Kasner universe
Kasner solved the Einstein field equations for a Bianchi Type $`I`$ universe with a vanishing energy-momentum tensor<sup>§</sup><sup>§</sup>§Due to the fact that a cosmological constant may be viewed as a vacuum energy, we will treat a cosmological constant as a part of the energy-momentum tensor. , $`T^{\mu \nu }=0`$ and spatial sections homeomorphic to the Euclidean three-space $`^3`$. The solution which now bears his name can be written as:
$$ds^2=dt^2+t^{\frac{2}{3}}\left[t^{\frac{4}{3}\mathrm{cos}(\gamma )}dX^2+t^{\frac{4}{3}\mathrm{cos}(\gamma \frac{2}{3}\pi )}dY^2+t^{\frac{4}{3}\mathrm{cos}(\gamma +\frac{2}{3}\pi )}dZ^2\right]$$
(1)
where $`\gamma `$ is an angular variable. If the three (complex) roots of the cubic
$$z^3\frac{64}{27}e^{3i\gamma }=0$$
(2)
are denoted by $`p_1`$, $`p_2`$ and $`p_3`$, the real parts of $`p_1,p_2`$ and $`p_3`$ are equal to the three exponents inside the square brackets in eq. 1. Due to eq. 2 the following will also be true:
$$p_1+p_2+p_3=p_1^2+p_2^2+p_3^2=0$$
(3)
The Kasner solutions are therefore a 1-parameter family of solutions parameterized by the angular variable $`\gamma `$. This solution space is therefore usually called the Kasner circle.
Let us denote the 2-parameter family parameterization of the flat Euclidean 3-space with metric
$$d\sigma _\gamma ^2(t)=t^{\frac{4}{3}\mathrm{cos}(\gamma )}dX^2+t^{\frac{4}{3}\mathrm{cos}(\gamma \frac{2}{3}\pi )}dY^2+t^{\frac{4}{3}\mathrm{cos}(\gamma +\frac{2}{3}\pi )}dZ^2$$
(4)
by $`(^3,d\sigma _\gamma ^2(t))`$. For a fixed $`X,Y`$ and $`Z`$ this family will be volume preserving and homogeneous but it has an anisotropic expansion (in $`t`$).
Despite that Kasner solved the field equations for a vanishing source tensor, the solutions for a mixture of dust and a vacuum energy can be brought onto a similar form to that of Kasner eq. 1.
## IV The general solution of the Bianchi Type $`I`$ universe with $`T^3`$ spatial sections and with dust and a cosmological constant
From we get the Hamiltonian
$$=\frac{N}{24}\left[e^{3\alpha }(p_\alpha ^2+p_+^2+p_{}^2)+48M+48\mathrm{\Lambda }e^{3\alpha }\right]$$
(5)
where $`M`$ is a constant defined by $`d^3x\sqrt{h}\rho _{dust}=2M`$. The canonical 1-form is given by:
$`\mathrm{\Theta }=p_\alpha d\alpha +p_+d\beta _++p_{}d\beta _{}+p_\theta d\theta +p_\varphi d\varphi +p_\psi d\psi `$We have here treated the dust field non-dynamically, i.e. solved the classical equations for co-moving dust and thereafter inserted the solutions into the action. The conjugate momentum for the dust field is therefore absent in eq. 5 (compare with )
The Einstein field equations are now equivalent to the set of Hamiltonian equations
$$\dot{p}_i=\frac{}{q_i},\dot{q}_i=\frac{}{p_i}$$
(6)
together with the constraint equation
$`=0`$First we will choose the gauge $`N=1`$. From the Hamiltonian equations of motion the following variables will be constants: The Euler angles $`(\theta ,\varphi ,\psi )`$ and their conjugated momenta $`(p_\theta ,p_\varphi ,p_\psi )`$ and the conjugated momenta $`p_\pm `$. Let us set
$`p_+=4a,p_{}=4b.`$and introduce the anisotropy parameter $`A^2=a^2+b^2`$. In addition two more constants $`C_\pm `$ appear when the equations for $`\dot{\beta }_\pm `$ are integrated. The constants $`(\theta ,\varphi ,\psi ,p_\theta ,p_\varphi ,p_\psi ,A,C_+,C_{})`$ have a clear geometrical meaning. These constants (for $`A0`$) specify a matrixFor an explicit expression of this matrix consult . $`𝐋GL^+(3)`$ such that the regular solid cube is mapped into the spaces $`(^3,d\sigma _\gamma ^2(t))`$ by the linear mapping $`l_𝐋(𝐱)\mathrm{𝐋𝐱}`$. The cube is mapped onto a parallelepiped with volume $`A=det(𝐋)`$. Introducing the volume element $`v=e^{3\alpha }`$, the constraint equation yields the equation for $`v`$:
$$\dot{v}^2=3\mathrm{\Lambda }v^2+3Mv+A^2$$
(7)
The constant $`M`$ will scale as $`A`$, so it is more useful to introduce a new constant $`m`$ by $`M=Am`$. The volume element $`v`$ will also scale as $`A`$ so we introduce the function $`V(t)`$ by $`v(t)=AV(t)`$.
### A The case $`\mathrm{\Lambda }>0`$
A singularity seems unavoidable since for $`v=0`$, $`\dot{v}=\pm A`$, the volume element starts off at the initial singularity or collapses at the final singularity at a speed $`A`$. Our interest will be mainly on the expanding solutions. Defining $`\kappa =\frac{m}{2\mathrm{\Lambda }}`$ and $`\omega ^2=3\mathrm{\Lambda }`$ we can write the solutions as:
$`V(t)={\displaystyle \frac{1}{\omega }}[\mathrm{sinh}\omega t+\kappa \omega (\mathrm{cosh}\omega t1)]`$where we have chosen $`V(0)=0`$.
The remaining equations can be integrated in a straight forward manner, and introducing
$`\mathrm{\Sigma }_+(t)={\displaystyle \frac{\frac{2}{\omega }(\mathrm{cosh}\omega t1)}{\mathrm{sinh}\omega t+\kappa \omega (\mathrm{cosh}\omega t1)}}`$the equations for $`\beta _\pm `$ can be solved to yield
$`\beta _+=`$ $`{\displaystyle \frac{1}{3}}\mathrm{sin}(\gamma {\displaystyle \frac{\pi }{6}})\mathrm{ln}\mathrm{\Sigma }_+(t)+C_+`$ (8)
$`\beta _{}=`$ $`{\displaystyle \frac{1}{3}}\mathrm{cos}(\gamma {\displaystyle \frac{\pi }{6}})\mathrm{ln}\mathrm{\Sigma }_+(t)+C_{}`$ (9)
The line-element for a dust-filled Bianchi Type $`I`$ with a positive cosmological constant is therefore
$$ds^2=dt^2+V(t)^{\frac{2}{3}}\left[\mathrm{\Sigma }_+(t)^{\frac{4}{3}\mathrm{cos}(\gamma )}dX^2+\mathrm{\Sigma }_+(t)^{\frac{4}{3}\mathrm{cos}(\gamma \frac{2}{3}\pi )}dY^2+\mathrm{\Sigma }_+(t)^{\frac{4}{3}\mathrm{cos}(\gamma +\frac{2}{3}\pi )}dZ^2\right]$$
(10)
whereHere the notation is a bit misleading(which it often is concerning angular variables). The “exact” differentials $`d𝐗`$ and $`d𝐱`$ are only exact in the covering space $`^3`$. On the torus $`d𝐱`$ can be a globally defined 1-form which will be closed but not exact i.e. it corresponds to a non-trivial element in the deRahm cohomology class of the torus. $`d𝐗=𝐋d𝐱`$ and $`0x,y,z1`$ are “angular” variables.
### B The case $`\mathrm{\Lambda }<0`$
Let us now consider the case where $`\mathrm{\Lambda }`$ is negative. Perhaps the simplest way to obtain the solution for $`\mathrm{\Lambda }<0`$ is to redefine $`\omega ^2=3|\mathrm{\Lambda }|`$, and $`\kappa =\frac{m}{2|\mathrm{\Lambda }|}`$, and perform the following mapping:
$`\omega `$ $`i\omega `$
$`\kappa `$ $`\kappa `$
to obtain for the volume function
$`V(t)={\displaystyle \frac{1}{\omega }}[\mathrm{sin}\omega t+\kappa \omega (1\mathrm{cos}\omega t)].`$We can then define
$`\mathrm{\Sigma }_{}(t)={\displaystyle \frac{\frac{2}{\omega }(1\mathrm{cos}\omega t)}{\mathrm{sin}\omega t+\kappa \omega (1\mathrm{cos}\omega t)}}`$The resulting line-element is then
$$ds^2=dt^2+V(t)^{\frac{2}{3}}\left[\mathrm{\Sigma }_{}(t)^{\frac{4}{3}\mathrm{cos}(\gamma )}dX^2+\mathrm{\Sigma }_{}(t)^{\frac{4}{3}\mathrm{cos}(\gamma \frac{2}{3}\pi )}dY^2+\mathrm{\Sigma }_{}(t)^{\frac{4}{3}\mathrm{cos}(\gamma +\frac{2}{3}\pi )}dZ^2\right]$$
(11)
This line element describes a universe that expands at first. After a while the cosmological term becomes dominant and turn the expanding phase into a contracting one, and it ends in a final singularity after an elapsed time $`t_F=\frac{2}{\omega }\left[\pi \mathrm{arctan}\left(\frac{2}{m}\sqrt{\frac{|\mathrm{\Lambda }|}{3}}\right)\right]`$.
### C The case $`\mathrm{\Lambda }=0`$, and a general form of the line element
The equations for $`\mathrm{\Lambda }=0`$ can be solved in a similar manner, and we will therefore just write down the result. Defining
$`\mathrm{\Sigma }_0(t)={\displaystyle \frac{t}{1+\frac{3}{4}mt}}`$the line element for $`\mathrm{\Lambda }=0`$ takes the form
$$ds^2=dt^2+\left(t+\frac{3}{4}mt^2\right)^{\frac{2}{3}}\left[\mathrm{\Sigma }_0(t)^{\frac{4}{3}\mathrm{cos}(\gamma )}dX^2+\mathrm{\Sigma }_0(t)^{\frac{4}{3}\mathrm{cos}(\gamma \frac{2}{3}\pi )}dY^2+\mathrm{\Sigma }_0(t)^{\frac{4}{3}\mathrm{cos}(\gamma +\frac{2}{3}\pi )}dZ^2\right]$$
(12)
Up to a simple rescaling these three line elements are those derived by Saunders. The $`\mathrm{\Lambda }=0`$ case was also derived by Stephani and the $`m=0`$ case by Grøn. Already now we can see the resemblance of these three cases and the Kasner line element. However, the line element can be brought into a more Kasner-like form by choosing another function $`N`$.
We define a new time variable, $`\tau `$, by:
$`t{\displaystyle _0^\tau }{\displaystyle \frac{dz}{(1\frac{3}{4}mz)^2\frac{3}{4}\mathrm{\Lambda }z^2}}=\{\begin{array}{cc}\mathrm{\Sigma }_{}^1(\tau )\hfill & ,\mathrm{\Lambda }<0\hfill \\ \mathrm{\Sigma }_0^1(\tau )\hfill & ,\mathrm{\Lambda }=0\hfill \\ \mathrm{\Sigma }_+^1(\tau )\hfill & ,\mathrm{\Lambda }>0\hfill \end{array}`$ (13)
where <sup>-1</sup> means the inverse function. Inserting this new time variable into the previous line elements, we see that all the three cases can be brought into a similar form. Thus we can write all the three line elements as:
$`\begin{array}{cc}\hfill ds^2=& {\displaystyle \frac{d\tau ^2}{\left((1\frac{3}{4}m\tau )^2\frac{3}{4}\mathrm{\Lambda }\tau ^2\right)^2}}\hfill \\ & +{\displaystyle \frac{1}{\left((1\frac{3}{4}m\tau )^2\frac{3}{4}\mathrm{\Lambda }\tau ^2\right)^{\frac{2}{3}}}}\tau ^{\frac{2}{3}}\left[\tau ^{\frac{4}{3}\mathrm{cos}(\gamma )}dX^2+\tau ^{\frac{4}{3}\mathrm{cos}(\gamma \frac{2\pi }{3})}dY^2+\tau ^{\frac{4}{3}\mathrm{cos}(\gamma +\frac{2\pi }{3})}dZ^2\right]\hfill \end{array}`$ (14)
As we see, the line element is now manifestly cast onto a Kasner form. The solution space is now explicitly decomposed into the following sequence:
$$\begin{array}{ccccccc}C& \stackrel{\begin{array}{c}SO\left(3\right)\times ^6\\ & & \\ l_𝐋\end{array}}{}& ^3& \stackrel{\begin{array}{c}S^1\\ & & \\ s_\gamma \left(t\right)\end{array}}{}& (^3,d\sigma _\gamma ^2(t))& \stackrel{\begin{array}{c}T^{\mu \nu }\\ & & \\ K\end{array}}{}& (\times \mathrm{\Sigma },ds^2)\end{array}$$
(15)
These mappings can be given the following interpretations:
1. Deformation of the cube. If we represent the unit cube $`C`$ by the embedding $`C\{(x^i),i=1,2,3|0(x^ix_0^i)1\}`$ for any $`(x_0^i)`$, then $`l_𝐋(𝐱)=\mathrm{𝐋𝐱}`$ for $`𝐋GL^+(3)SO(3)\times ^6`$.(See figure 1)
2. An explicit 2-parameter parameterization of the Euclidean 3-space with an anisotropic metric. Given $`(X,Y,Z)^3`$ the metric is given by eq. 4.
3. The Kasner foliation, given explicitly by: $`K(^3,d\sigma _\gamma ^2(t))=(\times \mathrm{\Sigma },ds^2)`$ where $`\mathrm{\Sigma }=^3`$ and $`ds^2=N(t)^2dt^2+V(t)^{\frac{2}{3}}d\sigma _\gamma ^2(t)`$. The $`N`$ and $`V`$ are determined by the Einstein field equations and as shown they are given by $`N(t)=\frac{1}{\left((1\frac{3}{4}mt)^2\frac{3}{4}\mathrm{\Lambda }t^2\right)}`$ and $`V(t)=tN(t)`$.
The solution space is now parameterized by these mappings, and the effect of choosing compact spatial sections is now explicitly shown. We have also factorized the solution space into a topological (or modular) sector (the $`SO(3)\times ^6`$ sector) and a Kasner sector (the Kasner circle). Contributions from the dust and the cosmological terms are also factored out. Different matter configurations only affect the last mapping $`K`$, all the other degrees of freedom affect only the mappings $`l_𝐋`$ and $`s_\gamma `$.
Interestingly, the Kasner limit, where $`A\mathrm{}`$ in such a way that $`lim_A\mathrm{}l_𝐋(C)=^3`$, is well defined. The topological sector becomes degenerate and effectively we have only a $`S^1`$ degree of freedom. On the other hand the FRW limit $`A0`$ is apparently badly defined. Again the topological sector becomes degenerate in addition to the $`S^1`$-sector, but we do not seem to uniquely reproduce the FRW solutions.
There is also one more interesting consequence of a non-trivial topology. For the Kasner universe ($`T^{\mu \nu }=0`$) the special case $`\gamma =0`$:
$`ds^2=dt^2+t^2dX^2+dY^2+dZ^3`$is through the transformation $`\stackrel{~}{T}=t\mathrm{cosh}X,\stackrel{~}{X}=t\mathrm{sinh}X`$ seen to be equal to the inside of the future lightcone in the $`(\stackrel{~}{T},\stackrel{~}{X})`$ Minkowski space. Thus it can be naturally expanded to a flat manifold containing the apparent singularity $`t=0`$. The compact case on the contrary can not. Assuming $`𝐋=A^{\frac{1}{3}}(\mathrm{𝐢𝐝})`$ the differential structure of the $`(\stackrel{~}{T},\stackrel{~}{X})`$ space is that of a cone. The singularity $`t=0`$ is now the locus of the cone. It therefore represents a true singularity in the $`C^{\mathrm{}}`$ structure of the maximal extended space (i.e. the cone including the locus). Interestingly we will actually know the true nature of the singularity. Taking the limit $`A\mathrm{}`$ to obtain the Kasner universe, topological considerations suggest that the lightcone in the “flat” $`(\stackrel{~}{T},\stackrel{~}{X})`$ ought to be identified as *one* point. The geometry of the space $`\gamma =0`$ is no longer globally flat, it is more like that of a cone. This explains perhaps why the “flat space” case $`\gamma =0`$ does not evolve as a flat space through the factorization 15.
## V The Wheeler-DeWitt Equation
In the previous sections we have solved the classical equations for a Bianchi Type $`I`$ universe. We will now quantizise the dynamical degrees of freedom according to the Wheeler-DeWitt procedure. The constraint equation will then turn into an operator equation on the wave function. The resulting equation is the so-called Wheeler-DeWitt (WD) equation. Let us also include a real scalar field $`\mathrm{\Phi }`$, which later will yield interesting back-reaction effects.
The WD equation for a Bianchi Type $`I`$ with a scalar field reads:
$`{\displaystyle \frac{1}{2}}(^2+V_E)\mathrm{\Psi }=0`$where
$`^2=`$ $`{\displaystyle \frac{e^{3\alpha }}{12}}\left(e^{\xi \alpha }{\displaystyle \frac{}{\alpha }}e^{\xi \alpha }{\displaystyle \frac{}{\alpha }}+_\beta ^2+12{\displaystyle \frac{^2}{\mathrm{\Phi }^2}}\right)`$
$`V_E=`$ $`V(\mathrm{\Phi })e^{3\alpha }+4M`$
Here $`\xi `$ represents some of the factor ordering ambiguity and $`_\beta ^2=\frac{^2}{\beta _+^2}+\frac{^2}{\beta _{}^2}`$ is the usual Laplace operator in $`\beta `$-space. To simplify the expressions we introduce the volume element
$`v=e^{3\alpha }`$ and do the rescaling: $`\stackrel{~}{\beta }_\pm =3\beta _\pm `$, $`\stackrel{~}{\mathrm{\Phi }}=\frac{\sqrt{3}}{2}\mathrm{\Phi }`$, $`\mu =\frac{16}{3}M`$ and $`\stackrel{~}{V}(\stackrel{~}{\mathrm{\Phi }})=\frac{4}{3}V(\mathrm{\Phi })`$. We now drop the tildes to avoid unnecessary writing. The WD equation then turns into (with $`\zeta =1\frac{\xi }{3}`$)
$$\left[v^2\frac{^2}{v^2}+\zeta v\frac{}{v}_\beta ^2\frac{^2}{\mathrm{\Phi }^2}+\mu v+V(\mathrm{\Phi })v^2\right]\mathrm{\Psi }(v,\beta _\pm ,\mathrm{\Phi },\varphi ,\theta ,\psi )=0$$
(16)
Since the WD equation does not involve any of the Euler angles, the $`SO(3)`$ sector can be treated separately.
### A The $`SO(3)`$-sector
Since $`SO(3)^3`$, this section of superspace is naturally equipped with an elliptic Riemannian structure. Due to the compactness of $`SO(3)`$ as a topological space there exists a complete countable set of orthogonal functions on this space. Thus the conjugated momenta are quantized. We may choose the eigenfunctions in such a way that they are characterized by three quantum numbers: $`(l,m,m^{})`$. These numbers are all integers and obeys: $`lm^{},ml`$ and $`l=0,1,2..`$. The eigenfunctions are not surprisingly the irreducible representation matrices of the group $`SO(3)`$, often written as $`D_{mm^{}}^{(l)}(\varphi ,\theta ,\psi )`$. These functions are given by:
$$D_{mm^{}}^{(l)}(\varphi ,\theta ,\psi )=e^{im\varphi }e^{im^{}\psi }d_{mm^{}}^{(l)}(\theta )$$
(17)
where
$`d_{mm^{}}^{(l)}(\theta )={\displaystyle \underset{\lambda }{}}{\displaystyle \frac{(1)^\lambda [(l+m)!(lm)!(l+m^{})!(lm^{})!]^{\frac{1}{2}}}{\lambda !(l+m\lambda )!(lm^{}\lambda )!(\lambda +m^{}m)!}}`$ (18)
$`\times \left(\mathrm{cos}{\displaystyle \frac{\theta }{2}}\right)^{2l2\lambda m^{}+m}\left(\mathrm{sin}{\displaystyle \frac{\theta }{2}}\right)^{2\lambda +m^{}m}`$ (19)
where the sum is only over integer $`\lambda `$ which makes sense. The algebraic properties of these functions now follow from the group theoretical properties of the Lie group $`SO(3)`$. We could for instance create a “Trace wave function”. Every point in $`^3`$ (or element in $`SO(3)`$) may be represented by a unit rotation vector $`\widehat{n}`$ in $`^3`$ and a rotation angle $`\alpha `$. Then the “Trace wave function” may be defined by:
$`\mathrm{\Psi }_{Tr}^{(l)}(\varphi ,\theta ,\psi )={\displaystyle \underset{m=l}{\overset{l}{}}}D_{mm}^{(l)}(\varphi ,\theta ,\psi )={\displaystyle \frac{\mathrm{sin}(l+\frac{1}{2})\alpha }{\mathrm{sin}\frac{\alpha }{2}}}`$
These functions have their maximum at the unit element, will be zero where $`\alpha =\frac{2\pi n}{2l+1}`$ where $`n`$ is an integer $`1n2l`$. From a classical quantum description if the Bianchi Type $`I`$ universe was in such a state, the fundamental cube would look as if it were “wobbling” around its identity mapping. It might suddenly do a quantum leap to a different orientation. Investigating the “Probability” amplitude $`|\mathrm{\Psi }_{Tr}^{(l)}|^2`$ we see that the classical description describes a shell model (fig. 2). There are $`2l`$ regions where the Bianchi universe could be. These regions are separated by forbidden regions. The most probable shell is the one containing the identity element. Let us therefore call this the ground state. This situation becomes even more evident in the limit $`l\mathrm{}`$. It might appear that these wave functions allow a geometry change. If a Bianchi Type $`I`$ universe was born in an “excited” state which is meta-stable, it could suddenly cross the barrier to the ground state. The state in the neighborhood of the identity element would be the most probable. Due to the representation decomposition
$`𝐃^{(l_1)}𝐃^{(l_2)}{\displaystyle \underset{l=|l_1l_2|}{\overset{l_1+l_2}{}}}𝐃^{(l)}`$the following will also be true:
$`\mathrm{\Psi }_{Tr}^{(l_1l_2)}=\mathrm{\Psi }_{Tr}^{(l_1)}\mathrm{\Psi }_{Tr}^{(l_2)}={\displaystyle \underset{l=|l_1l_2|}{\overset{l_1+l_2}{}}}\mathrm{\Psi }_{Tr}^{(l)}`$Thus the solution space spanned by these Trace wave functions could be generated by the two generators: $`\mathrm{\Psi }_{Tr}^{(0)}`$ and $`\mathrm{\Psi }_{Tr}^{(1)}`$. These Trace wave functions are of course only a particular set of functions which is by no means complete (although orthogonal) but drawn to attention only because of the algebraic beauty they possess. The famous orthogonality theorem for the matrix elements of irreducible representations will ensure that our functions $`D_{mm^{}}^{(l)}(\varphi ,\theta ,\psi )`$ are orthogonal over the Riemannian space $`^3`$. All of these identities follow from the Lie group properties of $`SO(3)`$ and there are of course a lot more than the ones mentioned here. Thus a lot of the properties of the Bianchi Type $`I`$ are directly related to the structure of the superspace, at least in this description. The fact that the superspace has this Lie group structure provides a lot of information.
### B Exact solutions for a zero-mass scalar field
For a Klein-Gordon field the scalar potential can be written as $`V(\mathrm{\Phi })=\lambda +m^2\mathrm{\Phi }^2`$, where $`\lambda `$ correspond to a cosmological constant. For this type of potential exact solutions for the WD equation is very difficult if not impossible to obtain. However, if we are dealing with a zero-mass scalar field $`m=0`$, exact solutions are possible to obtain. Firstly we can separate out the $`\beta _\pm `$ and $`\mathrm{\Phi }`$ dependent parts by introducing $`\mathrm{\Psi }(v,\beta _\pm ,\mathrm{\Phi })=F(v)e^{i(k_+\beta _++k_{}\beta _{})}e^{in\mathrm{\Phi }}`$. As we see $`k^2k_+^2+k_{}^2`$ corresponds to the classical anisotropy parameter $`A^2`$ i.e. the volume of the fundamental domain. Thus, these particular solutions are planar wave solutions where $`F(v)`$ satisfy the equation:
$$v^2\frac{d^2F}{dv^2}+\zeta v\frac{dF}{dv}+(k^2+n^2+\mu v+\lambda v^2)F=0$$
(20)
This is *Whittaker’s equation* and its solutions can be written as
$$F(v)=v^{a\frac{1}{2}}W_{L,p}(\pm 2i\lambda ^{\frac{1}{2}}v)$$
(21)
where $`a=\frac{\xi }{6}`$, $`L=\frac{i\mu }{2\lambda ^{\frac{1}{2}}}`$, $`p^2=a^2(k^2+n^2)`$ and $`W_{L,p}(z)`$ is the Whittaker function. These function have an essential singularity at infinity and have a branch point at the origin. This is due to the fact that the classical solutions have a physical singularity at vanishing 3-volume. In the absence of dust $`\mu =0`$, the equation 20 turn into a Bessel equation. Thus the solution for $`\mu =0`$ can be written as
$`F(v)=v^aZ_p(\lambda ^{\frac{1}{2}}v)`$where $`Z_p(z)`$ is one of the Bessel functions or a linear combination of them. This can also be seen from the identity
$`W_{0,p}(2z)=\sqrt{{\displaystyle \frac{2z}{\pi }}}K_p(z)`$The behavior of these solutions are not drasticly altered with the inclusion of a non-zero dust term. For simplicity’s sake we will set $`\mu =0`$ in the rest of this paper. In the further investigations the dust term is not essential and because the properties of the Bessel functions is better known we will simply drop the dust term. The solutions are therefore (a linear combination of) the Bessel functions.
In the case of a classical state that is not allowed, for instance if we have a negative cosmological constant and a large enough $`v`$, we would expect an exponential decrease of the wave function. For negative $`\lambda `$ we can write the solutions as a linear combination of the modified Bessel functions $`K_p(|\lambda |^{\frac{1}{2}}v)`$ and $`I_p(|\lambda |^{\frac{1}{2}}v)`$. However, $`I_p(|\lambda |^{\frac{1}{2}}v)e^{|\lambda |^{\frac{1}{2}}v}`$ for $`|\lambda |^{\frac{1}{2}}v|p|`$, so we have to abandon these solutions if we demand an exponential decrease for $`|\lambda |^{\frac{1}{2}}v|p|`$. The actual linear combination is a matter of boundary condition, which we will leave for another section.
### C A non-zero mass scalar field: Harmonic oscillator expansion
For a non-zero mass scalar field the WD equation is difficult if not impossible to solve exactly.
Let us try to find solutions of the form:
$$\mathrm{\Psi }(v,\mathrm{\Phi })=\underset{n=0}{\overset{\mathrm{}}{}}c_n(v)\psi _n(v,\mathrm{\Phi })$$
(22)
where $`\psi _n(v,\mathrm{\Phi })`$ are the eigenfunctions of the harmonic oscillator equation:
$$\left[\frac{1}{2}\frac{^2}{\mathrm{\Phi }^2}+\frac{1}{2}m^2v^2\mathrm{\Phi }^2\right]\psi _n=mv(n+\frac{1}{2})\psi _n$$
(23)
Thus, we can consider the wave equation $`\mathrm{\Psi }`$ as a path in the Hilbert space which is spanned by the basis $`\{|n\}`$ where $`|n`$ is the usual normalized eigenstate of the one-dimensional harmonic oscillator equation. In a coordinate representation these are given by:
$$\psi _n(v,\mathrm{\Phi })=\left(\frac{mv}{\pi }\right)^{\frac{1}{4}}\frac{1}{\sqrt{2^nn!}}e^{\frac{1}{2}mv\mathrm{\Phi }^2}H_n(\mathrm{\Phi }\sqrt{mv})$$
(24)
where $`H_n`$ is the $`n`$’th Hermite polynomial. Calculating the derivatives with respect to $`v`$:
$`v{\displaystyle \frac{}{v}}|n=`$ $`\left[{\displaystyle \frac{1}{4}}{\displaystyle \frac{1}{2}}mv\mathrm{\Phi }^2\right]|n+\sqrt{{\displaystyle \frac{n}{2}}}\sqrt{mv}\mathrm{\Phi }|n1`$ (25)
$`\begin{array}{cc}\hfill v^2{\displaystyle \frac{^2}{v^2}}|n=& \left[{\displaystyle \frac{3}{16}}{\displaystyle \frac{1}{4}}mv\mathrm{\Phi }^2+{\displaystyle \frac{1}{4}}m^2v^2\mathrm{\Phi }^4\right]|n\hfill \\ & \sqrt{{\displaystyle \frac{n}{2}}}(mv)^{\frac{3}{2}}\mathrm{\Phi }^3|n1\hfill \\ & +{\displaystyle \frac{1}{2}}\sqrt{n(n1)}mv\mathrm{\Phi }^2|n2\hfill \end{array}`$ (26)
where we have used the identity:
$`H_n^{}=2nH_{n1}`$Using the annihilation and creation operators $`a`$ and $`a^{}`$ defined by:
$`\begin{array}{cc}\hfill a=& \sqrt{{\displaystyle \frac{mv}{2}}}\mathrm{\Phi }{\displaystyle \frac{1}{\sqrt{2mv}}}{\displaystyle \frac{}{\mathrm{\Phi }}}\hfill \\ \hfill a^{}=& \sqrt{{\displaystyle \frac{mv}{2}}}\mathrm{\Phi }+{\displaystyle \frac{1}{\sqrt{2mv}}}{\displaystyle \frac{}{\mathrm{\Phi }}}\hfill \end{array}`$ (27)
we can write any power of $`\sqrt{mv}\mathrm{\Phi }`$ as
$`(mv)^{\frac{k}{2}}\mathrm{\Phi }^k={\displaystyle \frac{1}{\sqrt{2^k}}}\left(a+a^{}\right)^k`$Inserting eq. 22 into the original WD equation for a non-zero mass scalar field, and using eq. 23, 25, 26 and 27 we can get a differential equation which the coefficients $`c_n(v)`$ shall satisfy. If we instead introduce a new family of functions defined by $`d_n(v)=v^{\frac{\zeta }{2}}c_n(v)`$ the resulting equation may be written as:
$`\begin{array}{cc}\hfill 0=& v^2d_n^{\prime \prime }+\left(\lambda v^2+mv(2n+1)+k^2+{\displaystyle \frac{1}{8}}{\displaystyle \frac{\xi ^2}{36}}{\displaystyle \frac{1}{8}}n(n+1)\right)d_n\hfill \\ & {\displaystyle \frac{1}{2}}v\left(\sqrt{n(n1)}d_{n2}^{}\sqrt{(n+1)(n+2)}d_{n+2}^{}\right)\hfill \\ & +{\displaystyle \frac{1}{4}}\left(\sqrt{n(n1)}d_{n2}\sqrt{(n+1)(n+2)}d_{n+2}\right)\hfill \\ & +{\displaystyle \frac{1}{16}}\left(\sqrt{n(n1)(n2)(n3)}d_{n4}+\sqrt{(n+1)(n+2)(n+3)(n+4)}d_{n+4}\right)\hfill \end{array}`$ (28)
The infinite set of equations divides into two, one odd set of equations and one even set of equations, which involve respectively odd or even coefficients only. The Hilbert space is therefore divided into two: odd and even states. Now it is time for some approximations. Let us investigate the properties of the solutions in the limit $`v0`$. In the further calculations we will assume that the wave function $`\mathrm{\Psi }(v,\mathrm{\Phi })`$ is independent of $`\mathrm{\Phi }`$ in the limit $`v0`$ in the following sense:
> Given a compact interval $`I`$ and a $`\delta >0`$. Then there exists a $`ϵ>0`$ so that for $`0<v<ϵ`$, $`|\frac{\mathrm{\Psi }}{\mathrm{\Phi }}(v,\mathrm{\Phi }_0)|<\delta `$ $`\mathrm{\Phi }_0I`$.
This is exactly the right boundary condition for demanding that $`c_n(v)a_n(v)`$ as $`v0`$ where $`a_n(v)`$ is the coefficients of a function independent of $`\mathrm{\Phi }`$ (constant in $`\mathrm{\Phi }`$) in the harmonic oscillator expansion. Using the test function $`f=\left(\frac{mv}{4\pi }\right)^{\frac{1}{4}}`$ we see that:
$`|f={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\sqrt{(2n)!}}{2^nn!}}|2n`$Based on these results we can split the total wave function into an even and an odd part $`\mathrm{\Psi }=\mathrm{\Psi }_{odd}+\mathrm{\Psi }_{even}`$. We extract the solution for the constant function $`f`$ by defining a new family of coefficients by $`d_{2n}=\frac{\sqrt{(2n)!}}{2^nn!}_n`$. The equation for $`_n`$ will then be
$`\begin{array}{cc}\hfill 0=& v^2_n^{\prime \prime }+\left(\lambda v^2+mv(4n+1)+k^2+{\displaystyle \frac{1}{8}}{\displaystyle \frac{\xi ^2}{36}}{\displaystyle \frac{1}{2}}n(n+{\displaystyle \frac{1}{2}})\right)_n\hfill \\ & v\left(n_{n1}^{}(n+{\displaystyle \frac{1}{2}})_{n+1}^{}\right)\hfill \\ & +{\displaystyle \frac{1}{2}}\left(n_{n1}(n+{\displaystyle \frac{1}{2}})_{n+1}\right)\hfill \\ & +{\displaystyle \frac{1}{4}}\left(n(n1)_{n2}+(n+{\displaystyle \frac{1}{2}})(n+{\displaystyle \frac{3}{2}})_{n+2}\right)\hfill \end{array}`$ (29)
This equation can without difficulty be solved to first order in $`v`$. The result is:
$$_n=v^l\left(1\frac{1}{4}\frac{1}{1\pm \sqrt{\frac{\xi ^2}{36}k^2}}(4n+1)mv+𝒪(v^2)\right)$$
(30)
where $`l=\frac{1}{4}\pm \sqrt{\frac{\xi ^2}{36}k^2}`$. Assuming $`\xi <6`$ the real value of the first order term will be negative, i.e. to first order the higher exited modes in the expansion will decay compared to the ground state.
### D Backreaction Effects: The massless case
The preceding results describe an even wave function in $`\mathrm{\Phi }`$. The expectation value of the scalar field vanishes, but the expectation value of $`\mathrm{\Phi }^2`$ diverges as $`v0`$. This will have consequences for the effective Hamiltonian in the theory. Let us try to replace the energy momentum tensor, $`T^{\mu \nu }`$, by its expectation value $`<T^{\mu \nu }>`$ using the obtained results. This effectively means that the scalar field operator is replaced with its expectation value:
$`{\displaystyle \frac{^2}{\mathrm{\Phi }^2}}+m^2v^2\mathrm{\Phi }^2`$ $`{\displaystyle \frac{\mathrm{\Psi }\left|\left[\frac{^2}{\mathrm{\Phi }^2}+m^2v^2\mathrm{\Phi }^2\right]\right|\mathrm{\Psi }}{\mathrm{\Psi }|\mathrm{\Psi }}}`$
In the massless case the expectation value of the energy momentum tensor only contributes with a positive constant $`n^2`$. Keeping the volume of the universe fixed (i.e. keeping $`A^2`$ fixed), the term $`n^2`$ will reduce the radius of the Kasner circle (the “radius” in eq. 2). Note also that the equations 3 will still be true. Since the center of the Kasner circle represents an isotropic universe, we see that the inclusion of a massless scalar field effectively reduces the anisotropy of the universe.
### E Backreaction Effects: The Harmonic oscillator expansion
Using the harmonic oscillator expansion we can try to understand the effect of a non-zero mass scalar field on the classical equations. In the harmonic oscillator expansion we can write
$`\mathrm{\Psi }\left|\left[{\displaystyle \frac{^2}{\mathrm{\Phi }^2}}+m^2v^2\mathrm{\Phi }^2\right]\right|\mathrm{\Psi }=mv{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}|c_n|^2(2n+1)`$ (31)
In our case we shall use only the even part, but unfortuneatly the results obtained can not be used as they stand; every sum will diverge. We have to regularize the result so that the sums may be performed. Let us write:
$`|_n|^2v^{2\text{Re}(l)}\left(1{\displaystyle \frac{2}{\eta }}\kappa mv\right)^{\eta (n+\frac{1}{4})}`$where $`\kappa =\text{Re}\left(\frac{1}{1\pm \sqrt{\frac{\xi ^2}{36}k^2}}\right)`$ and $`\eta `$ is some unknown parameter which represents the uncertainty in the higher order contribution in the expansion (30). To first order, however, we see that this is exactly the previously obtained results. With this replacement the sums may be performed and the result is to first order in $`v`$:
$$\frac{\mathrm{\Psi }_{even}\left|\left[\frac{^2}{\mathrm{\Phi }^2}+m^2v^2\mathrm{\Phi }^2\right]\right|\mathrm{\Psi }_{even}}{\mathrm{\Psi }_{even}|\mathrm{\Psi }_{even}}mv\left(\frac{2\eta 1}{\eta }\right)+\frac{1}{\kappa }$$
(32)
Luckily the unknown parameter $`\eta `$ does not contribute to lowest order, so the validity of the regularization procedure which we used is strengthened. If we assume $`\xi =0`$ and we associate $`k^2`$ with the anisotropy parameter $`A^2`$, we see that we get an effective anisotropy parameter $`A_{eff}^2=2A^2+\frac{9}{16}`$. Comparing this to the classical solutions, an increase of the effective anisotropy parameter in the Hamiltonian constraint correspond to an increase of the speed of the volume element by $`\frac{A_{eff}}{A}`$ and a decrease of the anisotropy speed by the same amount.
Even though these considerations were very simple and only approximative, this indicates that the effect of a simple scalar field with mass is that it reduces the effective anisotropy. Comparing with the zero mass solutions, we see that a zero mass scalar field also has the same effect, it increases the anisotropy parameter but reduces the anisotropy. The result in the limit $`v0`$ for a non-zero mass is thus reasonable.
## VI Boundary Conditions
Concerning boundary conditions, quantum cosmology has mostly invoked two different boundary conditions. One is that of Hartle-Hawking, which is stated in terms of the path integral formalism. The Hartle-Hawking wave function $`\mathrm{\Psi }_{HH}`$ for a Bianchi Type $`I`$ universe with a positive cosmological constant has been derived by Duncan and Jensen. Duncan and Jensen also use three-torus as spatial sections, but they do not give any comments concerning the $`SO(3)`$ sector. Apparently the only reason they use compact spatial sections is to make the spatial integration in the action finite. In this paper we will however be more interested in the tunneling boundary proposal by Vilenkin.
### A The Bianchi Type $`I`$ minisuperspace
Essential in this discussion is the description of the Bianchi Type $`I`$ from a minisuperspace point of view. If we exclude a scalar field degree of freedom, the minisuperspace will be 6-dimensional. Topologically we can write the minisuperspace as $`M^3\times ^3`$ where $`^3SO(3)`$ and $`M^3`$ is the 3-dimensional Minkowski space. As the $`SO(3)`$-sector has already been discussed, we will only look at the evolution in the Minkowski sector. We will also choose the lapse to be unity $`N=1`$. The DeWitt metric on the Minkowski sector will then be:
$$ds^2=e^{3\alpha }(d\alpha ^2+d\beta _+^2+d\beta _{}^2)$$
(33)
Doing a conformal transformation we can map this metric onto a Penrose-Carter diagram.
Mapping the classical solutions (with no dust) into this conformal diagram we get a picture like figure 3. Notice that all the three different cases start off at past light-like infinity $`^{}`$. They all start off as a Kasner universe (straight line). They end up however, at different places. For a positive-valued cosmological constant the universe ends as a deSitter universe: moving along the $`t`$-axis to the time-like future $`i^+`$. The Kasner universe ($`\mathrm{\Lambda }=0`$) ends at the future light-like infinity $`^+`$, while the case $`\mathrm{\Lambda }<0`$ the universe collapses again in past light-like infinity $`^{}`$. The whole boundary *except* past time-like infinity $`i^{}`$ represents singular space-time geometries, and thus according to the tunneling boundary condition, is not allowed. However, from the calculations done in the previous chapters, there may be indications that the starting point $`^{}`$ is somewhat unstable concerning inclusion of a scalar field. Even a rather modestly behaved scalar field effectively reduces the speed of the universe near the initial singularity to below “the speed of light”. The reduction of the speed of a Kasner universe is shown in figure 4. This effectively changes the path’s end-points, so that the universe starts off at $`i^{}`$. It ought to be emphasized that this is not enough to ensure regularity of the universe at the starting point, but this may indicate that there exists a homotopy of time-like paths connecting the Kasner universe with the deSitter solution. These different solutions will be universes with different speeds in superspace. One might say that we have given the Kasner universe a mass, and the homotopy can be defined by the action of the 3-dimensional Lorentz group.
In the light of this discussion we change the formulation of the tunneling proposal *to include all universes which may be connected to universes which satisfy the regularity condition, by the action of the Lorentz group in superspace.*
### B The tunneling wave function
The tunneling wave function will have a current in superspace which points outwards at the singular boundaries of superspace. The conserved current
$`J={\displaystyle \frac{i}{2}}(\mathrm{\Psi }^{}\mathrm{\Psi }\mathrm{\Psi }\mathrm{\Psi }^{})`$which is analogous to the conserved Klein-Gordon current, states that its zero- component $`J^0`$ is a conserved entity. The zeroth component of the Klein-Gordon current can be interpreted as a charge, which locally can obtain both negative and positive values. Also the zeroth component superspace current can be both negative and positive. In light of the tunneling proposal, these correspond to expanding and contracting solutions.
Using the solutions to the WD-equation for a zero-mass scalar field, we get the “space” components of the superspace current to be:
$$J^\pm =\frac{i}{2}v^{\frac{2\xi }{3}}(\mathrm{\Psi }^{}_\pm \mathrm{\Psi }\mathrm{\Psi }_\pm \mathrm{\Psi }^{})=v^{\frac{2\xi }{3}}k_\pm |\mathrm{\Psi }|^2$$
(34)
At the singular boundary $`v\mathrm{}`$, we use the solutions for a positive cosmological constant in the big geometry limit. The two possible solutions are the Hankel functions $`H_p^{(1)/(2)}(v)\frac{1}{\sqrt{v}}e^{\pm iv}`$. The zero-component then becomes:
$$J^0|\lambda |^{\frac{1}{2}}v^{2(1\frac{\xi }{3})}|\mathrm{\Psi }|^2$$
(35)
The outgoing modes are therefore the second Hankel function $`H_p^{(2)}(v)`$ and the tunneling wave functions are are a linear combinations of the functions
$$\mathrm{\Psi }(v,\beta _\pm ,\mathrm{\Phi })=v^{\frac{\xi }{6}}H_p^{(2)}(\lambda ^{\frac{1}{2}}v)e^{i(k_+\beta _++k_{}\beta _{})}e^{in\mathrm{\Phi }}$$
(36)
The solution space of the WD equation will therefore be spanned by the solutions (36) multiplied by the solutions from the $`SO(3)`$-sector $`D_{mm^{}}^{(l)}`$.
### C Does the Tunneling boundary proposal predict an isotropic universe?
As we have constructed the Tunneling wave function we might wonder if the wave function predict an isotropic or an anisotropic universe. In the classical case we had trouble with reconstructing the FRW universe in the limit $`A0`$. The FRW universe appeared as if it was disconnected from the Bianchi Type $`I`$ solutions. This picture is drastically improved in the Quantum case. If we hold $`v`$ constant, that is keeping the volume of the universe constant, and let $`A`$ approach zero from a classical point of view we have to let the function $`V(t)`$ approach infinity. For a positive cosmological constant the limit $`V\mathrm{}`$ is indeed a isotropic FRW limit with finite spatial volume. As $`v`$ and $`A`$ are independent variables we could interpret the $`A0`$ a FRW limit. Inserting a scalar field we saw that it also reduced the anisotropy. The anisotropy was reduced as the ratio $`\frac{k^2}{k^2+n^2}`$. These two mechanisms for reducing the anisotropy affect the classical solutions in two different ways. Whereas the reduction of $`A`$ isotropize the universe through the modular and evolutionary degrees of freedom, the inclusion of a scalar field reduces the Kasner circle. Thus a scalar field could also work for isotropy reduction for the Kasner universe (with infinite spatial sections). $`A0`$ for the Kasner universe is however more subtle because apparently we do not have any clear geometric meaning of the entity $`A`$ in that case.
The $`k^2`$ enters the wave functions through the order of the Hankel function. The order $`p`$ is given by $`p^2=\frac{\xi ^2}{36}k^2n^2`$. The factor ordering parameter is assumed to be small, so we assume that $`p`$ is purely imaginary: $`p=ir`$ where $`r0`$. A small $`r`$ means an isotropic universe. From the integral representation of the Hankel functions we can write
$`H_{ir}^{(2)}(z)={\displaystyle \frac{ie^{\frac{r\pi }{2}}}{\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}e^{iz\mathrm{cosh}tirt}𝑑t`$Apparently the wave functions are suppressed by an exponential factor as $`r`$ grows. Thus the probability amplitude is exponentially damped in $`r`$: $`|\mathrm{\Psi }|^2e^{r\pi }`$. We do have to be a bit careful however because any linear combination of these wave functions is a solution. We might imagine that it had an exponentially growing prefactor in $`r`$. Let us try to estimate a reasonable prefactor by “normalizing” the wave functions over superspace. Since the variable $`v`$ will behave as a time variable in superspace we could also say that we divide by the “time average” of the particular solution. The “normalizing” integrals $`𝑑v|\mathrm{\Psi }(v)|^2`$ diverge for the Hankel functions. We will therefore do the following: we Wick-rotate the volume element, $`viv`$ so that the Hankel functions $`H_p^{(2)}`$ are “rotated” to the Bessel functions $`K_p`$. This function will under reasonable conditions be square integrable. We estimate the prefactor by demanding that the wave function is normalized after a Wick-rotation of the volume element. The Bessel functions are related through the following equations
$`K_p(z)`$ $`=`$ $`{\displaystyle \frac{\pi }{2}}ie^{\frac{\pi }{2}pi}H_p^{(1)}(iz)`$ (37)
$`\overline{H_p^{(2)}}(z)`$ $`=`$ $`H_p^{(1)}(\overline{z})`$ (38)
We also introduce the integration measure $`\sqrt{|G|}`$ which reflects the geometry of minisuperspace. The DeWitt metric (eq. 33) will yield $`\sqrt{|G|}=v^{\frac{1}{2}}`$ but if we assume<sup>\**</sup><sup>\**</sup>\**Thus these two integration measures coincide iff $`\xi =\frac{3}{2}`$. We would also mention that the integral is manageable for any power of $`v`$ (provided that the integral converges). $`\sqrt{|G|}=v^{1\frac{\xi }{3}}`$ the resulting integral become surprisingly simple:
$`I`$ $`=`$ $`{\displaystyle _0^i\mathrm{}}𝑑v\sqrt{|G|}|\mathrm{\Psi }|^2={\displaystyle \frac{4}{\pi ^2}}e^{\pi r}{\displaystyle _0^{\mathrm{}}}𝑑vv\left|K_{ir}(\lambda ^{\frac{1}{2}}v)\right|^2`$ (39)
$`=`$ $`{\displaystyle \frac{1}{\lambda \pi ^2}}{\displaystyle \frac{\pi r}{1e^{2\pi r}}}`$ (40)
Using these wave functions after “normalization” the probability amplitude decays as
$`|\mathrm{\Psi }|^2{\displaystyle \frac{e^{\pi r}}{2\pi r}}\left(1e^{2\pi r}\right)`$ (41)
These wave functions will peak for $`r=0`$ so they predict an isotropic universe. They will exponentially decay for larger values of $`r`$. Whether this “normalizing” function is the right weight function is so far only speculation, but if these functions are reasonable the wave functions predict an isotropic universe.
## VII Conclusion and Summary
In this paper we have discussed the Bianchi Type $`I`$ minisuperspace model with compact spatial sections. In the classical case this removed some pathologies. The classical solutions were all written down in a way so that the solution space was manifestly factorized into three parts: a topological sector, a Kasner sector and a matter sector. The Kasner solutions were obtained in the limit $`A\mathrm{}`$ where one easily could see that the modular sector becomes degenerate. Choosing compact spatial sections also gave a better understanding of the topology of the 4-space. For instance the apparently flat case $`\gamma =0=T^{\mu \nu }`$ of the Kasner universe was shown to have a conical singularity at $`t=0`$.
In the quantum mechanical case since the (true) phase space has a dimension equal to the dimension of the solution space, it is easier to give a precise meaning to the WD equation. This paper has also shown the necessity of introducing compact spatial sections when one wants to quantize the model, and provides a solution to the WD equation for $`T^3`$ spatial sections. Also a set of solutions in the $`SO(3)`$ sector have been given which had particularly nice algebraic properties. These solutions have a structure which describes a shell structure in the $`SO(3)`$ sector. It was also indicated how these solutions allow geometry changing solutions.
Inserting various matter configurations in general showed isotropization of the universe. The resulting tunneling wave functions appeared as if they peaked at small anisotropy. Under reasonable considerations we constructed a wave function which had an exponentially decaying probability amplitude for increasing anisotropy. This seems to agree with other authors.
It is quite clear that universe models with compact spatial sections will yield many interesting and surprising results. Since these possibilities only recently have come to physicist’s and cosmologist’s attention there is a *lot* of work still to be done. What we have shown in this paper is that even at the classical level, this may reveal interesting properties of cosmological models.
## Acknowledgments
First of all I would like to thank Ø. Grøn for great inspiration and for encouraging me to write this paper. I am also very grateful to Ø. Grøn and B. Steffensen for reading through the manuscript and making useful comments. Helpful conversations with J.M. Leinaas are also gratefully acknowledged. |
warning/0003/gr-qc0003095.html | ar5iv | text | # The quantum modes of the (1+1)-dimensional oscillators in general relativity
## 1 Introduction
In general relativity, the geometric models play the role of kinetics, helping us to understand the characteristics of the classical or quantum free motion on a given background. One of the simplest geometric models in (1+1) dimensions is that of the quantum relativistic oscillator (RO) defined as a free massive scalar particle on the anti-de Sitter static background . Recently, we have generalized this model to a family of quantum models of RO whose metrics are one-parameter deformations (i.e. conformal transformations) of the anti-de Sitter or de Sitter ones . As it is shown in Refs., the deformed anti-de Sitter metrics give the relativistic correspondents the usual nonrelativistic Pöschl-Teller (PT) problems while the deformed de Sitter metrics generate relativistic Rosen-Morse (RM) problems . A remarkable property of these RO is that all of them have as nonrelativistic limit just the usual nonrelativistic harmonic oscillator (NRHO) .
In these relativistic models the Klein-Gordon equation is analytically solvable in the same manner as the Schrödinger equation of the mentioned well-studied nonrelativistic problems. This allows one to study the RO by using the successful methods of supersymmetry and shape invariance with minimal changes requested by the specific form the Klein-Gordon equation . In this way one can derive the normalized energy eigenfunctions of the discrete energy spectrum and the form of the shift operators of the energy basis that are involved in the structure of the dynamical algebras .
Here we would like to present a systematic study of our family of RO based on the supersymmetry and shape invariance of the relativistic potentials, pointing out the main specific features of the PT and RM relativistic problems. We believe that this first example of a family of metrics generating analytically solvable quantum problems could be of interest for further investigations concerning the supersymmetry of other solvable relativistic quantum models in (3+1) dimensions or more .
We start in Sec.2 with a short review of the (1+1) relativistic scalar quantum mechanics constructed as the one-particle restriction of the theory of the scalar free field on curved space-time. We define the state space in the coordinate representation and we introduce the coordinate and momentum operators. In Sec.3 we present the relativistic PT and RM oscillators giving their energy spectra and the energy eigenfunctions up to normalization factors. The relativistic supersymmetry and the shape invariance of the relativistic PT and RM potentials are used in the next section for deriving the definitive form of the normalized energy eigenfunctions of the discrete energy spectra. The Sec.5. is devoted to the properties of the shift operators of the energy bases of our RO. Therein we recover the known shift operators of the PT models and we derive those of the RM models.
## 2 Relativistic quantum mechanics
It is well-known that the one-particle relativistic quantum mechanics cannot be constructed as an independent consistent theory because of some difficulties related to the probabilistic interpretation of the relativistic wave functions. The good theory of the relativistic quantum systems is in fact the quantum field theory where the second quantization guarantees a coherent probabilistic interpretation. In these conditions, the relativistic quantum mechanics, in the sense of general relativity, can be seen as the one-particle restriction of the quantum field theory on curved backgrounds. Herein the quantum modes of the scalar particle in external gravitational field are given by the particular solutions of the free Klein-Gordon equation. These may form a basis in the space of wave functions organized as a Hilbert space with respect to the scalar product derived from the expression of the conserved electric charge .
### 2.1 The Klein-Gordon equation in special frames
Let us consider a (1+1)-dimensional background with a static local chart (i.e., natural frame) of holonomic coordinates $`(u^0,u^1)(t,u)`$ where the metric tensor defined on the space domain $`D_u`$ is $`g_{\mu \nu }(u)`$, $`\mu ,\nu =0,1`$, and $`g=\mathrm{det}(g_{\mu \nu })`$. The one-particle quantum modes of a scalar field $`\varphi `$ of the mass $`m`$, minimally coupled with the gravitational field, are given by the Klein-Gordon equation
$$\frac{1}{\sqrt{g}}_\mu \left(\sqrt{g}g^{\mu \nu }_\nu \varphi \right)+m^2\varphi =0,$$
(1)
written in natural units with $`\mathrm{}=c=1`$. Since in the static charts the energy, $`E`$, is conserved, the Klein-Gordon equation has a set of fundamental solutions (of positive and negative frequencies),
$$\varphi _E^{(+)}(t,u)=\frac{1}{\sqrt{2E}}e^{iEt}U_E(u),\varphi ^{()}=(\varphi ^{(+)})^{},$$
(2)
which depend on the static energy eigenfunctions $`U_E`$. These may be orthonormal (in usual or generalized sense) with respect to the relativistic scalar product
$$U,U^{}=_{D_u}𝑑u\mu (u)U(u)^{}U^{}(u)$$
(3)
where
$$\mu =\sqrt{g}g^{00}$$
(4)
is the specific relativistic weight function of the scalar field. The Hilbert space of the square integrable functions with respect to this scalar product is denoted by $`^2(D_u,\mu )`$. Oviously, the set of the wave functions $`U_E`$ represents an usual or generalized basis in this space. This will be called the energy basis.
In the case of the static backgrounds, a change of the space coordinates does not change the quantum modes. On the other hand, it is known that in (1+1) dimensions any static background has a special natural frame where the metric is the conformal transformation of the Minkowski flat one. Starting with any natural frame $`(t,u)`$, the space coordinate of the special frame $`(t,x)`$ reads
$$x=\chi (u)=𝑑u\mu (u)+\mathrm{const}.,$$
(5)
where the constant assures the condition $`\chi (0)=0`$. In the special frame we have $`\stackrel{~}{g}_{00}(x)=\stackrel{~}{g}_{11}(x)`$ and $`\stackrel{~}{\mu }(x)=1`$. Therefore, the scalar product (3) becomes just the usual one of the Hilbert space $`^2(D)`$ where $`D`$ is the domain of the coordinate $`x`$ corresponding to $`D_u`$.
If we denote $`\stackrel{~}{g}_{00}=1+v`$, then we can write the line element of the special frame as
$$ds^2=[1+v(x)](dt^2dx^2)$$
(6)
while the Klein-Gordon equation takes the form
$$\left[\frac{d^2}{dx^2}+V_R(x)\right]U_E(x)=(E^2m^2)U_E(x)$$
(7)
where $`V_R=m^2v`$. We say that this is the relativistic potential since in the nonrelativistic limit $`V_R/2m`$ becomes just the usual potential of the corresponding Schrödinger equation.
### 2.2 Observables
The linear operators on $`^2`$ (denoted here using boldface) can be defined either by giving their matrix elements in a countable basis of $`^2`$ or as differential operators in the coordinate representation. The most general differential operator we use in an arbitrary frame $`(t,u)`$ has the form
$$(𝐃U)(u)=i\left[f(u)\frac{d}{du}+h(u)\right]U(u),$$
(8)
depending on two real functions $`f`$ and $`h`$. Its adjoint with respect to the scalar product (3) is
$$𝐃^{}=𝐃+i\left[\frac{1}{\mu }\frac{d(\mu f)}{du}2h\right]\mathrm{𝟏}$$
(9)
where $`\mathrm{𝟏}`$ is the identity operator. Hereby, we see that for $`h=_u(\mu f)/2\mu `$ the operator $`𝐃`$ is self-adjoint .
The consequence is that we can introduce a unique pair of self-adjoint coordinate and momentum operators defining their action in any natural frame $`(t,u)`$. One can easily verify that the following definitions,
$$(𝐗U)(u)=\chi (u)U(u),(𝐏U)(u)=i\frac{1}{\mu (u)}\frac{dU(u)}{du},$$
(10)
are satisfactory since the commutation relation
$$[𝐏,𝐗]=i\mathrm{𝟏}$$
(11)
is just the desired one. Obviously, in the special frame $`(t,x)`$ these operators have the same action as in the case of the Minkowski flat space-time, namely
$$(𝐗U)(x)=xU(x),(𝐏U)(x)=i\frac{dU(x)}{dx}.$$
(12)
Furthermore, one can put the Klein-Gordon equation (7) in operator form,
$$𝐇^2=m^2\mathrm{𝟏}+𝚫[V_R],$$
(13)
where $`𝐇`$ is the Hamiltonian operator defined by $`𝐇U_E=EU_E`$ and
$$𝚫[V]=𝐏^2+V(𝐗).$$
(14)
Thus we have obtained the main operators on $`^2`$. The whole algebra of observables is that freely generated by the operators $`𝐗`$ and $`𝐏`$, like in the Schrödinger picture of the nonrelativistic one-dimensional quantum mechanics.
## 3 Relativistic Oscillators
The new geometric models of RO we discuss here are simple systems of free test particles that move on static backgrounds simulating oscillations. This means that there are local charts of coordinates $`(t,u)`$ where an observer at $`u=0`$ moving along the direction $`_t`$ observes an oscillatory geodesic motion. These charts called proper natural frames have line elements ,
$$ds^2=g_{00}dt^2+g_{11}du^2=\frac{1+(1+\lambda )\omega ^2u^2}{1+\lambda \omega ^2u^2}dt^2\frac{1+(1+\lambda )\omega ^2u^2}{(1+\lambda \omega ^2u^2)^2}du^2,$$
(15)
depending on a real parameter $`\lambda `$. Thus one obtains a family of metrics which are conformal transformations either of the anti-de Sitter metric (as given in Ref.) or of the de Sitter one. The anti-de Sitter metric with $`\lambda =1`$ is also included in this family. A special case is that of $`\lambda =0`$ when we say that the line element
$$ds^2=(1+\omega ^2u^2)(dt^2du^2)$$
(16)
defines the normal RO. In Ref it is shown that the quantum models with $`\lambda 0`$ have countable energy spectra while for $`\lambda >0`$ the energy spectra are mixed, with a finite discrete sequence and a continuous part. All these models will be presented here in the special frames $`(t,x)`$ associated with the proper frames $`(t,u)`$ defined above. The advantage is that in the special frames our RO appear either as PT or as RM relativistic systems which can be analytically solved like those known from the nonrelativistic quantum mechanics.
### 3.1 The relativistic Pöschl-Teller models
Let us consider first the models with $`\lambda <0`$ when the metrics are conformal transformations of the anti-de Sitter one. We denote
$$\lambda =ϵ^2,\widehat{\omega }=ϵ\omega ,ϵ0$$
(17)
and calculate the space coordinate of the special frame. According to Eq.(5) we obtain
$$x=\frac{1}{\widehat{\omega }}\mathrm{arcsin}\widehat{\omega }u$$
(18)
while from Eq.(15) we write the line element in this frame,
$$ds^2=\left(1+\frac{1}{ϵ^2}\mathrm{tan}^2\widehat{\omega }x\right)(dt^2dx^2),$$
(19)
where the space domain is $`D=(\pi /2\widehat{\omega },\pi /2\widehat{\omega })`$ because of the event horizon at $`\pm \pi /2\widehat{\omega }`$. The relativistic potential,
$$V_{PT}(k,x)=\frac{m^2}{ϵ^2}\mathrm{tan}^2\widehat{\omega }x=\widehat{\omega }^2k(k1)\mathrm{tan}^2\widehat{\omega }x,$$
(20)
is a PT one depending on the new parameter
$$k=\sqrt{\frac{m^2}{ϵ^2\widehat{\omega }^2}+\frac{1}{4}}+\frac{1}{2}.$$
(21)
In the following we use $`k`$ instead of $`m`$, as the main parameter of the PT models that will be denoted from now by $`(k)`$.
The Klein-Gordon equation (7) of the model $`(k)`$ with the potential (20) can be written as
$$\left[\frac{1}{\widehat{\omega }^2}\frac{d^2}{dx^2}+\frac{k(k1)}{\mathrm{cos}^2\widehat{\omega }x}\right]U(x)=\nu ^2U(x)$$
(22)
where
$$\nu ^2=\frac{E^2}{\widehat{\omega }^2}\left(1\frac{1}{ϵ^2}\right)\frac{m^2}{\widehat{\omega }^2}=\frac{E^2}{\widehat{\omega }^2}+(1ϵ^2)k(k1).$$
(23)
Its solutions
$$U(x)\mathrm{sin}^{2s}\widehat{\omega }x\mathrm{cos}^{2p}\widehat{\omega }xF(s+p\frac{\nu }{2},s+p+\frac{\nu }{2},2s+\frac{1}{2},\mathrm{sin}^2\widehat{\omega }x),$$
(24)
are expressed in terms of Gauss hypergeometric functions whose parameters $`s`$ and $`p`$ are solutions of the equations $`2s(2s1)=0`$ and $`2p(2p1)=k(k1)`$. The wave functions (24) have good physical meaning only when $`F`$ is a polynomial selected by a suitable quantization condition (since otherwise $`F`$ is strongly divergent for $`x\pm \pi /2\widehat{\omega }`$). Therefore, we introduce the quantum number $`n_s`$ and impose
$$\nu =2(n_s+s+p),n_s=0,1,2,\mathrm{}.$$
(25)
In addition, we choose the boundary conditions of the regular modes given by $`2s=0,1`$ and $`2p=k`$. Then the energy levels
$$E_{k,n}^{}{}_{}{}^{2}=\widehat{\omega }^2[(k+n)^2+(ϵ^21)k(k1)]$$
(26)
depend only on the main quantum number, $`n=2n_s+2s`$, which takes even values if $`s=0`$ and odd values for $`s=1/2`$. Particularly for the anti-de Sitter model with $`ϵ=1`$ we recover the well-known result $`E_{k,n}=\omega (k+n)`$ .
The next step is to derive the concrete form of the normalized energy eigenfunctions corresponding to these energy levels. According to Eqs.(24) and (38), these are
$$U_{k,n}(x)=N_{k,n}\mathrm{sin}^{2s}\widehat{\omega }x\mathrm{cos}^k\widehat{\omega }xF(n_s,n_s+k+2s,2s+\frac{1}{2},\mathrm{sin}^2\widehat{\omega }x),$$
(27)
where the normalization constants $`N_{k,n}`$ might be calculated with the help of the scalar product of $`^2(D)`$. However, there is another efficient method based on supersymmetry and shape invariance giving directly the Rodrigues formula of the normalized eigenfunctions. This will be presented in the next section.
We specify that our PT models are well-defined for any $`k[1,\mathrm{})`$ since the limit $`k1`$ (when $`m0`$) has a good physical meaning. Indeed, in this case the massless particle remains confined to the rectangular infinite well of width $`\pi /\widehat{\omega }`$ having the equidistant energy levels
$$E_{1,n}=\widehat{\omega }(n+1),$$
(28)
corresponding to the normalized eigenfunctions
$$U_{1,n}(x)=\sqrt{\frac{2\widehat{\omega }}{\pi }}\mathrm{sin}(n+1)\left(\frac{\pi }{2}\widehat{\omega }x\right),n=0,1,2,\mathrm{}.$$
(29)
Notice that this is a pure relativistic model since its nonrelativistic limit does not make sense.
### 3.2 The relativistic Rosen-Morse models
For $`\lambda >0`$ the metrics of RO are conformal transformations of the de Sitter metric. Now we change the significance of $`ϵ`$ and put
$$\lambda =ϵ^2,\widehat{\omega }=ϵ\omega ,ϵ0.$$
(30)
Furthermore, from Eq.(5) we find
$$x=\frac{1}{\widehat{\omega }}\mathrm{arcsinh}\widehat{\omega }u$$
(31)
and from Eq.(15) we obtain the line element
$$ds^2=\left(1+\frac{1}{ϵ^2}\mathrm{tanh}^2\widehat{\omega }x\right)(dt^2dx^2)$$
(32)
in the special frame where the space domain is $`D=(\mathrm{},\mathrm{})`$. These metrics define relativistic RM models whose potentials,
$$V_{RM}(j,x)=\frac{m^2}{ϵ^2}\mathrm{tanh}^2\widehat{\omega }x=\widehat{\omega }^2j(j+1)\mathrm{tanh}^2\widehat{\omega }x,$$
(33)
depend on the parameter
$$j=\sqrt{\frac{m^2}{ϵ^2\widehat{\omega }^2}+\frac{1}{4}}\frac{1}{2}.$$
(34)
Like in the case of PT models, we consider that $`j`$ is the main parameter of the RM models, denoted by $`(j)`$.
Now the Klein-Gordon equation is
$$\left[\frac{1}{\widehat{\omega }^2}\frac{d^2}{dx^2}+\frac{j(j+1)}{\mathrm{cosh}^2\widehat{\omega }x}\right]U(x)=\widehat{\nu }^2U(x)$$
(35)
where
$$\widehat{\nu }^2=\frac{E^2}{\widehat{\omega }^2}+\left(1+\frac{1}{ϵ^2}\right)\frac{m^2}{\widehat{\omega }^2}=\frac{E^2}{\widehat{\omega }^2}+(1+ϵ^2)j(j+1).$$
(36)
The solutions
$$U(x)\mathrm{sinh}^{2s}\widehat{\omega }x\mathrm{cosh}^{2p}\widehat{\omega }xF(s+p\frac{\widehat{\nu }}{2},s+p+\frac{\widehat{\nu }}{2},2s+\frac{1}{2},\mathrm{sinh}^2\widehat{\omega }x),$$
(37)
depend on the parameters $`s`$ and $`p`$ which satisfy $`2s=0,1`$ and $`2p(2p1)=j(j+1)`$. Like in the nonrelativistic case, the relativistic RM models have mixed energy spectra with a finite discrete sequence and a continuous part .
The discrete levels arise from the quantization condition
$$\widehat{\nu }=2(n_s+s+p),n_s=0,1,2,\mathrm{}$$
(38)
One can show that the corresponding energy eigenfunctions are square integrable only when $`2p=j`$ and the quantum main number $`n=2n_s+2s`$ takes the values $`n=0,1,\mathrm{},n_{max}<j`$. Therefore, these are
$$U_{j,n}(x)=N_{j,n}\mathrm{sinh}^{2s}\widehat{\omega }x\mathrm{cosh}^j\widehat{\omega }xF(n_s,n_sj+2s,2s+\frac{1}{2},\mathrm{sinh}^2\widehat{\omega }x).$$
(39)
Hence it results that the discrete energy spectrum is finite having $`n_{max}+1`$ levels where $`n_{max}`$ is the highest integer smaller than $`j`$. This spectrum is included in the domain $`[m,m\sqrt{1+1/ϵ^2})`$ since the energy levels are
$$E_{j,n}^{}{}_{}{}^{2}=\widehat{\omega }^2[(nj)^2+(ϵ^2+1)j(j+1)],n=0,1,2\mathrm{}n_{max}.$$
(40)
The definitive form of the normalized energy eigenfunctions of the discrete spectrum will be calculated in the next section by using the shape invariance of the RM potentials.
The continuous spectrum cover the domain $`[m\sqrt{1+1/ϵ^2},\mathrm{})`$ where we have $`\widehat{\nu }=i|\widehat{\nu }|`$. The generalized energy eigenfuctions of this spectrum are tempered distributions of the form (37) with $`2s=0,1`$ and $`2p=j`$. We note that for $`m0`$ (when $`j0`$) the discrete spectrum disappears while the continuous one becomes $`[0,\mathrm{})`$. In this model the massless test particle move like in flat space-time and, therefore, it does not have nonrelativistic limit.
### 3.3 The normal RO and the nonrelativistic limit
Our family of RO is continuous in $`\lambda =0`$ . This means that the limits for $`ϵ0`$ of the PT and RM models must coincide. Indeed, according to Eqs.(18) and (31) we find that in this limit $`xu`$ while from Eqs.(21) and (34) it results that for any model with $`m0`$ we have $`k\mathrm{}`$, $`j\mathrm{}`$ but
$$\underset{ϵ0}{lim}ϵ^2k=\underset{ϵ0}{lim}ϵ^2j=\frac{m}{\omega }.$$
(41)
Furthermore, we can verify that the finite discrete spectra of the models with $`\lambda >0`$ become countable while the continuous spectra disappear in a such a manner that the RM models and the PT ones have the same limit which is just the normal RO (with $`\lambda =0`$). The special frame of this model coincides with the proper one where the metric is defined by Eq.(16). Therefore, the relativistic potential is
$$V_0(u)=\underset{ϵ0}{lim}V_{PT}(x)=\underset{ϵ0}{lim}V_{RM}(x)=m^2\omega ^2u^2$$
(42)
and the Klein-Gordon equation
$$\left[\frac{d^2}{du^2}+m^2\omega ^2u^2\right]U_n^{(0)}(u)=(E_{n}^{}{}_{}{}^{2}m^2)U_n^{(0)}(u)$$
(43)
gives the familiar energy eigenfunctions of the NRHO,
$$U_n^{(0)}=\left(\frac{m\omega }{\pi }\right)^{1/4}\frac{1}{\sqrt{n!2^n}}e^{m\omega ^2u^2/2}H_n(\sqrt{m\omega }u),$$
(44)
(where $`H_n`$ are Hermite polynomials), but relativistic energy levels,
$$E_{n}^{(0)}{}_{}{}^{2}=m^2+2m\omega (n+\frac{1}{2}).$$
(45)
In the nonrelativistic limit (defined by $`m/\omega \mathrm{}`$), the normal RO becomes the NRHO with the potential $`V_0/2m`$ and usual energy levels. The nonrelativistic limit of the other models, with $`m0`$ and $`\lambda 0`$, can be easily calculated if we observe that, according to Eqs.(21) and (34), this is equivalent with the limit $`\lambda 0`$ and, in addition, $`m\omega `$. Hereby it results that all the RO with $`m>0`$ have the same nonrelativistic limit like that of the normal RO of the mass $`m`$, namely the usual NRHO. On the other hand, it is interesting that in this way we can show that the parameter $`\lambda `$, or the parameter $`ϵ`$ related to it, does not have a direct nonrelativistic equivalent, since the terms involving $`\lambda `$ vanish in this limit.
## 4 Supersymmetry and Shape invariance
A relativistic supersymmetric quantum mechanics can be constructed in the same way as the nonrelativistic one. The main problem here is to find the operator which should play the role of Hamiltonian. We shall show that this is the operator (14) with a suitable translated potential.
### 4.1 Supersymmetry
Let us start with a (1+1)-dimensional relativistic model with the potential $`V_R`$ giving finite or countable energy spectrum. First we denote the energy levels by $`E_n^{()}`$ and the corresponding energy eigenfunctions by $`U_n^{()}`$. Then Eq.(7) in the special frame can be written as
$$𝚫[V_{}]U_n^{()}=d_n^{()}U_n^{()},n=0,1,2,\mathrm{}$$
(46)
where
$$V_{}=V_R(E_{0}^{()}{}_{}{}^{2}m^2)$$
(47)
and
$$d_n^{()}=E_{n}^{()}{}_{}{}^{2}E_{0}^{()}{}_{}{}^{2}.$$
(48)
Herein we have translated the spectrum of $`𝚫`$ in a such a manner to accomplish the condition $`d_0^{()}=0`$ we need for defining the superpotential
$$W(x)=\frac{1}{U_0^{()}}\frac{dU_0^{()}(x)}{dx}.$$
(49)
Then we have $`V_{}=W^2W^{}`$ (with the notation $`{}_{}{}^{}=_x`$) and the supersymmetric partner (superpartner) potential of $`V_{}`$ reads $`V_+=W^2+W^{}=V_{}+2W^2`$. Furthermore, we introduce the operator
$$𝐀=i𝐏+W(𝐗)$$
(50)
which satisfies
$$[𝐀,𝐀^{}]=2W^{}(𝐗)$$
(51)
and help us to write
$$𝚫[V_{}]=𝐀^{}𝐀,𝚫[V_+]=\mathrm{𝐀𝐀}^{}.$$
(52)
Now, like in the nonrelativistic case , we can convince ourselves that the spectrum of the eigenvalue problem
$$𝚫[V_+]U_n^{(+)}=d_n^{(+)}U_n^{(+)}$$
(53)
coincides with that of Eq.(46), apart of the eigenvalue $`d_0^{()}=0`$. Thus we have $`d_n^{(+)}=d_{n+1}^{()}`$, $`n=0,1,2,..`$, while the normalized eigenfunctions of $`𝚫[V_{}]`$ and $`𝚫[V_+]`$ satisfy
$$𝐀U_n^{()}=\eta \sqrt{d_n^{()}}U_{n1}^{(+)},𝐀^{}U_{n1}^{(+)}=\eta ^{}\sqrt{d_n^{()}}U_n^{()}$$
(54)
where $`\eta `$ is an arbitrary phase factor.
Hence, we can say that the (1+1) relativistic supersymmetric quantum mechanics has the same main features as the nonrelativistic one. It remains us to study the shape invariance of the relativistic potentials of the RO related through supersymmetry.
### 4.2 Shape invariance
Let us consider the PT model (k) and identify $`U_n^{()}U_{k,n}`$ and $`E_n^{()}E_{k,n}`$. Then the differences (48) are
$$d_n^{()}d_{k,n}=E_{k,n}^{}{}_{}{}^{2}E_{k,0}^{}{}_{}{}^{2}=\widehat{\omega }^2n(n+2k),$$
(55)
and from Eqs.(47) and (20) we obtain
$$V_{}(k,x)=V_{PT}(k,x)+m^2E_{k,0}^{}{}_{}{}^{2}=\widehat{\omega }^2[k(k1)\mathrm{tan}^2\widehat{\omega }xk],$$
(56)
On the other hand, the normalized ground-state eigenfunction calculated from Eq.(27),
$$U_{k,0}(x)=\left(\frac{\widehat{\omega }^2}{\pi }\right)^{\frac{1}{4}}\left[\frac{\mathrm{\Gamma }(k+1)}{\mathrm{\Gamma }(k+\frac{1}{2})}\right]^{\frac{1}{2}}\mathrm{cos}^k\widehat{\omega }x,$$
(57)
gives the superpotential $`W(k,x)=\widehat{\omega }k\mathrm{tan}\widehat{\omega }x`$ which allows us to find the superpartner of $`V_{}`$,
$$V_+(k,x)=V_{}(k,x)+2W(k,x)^2=\widehat{\omega }^2[k(k+1)\mathrm{tan}^2\widehat{\omega }x+k].$$
(58)
Moreover, with this superpotential the operator (50) reads
$$𝐀_k=i𝐏+\widehat{\omega }k\mathrm{tan}\widehat{\omega }𝐗=\mathrm{cos}^k\widehat{\omega }𝐗(i𝐏)\mathrm{cos}^k\widehat{\omega }𝐗$$
(59)
while from Eq.(51) we obtain
$$[𝐀_k,𝐀_k^{}]=2k\widehat{\omega }^2\mathrm{𝟏}+\frac{1}{2k}(𝐀_k+𝐀_k^{})^2.$$
(60)
Now we observe that the potentials $`V_{}(k)`$ and $`V_+(k)`$ are shape invariant since
$$V_+(k,x)=V_{}(k+1,x)+\widehat{\omega }^2(2k+1).$$
(61)
Consequently, we can identify $`U_n^{(+)}U_{k+1,n}`$ which means that the normalized energy eigenfunctions satisfy
$$𝐀_kU_{k,n}=\sqrt{d_{k,n}}U_{k+1,n1},𝐀_k^{}U_{k+1,n1}=\sqrt{d_{k,n}}U_{k,n}.$$
(62)
as it results from Eqs.(52) with $`\eta =1`$. Thus we have related the energy eigenfunctions of the model $`(k)`$ with those of its superpartner model, $`(k+1)`$. In general, we can write any normalized energy eigenfunction of the model $`(k)`$ as
$$U_{k,n}=\frac{1}{\widehat{\omega }^n\sqrt{n!}}\left[\frac{\mathrm{\Gamma }(n+2k)}{\mathrm{\Gamma }(2n+2k)}\right]^{\frac{1}{2}}𝐀_k^{}𝐀_{k+1}^{}\mathrm{}𝐀_{k+n1}^{}U_{k+n,0}.$$
(63)
where $`U_{k+n,0}`$ is the normalized ground-state eigenfunction of the model $`(k+n)`$ given by Eq.(57).
For the relativistic RM models we use the same method starting with the model $`(j)`$ and denoting $`U_n^{()}U_{j,n}`$ and $`E_n^{()}E_{j,n}`$. Then the differences (48) are
$$d_n^{()}d_{j,n}=E_{j,n}^{}{}_{}{}^{2}E_{j,0}^{}{}_{}{}^{2}=\widehat{\omega }^2n(2jn),$$
(64)
and, according to (33), we have
$$V_{}(j,x)=V_{RM}(j,x)+m^2E_{j,0}^2=\widehat{\omega }^2[j(j+1)\mathrm{tanh}^2\widehat{\omega }xj].$$
(65)
From Eq.(39) we find the normalized ground-state eigenfunction
$$U_{j,0}(x)=\left(\frac{\widehat{\omega }^2}{\pi }\right)^{\frac{1}{4}}\left[\frac{\mathrm{\Gamma }(j+\frac{1}{2})}{\mathrm{\Gamma }(j)}\right]^{\frac{1}{2}}\mathrm{cosh}^j\widehat{\omega }x,$$
(66)
giving the superpotential $`W(j,x)=\widehat{\omega }j\mathrm{tanh}\widehat{\omega }x`$. Hereby we obtain
$$V_+(j,x)=V_{}(j,x)+2W(j,x)^2=\widehat{\omega }^2[j(j1)\mathrm{tanh}^2\widehat{\omega }x+j].$$
(67)
Now the operator (50) reads
$$𝐀_j=i𝐏+\widehat{\omega }j\mathrm{tanh}\widehat{\omega }𝐗=\mathrm{cosh}^j\widehat{\omega }𝐗(i𝐏)\mathrm{cosh}^j\widehat{\omega }𝐗$$
(68)
while Eq.(51) gives
$$[𝐀_j,𝐀_j^{}]=2j\widehat{\omega }^2\mathrm{𝟏}\frac{1}{2j}(𝐀_j+𝐀_j^{})^2.$$
(69)
The potentials $`V_{}(j)`$ and $`V_+(j)`$ are shape invariant since
$$V_+(j,x)=V_{}(j1,x)+\widehat{\omega }^2(2j1).$$
(70)
Consequently, as in the previous case, we find that the normalized energy eigenfunctions satisfy
$$𝐀_jU_{j,n}=\sqrt{d_{j,n}}U_{j1,n1},𝐀_j^{}U_{j1,n1}=\sqrt{d_{j,n}}U_{j,n}$$
(71)
if we take $`\eta =1`$ in Eqs.(52). Thus we have obtained the relation between the sets of energy eigenfunctions of the superpartner models $`(j)`$ and $`(j1)`$. Moreover, we can also express the normalized eigenfunctions as
$$U_{j,n}=\frac{1}{\widehat{\omega }^n\sqrt{n!}}\left[\frac{\mathrm{\Gamma }(2j2n+1)}{\mathrm{\Gamma }(2jn+1)}\right]^{\frac{1}{2}}𝐀_j^{}𝐀_{j1}^{}\mathrm{}𝐀_{jn+1}^{}U_{jn,0}.$$
(72)
where now $`U_{jn,0}`$ is the normalized ground-state eigenfunction of the model $`(jn)`$ given by Eq.(66).
### 4.3 The normalized energy eigenfunctions
The normalization of the energy eigenfunctions of the PT models may be easily done in usual way but for the RM models there are some technical difficulties that can be avoided by using the previous results. Indeed, we observe that the Eqs.(63) and (72) are nothing else than the operator form of the Rodrigues formulas of the normalized eigenfunctions (in our phase convention with $`\eta =1`$). Therefore, it remains only to rewrite their expressions in usual form.
For the PT models we replace the operator (59) in Eq.(63) which takes the form
$`U_{k,n}(x)={\displaystyle \frac{(1)^n}{\widehat{\omega }^n\sqrt{n!}}}\left[{\displaystyle \frac{\mathrm{\Gamma }(n+2k)}{\mathrm{\Gamma }(2n+2k)}}\right]^{\frac{1}{2}}\mathrm{cos}^k\widehat{\omega }x{\displaystyle \frac{d}{dx}}{\displaystyle \frac{1}{\mathrm{cos}\widehat{\omega }x}}{\displaystyle \frac{d}{dx}}\mathrm{}`$
$`\mathrm{}{\displaystyle \frac{1}{\mathrm{cos}\widehat{\omega }x}}{\displaystyle \frac{d}{dx}}\mathrm{cos}^{k+n1}\widehat{\omega }xU_{k+n,0}(x).`$ (73)
Then, according to Eqs.(18) and (57), we obtain the final Rodrigues formula of the normalized energy eigenfunctions of the PT models in proper frames,
$`U_{k,n}(u)=\left({\displaystyle \frac{\widehat{\omega }^2}{\pi }}\right)^{\frac{1}{4}}{\displaystyle \frac{(1)^n}{\widehat{\omega }^n\sqrt{n!}}}\left[{\displaystyle \frac{\mathrm{\Gamma }(2k+n)\mathrm{\Gamma }(k+n+1)}{\mathrm{\Gamma }(2k+2n)\mathrm{\Gamma }(k+n+\frac{1}{2})}}\right]^{\frac{1}{2}}`$
$`\times (1\widehat{\omega }^2u^2)^{\frac{k1}{2}}{\displaystyle \frac{d^n}{du^n}}(1\widehat{\omega }^2u^2)^{k+n\frac{1}{2}}.`$ (74)
In the same way we can derive the Rodrigues formula for the normalized energy eigenfunctions of the RM models. By using the Eqs.(68) and (72) we find the normalized energy egenfunctions of the discrete spectrum in the proper frame,
$`U_{j,n}(u)=\left({\displaystyle \frac{\widehat{\omega }^2}{\pi }}\right)^{\frac{1}{4}}{\displaystyle \frac{(1)^n}{\widehat{\omega }^n\sqrt{n!}}}\left[{\displaystyle \frac{\mathrm{\Gamma }(2j2n+1)\mathrm{\Gamma }(jn+\frac{1}{2})}{\mathrm{\Gamma }(2jn+1)\mathrm{\Gamma }(jn)}}\right]^{\frac{1}{2}}`$
$`\times (1+\widehat{\omega }^2u^2)^{\frac{j+1}{2}}{\displaystyle \frac{d^n}{du^n}}(1+\widehat{\omega }^2u^2)^{j+n\frac{1}{2}}.`$ (75)
Of course, as it was expected, this formula gives square integrable functions only for $`nn_{max}`$. On the other hand, here the problem of ”normalization” of the generalized energy eigenfuntions of the continuous spectrum remains open since there are not yet efficient procedures for doing this.
Hence we have obtained the definitive formulas of the normalized energy eigenfunctions of our RO corresponding to the discrete energy levels. These can be written now in terms of Jacobi or Gegenbauer polynomials and even as associated Legendre functions but only when $`k`$ and $`j`$ are integer numbers. In the particular case of the PT model with $`k=1`$ the trigonometric form (29) can be derived from Eq.(4.3) by using the properties of the Tchebyshev polynomials.
In Sec.3.3 we have seen that the normal RO has the same energy eigenfunctions as the NRHO. Now, by taking into account that for large arguments, $`z`$, we have $`\mathrm{\Gamma }(z+a)/\mathrm{\Gamma }(z+b)z^{(ab)}`$ and by using Eqs.(41) we verify directly that
$$\underset{ϵ0}{lim}U_{k,n}(u)=\underset{ϵ0}{lim}U_{j,n}(u)=U_n^{(0)}(u).$$
(76)
Because of these properties we can say that the eigenfunctions (4.3) and (4.3) represent relativistic generalizations of the NRHO eigenfunctions, different from that of the algebraic method .
## 5 Shift operators
In our models only one main quantum number is involved and, therefore, in each model we must have a pair of shift operators, i.e. the rising and the lowering operators of the energy basis. In general, the shift operators are different from those of the supersymmetry apart the shift operators of the normal RO that are up to factors just those of the supersymmetry since this model is its own superpartner.
Let us start with this simplest case since here the energy eigenfunctions are similar to those of the NRHO. Consequently, we can take over the well-known results from the nonrelativistic theory defining the differential operators
$`(𝐚U)(u)`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2m\omega }}}\left({\displaystyle \frac{d}{du}}+m\omega u\right)U(u),`$ (77)
$`(𝐚^{}U)(u)`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2m\omega }}}\left({\displaystyle \frac{d}{du}}+m\omega u\right)U(u).`$ (78)
of the Heisenberg-Weyl algebra. Obviously, they are the desired shift operators which obey $`[𝐚,𝐚^{}]=\mathrm{𝟏}`$ giving us the operator of number of quanta $`𝐍=𝐚^{}𝐚`$ and
$$𝐗=\frac{1}{\sqrt{2m\omega }}(𝐚^{}+𝐚),𝐏=i\sqrt{\frac{m\omega }{2}}(𝐚^{}𝐚).$$
(79)
Moreover, it is natural to find that
$$\underset{ϵ0}{lim}𝐀_k=\underset{ϵ0}{lim}𝐀_j=\sqrt{2m\omega }𝐚.$$
(80)
For the models with $`\lambda 0`$ the shift operators differ from those of supersymmetry. They can be calculated directly by using the action of the supesymmetry operators and the form of the normalized energy eigenfunctions derived above. In the case of $`\lambda =ϵ^2`$, after a few manipulation, we find that the shift operators of the PT model $`(k)`$ can be defined in the proper frame as
$`(𝐀_{k,(+)}U_{k,n})(u)`$ $`=`$ $`{\displaystyle \frac{1}{\widehat{\omega }\sqrt{2k}}}\left[(1\widehat{\omega }^2u^2){\displaystyle \frac{d}{du}}+\widehat{\omega }^2u(k+n)\right]U_{k,n}(u),`$ (81)
$`(𝐀_{k,()}U_{k,n})(u)`$ $`=`$ $`{\displaystyle \frac{1}{\widehat{\omega }\sqrt{2k}}}\left[(1\widehat{\omega }^2u^2){\displaystyle \frac{d}{du}}+\widehat{\omega }^2u(k+n)\right]U_{k,n}(u).`$ (82)
Their shifting action is
$$𝐀_{k,(+)}U_{k,n}=C_{k,n}^{(+)}U_{k,n+1},𝐀_{k,()}U_{k,n}=C_{k,n}^{()}U_{k,n1},$$
(83)
where
$`C_{k,n}^{(+)}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2k}}}\left[{\displaystyle \frac{(2k+n)(k+n)}{k+n+1}}\right]^{\frac{1}{2}}\sqrt{n+1},`$ (84)
$`C_{k,n}^{()}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2k}}}\left[{\displaystyle \frac{(2k+n1)(k+n)}{k+n1}}\right]^{\frac{1}{2}}\sqrt{n}.`$ (85)
If we rewrite the action of the operators (81) and (82) in the special frame $`(t,x)`$ then we recover the result of Ref.. Furthermore, we can verify the commutation relation
$$[𝐀_{k,()},𝐀_{k,(+)}]U_{k,n}=\left(1+\frac{n}{k}\right)U_{k,n}$$
(86)
and the identity
$$2k𝐀_{k,(+)}𝐀_{k,()}U_{k,n}=n(2k+n1)U_{k,n}$$
(87)
which is just the Klein-Gordon equation in operator form . In the limit $`ϵ0`$ we have
$$\underset{ϵ0}{lim}𝐀_{k,(+)}=𝐚^{},\underset{ϵ0}{lim}𝐀_{k,()}=𝐚.$$
(88)
With the same procedure we find the shift operators of the RM model $`(j)`$ in the proper frame,
$`(𝐀_{j,(+)}U_{j,n})(u)`$ $`=`$ $`{\displaystyle \frac{1}{\widehat{\omega }\sqrt{2j}}}\left[(1+\widehat{\omega }^2u^2){\displaystyle \frac{d}{du}}+\widehat{\omega }^2u(jn)\right]U_{j,n}(u),`$ (89)
$`(𝐀_{j,()}U_{j,n})(u)`$ $`=`$ $`{\displaystyle \frac{1}{\widehat{\omega }\sqrt{2j}}}\left[(1+\widehat{\omega }^2u^2){\displaystyle \frac{d}{du}}+\widehat{\omega }^2u(jn)\right]U_{j,n}(u),`$ (90)
which have the action
$$𝐀_{j,(+)}U_{j,n}=C_{j,n}^{(+)}U_{j,n+1},𝐀_{j,()}U_{j,n}=C_{j,n}^{()}U_{j,n1},$$
(91)
where
$`C_{j,n}^{(+)}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2j}}}\left[{\displaystyle \frac{(2jn)(jn)}{jn1}}\right]^{\frac{1}{2}}\sqrt{n+1},`$ (92)
$`C_{j,n}^{()}`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2j}}}\left[{\displaystyle \frac{(2jn+1)(jn)}{jn+1}}\right]^{\frac{1}{2}}\sqrt{n}.`$ (93)
They obey the commutation rule
$$[𝐀_{j,()},𝐀_{j,(+)}]U_{j,n}=\left(1\frac{n}{j}\right)U_{j,n}$$
(94)
and give us the operator form of the Klein-Gordon equation for discrete levels,
$$2j𝐀_{j,(+)}𝐀_{j,()}U_{j,n}=n(2jn+1)U_{j,n}.$$
(95)
Finally we find that
$$\underset{ϵ0}{lim}𝐀_{j,(+)}=𝐚^{},\underset{ϵ0}{lim}𝐀_{j,()}=𝐚.$$
(96)
We must specify that the shift operators of the models with $`\lambda 0`$ have two important properties, namely: they are not pure differential operators on $`^2(D_u,\mu )`$ and, moreover, the raising and lowering operators are not adjoint with each other, i.e. $`𝐀_{k,(\pm )}(𝐀_{k,()})^{}`$ and similarly for the RM models.
## 6 Comments
In this article we have studied the quantum modes of a family of (1+1) RO by using the methods of a supersymmetric relativistic quantum mechanics similar with the well-known nonrelativistic one. This was possible since the form of the Klein-Gordon equation in the special frames is very close to that of the Schrödinger equation, allowing us to introduce the relativistic potentials involved in supersymmetry and to exploit their shape invariance.
However, our relativistic theory has some new interesting features due to the fact that the mass is not only involved in the formula of the energy levels but also plays here the role of a coupling constant. For this reason there are some kind of regularities leading to a very simple parametrization in a such a manner that for any pair of superpartner models we have either $`\mathrm{\Delta }k=\pm 1`$ or $`\mathrm{\Delta }j=1`$. Thus $`k`$ and $`j`$ simulate the behavior of quantum numbers even though they can not be considered as eigenvalues of self-adjoint operators . On the other hand, the models with superpartner potentials can be seen as having particles of different masses moving on the same background. The consequence is that the masses of the sets of superpartner PT or RM models appear as being quantized according to the formulas $`m_k^2=ϵ^2\widehat{\omega }^2k(k1)`$ and $`m_j^2=ϵ^2\widehat{\omega }^2j(j+1)`$ respectively. These remarkable properties helped us to easily write down the Rodrigues formulas of the normalized energy eigenfunctions of the discrete spectra and to find the corresponding shift operators.
Concluding we can say that our family of models brings together the main solvable problems with parity-symmetric potentials of the one-dimensional quantum mechanics, interpreted as relativistic oscillators in the sense that all these models (apart those with $`k=1`$ and $`j=0`$) lead to the NRHO in the nonrelativistic limit. |
warning/0003/quant-ph0003105.html | ar5iv | text | # Quantum Fluctuations of a Single Trapped Atom: Transient Rabi Oscillations and Magnetic Bistability
## 1 Introduction
Radiation matter interaction has been studied for a long time with atomic samples. It is known that density matrix theory is well suited to exhaustively describe the properties of fluorescence from macroscopic atomic ensembles, and that in this limit semiclassical theory gives excellent approximations, since phenomena related to the quantum nature of the light field are often hidden.
On the other hand, it has also been realized since more than two decades now that isolation of a single atom provides an opportunity to directly observe pure quantum properties of the interacting radiation-matter system. A celebrated example is the observation of photon antibunching, where initial experiments were carried out with extremely diluted atomic beams, and more precise investigations became possible when ion traps were employed to achieve long-term confinement of a single atomic particle .
For neutral atoms the confinement strength is reduced since trapping has to rely on forces derived from electric or magnetic dipole interaction. Therefore neutral atoms have only more recently become available for measurements at the microscopic level through the application of laser cooling techniques well established for macroscopic samples with many atoms. With laser light single neutral atoms can be trapped and observed for long times in analogy with ion traps.
One of the most powerful methods to observe quantum fluctuations in atom-radiation interaction is to measure photon correlations of the light emitted by a single atom. The quantum theoretical description of photodetection was put forward by Glauber in the early 60s, and very successfully applied to all appropriate experiments since. In the photon language the normalized intensity-intensity correlation function is the conditional probability to detect a second photon if a first one was detected a time $`\tau `$ before. This second order correlation function is defined by
$$g^{(2)}(\tau )=\frac{:\widehat{n}(t+\tau )\widehat{n}(t):}{\widehat{n}(t)\widehat{n}(t)},$$
(1)
where $`\widehat{n}(t)=\widehat{a}^{}(t)\widehat{a}(t)`$ is the photon number operator constructed from field operators $`\widehat{a}^{}`$, $`\widehat{a}`$, and where $`:\widehat{n}(t+\tau )\widehat{n}(t):`$ denotes normal ordering of field operators. At sufficiently long delays $`\tau `$ all possible correlations have decayed, and hence $`:\widehat{n}(t+\tau )\widehat{n}(t):\widehat{n}(t)\widehat{n}(t)`$. It is thus often informative to discuss the deviation of $`g^{(2)}(\tau )`$ from unity, or the quantity $`g^{(2)}(\tau )1`$ as a measure of the fluctuation strength of the system, called *contrast* in the following.
For auto correlations (1) all classical fields must obey $`g^{(2)}(0)1`$, while photon antibunching shows $`g^{(2)}(0)=0`$, making it a prime example of quantum field fluctuations. In an intuitive interpretation it is said that the first detected photon projects an atom in its ground state. From this initial state the atom then relaxes back to its equilibrium state on the time scale of its radiative decay. In this sense the observation of this fluctuation is ’measurement induced’.
In our work we are additionally interested in the case where different atomic transitions can be distinguished by their polarization properties. We therefore analyze cross correlations of orthogonal polarization states $`\beta `$ and $`\alpha `$ of the resonance fluorescence from a single atom,
$$g_{\alpha \beta }^{(2)}(\tau )=\frac{:\widehat{n}_\alpha (t+\tau )\widehat{n}_\beta (t):}{\widehat{n}_\alpha (t)\widehat{n}_\beta (t)}.$$
(2)
A single trapped Cesium atom shows very strong polarization correlations in its resonance fluorescence. While this effect is very pronounced for correlations between orthogonal circular polarizations of the detected photons, it vanishes for linearly polarized photons even in a light field with linear polarization at every place. It is the purpose of this work to show that the strong contrast in $`g_{\alpha \beta }^{(2)}(\tau )`$ that we observe for orthogonally circular polarization states in the fluorescence of a single trapped atom can be interpreted as a direct consequence of the atomic orientation or magnetization undergoing spontaneous or quantum fluctuations.
## 2 Photon antibunching and transient Rabi oscillations
All classical fields have auto correlations $`g^{(2)}(0)g^{(2)}(\tau )`$, and a value $`g^{(2)}(0)1<0`$ is classically forbidden (for classical fields one should replace all operators $`\widehat{n}(t)`$ in (1) by their classical counterparts, intensities $`I(t)`$). This enhanced probability to detect two photons simultaneously is called ’photon bunching’ and was observed as early as in 1956 from a usual thermal light source.
For a single-atom fluorescence, however, $`g^{(2)}(\tau )`$ vanishes identically for $`\tau =0`$ which is a reflection of the fact that one can find at most one photon in the field mode of interest and can never detect two photons simultaneously. This phenomenon is called ’photon antibunching’ and is regarded as an important manifestation of the quantum nature of light.
Since the emitted light field reflects the evolution of the atomic dipole moment the correlation function $`g^{(2)}(\tau )`$ visualizes the internal dynamics of the observed atom for $`\tau >0`$. The state of an excited atom evolves continuously in the absence of a measurement, but theory predicts a sudden projection to the ground state when a photon is detected.
This measurement ’triggers’ the atom to the initial conditions $`\rho _{gg}(0)=1`$ and $`\rho _{ee}(0)=0`$, where $`\rho _{gg}(t)`$ and $`\rho _{ee}(t)`$ represent the population of the ground and excited atomic state, respectively. At that instant the coherent evolution starts again from the values $`\rho _{gg}(0)=1`$ and $`\rho _{ee}(0)=0`$ and will be interrupted by the next spontaneous emission. The normalized probability for detecting a second spontaneously emitted photon is now proportional to the population of the excited atomic state $`\rho _{ee}`$ according to
$$g^{(2)}(\tau )=\rho _{ee}(\tau )/\rho _{ee}(\mathrm{}).$$
(3)
Since the emission times are random, after averaging over many evolution trajectories the measured $`g^{(2)}(\tau )`$ shows directly relaxation of the system back to the equilibrium state after its wave function has collapsed due to the first photon detection.
In fig. 1 we show $`g^{(2)}(\tau )`$ of resonance fluorescence from a single atom stored in a magneto-optical trap (MOT) with clearly observable photon antibunching and transient oscillations. As we already reported previously , the observed transient oscillations in the population of the excited state in Fig. 1 corresponding to coherent excitation and deexcitation cycles (Rabi oscillations) can be surprisingly well described by a simple model (solid line in Fig. 1) of a two-level atom (similar to $`F=0F=1`$ transition) in spite of the complicated multilevel structure of the Cesium atom and light interference pattern (see below). This observation suggests that due to optical pumping a trapped atom spends most of its time in the magnetic substate that interacts most strongly with the local field and is forced to behave, to a good approximation, like a two-level system. For $`\tau `$ larger than the life time of the excited state the correlations die out due to the fluctuations of the vacuum field.
## 3 Orientation dynamics of the atom revealed by polarization correlations
An atom trapped in a MOT and moving through the light interference pattern experiences various intensities and polarizations at different places. The polarization of the resonance fluorescence is determined by the atomic interaction with the local light field and changes on the time scale of atomic transport over an optical wavelength $`\lambda `$. Thus, in addition to correlations of the total intensity one expects also polarization effects, that is correlations $`g_{\alpha \beta }^{(2)}(\tau )`$ measured between any polarization components $`\alpha `$ and $`\beta `$ which should strongly depend on the atomic motion and the light-field topography.
The light field of the MOT is formed by three mutually orthogonal pairs of counterpropagating laser beams with $`\sigma ^+`$ and $`\sigma ^{}`$ polarization. A pair of two circularly polarized laser beams with the same handedness produces a local polarization that is linear everywhere with a direction of polarization that rotates a full turn every half wavelength. With two additional ’polarization screws’ for other directions along with the relative phases $`\varphi `$ and $`\psi `$ between these standing waves one obtains for the total electric field in a 3D MOT<sub>ϕψ</sub>
$$\stackrel{}{E}=(\mathrm{sin}kz+\mathrm{sin}kye^{i\psi })\widehat{x}+(\mathrm{cos}kz+\mathrm{cos}kxe^{i\varphi })\widehat{y}+(\mathrm{sin}kxe^{i\varphi }+\mathrm{cos}kye^{i\psi })\widehat{z}.$$
(4)
We have already reported on strong correlations between circularly polarized photons and vanishing correlations between linear polarization components observed in the resonance fluorescence of a single atom trapped in a standard MOT light field configuration . This result seems to be intuitive only for specific choices of time phases: for example, $`\varphi =90^{},\psi =0`$ yields an ’antiferromagnetic’ light-field structure with alternating right- and left-hand circular polarizations at points of deepest light shift potential, see . However, in a standard MOT the phases $`\varphi `$ and $`\psi `$ change randomly due to acoustic jitter and thermal drifts. Thus the trapping light field has no well defined polarization state and the correlations are averaged over all possible values of $`\varphi `$ and $`\psi `$.
Since the MOT light field topography (4) strongly depends on the relative time-phases of the three contributing standing waves, we have chosen a setup where $`\varphi `$ and $`\psi `$ are intrinsically stable . The concept uses a single standing wave which is multiply folded and brought into triple intersection with itself. The phases $`\varphi `$ and $`\psi `$ can be adjusted by means of Faraday rotators. Details of this approach have been published elsewhere .
If one models an atom by a ”classical emitter” ($`F=0F=1`$ transition or steady-state density matrix of a multi-level atom) then its induced dipole moment will be proportional to the local light field. This idea is often used for interpretation of polarization correlations in the fluorescence of a large number of laser-cooled atoms . As we will show this description fails completely in the case of a single atom.
The most interesting case occurs for $`\varphi =\psi =0`$. In this situation the three standing waves oscillate synchronously and thus the interference light field has a linear polarization at every point and lacks handednees completely. In this case the model of a classical emitter predicts strong correlations between orthogonal linear polarization components and relatively weak correlations between circular components in clear contradiction with experimental results (see below).
## 4 Experimental Setup
For polarization-sensitive correlation measurements we have trapped individual neutral atoms in a standard six-beam MOT with the only exception that the phases and hence the light field topography is fully controlled . At a quadrupole field gradient of 12.5 G/cm the storage volume extends over approximately 100 $`\mu `$m. In order to trap small, countable numbers of atoms the loading rate from the background atomic vapour into the trap is kept very low. This is achieved on the one hand by lowering the Cesium partial pressure to $`10^{15}`$ mbar (at a base pressure of $`510^{10}`$ mbar) and on the other hand by using trapping laser beams of diameter 4 mm only. The average number $`N`$ of trapped atoms (typically between 1 and 5 in this experiment) can be easily adjusted by variation of the Cesium pressure. Although $`N`$ also depends on the trapping laser intensity $`I`$ and detuning $`\delta `$ of the trapping laser from atomic resonance, we are able to observe trapping of individual atoms over a wide range of parameters: $`0.3I_0<I<3.6I_0`$ per laser beam and $`5.2\mathrm{\Gamma }<\delta <1.1\mathrm{\Gamma }`$. The natural linewidth and the saturation intensity of the cooling transition are $`\mathrm{\Gamma }=2\pi \times 5.2`$ MHz and $`I_0=1.1`$ mW/cm<sup>2</sup>, respectively.
The atomic resonance fluorescence is due to excitation by the trapping laser field only. Fluorescent light is collected from a $`5\%`$ solid angle by a lens and then splitted into orthogonal polarization states by means of polarizing optics (see Fig. 2) which also directs the corresponding light onto two avalanche photo diodes (APD). The APDs are operated in single photon counting mode and achieve a photon detection efficiency of 47 % at a dark count rate of 10 s<sup>-1</sup>. The average photon count rate for an individual atom lies in the range 3-10 kHz in our experiments, depending on laser intensity and detuning. Observation direction is in the xy-plane at 45 to the laser beams (the $`z`$-axis is the symmetry axis of the MOT quadrupole magnetic field).
Usually, a measurement of the cross correlation function along with the total intensity-intensity correlation provides complete information: the corresponding auto correlation can be inferred from the sum rules for orthogonal polarization components . For example for circular components one has $`g_{++}^{(2)}(\tau )+g_+^{(2)}(\tau )=2g^{(2)}(\tau )`$. A cross correlation measurement ($`+`$) can be carried out 4 times faster than the corresponding auto correlation measurement ($`++`$) where only one half of the total fluorescence is detected.
A computer registers the arrival times of all photons from the two APD channels with 100 ns time resolution and with 700 ns dead time in each channel. The entire experimental information accessible in the set-up is stored and thus can be processed afterwards by correlation analysis through numerical multi-stop procedures, which completely eliminate systematic errors such as photon pile-up introduced by single-stop methods traditionally used in Hanbury Brown & Twiss type experiments .
Atoms are randomly loaded from background vapour and randomly lost due to collisions with background gas. But since individual atom arrival and departure events are easily located within 1 ms, it is straightforward to determine the instantaneous number of atoms from the average count rate. Note that the number of trapped atoms fluctuates on the second time scale . This enables us to separate all data from a single experimental run into different classes with the number of trapped atoms as a parameter - the data for different atom numbers are therefore obtained under identical experimental conditions. It is also easily possible to distinguish correlations of the fluorescence of trapped individual atoms from uncorrelated background of detection events due to stray light or fluorescence from thermal, untrapped atoms. As a consequence all measured correlation functions can always be unequivocal normalized.
## 5 Experimental Results
Measured second order correlation functions for orthogonal polarization components in the fluorescence of a single atom in the MOT<sub>00</sub> are shown in fig. 3. Within our experimental uncertainties correlations are completely absent for linear polarization components of the fluorescence, in baffling contrast with the result for circularly polarized components.
For a single atom the circular correlation contrast reaches s values up to 62%. This very strong correlation typically relaxes within a few $`\mu `$s. Since photon antibunching and Rabi oscillations at ns time scales are not resolved in these measurements, we have assumed for a simple analysis the relaxation process of the polarization correlations to be exponential and extracted a single relaxation time constant $`\tau _r`$. The intensity-intensity auto correlation $`g^{(2)}(\tau )`$ also shows a contrast of about 30% due to intensity modulations of the MOT light field, but the relaxation time constant of about 0.6 $`\mu `$s in fig. 4 is significantly shorter than for the circular cross correlations .
Furthermore, we have experimentally verified that $`\tau _r`$ does not depend on the number of trapped atoms and that the contrast of the correlation function is proportional to the inverse number of atoms, $`N^1`$ (see fig. 5). Thus, as one would expect, we deal with a pure single-atom effect.
## 6 Discussion
In order to interpret the polarization properties of the resonance fluorescence of atoms driven by a light field with linear polarization we begin by considering an atom at rest. As this light field does not favor any of the two circular polarization states, one can assume that prior to the detection of the first photon the distribution of atomic magnetic sublevels is symmetric, $`m=0`$. If we suppose for simplicity that the local light field consists of equal parts of both orthogonal circular polarizations only, it is clear that the $`m=0`$ state is unstable . At the level of an individual particle this equilibrium state can be distorted by the observation of a single circularly polarized photon, which projects the atom into its ground state, breaks the symmetry of the Zeeman substate population (fig. 6), and creates an imbalance in the interaction strengths with both circular polarization components. The next absorption will be preferentially further enhance the asymmetry. The imbalance in the interaction strengths rapidly grows with $`m`$ leading to fast pumping into one of the outmost Zeeman states $`m=\pm F`$. The ratio of the interaction strengths in these stretched states reaches the value $`(2F+1)(F+1)`$ making them very stable for large $`F`$ even in the presence of the polarization component in the light field driving $`\mathrm{\Delta }m=0`$ transitions. The atom in this oriented state prefers to radiate into the same polarization state as the first detected photon, resulting in anticoincidences in the cross correlation for orthogonal circular polarizations.
Note that this effect is a specific feature of $`FF+1`$ transitions and does not occur for $`FF`$ or $`FF1`$ transitions. It can be regarded as bistability, since an emission of a circularly polarized photon makes the atom more likely to be pumped in the corresponding outmost Zeeman state. As both stretched states $`m=+F`$ and $`m=F`$ are equivalent in the presence of a linear polarized light field, this effect is principally unobservable in an atomic ensemble. However, if one adds some circular polarization component to the driving light field a spontaneous spin polarization of a macroscopic sample can be observed as recently demonstrated in by optically pumping on the $`F=3F=4`$ hyperfine component of the $`D_1`$-line of Cs.
Although the real 3D-situation in our experiment is much more complicated than the simple model presented above, our interpretation is furthermore supported by the following considerations: As noted in , the condition for magnetic bistability in a linearly polarized light field is given by $`|90^{}\beta |<45^{}`$, where $`\beta `$ is the angle between the magnetic field (quantization axis) and the light polarization. In the MOT<sub>00</sub> there are 8 intensity antinodes in a unit cell with directions of the local linear polarization at these points coinciding with the diagonals of the coordinate system $`(\pm 1,\pm 1,\pm 1)`$. It is easy to see that the bistability condition is fulfilled for the majority of points in the MOT magnetic quadrupole field $`B(x,y,2z)`$.
However, in our case the stretched states are not intrinsically stable. While radiation pressure forces are balanced for an aligned atom with $`m`$=0, they are unbalanced for an oriented atom since the local linearly polarized light field is created by (at least two) counterpropagating laser beams with orthogonal circular polarizations. The imbalance in the light forces created by atomic orientation thus causes acceleration, or heating which is again damped by the usual laser friction forces . Thus for our experiment we must acknowledge that the observation of a circularly polarized photon not only redefines atomic orientation but also its mechanical status: Internal and external atomic degrees of freedom are inextricably entangled.
The measured relaxation time constant of $`g_\mathrm{}r^{(2)}(\tau )`$ indeed depends strongly on the atom-light field interaction which also governs atomic motion in the trap. The interaction strength is measured by the light shift parameter $`\mathrm{\Lambda }`$ corresponding to the maximum energy shift of the atomic energy levels. Under our experimental parameters sub-Doppler cooling leads to atomic temperatures proportional to the light shift, thus $`T_{kin}\mathrm{\Lambda }`$ .
In fig.7 we show the measured dependence of the time constant $`\tau _r`$ as a function of $`\mathrm{\Lambda }`$. As expected we find good agreement of our experimental data with the functional relationship $`\tau _r1/\overline{v}1/\sqrt{T_{kin}}1/\sqrt{\mathrm{\Lambda }}`$. Here $`\overline{v}`$ denotes the average atomic velocity. This means that relaxation of the spontaneous atomic magnetization is determined by atomic motion through the light field.
We can carry out our analysis one step further if we assume that the characteristic length over which relaxation takes place is $`\tau _r\overline{v}`$ = $`\lambda /2`$, the spatial period of the MOT light field. It is then straightforward to evaluate characteristic kinetic temperatures from our correlation measurements to be in the range between 10-68 $`\mu `$K for the $`\mathrm{\Lambda }`$ range shown in fig. 7, in good agreement with previous measurements .
## 7 Summary
Photon correlations observed in the resonance fluorescence of a single atom provide a direct access to the internal atomic dynamics. We have shown two examples of resolved quantum dynamics of an isolated atom stored in a magneto-optical trap. Beyond observation of the well-known phenomenon of transient Rabi oscillations (usually connected with photon antibunching), we have observed fluctuations of the atomic magnetic orientation by measuring photon correlations between orthogonal polarization components.
Using a simple model we have given evidence for the following dynamical processes causing strong circular cross correlations in resonance fluorescence: Spontaneous emission of circularly polarized photons causes instantaneous orientation. Subsequent photons are preferentially absorbed and emitted with identical polarization. This memory effect leading to correlated absorption of photons with equal polarization and thus to increased momentum diffusion of an atom with a multi-level structure has been discussed in . In our experiments we can clearly isolate this effect by observation of an anticorrelation of circularly polarized photons successively emitted with opposite handedness.
Subsequent optical pumping induced by atomic motion in the light field causes relaxation of the orientation clearly seen in the photon correlations. In a sense we have seen the elementary sub-Doppler cooling and heating forces in a $`\sigma ^+\sigma ^{}`$-molasses at work.
We thank Svenja Knappe for providing some basic techniques in an early stage of the experiment. We are also pleased to acknowledge Frans E. van Dorsselaer and Gerhard Nienhuis for fruitful discussions and sharing their insight into elementary atomic processes. This work is supported by the Deutsche Forschungsgemeinschaft (DFG). |
warning/0003/cond-mat0003205.html | ar5iv | text | # Toy model for the mean-field theory of hard-sphere liquids
## I Introduction
The theory of classical liquids received recently an important stimulus from the the theory of structural glasses . In the pioneering series of papers by Kirkpatrick and Thirumalai the possible connection of structural-glass transition with the spin-glass transition in $`p`$-spin models was put forward. The analogy was then developed e. g. in the problem of minimally correlated sequences, which was shown to possess a glassy behavior without quenched randomness .
However, it would be desirable to put this analogy further. One of the difficulties occurring in structural glasses, when compared to spin glasses, comes from the absence of any kind of analytically solvable mean-field version of the model. In the case of spin glass, the role is played by the fully-connected Ising model, which is solvable very easily. The disordered version is the well-known Sherrington-Kikpatrick model , whose understanding is now very close to be complete .
On the contrary, no analytical solution of a “mean-field” liquid is known, as far as we know. It can even sound not very reasonable to speak about a mean-field liquid, because the relevant high-density phase is characterized by strong short-range correlations, which can hardly be replaced by an effective medium. So, the meaning of the mean field should be better specified. In our investigation, we will understand by mean-field the situation which occurs in very high dimension, $`D\mathrm{}`$. The purpose of the present work is to introduce a simple model of a liquid, which is analytically solvable in the limit of infinite dimension, at least in a certain well-defined range of densities.
## II simplified HNC equations
We consider a liquid composed of hard spheres with the diameter 1. There is only one independent state variable, which is the spatial density of particles $`\rho `$.
The configuration of the liquid is described by the radial pair distribution function $`g(r)=h(r)+1`$. In the hypernetted chain (HNC) approximation we have a closed set of equations for the correlation function $`h(r)`$
$`h(r)+1=\mathrm{exp}(W(r)\beta U(r))`$ (1)
(2)
$`\widehat{W}(p)={\displaystyle \frac{\rho \widehat{h}^2(p)}{1+\rho \widehat{h}(p)}}.`$ (3)
The potential is $`U(r)=0`$ for $`r>1`$ and $`U(r)=\mathrm{}`$ for $`r<1`$. These equations can be interpreted as conditions for minimization of the free energy functional
$`\left[h\right]=`$ (4)
$`\rho ^2{\displaystyle drr^{D1}((h(r)+1)(\mathrm{ln}(h(r)+1)1+U(r))+1)}`$ (5)
$`+{\displaystyle \frac{1}{(2\pi )^D}}{\displaystyle dpp^{D1}L_3(\rho \widehat{h}(p))}`$ (6)
with $`L_3(x)=\mathrm{ln}(1+x)+x\frac{x^2}{2}`$. The function $`L_3(x)`$ has the following behavior: $`L_3(x)\mathrm{}`$ for $`x1`$ and $`L_3(x)x^3/3`$ for $`x1`$.
Our main approximation will consist in replacement the function $`L_3(x)`$ by $`L_{\mathrm{}}(x)`$, where $`L_{\mathrm{}}=\mathrm{}`$ for $`x<1`$ and $`L_{\mathrm{}}=0`$ otherwise. The motivation for this approximation is that we suppose that the main efffect of $`L_3(x)`$ is to forbid the region, where $`\rho \widehat{h}(p)>1`$. Then, minimization of the free energy functional amounts to satisfying the conditions
$`\rho \widehat{h}(p)`$ $``$ $`1`$ (7)
$`h(r)`$ $``$ $`1`$ (8)
$`h(r)`$ $`=`$ $`1\mathrm{for}r<1`$ (9)
which are in fact the minimum physical requirements for any correlation function $`h(r)`$. In this sense we are building a “minimum” model of a liquid. In addition to the constraints (8) we require that the function $`h(r)`$ depends continuously on the density. The absence of a solution which would be continuous in density would be a signal of a phase transition. This is, however, not found in the present calculations.
In 3D we can compute the function $`h(r)`$ numerically by increasing slowly the density $`\rho `$ and adjusting iteratively the function $`h(r)`$ so as the conditions (8) are satisfied. The resulting pair distribution function $`g(r)=h(r)+1`$ is shown in the Fig. 1. The Fourier transform $`\widehat{h}(p)`$ for $`\rho =1.2`$ is shown in Fig. 2.
We can see that for densities up to about $`\rho =1`$ the pair distribution function agrees qualitatively with the well-known results of HNC approximation or numerical simulations (see ). However, at about $`\rho =1.2`$ the behavior changes. A gap opens between the principal peak at $`r=1`$ and the secondary peak at $`r2`$. the gap broadens with increased density and at about $`\rho 1.5`$ a second gap occurs around $`r2.2`$. We observed, that further compression leads to the occurrence of a third gap separating the peaks at $`r1.6`$ and $`r2`$. We expect that continued increase of the density results in increased number of gaps.
The presence of the gaps is an artifact of the approximation. In reality the values of $`g(r)`$ will not be strictly zero, but small.
From the value of the radial distribution function at $`r=1`$ the pressure can be computed and the resulting equation of state is shown in Fig. 3, together with the results obtained by solving the HNC equations (2) and the formula computed in the Percus-Yevick (PY) approximation . We can see that our model behaves qualitatively in the same way as the other approximations, even though quantitative agreement is poor. On the other hand, the equation of state of our model does not differ from either HNC or PY approximation more than these two approximations differ one from the other.
We can see from these results, that the present approach in 3D gives at least qualitatively sensible results. However, our aim is not to provide a new approximation for real three-dimensional liquids, but a model which describes reasonably well the qualitative features of a liquid and is soluble in the limit of infinite spatial dimension. This will be done in the next section.
## III Solution of the model in high dimension
In this section we will investigate a $`D`$-dimensional version of the model, with $`D=2N+3`$ and $`N\mathrm{}`$. The main quantity of interest is again the correlation function $`h(r)=g(r)1`$. We will suppose that it has the form $`h(r)=h_0(r)+\overline{h}(r)`$ where $`h_0(r)=\theta (1r)`$ and $`\overline{h}(r)=0`$ for $`r<1`$.
The pressure is directly related to the value of that $`r=1`$, more precisely to $`lim_{r1^+}h(r)=\overline{h}(1^+)`$ (see )
$$\frac{1}{kT}P=\rho +\frac{1}{2}V_D\rho ^2(1+\overline{h}(1^+)).$$
(10)
In the following, we will use rescaled quantities. We will use $`\overline{\rho }=\rho V_D`$ for the density, while for the pressure we use $`\overline{P}=\frac{V_DP}{kT}`$.
In order to check the conditions (8) we should know the properties of the Fourier transform in high dimension. The relevant formulae used in this section are given in the Appendix A.
In the zeroth approximation $`\widehat{h}(\widehat{p})=V_D\mathrm{\Psi }(\widehat{p})`$, which is correct as long as $`\rho <1/V_D`$. So, for $`\overline{\rho }<1`$ we have $`\overline{h}(r)=0`$ and the pressure is given by first virial correction,
$$\overline{P}=\overline{\rho }+\frac{1}{2}\overline{\rho }^2.$$
(11)
For $`\overline{\rho }>1`$ we rewrite $`h`$ in the form
$$h(r)=h_0^{}(r)+\overline{h}^{}(r)$$
(12)
where still $`\overline{h}^{}=\overline{h}(r)`$ for $`r>1`$, but $`\overline{h}^{}(r)`$ is continuous for all $`r`$. The Fourier transform of $`h_0^{}`$ is easy to compute, if we know the value $`h^{}(1^{})=1A`$, where $`A=\overline{h}(1^+)=\widehat{h}_1(1)`$. Then
$$\widehat{h_0^{}}(\widehat{p})=(1+A)V_D\mathrm{\Psi }_0(\widehat{p}).$$
(13)
In order to ensure $`\widehat{h}(\widehat{p})1/\rho `$ everywhere, we set
$$\widehat{\overline{h}^{}}(\widehat{p})=\theta _1\left((A+1)V_D\mathrm{\Psi }_0(\widehat{p})\frac{1}{\rho }\right).$$
(14)
(We denote $`\theta _1(x)=x`$ for $`x>0`$, $`\theta _1(x)=0`$ for $`x0`$.)
The relevant quantity is
$$\rho _1=\frac{\mathrm{ln}\overline{\rho }}{N}.$$
(15)
Indeed, one particle occupies space $`2^DV_D`$, so an absolute upper bound for the density is $`\rho _1<\mathrm{ln}4`$.
Let us suppose that $`\widehat{p}_c<1`$. This condition restrict the range of densities investigated to a certain interval, which will be found in what follows.
For $`\widehat{p}<1`$ the function $`\mathrm{\Psi }_0(\widehat{p})=\mathrm{exp}(N\varphi _0(\widehat{p}))`$ is monotonously decreasing and we have the following equation for $`\widehat{p}_c`$.
$$\varphi _0(\widehat{p}_c)=\rho _1\frac{\mathrm{ln}(A+1)}{N}$$
(16)
When computing the inverse Fourier transform of $`\widehat{\overline{h}^{}}(\widehat{p})`$ we need only the behavior around the point $`\widehat{p}_c`$, which is
$$\widehat{\overline{h}^{}}(\widehat{p})V_D\mathrm{\Psi }_0^{}(\widehat{p}_c)\widehat{p}_c(1\frac{\widehat{p}}{\widehat{p}_c})$$
(17)
and gives, using (47)
$$\widehat{h}_1(r)=\left(\frac{N}{2\pi }\right)\frac{V_D^2\widehat{p}_c^{D+1}(\mathrm{\Psi }_0^{}(\widehat{p}_c))(A+1)}{D+1}\mathrm{\Psi }_0(\widehat{p}_cr)$$
(18)
We obtain immediately the equation for $`A`$
$$A=\left(\frac{N}{2\pi }\right)^D\frac{N}{D+1}V_D^2\widehat{p}_c^{D+1}\varphi _0^{}(\widehat{p}_c))(A+1)\mathrm{\Psi }_0^2(\widehat{p}_c).$$
(19)
In the interval $`\widehat{p}_c<1`$ we have
$$\varphi _0^{}(\widehat{p}_c)=\frac{\widehat{p}_c}{1+\sqrt{1\widehat{p}_c^2}}<\widehat{p}_c.$$
(20)
Hence, if we suppose that a solution such that $`A<1`$ exists, we have
$$A\left(\frac{N}{2\pi }\right)^DV_D^2\widehat{p}_c^{D+2}\mathrm{\Psi }_0^2(\widehat{p}_c)\mathrm{exp}(2NK(\widehat{p}_c))$$
(21)
where
$$K(\widehat{p})=\mathrm{ln}(1+\sqrt{1\widehat{p}^2})\sqrt{1\widehat{p}^2}\mathrm{ln}\widehat{p}.$$
(22)
We have $`K(1)=0`$ and $`K(\widehat{p})=\sqrt{1\widehat{p}^2}/\widehat{p}<0`$, so $`K(\widehat{p})>0`$ for $`\widehat{p}<1`$. For fixed $`\widehat{p}_c<1`$ and $`N\mathrm{}`$ we have $`A1`$ and therefore we can neglect $`A`$ in the equations (16) and (19).
So, we can conclude that in the range of densities $`\rho _1<\rho _{1c}=\varphi _0(1)=1\mathrm{ln}(2)=0.3068\mathrm{}`$ the following equation for $`\widehat{p}_c`$ holds
$$\rho _1=\mathrm{ln}(1+\sqrt{1\widehat{p}_c^2})\sqrt{1\widehat{p}_c^2}+1\mathrm{ln}2.$$
(23)
The solution of the latter equation can be easily obtained in the form of power series. We show here only first several terms. The expansion up to order 16 is given in the Appendix B and the graph is shown in Fig. 4.
$$\widehat{p}_c^2=4\rho _12\rho _{1}^{}{}_{}{}^{2}\frac{2}{3}\rho _{1}^{}{}_{}{}^{3}\frac{5}{6}\rho _{1}^{}{}_{}{}^{4}O(\rho _{1}^{}{}_{}{}^{5}).$$
(24)
From the solution Eq. (23) we can compute the pressure. We write $`\overline{P}=\overline{\rho }+\frac{1}{2}\overline{\rho }^2+P_c`$ and re-scale the correction as $`P_c=\mathrm{exp}(NP_1)`$. We obtain
$$P_1=22\mathrm{ln}2+\mathrm{ln}\widehat{p}_c^2$$
(25)
We can see that the density $`\rho _{1t}=\varphi _0\left(\frac{2}{\mathrm{e}}\right)=0.14676\mathrm{}`$ separates two regimes. For $`\rho _1<\rho _{1t}`$ the correction $`P_c`$ to the lowest virial becomes negligible for large $`N`$, while for $`\rho _1>\rho _{1t}`$ the correction diverges for $`N\mathrm{}`$. The graph of the function $`P_1(\rho _1)`$ is shown in Fig. 5.
For the correlation function in the interval $`r\widehat{p}_c<1`$ we have $`h(r)=\mathrm{exp}(Nh_1(r))`$, where
$`h_1(r)=`$ (26)
$`12\mathrm{ln}2+\mathrm{ln}\widehat{p}_c^2\rho _1+`$ (27)
$`+\sqrt{1\widehat{p}_c^2r^2}\mathrm{ln}(1+\sqrt{1\widehat{p}_c^2r^2}).`$ (28)
and for $`r\widehat{p}_c>1`$ we can use the following scaling $`\widehat{h}(r)=\mathrm{exp}(Nh_1(r))\mathrm{cos}(Nh_2(r))`$ where, using the expressions (42) and (43), we obtain
$$h_1(r)=12\mathrm{ln}2+\mathrm{ln}\widehat{p}_c\rho _1\mathrm{ln}r$$
(29)
$$h_2(r)=\sqrt{\widehat{p}_c^2r^21}\mathrm{arctan}\sqrt{\widehat{p}_c^2r^21}.$$
(30)
We can also see that $`|h_1(r)1`$ for $`\rho _1<\rho _{1c}`$, so that $`g(r)>0`$ for $`r>1`$ and the gaps in $`g(r)`$, discussed in the last section do not occur. However, when $`\rho _1`$ approaches $`\rho _{1c}`$ the absolute value of $`h_1(r)`$ can be of order $`O(1)`$ and a gap can appear at the density $`\rho _{1c}`$. The detailed investigation of this process and the behavior of the model for $`\rho _1>\rho _{1c}`$ remains an open question.
## IV Conclusions
We investigated a simple model for a hard-sphere liquid. By numerical solution in 3 dimensions, we found a qualitatively realistic behavior. The results for the equation of state are compatible with the hypernetted chain and Percus-Yevick approximations.
We solved the model analytically in the limit of large spatial dimension. We found that two scales of density and pressure appear, which corresponds to two regimes of density. For $`\overline{\rho }<1`$ the equation of state is given by first virial correction (11), while for $`\overline{\rho }>1`$ the quantity relevant to the further virial corrections is $`\rho _1=\mathrm{ln}\overline{\rho }/N`$ and the pressure correction itself scale as $`P_1=\mathrm{ln}P_c/N`$ (25). We have found the solution in the interval $`0<0\rho _1<\rho _{1c}=0.3068\mathrm{}`$. Two regimes are present within this interval. For $`\rho _1<\rho _{1t}=0.14676\mathrm{}`$ the correction $`P_c`$ vanishes for large $`N`$, while for $`\rho _1<\rho _{1t}`$ it diverges for large $`N`$.
It should be expected, that the presence of presence of gaps in $`g(r)`$ will lead to qualitatively different behavior for densities higher than $`\rho _{1c}`$.
###### Acknowledgements.
One of us (F.S.) wishes to thank to the INFN section of Rome University “La Sapienza” for financial support and kind hospitality. This work was supported by the project No. 202/00/1187 of the Grant Agency of the Czech Republic.
## Appendix A
Here we derive the formula for the Fourier transform in high dimension.
Fourier transform in $`D=2N+3`$ dimensions is ($`x`$ and $`k`$ are $`D`$-dimensional vectors)
$$\widehat{f}_D(k)=\mathrm{d}^Dx\mathrm{e}^{\mathrm{i}xk}f_D(x)$$
(31)
$$f_D(x)=(2\pi )^D\mathrm{d}^Dk\mathrm{e}^{\mathrm{i}xk}\widehat{f}_D(k).$$
(32)
We suppose that the functions depend only on the radial coordinate, $`f(r)=f_D(x)`$ for $`r=|x|`$ and $`\widehat{f}(p)=\widehat{f}_D(k)`$ for $`p=|k|`$. After rescaling $`\widehat{p}=p/N`$ we will finally have
$`\widehat{f}(\widehat{p})=`$ (33)
$`C_N{\displaystyle _0^{\mathrm{}}}dr{\displaystyle _1^1}dz\left[r^{2(1+\frac{1}{N})}(1z^2)\mathrm{e}^{\mathrm{i}\widehat{p}rz}\right]^Nf(r)`$ (34)
where the constant $`C_N`$ is fixed by condition that for $`f(r)=\theta (1r)`$ we have $`\widehat{f}(0)=V_D`$ with $`V_D`$ volume of the $`D`$-dimensional sphere,
$$V_D=\frac{2\pi ^{D/2}}{D\mathrm{\Gamma }(\frac{D}{2})}\left(\frac{\mathrm{e}\pi }{N}\right)^N.$$
(35)
Similarly for the inverse Fourier transform we will have
$`f(r)=`$ (36)
$`\widehat{C}_N{\displaystyle _0^{\mathrm{}}}d\widehat{p}{\displaystyle _1^1}dz\left[\widehat{p}^{2(1+\frac{1}{N})}(1z^2)\mathrm{e}^{\mathrm{i}\widehat{p}rz}\right]^N\widehat{f}(\widehat{p})`$ (37)
with coefficient
$$\widehat{C}_N=C_N\left(\frac{N}{2\pi }\right)^D.$$
(38)
The calculation of the Fourier transform can be performed by the saddle-point method. The essential result is the Fourier transform of the surface of unit sphere, $`f(r)=\delta (r1)`$. We obtain $`\widehat{f}(\widehat{p})\mathrm{\Psi }(\widehat{p})`$ where
$$\mathrm{\Psi }(\widehat{p})=\mathrm{\Psi }_0(\widehat{p})=\mathrm{exp}(N\varphi _0(\widehat{p}))$$
(39)
for $`\widehat{p}<1`$ and
$$\mathrm{\Psi }(\widehat{p})=\mathrm{\Psi }_1(\widehat{p})=\mathrm{exp}(N\varphi _1(\widehat{p}))\mathrm{cos}(N\varphi _2(\widehat{p}))$$
(40)
for $`\widehat{p}>1`$.
The explicit form of the functions $`\varphi _0,\varphi _1,\varphi _2`$ is given by
$`\varphi _0(\widehat{p})=1\mathrm{ln}2+\mathrm{ln}(1+\sqrt{1\widehat{p}^2})\sqrt{1\widehat{p}^2}`$ (41)
$`\varphi _1(\widehat{p})=1\mathrm{ln}2+\mathrm{ln}\widehat{p}`$ (42)
$`\varphi _2(\widehat{p})=\sqrt{\widehat{p}^21}\mathrm{arctan}\sqrt{\widehat{p}^21}.`$ (43)
Note that $`\mathrm{\Psi }(0)=1`$.
From here we can deduce the following Fourier transforms ($`\theta (x)=1`$ for $`x>0`$ and $`\theta (x)=0`$ for $`x<0`$). For $`f(r)=A\theta (r_0r)`$:
$$\widehat{f}(\widehat{p})=AV_Dr_0^D\mathrm{\Psi }(\widehat{p}r_0).$$
(44)
For $`f(r)=A(1r/r_0)\theta (r_0r)`$:
$$\widehat{f}(\widehat{p})=\frac{AV_Dr_0^D}{D+1}\mathrm{\Psi }_0(\widehat{p}r_0).$$
(45)
Because the inverse Fourier transform has the same form and differs only in the factor $`\widehat{C}_N`$ instead of $`C_N`$, we can also immediately write for $`\widehat{f}(\widehat{p})=A\theta (\widehat{p}_c\widehat{p})`$:
$$f(r)=\left(\frac{N}{2\pi }\right)^DAV_D\widehat{p}_c^D\mathrm{\Psi }(\widehat{p}_cr)$$
(46)
and for $`\widehat{f}(\widehat{p})=A(1\widehat{p}/\widehat{p}_c)\theta (\widehat{p}_c\widehat{p})`$
$$f(r)=\left(\frac{N}{2\pi }\right)^D\frac{AV_D\widehat{p}_c^D}{D+1}\mathrm{\Psi }(\widehat{p}_cr).$$
(47)
## Appendix B
Using Maple V we get the following expansion for the solution of equation (23).
$`\widehat{p}_c^2=`$ (48)
$`4\rho _12\rho _{1}^{}{}_{}{}^{2}2/3\rho _{1}^{}{}_{}{}^{3}5/6\rho _{1}^{}{}_{}{}^{4}{\displaystyle \frac{41}{30}}\rho _{1}^{}{}_{}{}^{5}{\displaystyle \frac{469}{180}}\rho _{1}^{}{}_{}{}^{6}`$ (49)
(50)
$`{\displaystyle \frac{6889}{1260}}\rho _{1}^{}{}_{}{}^{7}{\displaystyle \frac{24721}{2016}}\rho _{1}^{}{}_{}{}^{8}{\displaystyle \frac{2620169}{90720}}\rho _{1}^{}{}_{}{}^{9}{\displaystyle \frac{64074901}{907200}}\rho _{1}^{}{}_{}{}^{10}`$ (51)
(52)
$`{\displaystyle \frac{1775623081}{9979200}}\rho _{1}^{}{}_{}{}^{11}{\displaystyle \frac{1571135527}{3421440}}\rho _{1}^{}{}_{}{}^{12}`$ (53)
(54)
$`{\displaystyle \frac{1882140936521}{1556755200}}\rho _{1}^{}{}_{}{}^{13}{\displaystyle \frac{70552399533589}{21794572800}}\rho _{1}^{}{}_{}{}^{14}`$ (55)
(56)
$`{\displaystyle \frac{2874543652787689}{326918592000}}\rho _{1}^{}{}_{}{}^{15}{\displaystyle \frac{25296960472510609}{1046139494400}}\rho _{1}^{}{}_{}{}^{16}.`$ (57) |
warning/0003/cond-mat0003371.html | ar5iv | text | # Unquenched large orbital magnetic moment in NiO
\[
## Abstract
Magnetic properties of NiO are investigated by incorporating the spin-orbit interaction in the LSDA + $`U`$ scheme. It is found that the large part of orbital moment remains unquenched in NiO. The orbital moment contributes about $`\mu _L=0.29\mu _B`$ to the total magnetic moment of $`M=1.93\mu _B`$, as leads to the orbital-to-spin angular momentum ratio of $`L/S=0.36`$. The theoretical values are in good agreement with recent magnetic X-ray scattering measurements.
\]
Electronic and magnetic structures of late $`3d`$ transition metal (TM) mono-oxides (MnO, FeO, CoO, and NiO) have been extensively investigated over the last decades . Conventional band theory with the local spin density approximation (LSDA) fails to describe the electronic structures of the compounds. In the LSDA calculations, the energy gaps of MnO and NiO are underestimated. Even worse, FeO and CoO are predicted to be metallic which are, on the contrary, large gap insulators in nature. The main problem in the LSDA is the use of mean field-type exchange-correlation functional which is improper to describe localized $`3d`$ electrons.
There have been several theoretical efforts to cure the deficiencies in the LSDA, for example, the self-interaction correction (SIC) scheme , the GW approximation (GWA) , and the LSDA + $`U`$ method . The SIC-LSDA is in the line of the extended LSDA by removing unphysical electron interaction with itself. In the GWA, the quasi-particle energy is obtained through the self-energy calculation to the lowest order in the screened Coulomb interaction $`W`$. The GWA method applied to NiO gives a rather good description of the energy gap size . Computational load, however, is very heavy in the GWA. The LSDA + $`U`$ method overcomes the failure of the LSDA by incorporating the on-site Coulomb correlation $`U`$ of the multiband Hubbard model-like. In this method, localized $`3d`$ electrons are treated separately from delocalized $`sp`$ electrons. As a result, all the $`3d`$ TM mono-oxides in the LSDA + $`U`$ are obtained as insulators with well developed energy gaps which are comparable to experimental values . In this way, the energy gap problem in $`3d`$ TM mono-oxides is considered to be solved by various calculational methods.
Due to the outermost characteristics of TM $`3d`$ electrons, atomic $`3d`$ orbitals are greatly deformed in solids by the crystal field and/or band hybridization effects. Hence, the orbital moment of $`3d`$ TM ion is usually quenched in solids, because it originates from atomic nature of involved atomic elements. For example, all the $`3d`$ ferromagnetic transition metals of Fe, Co, and Ni show negligible orbital magnetic moments in the range of $`\mu _L0.1\mu _B`$ . It is, however, expected that the strong Coulomb correlation in $`3d`$ TM mono-oxides preserves the orbital moment of localized $`3d`$ electrons by reducing the ligand crystal field effects at metal ion sites.
In fact, for CoO, it is easily conceived that the orbital moment is only partially quenched because the measured magnetic moment $`M=3.4\mu _B`$ simply exceeds the spin magnetic moment alone and the minority $`t_{2g}`$ band is occupied only two-thirds of its available states. Therefore, the existence of orbital moment in CoO has been stressed many times . It is also shown that the LSDA + $`U`$ method gives a good description of magnetic structure of CoO with a large orbital moment of $`\mu _L1\mu _B`$ . On the other hand, in the case of NiO, measured magnetic moments are in the range of $`M=1.772.2\pm 0.2\mu _B`$ , which are comparable to the spin only moment of isolated Ni<sup>2+</sup> ion. Therefore, the magnetic moment in NiO has been fully attributed to the spin moment and the orbital moment is expected to be completely quenched.
However, recent magnetic X-ray scattering measurement indicates that the orbital-to-spin angular momentum ratio in NiO is as large as $`L/S=0.34`$, far from fully quenched orbital moment. In the nonresonant magnetic X-scattering, the separation of spin and orbital moment is possible, because the spin and orbital moment densities have different geometrical prefactors in the scattering cross section that can be adjusted by changing either the scattering geometry or the X-ray polarization. In Ref. , the orbital moment in NiO is extracted by the polarization analysis of nonresonant magnetic-scattering intensities. This method has evidenced a large contribution of the orbital moment to the total magnetic moment. They also found that the spin and orbital moments in NiO are collinear.
Magnetic moment is one of the basic ground state quantities which should be provided by an appropriate band method. If the orbital moment is so large in NiO, all the previous attempts which tried to directly compare calculated spin moments with experimental moments are under a mistake. The linearized muffin-tin orbital (LMTO) calculation for NiO with an orbital-polarization (OP) correction (LSDA + OP) in a crystal-field basis yields the spin and orbital magnetic moments of $`\mu _S=1.43\mu _B`$ and $`\mu _L=0.12\mu _B`$, respectively . Although the total magnetic moment is significantly improved to $`M=1.55\mu _B`$ with the OP correction, as compared to $`M=1.23\mu _B`$ from the LSDA, it is still smaller than the experimental measurements and the ratio of $`L/S=0.17`$ is only half of the magnetic X-ray scattering value . The SIC-LSDA method yields a large orbital moment of $`\mu _L=0.27\mu _B`$, which, however, was suspected as a methodological artifact by other authors . In more recent study, the orbital moment is simply neglected .
To determine the size of orbital moment in NiO, we have performed the LSDA + $`U`$ calculations using the LMTO band method within the atomic sphere approximation (ASA). The incorporation of the spin-orbit coupling into the LSDA + $`U`$ method is known to give right orbital polarization in strongly correlated electron systems . The LSDA + $`U`$ Hamiltonian is given by
$$_{\mathrm{L}SDA+U}=_{\mathrm{L}SDA}_{\mathrm{d}c}+_U$$
(1)
where the first term in the right hand side is the LSDA Hamiltonian and the second term is the double counting correction for the third term $`_U`$. With the Coulomb interaction $`U`$ and exchange interaction $`J`$ parameters, one can write $`_{\mathrm{d}c}`$ and $`_U`$, respectively, as
$`_{\mathrm{d}c}`$ $`=`$ $`{\displaystyle \frac{1}{2}}UN(N1){\displaystyle \frac{1}{2}}J{\displaystyle \underset{\sigma }{}}N^\sigma (N^\sigma 1),`$ (2)
$`_U`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\{m\},\sigma }{}}V(mm^{};m^{\prime \prime }m^{\prime \prime \prime })n_{mm^{\prime \prime }}^\sigma n_{m^{}m^{\prime \prime \prime }}^\sigma `$ (3)
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\{m\},\sigma }{}}\left[V(mm^{};m^{\prime \prime }m^{\prime \prime \prime })V(mm^{};m^{\prime \prime \prime }m^{\prime \prime })\right]`$ (5)
$`\times n_{mm^{\prime \prime }}^\sigma n_{m^{}m^{\prime \prime \prime }}^\sigma `$
where $`n_{mm^{}}^\sigma `$ is the $`d`$ occupation number matrix of spin $`\sigma `$ and $`N^\sigma =\mathrm{T}r(n_{mm^{}}^\sigma )`$, $`N=N^++N^{}`$. We relate the screened Coulomb interaction $`V(mm^{};m^{\prime \prime }m^{\prime \prime \prime })`$ with the Slater integral $`F^k`$;
$`V(mm^{};m^{\prime \prime }m^{\prime \prime \prime })={\displaystyle \underset{k=0}{\overset{2l}{}}}c^k(lm,lm^{\prime \prime })c^k(lm^{},lm^{\prime \prime \prime })F^k,`$ (6)
where $`c^k(lm,lm^{})`$ is a Gaunt coefficient. For $`3d`$ electrons, three Slater integrals of $`F^0,F^2`$, and $`F^4`$ are involved in the calculation. Among those, the ratio of $`F^4/F^2`$ is known to be constant $`0.625`$ for most $`3d`$ TM atoms . Hence, the actual number of parameters is reduced from three $`F^k`$’s to two of $`U`$ and $`J`$, which are given by $`U=F^0`$ and $`J=(F^2+F^4)/14`$, respectively. To determine the orbital moment, the spin-orbit coupling is simultaneously included in the self-consistent variational loop . We have assumed the antiferromagnetic ordering state of type-II, and the experimental lattice constant of $`a=7.893`$ a.u. is used.
In Fig. 1, magnetic moment behavior is shown with varying the lattice constant in the LSDA. Both the spin and orbital magnetic moments increase monotonically with increasing the lattice constant. This feature is understandable in view of that the crystal field strength at Ni sites and the hybridization between Ni $`3d`$ and O $`2p`$ bands are reduced with volume expansion. At the experimental lattice constant of $`a=7.893`$ a.u., the spin and orbital magnetic moments are obtained as $`\mu _S=1.12\mu _B`$ and $`\mu _L=0.15\mu _B`$, respectively, which are consistent with existing results . Both the spin and orbital magnetic moments are larger in NiO than in fcc Ni which has $`\mu _S=0.63\mu _B`$ and $`\mu _L=0.06\mu _B`$. This suggests that $`3d`$ electrons are more localized in NiO than in Ni metal. Although the ratio of $`L/S=0.27`$ in the LSDA is only slightly smaller than $`L/S=0.34`$ in the experiment (see Table. I), the spin and orbital polarizations in the LSDA are not large enough because of underestimation of the Coulomb correlation between $`3d`$ electrons.
Figure 2 shows the Coulomb correlation effects on the magnetic moments obtained by the LSDA + $`U`$ method. We have used the exchange parameter of $`J=0.89`$ eV which is comparable to literature value . Once the strong Coulomb interaction is introduced between Ni $`3d`$ electrons, the magnetic moments increase substantially. The role of the Coulomb interaction is significant to localize Ni $`3d`$ electrons. With $`U=8.0`$ eV found in Ref. and Ref. , we have obtained the spin and orbital magnetic moments of $`\mu _S=1.64\mu _B`$ and $`\mu _L=0.29\mu _B`$, respectively. The total magnetic moment $`M=1.93\mu _B`$ and the ratio $`L/S=0.36`$ for $`U=8.0`$ eV are in good agreement with the experimental data. The change of the Coulomb parameter $`U`$ by $`1.0`$ eV results in increments of both the magnetic moments and the ratio $`L/S`$ by $`\mathrm{\Delta }\mu _S\mathrm{\Delta }\mu _L0.01\mu _B`$ and $`\mathrm{\Delta }(L/S)0.01`$, respectively. The overall total magnetic moment difference and the ratio $`L/S`$ difference are only $`\mathrm{\Delta }M=0.12\mu _B`$ and $`\mathrm{\Delta }(L/S)=0.05`$ inbetween $`U=6.0`$ and $`10`$ eV. Thus, all the magnetic moments and the ratio $`L/S`$ are rather insensitive to the Coulomb interaction $`U`$ within the employed parameter range.
Calculated magnetic moments are summarized in Table. I in comparison with experimental data and previous results by other calculational methods. As mentioned above, the sizes of magnetic moments are underestimated in the LSDA. The total magnetic moment of $`M=1.27\mu _B`$ in the LSDA is much smaller than $`M=1.772.2\pm 0.2\mu _B`$ in experiments . In the LSDA + OP, the spin moment is significantly improved to $`\mu _S=1.43\mu _B`$ , which, however, is still smaller than the experimental values. More serious is the decrease of orbital moment in the LSDA + OP, which unfavorably makes the consistency between the theoretical and experimental $`L/S`$ ratio become worse. Both the SIC-LSDA ($`M=1.80\mu _B`$) and the LSDA + $`U`$ ($`M=1.93\mu _B`$) give values in agreement with experimental data. But it can be concluded that the LSDA + $`U`$ results with $`U=8.0`$ eV and $`J=0.89`$ eV, among various calculations, are especially in best agreement with experimental measurements. The ratio of $`L/S=0.36`$ in the LSDA + $`U`$ is also consistent with $`L/S=0.34`$ in the magnetic X-ray scattering measurement .
In conclusion, we have found using the LSDA + $`U`$ calculations that the orbital moment in NiO is not fully quenched, which is as large as $`\mu _L=0.29\mu _B`$. Both the total magnetic moment of $`M=1.93\mu _B`$ and the orbital-to-spin angular momentum ratio of $`L/S=0.36`$ for $`U=8.0`$ eV are close to experimental values. The Coulomb correlation and the spin-orbit coupling are crucial to get right magnetic polarization in NiO.
Acknowledgements$``$ The authors would like to thank K.B. Lee for helpful discussions. This work was supported by the KOSEF (1999-2-114-002-5) and in part by the Korean MOST-FORT fund. |
warning/0003/hep-th0003138.html | ar5iv | text | # Dirac Strings and Monopoles in the Continuum Limit of SU(2) Lattice Gauge Theory. ITEP-TH-7/00 MPI-PhT/2000-12
## Introduction
While perturbative Yang-Mills theories appear to be understood beyond any doubt, non-perturbative physics is much more challenging at the moment. Moreover, the main source of knowledge in the non-perturbative domain is the lattice gauge theories. In particular, there exist rich data supporting the idea of the quark confinement through the magnetic monopole condensation (for review, see, e.g. ).
Any analytical treatment of magnetic monopoles in the continuum limit represents apparent difficulties because of singularities in the gauge potential $`A_\mu `$. Indeed, such singularities are displayed already by the original Dirac monopole:
$$A_\mu dx_\mu =\frac{1}{2}(1+\mathrm{cos}\theta )d\phi ,$$
(1)
or in the component form in the spherical coordinates:
$$A_\theta =A_r=0,A_\phi =\frac{1}{2}\frac{(1+\mathrm{cos}\theta )}{r\mathrm{sin}\theta }.$$
(2)
The singularity along the line $`\theta =0`$ represents the Dirac string, while the singularity at $`r0`$ corresponds to a singular magnetic filed, $`𝐇𝐫/r^3`$. In non-Abelian theories with Higgs mechanism the singularities are resolved and there exists the famous ’t Hooft-Polyakov solution with finite energy. In a particular gauge, the corresponding potential is given by
$$A_i^a=f(r)\frac{\epsilon _{aik}r_k}{r^2}$$
(3)
where $`a`$ is the color index, $`a=1,2,3`$ and $`f(r)0`$ as $`r0`$ while $`f(r)1`$ as $`r\mathrm{}`$.
In pure gauge theories, there are no monopole solutions with finite energy. To reconcile this with observation of monopoles on the lattice, one considers dual gauge theories which serve as infrared limit of QCD . In its simplest version, the theory is build on an octet of dual gluons and three octets of scalar (Higgs) fields. In this paper, we would stick to consideration of monopoles within the fundamental QCD. The reason is that the monopoles on the lattice are defined beginning from elementary cubes, i.e. at smallest distances available. Our guiding principle is to reexamine the continuum limit by confronting the treatment of the monopole-associated singularities on the lattice and in the continuum.
In the lattice formulation, the singularities due to the Dirac string and at $`r0`$ are treated differently. As was emphasized first by Polyakov , the Dirac strings are allowed, i.e. cost no action in the lattice compact U(1) theory. As for the $`r0`$ singularity, it introduces in this case a physical divergence in the action. The suppression due to this divergence is overcome, however, by the entropy factor when the coupling constant $`g`$, included into the definition of $`A_\mu `$ above, is of order of unity.
In the non-Abelian gauge models the relation of monopoles to the action is much more obscure, as far as analytical results are concerned. Moreover, one of the most important steps in introducing monopoles is a pure topological definition which makes no reference to the associated non-Abelian action . In this formulation, monopoles are related to topology of gauge fixing. Namely, if the gauge is fixed (up to U(1) rotations) by directing a color vector $`h^a`$ in, say, the third direction, then the fixation fails at the points where all the components $`h^a`$ vanish. Moreover, one can prove that such points belong to monopole trajectories. The function of $`h^a`$ can be played by any vector, for example, by a particular Lorenz component of the gluonic field-strength tensor, say, $`F_{12}^a`$. Vanishing of $`h^a`$ has no direct effect on the action. Other Abelian projections revealing monopoles are also known, the most famous one seems to be the Maximal Abelian gauge (for review and references see ).
Monopoles which condense in the confining phase have magnetic charges $`|Q_M|=2`$, the same as the ’t Hooft-Polyakov monopole (3). In the corresponding U(1) projection the associated Dirac string does not introduce infinities because of the compactness of the U(1) subgroups of SU(2), see the discussion above. On the other hand the Dirac strings associated with $`|Q_M|=1`$ monopoles are not allowed in the QCD vacuum since in the continuum limit they have infinite energy. However, one can introduce $`|Q_M|=1`$ monopoles as external objects via the ’t Hooft loop operator .
In this paper, we consider magnetic monopoles in the continuum provided that the continuum is understood as the limiting case of lattice theories. First, we generalize the treatment of Dirac strings within the lattice compact QED to the non-Abelian case. As expected, the lattice formulation of the non-Abelian theories corresponds to non-observability of the Dirac strings, defined in a particular way. To substitute for their effect in the continuum, one allows for certain singular potentials. Thus, we argue that the standard continuum formulation is to be modified in a certain way to allow for the Dirac strings.
It is amusing that once the Dirac strings are admitted into the continuum limit the $`|Q_M|=2`$ monopoles cost no action either. Namely, we construct an explicit solution with zero action for a Dirac strings with open ends. In this respect the non-Abelian theories differ radically from their Abelian counterpart where the end points of the Dirac strings represent monopoles (1) with divergent action. It might worth emphasizing that the Abelian part of the fields in the no-action solution does correspond to the standard Abelian monopoles and it is the commutator term in the field-strength tensor which allows to nullify the non-Abelian action. This is in the correspondence with the instability of a single $`|Q_M|=2`$ monopole with nonzero action in the non-Abelian pure gauge theory which is known since long .
The explicit monopole-pair solution with no action mentioned above is obtained in empty, or perturbative vacuum. We check that quantum fluctuations around this zero-action solution do not distinguish it from the perturbative vacuum either. Therefore the modified continuum version corresponding to the limit of the lattice theories brings no change in the perturbative domain as compared with the standard Lagrangian theory.
We then imitate non-perturbative vacuum of QCD by introducing nonsingular background fields, $`F_{\mu \nu }^{soft}\mathrm{\Lambda }_{QCD}^2`$. Then there still exist Dirac strings with zero action whose color orientation is aligned with that of the background field. On the other hand, introduction of monopoles a la ’t Hooft (see above) is related to some singular gauge transformations with their own color orientation. As a result, monopoles in the physical vacuum are associated, generally speaking, with an action of order $`L\mathrm{\Lambda }_{QCD}`$ where $`L`$ is the length of the monopole world trajectory.
Finally, the same techniques as used to construct invisible Dirac strings in the continuum limit produce a continuum analog for the ’t Hooft loop operator. It shares the basic properties of the ’t Hooft loop operator and allows to formulate new predictions for the intermonopole potential. At short distances the ’t Hooft loop describes the Coulomb-like interaction of the monopoles with $`|Q_M|=1`$, Ref. . We fix the coefficient at front of this Coulombic term. At larger distances the $`|Q_M|=1`$ monopoles, introduced via the ’t Hooft loop interact with the $`|Q_M|=2`$ monopoles of the medium. We describe this interaction within the effective Abelian Higgs model (for review and references see ), which uniquely fixes the Yukawa-like behavior of the intermonopole potential. We include also consideration of the ’t Hooft loop at high temperatures where the Debye screening becomes essential.
The outline of the paper is as follows. In Section 1 we show that symmetries of the lattice and standard continuum actions of gluodynamics are different. We propose the modified continuum action which allows for the Dirac strings. In Sections 2, 3 monopole configurations within the new approach are considered. In Section 4 we introduce the ’t Hooft loop operator in the continuum. In Section 5 the predictions for the ’t Hooft loop are formulated. Our conclusions are summarized in the last section.
## 1 Dirac Strings in SU(2) Gauge Theory.
As is mentioned in the Introduction, the lattice formulation of the compact photodynamics gives a version of the U(1) gauge theory with unobservable Dirac strings. In this section we develop a generalization of this construction to the case of SU(2) gauge model.
The general one–plaquette action of SU(2) lattice gauge theory (LGT) can be represented as:
$$S_{lat}(U)=\frac{4}{g^2}\underset{p}{}S_P(1\frac{1}{2}TrU[p]),$$
(4)
where $`g`$ is the bare coupling, $`p`$ is the boundary of an elementary plaquette $`p`$, the sum is taken over all $`p`$, $`U[p]`$ is the ordered product of link variables $`U_l`$ along $`p`$. To have the correct naive continuum limit the function $`S_P`$ should obey the condition $`lim_{x0}S_P(x)=x+\mathrm{}`$. In particular, if $`S_P(x)=x`$ then (4) is the standard Wilson action. The exponent of the lattice field strength tensor $`F_p`$ defines $`U[p]`$:
$$U[p]=e^{i\widehat{F}_p}=\mathrm{cos}[\frac{1}{2}|F_p|]+i\tau ^an_p^a\mathrm{sin}[\frac{1}{2}|F_p|],$$
(5)
where $`\widehat{F}=F^a\tau ^a/2`$, $`|F|=\sqrt{F^aF^a}`$ and we define $`n_p^a=F_p^a/|F_p|`$ for $`|F_p|0`$, $`n^a`$ is an arbitrary unit vector for $`|F_p|=0`$. Sometimes we also use the vector-like notations $`\stackrel{}{F}`$ instead of $`F^a`$. The lattice action (4) depends only on $`\mathrm{cos}[\frac{1}{2}|F_p|]`$. Therefore the action of the SU(2) LGT possesses not only the usual gauge symmetry, but allows also for the gauge transformations which shift the field strength by $`4\pi k`$, $`|F_p||F_p|+4\pi k`$, $`kZ`$:
$$\begin{array}{ccc}\begin{array}{ccc}e^{i\widehat{F}_p}\hfill & =\hfill & \mathrm{exp}\{i|F_p|\widehat{n}_p\}=\hfill \\ & =\hfill & \mathrm{exp}\{i(|F_p|+4\pi )\widehat{n}_p\}=\mathrm{exp}\{i(F_p^a+4\pi n_p^a)\tau ^a/2\},\hfill \end{array}& & \end{array}$$
(6)
Thus the symmetry inherent to the lattice formulation can be represented as:
$$F_p^aF_p^a+4\pi n_p^a,\stackrel{}{F}_p\times \stackrel{}{n}_p=0,\stackrel{}{n}_p^2=1.$$
(7)
The symmetry (7) is absent in the conventional continuum action, $`(F_{\mu \nu }^a)^2d^4x`$ and therefore the continuum limit of SU(2) LGT is different from the commonly accepted SU(2) gluodynamics at least in this respect. Below we explore the consequences of Eq. (7) for the continuum theory.
In the continuum limit $`n_p^a`$ becomes a singular two-dimensional structure $`{}_{}{}^{}\mathrm{\Sigma }_{\mu \nu }^{a}=\frac{1}{2}\epsilon _{\mu \nu \lambda \rho }\mathrm{\Sigma }_{\lambda \rho }^a`$ which is a generalization of the Dirac strings in the compact electrodynamics and which transforms in the adjoint representation of the gauge group. Consider first a special class of the gauge potentials which may be gauge transformed to pure Abelian fields, $`A_\mu ^a=\delta ^{a,3}A_\mu `$. For such fields the action of the SU(2) gluodynamics coincides with the action of the compact U(1) gauge model, up to the ghost terms. Therefore in this gauge $`\mathrm{\Sigma }_{\mu \nu }^a=\delta ^{a,3}\mathrm{\Sigma }_{\mu \nu }`$, where $`\mathrm{\Sigma }_{\mu \nu }`$ is nothing else but the Dirac string:
$$\mathrm{\Sigma }_{\mu \nu }=d^2\sigma \sqrt{g}t_{\mu \nu }(\sigma )\delta ^{(4)}(x\stackrel{~}{x}(\sigma )),$$
(8)
with the world-sheet coordinates $`\stackrel{~}{x}(\sigma )`$ parameterized by $`\sigma _\alpha `$, $`\alpha =1,2`$:
$$t_{\mu \nu }(\sigma )=\frac{1}{\sqrt{g}}\epsilon ^{\alpha \beta }_\alpha \stackrel{~}{x}_\mu _\beta \stackrel{~}{x}_\nu ,t_{\mu \nu }^2=2,g(\sigma )=\mathrm{Det}[_\alpha \stackrel{~}{x}_\mu _\beta \stackrel{~}{x}_\mu ].$$
(9)
Thus for general gauge potentials
$$\mathrm{\Sigma }_{\mu \nu }^a=d^2\sigma \sqrt{g}t_{\mu \nu }^a(\sigma )\delta ^{(4)}(x\stackrel{~}{x}(\sigma )).$$
(10)
The second equality in (7) requires that
$$\stackrel{}{t}_{\mu \nu }(\sigma )\times {}_{}{}^{}\stackrel{}{F}_{\mu \nu }^{}(\stackrel{~}{x})=0,$$
(11)
where the continuum field strength tensor $`\widehat{F}_{\mu \nu }=_\mu \widehat{A}_\nu _\nu \widehat{A}_\mu i[\widehat{A}_\mu ,\widehat{A}_\nu ]`$. Eq. (11) determines the color structure of $`t_{\mu \nu }^a`$:
$$t_{\mu \nu }^a(\sigma )=t_{\mu \nu }(\sigma )n^a(\sigma ),n^a(\sigma )=(t{}_{}{}^{}F_{}^{a})\left[(t{}_{}{}^{}F_{}^{b})^2\right]^{1/2},$$
(12)
where $`(tF^a)t_{\mu \nu }(\sigma )F_{\mu \nu }^a(\stackrel{~}{x})`$ and $`n^a`$ is normalized as $`\stackrel{}{n}^2=1`$. On the set of points where $`(t{}_{}{}^{}F_{}^{a})=0`$ the direction of $`n^a(\sigma )`$ is arbitrary. Therefore, in the general case $`\mathrm{\Sigma }_{\mu \nu }^a`$ is given by
$`\mathrm{\Sigma }_{\mu \nu }^a={\displaystyle d^2\sigma \sqrt{g}t_{\mu \nu }(\sigma )n^a(\sigma )\delta ^{(4)}(x\stackrel{~}{x}(\sigma ))},`$
(13)
$`\stackrel{}{n}^{\mathrm{\hspace{0.33em}2}}(\sigma )=1,\stackrel{}{n}(\sigma )\times (t_{\mu \nu }(\sigma ){}_{}{}^{}\stackrel{}{F}_{\mu \nu }^{}(\stackrel{~}{x}))=0.`$
and the continuum analog of the lattice symmetry Eq. (6,7) is:
$$F_{\mu \nu }^aF_{\mu \nu }^a+4\pi {}_{}{}^{}\mathrm{\Sigma }_{\mu \nu }^{a},$$
(14)
Note that we do not claim that the only string-like singularities which may exist in the continuum limit of SU(2) LGT are of the type (13,14). Indeed, there are known examples of various Abelian gauges (see and references therein) in which Abelian monopoles and Dirac strings naturally arise. String singularities in these gauges are of the type (13), but their color orientation is different. Therefore the strings (13) are not the most general. Nevertheless, we claim that only the strings (13) produce no additional action. In other words the action of the SU(2) LGT in the continuum limit calculated with $`F_{\mu \nu }^a`$ and $`F_{\mu \nu }^a+4\pi {}_{}{}^{}\mathrm{\Sigma }_{\mu \nu }^{a}`$ is the same only if $`\mathrm{\Sigma }_{\mu \nu }^a`$ is given by (13). We shall come back to discuss this issue in Section 3.
The action of SU(2) gluodynamics which possesses the additional symmetry (14) can be formally represented as:
$$Z=𝒟A\mathrm{exp}\left\{S(F)\right\},$$
(15)
$$S(F)=\mathrm{log}𝒟\mathrm{\Sigma }\mathrm{exp}\left\{\frac{1}{4g^2}d^4x\left[F_{\mu \nu }^a+4\pi {}_{}{}^{}\mathrm{\Sigma }_{\mu \nu }^{a}\right]^2\right\},$$
(16)
where the integration is over all possible surfaces (13). The expressions (15,16) are only formal since, as we show in Section 2 it is impossible to separate rigorously the measure $`𝒟\mathrm{\Sigma }`$ from the gauge degrees of freedom in $`𝒟A`$. Nevertheless, the Eq. (15,16) is a good starting point for the analysis of the next section. Note that the action (16) is invariant under smooth SU(2) gauge transformations since vector $`n^a`$ transforms in the same way as $`F_{\mu \nu }^a`$ does. By construction, this action is also invariant under transformations (13,14) which correspond to the lattice symmetry relations (6,7).
Note also that for self-intersecting surface $`\mathrm{\Sigma }_{\mu \nu }`$, Eq. (13), the world-sheet vector field $`n^a(\sigma )`$ is generally multi-valued as function of $`\stackrel{~}{x}`$. Furthermore, for the non-orientable surfaces the field $`n^a(\sigma )`$ cannot be defined smoothly everywhere on $`\mathrm{\Sigma }`$. To avoid these complications we consider only the orientable surfaces without self-intersections. This reservation is specific for Dirac strings in the non-Abelian case.
## 2 Monopoles.
We proceed now to consider Dirac strings with open ends. The end points can be associated, as usual, with monopoles. If one follows only the Abelian-like part of the field strength tensors, these are standard Dirac monopoles. However, the full non-Abelian action is no longer bounded in terms of the Abelian magnetic field and we will present a zero-action solution for open Dirac strings. Therefore contrary to the Abelian models the open Dirac strings in SU(2) gluodynamics are the gauge copies of the vacuum $`A=0`$. Moreover, we show that this result is also valid at the one loop level and hence the Dirac strings (13) do not change the perturbation theory.
### 2.1 String Independence.
Consider the partition function (15,16) in case of a single surface $`\mathrm{\Sigma }_{\mu \nu }^a`$:
$$Z[\mathrm{\Sigma }]=𝒟A\mathrm{exp}\left\{\frac{1}{4g^2}d^4x\left[F_{\mu \nu }^a+4\pi q{}_{}{}^{}\mathrm{\Sigma }_{\mu \nu }^{a}\right]^2\right\},$$
(17)
where the constant $`q`$ is equal to unity in (15,16). Varying the gauge fields $`A`$ we get the classical equations of motion:
$$D_\nu \left(\widehat{F}_{\mu \nu }(A)+4\pi q{}_{}{}^{}\widehat{\mathrm{\Sigma }}_{\mu \nu }^{}\right)=0,$$
(18)
which should be supplemented by Bianchi identities:
$$D_\nu {}_{}{}^{}\widehat{F}_{\mu \nu }^{}=0.$$
(19)
Note that Eq. (18) is consistent with the covariant conservation of electric currents:
$`D_\mu D_\nu \widehat{F}_{\mu \nu }=4\pi qD_\mu D_\nu {}_{}{}^{}\widehat{\mathrm{\Sigma }}_{\mu \nu }^{}{\displaystyle }d^2\sigma _{\mu \nu }{}_{}{}^{}[D_\mu ,D_\nu ]\widehat{n}(\sigma )\delta ^{(4)}(x\stackrel{~}{x}(\sigma ))=0.`$
where the last equality is due to (13).
To appreciate the meaning of eq.(18) let us confine ourselves for the moment to the fields $`A_\mu ^a=Q^aA_\mu `$ with a constant color direction $`Q^a`$. Then $`n^aQ^a`$ and Eq. (18) becomes:
$$_\nu \left(_{[\mu }A_{\nu ]}\right)=4\pi q_\nu {}_{}{}^{}\mathrm{\Sigma }_{\mu \nu }^{}.$$
(20)
The solution of this equation in the Landau gauge,
$$A_\mu ^a=Q^a4\pi q\frac{1}{\mathrm{\Delta }}_\nu {}_{}{}^{}\mathrm{\Sigma }_{\mu \nu }^{},$$
(21)
corresponds to the gauge potential of an Abelian monopole current $`\mathrm{\Sigma }`$ embedded into the SU(2) group. Thus $`\mathrm{\Sigma }_{\mu \nu }`$ is the Dirac string worldsheet.
Let us show that the shape of $`\mathrm{\Sigma }`$ is irrelevant, that is the surface $`\mathrm{\Sigma }`$ can be shifted by a gauge transformation provided that the boundary $`\mathrm{\Sigma }`$ is fixed. Assuming that the orientable surface (13) has no self-intersections, we may write $`\mathrm{\Sigma }_{\mu \nu }^a=n^a\mathrm{\Sigma }_{\mu \nu }`$. Consider then a closed non self-intersecting surface $`𝒮`$ on which the vector field $`n^a(\sigma )`$ is a single valued function of $`\stackrel{~}{x}`$. We can define the field $`n^a(x)`$, $`\stackrel{}{n}^{\mathrm{\hspace{0.33em}2}}=1`$, in the whole space-time in such a way that
$$n^a(x)=n^a(\stackrel{~}{x})\text{for}x𝒮.$$
(22)
Note that the definition of the vector $`n^a(x)`$ is not unique, but this is irrelevant for our analysis.
Consider the following gauge transformation matrix:
$`\mathrm{\Omega }(𝒱_𝒮)`$ $`=`$ $`\mathrm{exp}\{i\alpha (𝒱_𝒮,x)\stackrel{}{n}(x)\stackrel{}{\tau }\},`$ (23)
$`\alpha (𝒱_𝒮,x)`$ $`=`$ $`2\pi q{\displaystyle \underset{\mathrm{}}{\overset{x}{}}}V_\mu 𝑑x_\mu ,V_\mu ={\displaystyle \underset{𝒱_𝒮}{}}\left({}_{}{}^{}d_{}^{3}\zeta \right)_\mu \delta ^{(4)}(x\zeta ),`$ (24)
where $`V_\mu `$ is a characteristic function of the volume $`𝒱_𝒮`$ bounded by the surface $`𝒮`$. The first integral in (24) is taken along any path $`C_x`$ connecting infinity with the point $`x`$. Under the general gauge transformation $`\mathrm{\Omega }`$ the field strength tensor transforms as:
$$\widehat{F}_{\mu \nu }(A^\mathrm{\Omega })=\mathrm{\Omega }^+\widehat{F}_{\mu \nu }(A)\mathrm{\Omega }+i\mathrm{\Omega }^+[_\mu ,_\nu ]\mathrm{\Omega }=\mathrm{\Omega }^+\widehat{F}_{\mu \nu }(A)\mathrm{\Omega }+\widehat{F}_{\mu \nu }(i\mathrm{\Omega }^+\mathrm{\Omega }).$$
(25)
Straightforward calculations show that for $`\mathrm{\Omega }`$ defined by Eq. (23)
$`F_{\mu \nu }^a(i\mathrm{\Omega }^+\mathrm{\Omega })`$ $`=`$ $`2n^a[_\mu ,_\nu ]\alpha `$
$`\left(\mathrm{sin}[2\alpha ]\delta ^{ac}+(1\mathrm{cos}[2\alpha ])\epsilon ^{abc}n^b\right)[_\mu ,_\nu ]n^c.`$
Note that the function $`\alpha (x)`$ takes only two values, $`0`$ and $`2\pi q`$. Therefore for integer or the half-integer valued charge $`q`$ we have:
$$F_{\mu \nu }^a(i\mathrm{\Omega }^+\mathrm{\Omega })=2n^a[_\mu ,_\nu ]\alpha =4\pi qn^a_{[\mu }V_{\nu ]}=4\pi qn^a{}_{}{}^{}𝒮_{\mu \nu }^{}.$$
(27)
Thus the gauge transformation considered adds a closed surface $`{}_{}{}^{}𝒮_{\mu \nu }^{}`$ to the field strength tensor, $`\widehat{F}(A^\mathrm{\Omega })=\mathrm{\Omega }^+\widehat{F}(A)\mathrm{\Omega }4\pi q{}_{}{}^{}\widehat{𝒮}`$, $`\widehat{𝒮}=\widehat{n}𝒮`$. It is easy to see that the color structure of the surface $`\mathrm{\Sigma }_{\mu \nu }^a`$ was inessential in our analysis. Indeed, one may perform exactly the same transformations with arbitrary $`n^a(\sigma )`$, $`\stackrel{}{n}^{\mathrm{\hspace{0.33em}2}}=1`$, instead of (13). Therefore the orientable non self-intersecting surface $`\mathrm{\Sigma }`$ in Eq. (17) with arbitrary color orientation can be deformed by the singular gauge transformation provided that $`q=0,\pm \frac{1}{2},\pm 1,\mathrm{}`$.
We see that the situation looks similar to the Abelian case where the shape of the Dirac strings is inessential and can be changed by a gauge transformation so that only the end points of the strings have a physical meaning: they are identified with monopoles. However despite of this similarity the Yang–Mills theory is different in some respects. In particular, for the examples considered below the boundaries of the strings (13) have zero action thus being a pure gauge artifacts<sup>1</sup><sup>1</sup>1 Note that in the standard Yang–Mills theory these configurations have infinite action and therefore are not important. .
### 2.2 Open Strings With Zero Action.
Consider the following gauge transformation matrix:
$$\mathrm{\Omega }_1=\left(\begin{array}{cc}\mathrm{cos}\frac{\theta }{2}e^{i\phi }& \mathrm{sin}\frac{\theta }{2}\\ & \\ \mathrm{sin}\frac{\theta }{2}& \mathrm{cos}\frac{\theta }{2}e^{i\phi }\end{array}\right),\mathrm{\Omega }_1^+\tau ^3\mathrm{\Omega }_1=\widehat{x}^a\tau ^a,$$
(28)
defined in the time-slice $`t=0`$; $`\theta `$, $`\phi `$ are polar and azimuthal angles. Performing the gauge transformation on the pure vacuum configuration $`A=0`$ one gets:
$${}_{}{}^{}F_{\mu \nu }^{a}(A^{\mathrm{\Omega }_1})=4\pi \delta _{0,[\mu }\delta _{\nu ],3}\delta ^{a,3}\mathrm{\Theta }(z)\delta (x)\delta (y).$$
(29)
Therefore the singular gauge transformation (28) produces singular $`F_{\mu \nu }^a`$ as well, but the singularity in (29) is of allowed type (14) with time independent surface $`\mathrm{\Sigma }`$ directed along $`\tau ^3`$ in the color space. On the other hand, in the gauge transformed potentials one finds an Abelian monopole which is double charged in terms of the minimal Dirac quantization condition (cf. Eq. (1)):
$$\begin{array}{ccc}A^3& =& A_\mu ^3dx_\mu =(1+\mathrm{cos}\theta )d\phi ,\hfill \\ \text{}A^+& =& (A_\mu ^1+iA_\mu ^2)dx_\mu =e^{i\phi }(d\theta i\mathrm{sin}\theta d\phi ).\hfill \end{array}$$
(30)
The interpretation of (29,30) is as follows. In the U(1) case due to the magnetic flux conservation the Dirac string terminates at an Abelian monopole with the magnetic field $`|𝐇|1/r^2`$. In the SU(2) gauge model the Abelian string with net flux $`4\pi `$ may disappear into the vacuum. Although we still have the conservation of Abelian flux, this does not imply any bound on the action. In fact, because of the nontrivial components $`A^\pm `$ the full SU(2) action is zero. The zero action of the configuration (30) is due to the cancellation between the Abelian-like and commutator pieces in $`F_{\mu \nu }^a`$. Note that already in Ref. it was shown that the double charged Abelian monopoles being immersed into SU(2) gauge group are unstable against fluctuations of non-Abelian components of gauge fields.
Proceed now to generalizing (28) to the case of finite Dirac string. Consider the potential $`A_\mu `$ which in U(1) theory represents the monopole–antimonopole pair located at $`x,y=0`$, $`z=\pm R/2`$:
$$A_\mu dx_\mu =\frac{1}{2}\left(\frac{z_+}{r_+}\frac{z_{}}{r_{}}\right)d\phi =A_D(z,\rho )d\phi ,0A_D(z,\rho )1$$
(31)
$$z_\pm =z\pm R/2,\rho ^2=x^2+y^2,r_\pm ^2=z_\pm ^2+\rho ^2,$$
(32)
and the following gauge transformation matrix
$$\mathrm{\Omega }_2=\left(\begin{array}{cc}e^{i\phi }\sqrt{A_D}\hfill & \sqrt{1A_D}\hfill \\ & \\ \sqrt{1A_D}\hfill & e^{i\phi }\sqrt{A_D}\hfill \end{array}\right).$$
(33)
It is easy to check that (33) when applied to the vacuum $`A=0`$ produces a string of the type (13,14) which begins and terminates at the points $`\rho =0`$, $`z=z_\pm `$, respectively:
$${}_{}{}^{}F_{\mu \nu }^{a}=4\pi \delta _{0,[\mu }\delta _{\nu ],3}\delta ^{a,3}\mathrm{\Theta }(R/2|z|)\delta (x)\delta (y).$$
(34)
The corresponding gauge potential $`A=i\mathrm{\Omega }_2^+\mathrm{\Omega }_2`$ contains Abelian monopole and antimonopole located at the ends of the string, $`A_\mu ^3dx_\mu =2A_D(z,\rho )d\varphi `$.
The described above monopole configurations might be related to the monopoles common to the the lattice Abelian projections . In a way, it is a consequence of the asymptotic freedom alone. Indeed, by the monopole one understands field configurations which in their Abelian part look like the standard Dirac monopole (1). The action associated with the Abelian monopole is linearly divergent in the ultraviolet,
$$S_{Abelian}(ag^2(a))^1,$$
where $`a`$ is an ultraviolet cut off, say, the lattice spacing. At first sight, on the background of this linear divergence the logarithmic behavior of the coupling is not important at all. However, it was shown in Ref. that the $`a^1`$ factor in the action can be overcome by the entropy since it is proportional to an exponential of the length of monopole trajectories measured in the same units of $`a`$. As a result, the value of the coupling is becoming crucial and the Abelian-like monopoles can be abundant in the vacuum only if the coupling is of order unit, $`g^21`$. Which is inconsistent with the asymptotic freedom of the gluodynamics. The only way out is to have the non-Abelian field strength vanishing at short distances, $`F_{\mu \nu }^a0`$ at $`r0`$. In other words, the cancellation of the Abelian-like and commutator terms in the field strength tensor should be exact at short distances. The latter condition is satisfied by (30) which appears to be the unique monopole solution at short distances
The monopole structure at short distances can be studied directly on the lattice. At the distances available so far, the monopoles in SU(2) LGT are associated with a sizeable excess in the action, although the excess is substantially smaller than it would be in the pure Abelian case . Further measurements at smaller distances would be very interesting.
### 2.3 Quantum Corrections.
The examples presented above show that an arbitrary (non self-intersecting) string (13) may be considered as a result of combined gauge transformations of the type (23), (28), (33). Moreover, in the case of trivial background $`\widehat{F}(A)=0`$ such a singular gauge transformations are allowed and produce no action. A crucial question is whether the strings (13) are equivalent to gauge transformations when quantum fluctuations are included. Of course, if it were not so that the gauge transformations considered are singular, there would be no doubt that the quantum corrections do not destroy equivalence of the two field configurations related by a gauge transformation. But because of the presence of singularities we performed an explicit analysis of the quantum corrections. The result is that the quantum corrections do not distinguish between the standard perturbative vacuum and the zero-action field configuration presented in the preceding section.
For the sake of definiteness we consider a straight Dirac string in the partition function (17)
$$Z[\mathrm{\Sigma }]=𝒟A\mathrm{exp}\left\{\frac{1}{4g^2}d^4x\left[F_{\mu \nu }^a+4\pi {}_{}{}^{}\mathrm{\Sigma }_{\mu \nu }^{a}\right]^2\right\},$$
(35)
$$\mathrm{\Sigma }_{\mu \nu }^a=4\pi \delta _{0,[\mu }\delta _{\nu ],3}\delta ^{a,3}\mathrm{\Theta }(z)\delta (x)\delta (y).$$
(36)
The ”classical” solution of the field equations is the pure gauge configuration (30) $`A^{cl}=i\mathrm{\Omega }^+\mathrm{\Omega }`$ where $`\mathrm{\Omega }`$ is given by (28) and the ”classical” action is $`S^{cl}=0`$. Expanding the action up to the second order in small perturbations $`A=A^{cl}+a`$ one finds that in the background gauge $`D_\mu (A^{cl})a_\mu =0`$ Eq. (35) becomes:
$$Z[\mathrm{\Sigma }]=\mathrm{Det}^1[D^2(A^{cl})]$$
(37)
since in the present case the Pauli paramagnetic term is zero. With conventional normalization to the perturbative vacuum to vacuum amplitude the question whether the string $`\mathrm{\Sigma }_{\mu \nu }^a`$ is relevant on quantum level, is equivalent to exploring the spectrum of the operator $`D^2(A^{cl})`$:
$$D^2(A^{cl})=M_1+M_2+M_3$$
(38)
$$M_1^{ab}=\delta ^{ab}\stackrel{}{}^2M_2^{ab}=2\epsilon ^{akb}\stackrel{}{A}^k\stackrel{}{}M_3^{ab}=\epsilon ^{akb}\stackrel{}{}\stackrel{}{A}^k+\stackrel{}{A}^a\stackrel{}{A}^b\delta ^{ab}\stackrel{}{A}^k\stackrel{}{A}^k$$
(39)
where superscripts denote the color indices and vector notations are used for spatial components of $`A^{cl}`$. Using the explicit form (28) one finds that the non-zero elements of the antisymmetric matrix $`M_2`$ are
$$\begin{array}{ccc}\hfill M_2^{12}& =& \frac{2}{r^2}\frac{1+\mathrm{cos}\theta }{\mathrm{sin}^2\theta }_\phi \hfill \\ \hfill \text{}M_2^{13}& =& \frac{2}{r^2}\left(\mathrm{cos}\phi _\theta \frac{\mathrm{sin}\phi }{\mathrm{sin}\theta }_\phi \right)\hfill \\ \hfill \text{}M_2^{23}& =& \frac{2}{r^2}\left(\mathrm{sin}\phi _\theta +\frac{\mathrm{cos}\phi }{\mathrm{sin}\theta }_\phi \right)\hfill \end{array}$$
(40)
where $`\theta `$ and $`\phi `$ are the polar and azimuthal angles, respectively. In the same coordinate system the matrix $`M_3`$ is given by
$$M_3=\frac{2}{r^2}\frac{1+\mathrm{cos}\theta }{\mathrm{sin}^2\theta }\left[\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ \mathrm{cos}\phi \mathrm{sin}\theta & \mathrm{sin}\phi \mathrm{sin}\theta & 1\mathrm{cos}\theta \end{array}\right]$$
(41)
It is convenient to perform the transformation $`D^2(A^{cl})RD^2(A^{cl})R^1`$ where the matrix $`R`$ transforms to the spherical basis:
$$R=\left[\begin{array}{ccc}\mathrm{cos}\phi \mathrm{sin}\theta & \mathrm{sin}\phi \mathrm{sin}\theta & \mathrm{cos}\theta \\ \mathrm{cos}\phi \mathrm{cos}\theta & \mathrm{sin}\phi \mathrm{cos}\theta & \mathrm{sin}\theta \\ \mathrm{sin}\phi & \mathrm{cos}\phi & 0\end{array}\right].$$
(42)
One finds that in the new basis:
$$D^2(A^{cl})=\stackrel{}{}^{\mathrm{\hspace{0.33em}2}}+\frac{1}{r^2}\left[\begin{array}{ccc}0& 0& 0\\ \text{}0& \frac{1}{\mathrm{sin}^2\theta }& \frac{2}{\mathrm{sin}^2\theta }_\phi \\ \text{}0& \frac{2}{\mathrm{sin}^2\theta }_\phi & \frac{1}{\mathrm{sin}^2\theta }\end{array}\right].$$
(43)
Next, introduce
$$a_0=a_ra_\pm =\frac{i}{\sqrt{2}}(a_\theta ia_\phi ),$$
(44)
then the operator $`D^2(A^{cl})`$ becomes diagonal:
$$D^2(A^{cl})=\stackrel{}{}^{\mathrm{\hspace{0.33em}2}}+\frac{1}{r^2}\mathrm{diag}[0,\frac{1}{\mathrm{sin}^2\theta }(12i_\phi ),\frac{1}{\mathrm{sin}^2\theta }(1+2i_\phi )]$$
(45)
Once $`D^2(A^{cl})`$ is brought to the diagonal form, the direct calculation shows that the spectrum of (45) is identical to that of free Laplacian $`\stackrel{}{}^{\mathrm{\hspace{0.33em}2}}`$. Therefore, the quantum fluctuations do not distinguish the string, Eq. (35,36), from the perturbative vacuum. Thus the modified theory (15,16) is perturbatively equivalent to the conventional gluodynamics.
## 3 Strings in General Background.
We have shown that in the perturbation theory both closed and open Dirac strings with arbitrary color orientation carry no action and are thus pure gauge artifacts. On one hand, this conclusion is welcome since it shows that the continuum limit as understood in this paper perturbatively is the same as the standard continuum limit. And, indeed, there are no doubts in the validity of the standard perturbation theory. On the other hand, if it were so that the singular fields admitted now into the continuum formulation are not associated with any action at all then the new formulation would be equivalent to the standard one.
In this section we address this issue on a non-perturbative level and imitate the non-perturbative fields by a smooth background. The crucial observation then is that only the Dirac strings with the proper color alignment (13) cost no action in the continuum limit, while the Dirac strings associated with the monopoles defined a la ’t Hooft do not satisfy this constraint.
Consider as an example the class of Abelian gauges of Ref. which are defined by the requirement that some adjoint operator $`h^a`$ is to be directed along $`\tau ^3`$ in the color space. This operator may be arbitrary in principle, but for the given $`h^a`$ the remaining gauge freedom consists of U(1) rotations around $`\tau ^3`$:
$$\begin{array}{ccc}& \mathrm{\Omega }^+\widehat{h}\mathrm{\Omega }h^3\tau ^3,& \\ & & \\ \mathrm{\Omega }=\stackrel{~}{\mathrm{\Omega }}H,& \stackrel{~}{\mathrm{\Omega }}G/U(1),& HU(1).\end{array}$$
(46)
This gauge condition is not defined on the set of points where $`h^a=0`$ which in four dimensions defines the monopole trajectory. Clearly enough, the Dirac strings associated with the monopoles are oriented along the third direction in the color space. Thus, there exist now two different directions in the color space determined by the background field and through the gauge fixation inherent to the definition of the monopoles.
Note that while the boundary of the string singularity is fixed by the equations $`h^a=0`$, the actual position of the string may be changed with a suitable choice of $`H`$. Indeed, the gauge transformation (LABEL:tHooft-gauge-transformation) gives rise to the Dirac string:
$$\widehat{F}(A^\mathrm{\Omega })=H^+\stackrel{~}{\mathrm{\Omega }}^+\widehat{F}(A)\stackrel{~}{\mathrm{\Omega }}H+iH^+[,]H+iH^+\left(\stackrel{~}{\mathrm{\Omega }}^+[,]\stackrel{~}{\mathrm{\Omega }}\right)H.$$
(47)
The both terms $`H^+[,]H`$ and $`\stackrel{~}{\mathrm{\Omega }}^+[,]\stackrel{~}{\mathrm{\Omega }}`$ are proportional to $`\tau ^3`$. Let us stress that the freedom to choose the U(1) matrix $`H`$ allows to shift the position of the string in the ordinary space. Simultaneously, the background field is also transformed and one may not say that shifting the Dirac string brings no change in the action. However, since there is no spontaneous breaking of the color symmetry, the dependence on the position of the string drops off after integrating over all the background fields.
Note that similar considerations apply to the ’t Hooft loop operator which we consider in the next section. Indeed, the definition of the ’t Hooft loop operator as well as its value in a given background are string dependent. But the freedom to shift the position of the string in the path integral approach guarantees that no physical result depends on the string position.
It is amusing to note that the present considerations provides with a general framework to understand the correlation between instantons and monopoles which has been discussed in various contexts recently (see, e.g., ). Indeed, background fields in the physical vacuum are described realistically by instantons (see for a review). On the other hand , monopoles, as is argued above, are meaningful only in the presence of background fields.
## 4 The ’t Hooft Loop in the Continuum Limit.
In this Section we show that the construction presented above allows to define and study the properties of the ’t Hooft loop operator in the path integral formalism.
The ’t Hooft loop in SU(2) LGT with the one-plaquette action (4) has the following form:
$$H_{lat}(\mathrm{\Sigma }_j)=\mathrm{exp}\{\frac{4}{g^2}\underset{p{}_{}{}^{}\mathrm{\Sigma }_{j}^{}}{}\left[S_P\right(1\frac{1}{2}TrU[p])S_P(1+\frac{1}{2}TrU[p])]\},$$
(48)
where $`{}_{}{}^{}\mathrm{\Sigma }_{j}^{}`$ is the set of the plaquettes dual to the surface $`\mathrm{\Sigma }_j`$ with the boundary $`j`$. In the path integral formulation the ’t Hooft loop effectively changes the sign of the plaquette variables $`U[p]`$ belonging to $`{}_{}{}^{}\mathrm{\Sigma }_{j}^{}`$: $`U[p]U[p]`$. To define the ’t Hooft loop in the continuum we consider the path integral, Eq. (17), with an open orientable non self-intersecting surface $`\mathrm{\Sigma }_j^a=n^a\mathrm{\Sigma }_j`$, $`\mathrm{\Sigma }_j=j`$, multiplied by the Wilson loop $`W_J(𝒞)`$:
$`Z(𝒞,\mathrm{\Sigma }_j)`$ $`=`$ $`{\displaystyle 𝒟A\mathrm{exp}\left\{\frac{1}{4g^2}d^4x\left[F_{\mu \nu }^a+4\pi q{}_{}{}^{}\mathrm{\Sigma }_{j\mu \nu }^{a}\right]^2\right\}W_J(𝒞)},`$ (49)
$`W_J(𝒞)`$ $`=`$ $`Tr\text{P}\mathrm{exp}\left\{i{\displaystyle \underset{𝒞}{}}T_J^aA_\mu ^a𝑑x_\mu \right\},`$ (50)
where $`T_J^a`$ are the generators of SU(2) in the representation $`J`$. It is convenient to use the following integral representation :
$`W_J(𝒞)={\displaystyle 𝒟\omega \mathrm{exp}\left\{iJ\underset{𝒞}{}\left[A_\mu ^\omega \right]^3𝑑x_\mu \right\}},\left[A_\mu ^\omega \right]^3=Tr\left[\tau ^3\omega ^+(A_\mu +i_\mu )\omega \right],`$
where the path integral is over all gauge transformations $`\omega `$ of the potential $`A`$ on the contour $`𝒞`$. Therefore
$$Z(𝒞,\mathrm{\Sigma }_j)=𝒟A𝒟\omega \mathrm{exp}\{S(A,\mathrm{\Sigma }_j)+iJ\underset{𝒞}{}\left[A_\mu ^\omega \right]^3𝑑x_\mu \}.$$
(51)
The action $`S(A,\mathrm{\Sigma }_j)`$ and the measure of integration $`𝒟A`$ are gauge invariant. The gauge transformation $`AA^{\omega ^1}`$ defined on $`𝒞`$ allows to factorize the integral $`𝒟\omega `$:
$$Z(𝒞,\mathrm{\Sigma }_j)=𝒟\omega 𝒟A\mathrm{exp}\{S(A,\mathrm{\Sigma }_j)+iJ\underset{𝒞}{}A_\mu ^3𝑑x_\mu \}.$$
(52)
The expression (52) has the following meaning: if there is no gauge fixing in the path integral (49) the Wilson loop may be calculated exactly by restricting the gauge potential $`A`$ to diagonal U(1) subgroup of SU(2) .
Now we deform the surface $`\mathrm{\Sigma }_j`$ spanned on the contour $`j`$ to another orientable non self-intersecting surface $`\mathrm{\Sigma }_j^{}`$ spanned on the same contour. We also consider the field $`n^a(\sigma )`$ defined on $`\mathrm{\Sigma }_j^{}`$ according to (13). Then the closed surface $`𝒮=\mathrm{\Sigma }_j\mathrm{\Sigma }_j^{}`$, which bounds the 3-volume $`𝒱_𝒮=𝒱_{\mathrm{\Sigma }_j\mathrm{\Sigma }_j^{}}`$, has no self-intersection points and there exists a vector field $`n^a(x)`$, defined in the whole space–time,
$$\begin{array}{ccc}n^a(x)=n^a(\stackrel{~}{x})& \text{for}& x\mathrm{\Sigma }_j,\\ n^a(x)=n^a(\stackrel{~}{x})& \text{for}& x\mathrm{\Sigma }_j^{}.\end{array}$$
(53)
In particular, $`n^a(x)`$ is defined on the contour $`𝒞`$ and there exists an SU(2) matrix $`h𝒞`$, such that:
$$\left[h\sigma ^3h^+\right]^a=n^a(x)x𝒞.$$
(54)
After the gauge transformation $`AA^h`$ Eq. (52) becomes:
$$Z(𝒞,\mathrm{\Sigma }_j)=𝒟\omega 𝒟A\mathrm{exp}\{S(A,\mathrm{\Sigma }_j)+iJ\underset{𝒞}{}(n^aA^a+[ih^+h]^3)$$
(55)
Consider now the additional gauge transformation $`\mathrm{\Omega }(𝒱_𝒮)`$ (23,24) with $`𝒱_𝒮=𝒱_{\mathrm{\Sigma }_j\mathrm{\Sigma }_j^{}}`$. A straightforward calculation gives
$$n^a\left[A_\mu ^\mathrm{\Omega }\right]^a=n^aA_\mu ^a4\pi qV_\mu .$$
(56)
As is shown in Section 2.1 for integer and half-integer charges $`q`$ this gauge transformation shifts $`\mathrm{\Sigma }_j`$ to $`\mathrm{\Sigma }_j^{}`$:
$$S(A^\mathrm{\Omega },\mathrm{\Sigma }_j)=S(A,\mathrm{\Sigma }_j^{}).$$
(57)
For self–consistency of the theory, the Wilson loop is to be invariant under the gauge transformations. If we apply the transformation (56,57) to $`Z(𝒞,\mathrm{\Sigma }_j)`$, see Eq. (49), we get:
$$\mathrm{\Omega }(𝒱_𝒮):Z(𝒞,\mathrm{\Sigma }_j)Z(𝒞,\mathrm{\Sigma }_j𝒮)e^{i\mathrm{\hspace{0.33em}4}\pi qJ(𝒞,𝒮)},$$
(58)
where $`(𝒞,𝒮)`$ is the 4D linking number between the closed contour $`𝒞`$ and the closed surface $`𝒮`$:
$`(𝒞,𝒮)={\displaystyle \underset{𝒮}{}}\left({}_{}{}^{}d_{}^{2}\sigma \right)_{\mu \nu }{\displaystyle \underset{𝒞}{}}𝑑x_\nu _\mu \mathrm{\Delta }^1(\stackrel{~}{x}(\sigma )x).`$ (59)
Since $`ZZ`$ and $`J`$ takes integer and half–integer values the independence of the Wilson loop on the gauge transformations $`\mathrm{\Omega }`$ implies the quantization condition:
$`qZZ.`$ (60)
This equation is a direct analog of the Dirac quantization condition in electrodynamics. Physically it means that the electrically charged particle introduced by Wilson loop does not scatter on the Dirac string $`𝒮`$.
Now we show that the ’t Hooft loop operator $`H(\mathrm{\Sigma }_j)`$ is given by:
$`H(\mathrm{\Sigma }_j)=\mathrm{exp}\left\{S(\widehat{F})S(\widehat{F}+2\pi {}_{}{}^{}\widehat{\mathrm{\Sigma }}_{j}^{})\right\},`$ (61)
where the surface $`\mathrm{\Sigma }_j^a`$ is bounded by the contour $`j`$ and is given by Eq. (13), the action $`S`$ is defined by Eq. (16). Indeed, the transformation (58), when applied to the quantum average of the product of the fundamental, $`J=1/2`$, Wilson loop and operator (61), gives:
$$<H(j,\mathrm{\Sigma }_j)W_{1/2}(𝒞)>=<H(j,\mathrm{\Sigma }_j^{})W_{1/2}(𝒞)e^{i\pi (𝒞,\mathrm{\Sigma }_j\mathrm{\Sigma }_j^{})}>.$$
(62)
This formula proves that the operator $`H`$ is the ’t Hooft loop operator since it is in accordance with relations given in Refs. .
## 5 Predictions for the ’t Hooft loop.
In this Section we consider the rectangular $`T\times R`$ time-like contours $`j`$, Eq. (61), with $`TR`$. Then the expectation value of the ’t Hooft loop operator is
$$<H(\mathrm{\Sigma }_j)>=<H(T,R)>e^{TV_{m\overline{m}}(R)},$$
(63)
where by analogy with the Wilson loop we refer to the quantity $`V_{m\overline{m}}(R)`$ as to the intermonopole (monopole–antimonopole) potential. It is worth emphasizing that the potential $`V_{m\overline{m}}`$ corresponds to the $`|Q_M|=1`$ monopoles while in Ref. the monopole–antimonopole potential with the charge $`|Q_M|=2`$ has been studied. These double charged monopoles are identified with the Abelian monopoles in Abelian projections.
Below we formulate predictions for the ’t Hooft loop operator, Eq. (61), and its expectation value, Eq. (63). In particular, we show that the ’t Hooft loop operator inserts the pair of $`|Q_M|=1`$ monopoles which are pure Abelian in the Maximal Abelian gauge. This fact allows to fix the short distance asymptotic of the intermonopole potential. We argue then that this potential at larger distances at zero and high temperatures is of Yukawa type. We also find the screening mass in both cases and compare it with the masses measured on the lattice . Our estimates turn to be in agreement with the numerical data.
### 5.1 Intermonopole Potential at Small Distances.
Consider the potential $`V_{m\overline{m}}(R)`$ at small distances for the monopole–antimonopole pair introduced by the operator $`H(T,R)`$. The definition (61) shows that we have enough gauge freedom to take $`\mathrm{\Sigma }_j^a=\delta ^{a,3}\mathrm{\Sigma }_j`$ on the non self-intersecting surface $`\mathrm{\Sigma }_j`$. Then at the classical level the solution of the corresponding equations of motion is :
$$\begin{array}{c}A_\mu ^3dx_\mu =\frac{1}{2}\left(\frac{z_+}{r_+}\frac{z_{}}{r_{}}\right)d\phi ,A_\mu ^{1,2}=0,\\ \text{}z_\pm =z\pm R/2,\rho ^2=x^2+y^2,r_\pm ^2=z_\pm ^2+\rho ^2,\end{array}$$
(64)
and represents the Abelian monopole-antimonopole pair separated by the distance $`R`$. Since the monopoles in (64) have minimal allowed magnetic charge $`q=1/2`$ (see Section 2.1), at the classical level the intermonopole potential is given by:
$$V_{m\overline{m}}(R)=\frac{\pi }{g^2R}=\pi ^2\beta \frac{1}{4\pi R},\beta =\frac{4}{g^2}.$$
(65)
Note that the statement on the Coulombic nature of the intermonopole potential at short distances is well known . However, the fixation of the coefficient in front of $`1/R`$ is new, to the best of our knowledge<sup>2</sup><sup>2</sup>2 The same coefficient is derived in Ref. , which appeared on the day of submission of the present paper. .
Since the potential (65) was obtained for pure Abelian fields, we still have to prove that the general solution with minimal energy in SU(2) gluodynamics is indeed a gauge rotation of (64). A straightforward way to test the Eq. (65) is to investigate the problem numerically. We have calculated the expectation value of the ’t Hooft loop in the standard SU(2) lattice gauge theory in the limit $`\beta \mathrm{}`$. Technically this limit is realized with the help of the so–called cooling procedure which was used to minimize the expectation value of the ’t Hooft loop with respect to the classical lattice equations of motion. Our calculations have been performed on the three-dimensional $`24^3`$ lattice with periodic boundary conditions, which is adequate to consider the static monopole–antimonopole pair. We minimized the ’t Hooft loop operator, which creates static monopole and antimonopole separated by the distance $`R`$. We have fitted our data for the monopole–antimonopole potential by:
$$V_{m\overline{m}}^{lat.}(R)=\pi ^2\beta \mathrm{\Delta }_{lat.}^1(R),$$
(66)
where $`\mathrm{\Delta }_{lat.}^1(R)`$ is the three-dimensional lattice Coulomb potential. Eq. (66) is the lattice regularization of the continuum expression (65). Note that the lattice and continuum potentials drastically differ from each other and this is of crucial importance in fitting the lattice data: the potential (66) is regular at $`R=0`$ contrary to (65).
Our numerical calculations confirmed the behavior (66) with accuracy $`2\%`$. We also observed that after the cooling procedure the fields are Abelian up to a gauge transformation. In more detail, we found that in the Maximal Abelian gauge the gauge fields are diagonal and consist of the Abelian monopoles located at the boundary of the string $`\mathrm{\Sigma }_j`$, Eq. (61). Therefore the classical limit of the state created by non-Abelian ’t Hooft loop is the Abelian monopole–antimonopole pair.
Moreover, once the result (65) is established classically, the effect of the quantum corrections is also known on general grounds. Namely, the effect of the quantum corrections should be reduced to the replacement of the bare coupling by the running one, $`g^2g^2(R)`$. Although the result is easy to guess, its derivation might look rather mysterious. Indeed, we have now both non-Abelian magnetic monopoles as external objects and ordinary gluons as virtual particles. At first sight we need both the standard and dual formulations of the gluodynamics to describe interaction both with magnetic and electric charges. While in case of U(1) gauge theories such a formulation is well known , it is absent in case of non-Abelian theories. Thus, we seem to know how the coupling runs although do not know, whose coupling is it!
We think that the resolution of the paradox is in the Abelian nature of the $`|Q_M|=1`$ monopoles established above. Indeed, the classical considerations allow us to fix vertices, or the Lagrangian. The exact Abelian nature of the monopoles implies that once we choose an Abelian gauge fixing only neutral bosons (diagonal gluons) interact with the monopoles $`|Q_M|=1`$. The charged vector bosons are still manifested through the loops. Thus, the situation is similar to the U(1) case with inclusion of the effect of virtual charged particles. As for the virtual monopoles, their effect can be neglected since the monopoles $`|Q_M|=1`$ are infinitely heavy in the continuum limit. There is no much difficulty to deal with this problem and one can check that indeed the effect of the loops is the running of the coupling $`g^2`$. The details of the U(1) case can be found in the review in Ref. , see also the recent paper . As for the perturbative calculations in non-Abelian theories in the Abelian projections, they can be found in Ref.
### 5.2 Abelian Dominance and Intermonopole Potential.
Next we discuss the monopole–antimonopole potential at larger distances. The basic idea is to apply the Abelian Dominance hypothesis . Indeed, as has been shown above the ’t Hooft loop operator inserts the $`|Q_M|=1`$ monopole pair in the vacuum of SU(2) gauge theory. Moreover, in the Maximal Abelian gauge these monopoles become a pure Abelian objects. Therefore it is natural to expect that in this particular gauge the dominant contribution to the potential (63) is due to the interaction with Abelian fields. In the Maximal Abelian gauge the vacuum of zero temperature SU(2) gluodynamics is a dual superconductor where, instead of condensate of Cooper pairs, there exists a monopole condensate. The principle of Abelian Dominance assumes that long distance properties of gluodynamics might be explained in terms of the interaction with the monopole condensate (for reviews see, e.g., ).
Following this logic, we expect that at the zero temperatures the monopole–antimonopole potential is:
$$V_{m\overline{m}}(R)=\frac{\pi }{g^2}\frac{e^{\mu R}}{R}$$
(67)
$$V_{m\overline{m}}^{lat.}(R)=\beta \pi ^2(\mathrm{\Delta }+\mu ^2)_{lat.}^1(R)$$
(68)
where $`\mu `$ is the dual photon mass $`m_V`$ and $`(\mathrm{\Delta }+\mu ^2)_{lat.}^1`$ is the three-dimensional lattice Yukawa potential. The recent numerical investigation of the ’t Hooft loop in SU(2) lattice gauge theory agrees with Eq. (67). The value of $`\mu 3.24(42)\sqrt{\sigma }`$ obtained in Ref. is quite close to the dual photon mass $`m_V1\mathrm{GeV}=2.3\sqrt{\sigma }`$ found in Ref. . Let us also note that we would not identify directly the mass $`\mu `$ in Eq. (67) with a glueball mass. Indeed, the definition of the ’t Hooft loop is highly nonlocal and includes a Dirac string with infinite action. Therefore, the validity of the dispersive relations is questionable in this case. Note, however, that $`\mu `$ in Eq. (67) coincides with the $`0^{++}`$ glueball mass in the strong coupling expansion . If this result is valid also in the weak coupling limit, then the Abelian Dominance is reduced to the prediction that the dual photon mass $`m_V`$ coincides with the $`0^{++}`$ glueball mass. Comparison of numerical results for the Yukawa mass $`\mu `$ with glueball masses can be found in .
Note that the prediction (67) is highly non-trivial in fact. Indeed the $`|Q_M|=1`$ monopoles are so to say fundamental monopoles which look as Abelian monopoles at short distances and are associated for this reason with an infinite action. They are introduced therefore as external objects via the ’t Hooft loop, similar to introduction of infinitely heavy quarks via the Wilson loop. The $`|Q_M|=2`$ monopoles, on the other hand, have a finite action and their description as a fundamental objects seems to be granted only at large distances. This could be manifested, in particular, through existence of an intermediate region between the distances where the Coulombic and Yukawa pictures apply. In other words, the coefficient in front of the Coulombic term could have not matched the coefficient in front of the Yukawa-like potential. However, existing data about the ’t Hooft loop indicate that the matching is exact, within the error bars. In other words, the dual Abelian Higgs model of QCD vacuum works already at smallest distances available on the lattice. Similar conclusions can be drawn in fact from the studies of the heavy quark potential induced by monopoles and from description of the structure of the flux string , for a review see .
### 5.3 Finite Temperatures.
The authors of Ref. have also performed numerical calculations of the ’t Hooft loop at finite temperatures, and determined the dependence of the Yukawa mass $`\mu `$ on the temperature. To provide a theoretical framework for the behavior of the ’t Hooft loop at high temperatures we can use again the idea of the Abelian Dominance.
In more detail, we estimate the screening mass $`\mu `$ using the fact that the Abelian model which corresponds to the high temperature SU(2) gluodynamics is the 3D compact U(1) theory. Therefore the intermonopole potential at high temperatures is essentially given by (67,68), with $`\mu `$ now being the Debye mass :
$$m_D^2=16\pi \frac{\rho }{e_3^2},$$
(69)
where $`\rho `$ is the density of Abelian monopoles and $`e_3`$ is the corresponding three-dimensional coupling constant. To estimate the temperature dependence of $`m_D`$ we use the numerical results of Ref. , where the density of Abelian monopoles was obtained<sup>3</sup><sup>3</sup>3 Note that the original result of Ref. for the lattice monopole density is: $`\rho _{lat.}=0.50(1)\beta _G^3`$, where $`\beta _G^3`$ is a three dimensional coupling constant which is expressed in terms of the 3D electric charge $`e_3`$ and lattice spacing $`a`$ as $`\beta _G^3=4/(ae_3^2)`$. The physical density $`\rho `$ of monopoles is given by $`\rho =\rho _{lat.}a^3`$ which can easily be transformed into Eq. (70). :
$$\rho =2^7(1\pm 0.02)e_3^6,$$
(70)
Therefore
$$m_D=1.11(2)e_3^2.$$
(71)
Moreover, at high temperatures we can use the dimensional reduction formalism and express the 3D coupling constant $`e_3`$ in terms of the 4D Yang–Mills coupling $`g`$. At the tree level one has
$$e_3^2(T)=g^2(\mathrm{\Lambda },T)T,$$
(72)
where $`g(\mathrm{\Lambda },T)`$ is the running coupling calculated at the scale $`T`$,
$$g^2(\mathrm{\Lambda },T)=\frac{11}{12\pi ^2}\mathrm{log}\left(\frac{T}{\mathrm{\Lambda }}\right)+\frac{17}{44\pi ^2}\mathrm{log}[2\mathrm{log}\left(\frac{T}{\mathrm{\Lambda }}\right)],$$
(73)
and $`\mathrm{\Lambda }`$ is a dimensional constant which can be determined from lattice simulations.
At present the lattice measurements of the $`\mathrm{\Lambda }`$ parameter are not very precise. We use the results of two particular calculations. Namely, in Ref. the lattice data for the gluon propagator have been used to determine the so–called ”magnetic mass” in high temperature SU(2) gluodynamics. These measurements imply the following value of $`\mathrm{\Lambda }`$:
$$\mathrm{\Lambda }=0.262(18)T_c=0.197(14)\sqrt{\sigma },$$
(74)
where $`T_c`$ is the temperature of the deconfinement phase transition, $`T_c0.75\sqrt{\sigma }`$. In Ref. , on the other hand, the spatial string tension has been calculated and the corresponding value of $`\mathrm{\Lambda }`$ turned to be three times smaller:
$$\mathrm{\Lambda }=0.076(13)T_c=0.057(10)\sqrt{\sigma }.$$
(75)
Collecting Eqs. (71)-(75) we get predictions for the Debye mass which are shown in Table 1 along with the values of mass $`\mu `$ obtained numerically in Ref. . One can clearly see that the predictions and the numerical results are in agreement within the theoretical uncertainties. There are at least three sources of these uncertainties. First, the value of $`\mathrm{\Lambda }`$ is not determined precisely as we already noted. Second, we have used the dimensional reduction which is supposed to work well only at asymptotically high temperatures, while only one value $`T=3.676\sqrt{\sigma }5T_c`$ in the Table 1 may be considered as high enough. Third, as we already noted the lattice and continuum Yukawa interactions are substantially different. For example, we may treat the ’t Hooft loop quantum average studied in Ref. as a two–point correlator in three spatial dimensions. Then we may use the results of Ref. and relate the value of $`\mu `$ obtained with the use of the continuum propagator to the correct value, $`\mu ^{\mathrm{correct}}\frac{2}{a}\mathrm{ArcSinh}\left(\frac{\mu a}{2}\right)`$. If we apply this correction to the values of $`\mu `$ in Table 1 then for $`a\mu =2.29(55)`$ and $`T=3.676\sqrt{\sigma }`$ the correction is essential. Indeed, we obtain: $`\mu ^{\mathrm{correct}}14.8(3.5)`$, which is quite close to our prediction with $`\mathrm{\Lambda }=0.197\sqrt{\sigma }`$.
## Conclusions
We have tried to formulate a theoretical framework which would allow for monopoles in the continuum version of non-Abelian gauge theories. Indeed, monopoles nowadays are very common field configuration on the lattice. In the continuum, on the other hand, monopoles appear to be associated with singular fields and divergent action.
The key element to introduce monopoles in the continuum is to allow for Dirac strings. While naively the action associated with the Dirac strings is infinite, they cost no action at all in the compact U(1) gauge model . Thus, within the continuum formulation, one has to postulate that there are certain singular fields which cost no action as well. An alternative representation for the singular fields are Dirac sheets (see, e.g., Eq. (8) above). In the non-Abelian case, we argued that the continuum version should admit certain singular or stringy fields without any change in the action. One can say that the Dirac strings which cost no action are aligned in the color space with the background, or regular fields.
Once the Dirac strings are admitted into the continuum version of gluodynamics, the end points of the strings, or monopoles, cost in the perturbative vacuum no action either. This is true both classically and with account of quantum corrections. And this is in distinction from the U(1) case where the end points are monopoles with an ultraviolet divergent action. As a result, although the modified continuum version appears very different from the standard one since it allows for singular potentials inversely proportional to the coupling, perturbatively the two theories are in fact equivalent. Thus, at this point the problem seems to be the other way around. Namely, there is no difficulty any longer to introduce fields which look as monopoles in terms of Abelian fields but cost no action and appear as gauge artifacts once the full spectrum of the non-Abelian degrees of freedom is taken into account.
The difference between the two formulations becomes manifest once the gauge is fixed a la ’t Hooft and background non-perturbative fields are introduced. The point is that in presence of the background field only those Dirac strings which are parallel to the background in the color space are non observable. On the other hand, the definition of the monopoles in terms of the topology of the gauge fixing introduces Dirac strings which do not satisfy this condition. As a result, the action associated with the monopoles is not vanishing any longer. And the monopoles do emerge as possible fluctuations with finite action which are present in the continuum theory modified to incorporate Dirac strings. It is worth emphasizing that upon integration over the background fields the monopole action does not depend on the position of the Dirac string but only on the monopole trajectory.
The machinery to prove the independence on the position of the Dirac string is also all what is needed to introduce a continuum analog of the ’t Hooft loop operator . The continuum formulation of the ’t Hooft loop is one of the central points of this paper. Furthermore, we were able to derive both rigorous and model-dependent results for the behavior of the ’t Hooft loop at zero and high temperature SU(2) gluodynamics.
In terms of physical applications, the picture developed explains in generic terms correlation between instantons and monopoles (for discussion see, e.g., ). Also, it was demonstrated that while perturbatively the modified theory allowing for the Dirac strings is equivalent to the standard one, non-perturbatively they are different. This might explain a kind of mystery with the non-perturbative $`1/Q^2`$ corrections from short distances which seem to exist phenomenologically but evade, so far, theoretical understanding within the standard framework (for reviews and further references see ).
## Acknowledgments
M.N.Ch. and M.I.P. acknowledge the kind hospitality of the staff of the Max-Planck Institut für Physik (München), where the work was initiated. The authors are grateful to V.A. Rubakov for useful discussions. Work of M.N.C., F.V.G. and M.I.P. was partially supported by grants RFBR 99-01230a, RFBR 96-1596740 and INTAS 96-370. |
warning/0003/nlin0003026.html | ar5iv | text | # Response to Comments on “Simple Measure for Complexity”
## Abstract
We respond to the comment by Crutchfield, Feldman and Shalizi and that by Binder and Perry, pointing out that there may be many maximum entropies, and therefore “disorders” and “simple complexities”. Which ones are appropriate depend on the questions being addressed. “Disorder” is not restricted to be the ratio of a nonequilibrium entropy to the corresponding equilibrium entropy; therefore, “simple complexity” need not vanish for all equiibrium systems, nor must it be nonvanishing for a nonequilibrium system.
We are pleased that our contribution on a “simple measure for complexity” (hereafter referred to as SDL) is of sufficient interest to have generated two comments, one by Crutchfield, Feldman and Shalizi (CFS) and another by Binder and Perry (BP). In SDL we proposed
$$\mathrm{\Gamma }_{\alpha \beta }\mathrm{\Delta }^\alpha \mathrm{\Omega }^\beta ,$$
(1)
$$\mathrm{\Delta }S/S_{max},\mathrm{\Omega }1\mathrm{\Delta }.$$
(2)
as a “simple measure for complexity”. $`\alpha `$ and $`\beta `$ are (constant) parameters, $`S`$ is the Boltzmann-Gibbs-Shannon entropy , and $`S_{max}`$, the maximum entropy. $`\mathrm{\Delta }`$ was introduced earlier by one of us as a measure for disorder, and $`\mathrm{\Omega }`$ is referred to as “order” .
CFS raise several points:
* Since $`S_{max}`$ is the equilibrium entropy, $`\mathrm{\Delta }`$ and $`\mathrm{\Gamma }_{\alpha \beta }`$ vanish for all equilibrium systems, and neither can “distinguish between two-dimensional Ising systems at low temperature, high temperature, or the critical temperature …\[nor\] …between the many different kinds of organization observed in equilibrium.”
* $`\mathrm{\Gamma }_{\alpha \beta }`$ “is over-universal in the sense that it has the same dependence on disorder for structurally distinct systems.”
* The “statistical complexity” $`C_\mu `$ of one-dimensional spin systems is not the same as the entropy of noninteracting spins. The identification in SDL of $`C_\mu `$ with the disorder of noninteracting spins is incorrect.
* SDL mentions “thermodynamic depth” as a complexity measure with a convex dependence on disorder, whereas Crutchfield and Shalizi have shown that it is an increasing function of disorder.
The comment of BP is more specific: $`\mathrm{\Gamma }_{\alpha \beta }`$ does not capture all aspects of complexity; in particular, “it does not describe the transition from regular to indexed languages observed at the period-doubling accumulation points of quadratic maps.”
We welcome the two comments, as well as the opportunity to respond to them and clarify the work presented in SDL. Let us first note that we have only a limited interest in terminology, and if someone does not like our use of the word “complexity” for the expression defined in eq. 1, let them call it the “lambda-function” or invent another term. The important thing is to have a clear definition of the terms one is using. For this reason, we will mostly refrain from calling $`\mathrm{\Gamma }_{\alpha \beta }`$ “complexity” in this response.
* CFS have understood $`S_{max}`$ to always be the equilibrium entropy, from which it follows that $`\mathrm{\Delta }`$ and $`\mathrm{\Gamma }_{\alpha \beta }`$ vanish for all equilibrium systems. This is a misinterpretation of SDL, perhaps due to our choice of a nonequilibrium system to illustrate the case where the entropy of the equiprobable distribution may not be the appropriate $`S_{max}`$ and our statement that one can interpret $`\mathrm{\Gamma }_{\alpha \beta }`$ as the product of “order” and “distance from equilibrium”. We did not write that “$`S_{max}`$ is taken to be the equilibrium entropy of the system …for all …systems.” . Neither “disorder” $`\mathrm{\Delta }`$ nor $`\mathrm{\Gamma }_{\alpha \beta }`$ is restricted to this interpretation. A perusal of our other work will yield examples additional to those in SDL where $`S_{max}`$ is not the equilibrium entropy of a nonequilibrium system.
In fact, it is possible to have more than one $`S_{max}`$, depending on the question(s) being addressed.
+ Take the entropy of the universe as largely due to the black body radiation background. The maximum conceivable entropy can be constructed by taking all the matter in the universe to make one black hole, yielding a very small “disorder” (see, e.g. ). In a sense that is an ultimate equilibrium entropy.
+ The absolute maximum entropy possible is usually taken to be the that of the equiprobable distribution.
+ For a nonequilibrium system, one could take the entropy of the equilibrium system with the same number of particles, total energy, …as the maximum entropy .
+ The one-dimensional Ising system (two-state spins, only nearest neighbor interactions) provides a simple example where different $`S_{max}`$’s are appropriate for answering different questions. The entropy is a function of the interaction parameter $`J`$, the external field $`B`$ and temperature $`T`$: $`S(B,J,T)`$. The case of vanishing external field and vanishing interaction yields the equiprobable distribution and the absolute maximum entropy: $`S(B=0,J=0,T)`$ . The absolute “disorder”, that with reference to $`S(B=0,J=0,T)`$, is then
$$\mathrm{\Delta }=S(B,J,T)/S(B=0,J=0,T).$$
(3)
How much of the reduction of $`S(B,J,T)`$ compared to $`S(B=0,J=0,T)`$ is due to the interaction between spins? To answer this question we find the maximum of $`S(B,J,T)`$ with respect to $`J`$ under the condition of constant net magnetization $`M`$ (since, even for the paramagnet, the entropy varies with $`M`$). As expected, the entropy is maximum in the case of vanishing interaction, $`J=0`$. Thus, the maximum entropy under the constraint of constant net magnetization is $`S(B_0,J=0,T)`$, where $`B_0`$ is the value of the external field such that $`M(B_0,J=0,T)=M(B,J,T)`$. We now have a second “disorder”:
$$\mathrm{\Delta }_{J=0}=S(B_0,J=0,T)/S(B=0,J=0,T).$$
(4)
This is the absolute “disorder” of the paramagnet with the same net magnetization as the Ising system with nonvanishing $`J`$. Since we have three entropies, $`S(B,J,T)S(B_0,J=0,T)S(B=0,J=0,T)`$, we can introduce a third “disorder”:
$$\widehat{\mathrm{\Delta }}=S(B,J,T)/S(B_0,J=0,T),$$
(5)
which is the “disorder” of the Ising system with respect to the paramagnet with the same net magnetization. The three “disorders” are related by $`\mathrm{\Delta }=\widehat{\mathrm{\Delta }}\mathrm{\Delta }_{J=0}`$.
The point is that there are many possible $`S_{max}`$’s, and therefore “disorders”, “orders” and “complexities” $`\mathrm{\Gamma }_{\alpha \beta }`$, even for equilibrium systems. Which one(s) are appropriate depends on the question(s) being addressed. It is not in general true that $`\mathrm{\Delta }`$ is identically $`1`$ for equilibrium systems; therefore neither “order” nor “complexity” must vanish at equilibrium. When CFS write that as a consequence of $`S_{max}`$’s being taken as the equilibrium entropy (for nonequilibrium systems) neither $`\mathrm{\Delta }`$ nor $`\mathrm{\Gamma }_{\alpha \beta }`$ can “distinguish between two-dimensional Ising systems at low temperature, high temperature, or the critical temperature …\[n\]or …between the many different kinds of organization observed in equilibrium”, this is an overly restrictive interpretation of $`\mathrm{\Delta }`$ and $`\mathrm{\Gamma }_{\alpha \beta }`$.
For a paramagnet “pumped out of equilibrium”, one could interpret $`S`$ and $`S_{max}`$ as the nonequilibrium entropy and the equilibrium entropy *under the appropriate constraints* , respectively. However, here again, CFS misinterpret our work. They wish to argue that since the pumped state is out of equilibrium, we would assign a nonzero level of complexity to this state. This is not true for this simple case of a paramagnet, for which the entropy can be written simply in terms of the total number of spins and the net magnetization. The key is to realize that the appropriate constraints here are just the total number of spins and the net magnetization; otherwise the nonequilibrium entropy could be greater than the equilibrium entropy. Since the total number of spins and the net magnetization must then be the same for the equilibrium and the nonequilibrium case, the entropies are the same, and we have maximum “disorder” and vanishing $`\mathrm{\Gamma }_{\alpha \beta }`$. Incidentally, we have never maintained that $`1\mathrm{\Delta }`$, where $`\mathrm{\Delta }`$ is the ratio of a nonequilibrium entropy to the corresponding equilibrium entropy, could distinguish different equilibrium distributions. To do this, one needs some of the various equilibrium disorders, as pointed out above.
* CFS write that $`\mathrm{\Gamma }_{\alpha \beta }`$ is “over-universal in the sense that it has the same dependence on disorder for structurally distinct systems.” We assume they mean that $`\mathrm{\Gamma }_{\alpha \beta }`$ always has the same dependence on “disorder” (given $`\alpha `$ and $`\beta `$). We agree with this as far as it goes; it follows from the definition of $`\mathrm{\Gamma }_{\alpha \beta }`$. It is rather superficial though, and the question arises as to which $`\mathrm{\Gamma }_{\alpha \beta }`$ and which “disorder” are meant. If $`\mathrm{\Gamma }_{\alpha \beta }`$ is calculated from one “disorder” and its dependence on another “disorder” investigated, $`\mathrm{\Gamma }_{\alpha \beta }`$ may well have a variable dependence on “disorder.” In Fig. 4 of SDL, we investigated $`\mathrm{\Gamma }_{11,J=0}`$ as a function of $`\mathrm{\Delta }`$ for one-dimensional Ising systems. This relation varies with $`J`$. Thus, it is not generally true that $`\mathrm{\Gamma }_{\alpha \beta }`$ “has the same dependence on disorder for structurally distinct systems.” One has to be careful to clearly state which “disorder” and which “complexity” one is dealing with. If one does so, then $`\mathrm{\Gamma }_{\alpha \beta }`$ may show different dependencies on “disorder” for “structurally distinct systems” and is not “over-universal” in the sense used here.
Our calculation of $`\mathrm{\Gamma }_{11,J=0}`$ as a function of $`\mathrm{\Delta }`$ is analogous to Crutchfield and Feldman’s calculation of “statistical complexity” $`C_\mu `$ and “excess entropy” $`E`$, or “effective measure complexity” , again for one-dimensional Ising systems. In our interpretation of their results, they found $`C_\mu `$, to within a multiplicative constant, to be $`\mathrm{\Delta }_{J=0}`$ (we will return to this point below), and $`E`$, again to within a multiplicative constant, to be $`\mathrm{\Delta }_{J=0}\mathrm{\Delta }`$. They then investigated the dependence of $`C_\mu `$ and $`E`$ on $`\mathrm{\Delta }`$ and found that these dependencies vary with $`J`$. From an “order”-“disorder” point of view, what they have investigated is the dependence of “disorder” under one set of conditions ($`J=0`$) on “disorder” under another set of conditions ($`J0`$), or in the case of $`E`$, the difference between these two “disorders” on one of the “disorders”.
* In SDL we identified $`C_\mu `$ for the one-dimensional Ising system, to within a multiplicative constant, with the “disorder” of that system in the absence of interactions between spins. CFS object to this on two grounds, the first of which is dimensional inconsistency. This is not the place to get into a discussion of the proper units for entropy; let us just reiterate – to within a multiplicative constant. More seriously, CFS maintain that we “conflate” the definition of $`C_\mu `$ with eq. (8) of , which is only valid within a limited range. Actually, we were not referring to that equation to identify $`C_\mu `$, but rather to identify the excess entropy $`E`$. Be that as it may, our identification of $`C_\mu `$ with $`\mathrm{\Delta }_{J=0}`$ in SDL is restricted to one-dimensional spin systems, which is what they treat in and we treat in SDL. We were not “conflating”, but referring to their results for one-dimensional spin systems. They disagree with this, too, when they say that although $`C_\mu =H(1)`$, $`H(1)`$ is not the same as the entropy of noninteracting spins. However, on pg. 1240 of they write: “For a NN system, Eq. (6) is equivalent to saying that $`C_\mu =H(1)`$, the entropy associated with the value of one spin.” Earlier, pg. 1239, they used the phrase “isolated-spin uncertainty $`H(1)`$”. Since an isolated spin can not be subject to interactions with neighboring spins, to our reading, they themselves have stated that $`H(1)`$ is the entropy of a spin subject to no interactions, and thus, to within a multiplicative constant, just $`\mathrm{\Delta }_{J=0}`$. (Note that according to the identification of $`C_\mu `$ with $`H(1)`$ breaks down for the paramagnetic case and the high temperature limit; we exclude these cases, too, of course.)
* Does “thermodynamic depth” show a convex dependence on “disorder”, as we stated in SDL, or does it increase monotonically with “disorder” ? Our statement was based on the original exposition by Lloyd and Pagels and other discussions (e.g. ). The results of Crutchfield and Shalizi would indeed seem to indicate that thermodynamic depth is a monotonically increasing function of “disorder”, given their insistence that the choice of states to be made should be the “causal states” of “$`ϵ`$ \- machines”. In particular, they object to “judiciously redefining” (pg. 277) the “appropriate set” of macroscopic states. However, we believe the situation may not be so simple. First of all, they write (pg. 278): “It is certainly not desirable to conflate a process’s complexity with the complexity of whatever apparatus connects the process to the variables we happen to have seized upon as handles”. This argument ignores the fact that the only access we have to real systems is through measurement. The situation would seem to be reminiscent of the endo-exophysics distinction (see, e.g., ), at least superficially. Crutchfield and Shalizi take more of a endophysical point of view, while it would seem that Lloyd and Pagels take a more exophysical approach. Along a similar vein, the argument of Crutchfield and Shalizi seems to ignore the problem of frames of reference. For example, Andresen and Gordon and Spirkl and Ries have shown that a necessary condition for minimum entropy production in a continuous time system is a constant rate of entropy production in eigen time. This can be described as the instantaneous internal time scale of the system and is different from clock or wall time except for linear systems. In other words, for nonlinear systems the rate of entropy production will not be constant for an external observer, but only to the (nonlinear) system as it sees itself. In any case, “thermodynamic depth” would seem to be a convex measure of “complexity” in some cases. “Back of the envelope” calculations of “thermodynamic depth”, taken as the difference between a coarse-grained entropy and a fine-grained entropy , for a one-dimensional Ising ferromagnet indicate a convex dependence on “disorder”. However, this is not the case for an antiferromagnet with sufficiently negative $`J`$. Thus, “thermodynamic depth” may not qualify as either a convex or a monotonic complexity measure, or it may be either depending on the particulars of the system being investigated.
* We agree with Binder and Perry (BP) that results obtained on the basis of $`\mathrm{\Gamma }_{\alpha \beta }`$ should be carefully interpreted and complemented with results based on other measures, if possible. However, we are not so sure that all of their statements are completely accurate. For example, they argue that “effective measure complexity” will “\[c\]ertainly …pick up the non-regularity of a language”, but that our measure will not in the logistic map. However, a comparison of Fig. 10 of and Fig. 3 of SDL show that, as stated in SDL, “major maxima as well as less major ones occur at the same values of $`r`$”, although the relative values of the maxima differ.
Perhaps the main point of contention is that BP desire a“complexity” measure which can become infinite, whereas we purposefully constructed the measure in SDL so that it would not have this property (for $`\alpha ,\beta 0`$). Our reasoning is similar to that which argues that $`S/S_{max}`$ is, for certain purposes, a “better” measure for“disorder” than is the entropy $`S`$ .
It would also seem that BP, like CFS, may have taken the definition of $`\mathrm{\Gamma }_{\alpha \beta }`$ too literally in that they may not have realized that there may be several different $`S_{max}`$’s, and therefore $`\mathrm{\Delta }`$’s, and therefore $`\mathrm{\Gamma }_{\alpha \beta }`$’s for a given system.
Nonetheless, we would like to reiterate that we stated only that $`\mathrm{\Gamma }_{11}`$ behaves similarly to “effective measure complexity” for the logistic map, that it was not clear to us why this is so, and we do not know the breadth of systems for which this will be the case.
There is a plethora of proposed complexity measures in the literature, all trying to capture some aspects of what we mean when we say that something is complex. Among ones that we ourselves find attractive are “thermodynamic depth”, “statistical complexity” and “effective measure complexity”. However, none of these nor any of the others capture all aspects of “complexity”. This is made explicit by the statement by CFS “that a useful role for statistical complexity measures is to capture the structure – patterns, organization, regularities, symmetries – intrinsic to a process”. We have nothing against this statement, unless one interprets “a useful role” as “ the only useful role”, or one means that a measure of complexity must be a statistical complexity measure. There are many useful roles for complexity measures. Perhaps at some time a consensus will arise; but until that time, we believe that there is a need for various approaches to complexity.
The situation becomes even more confused, when one realizes that even seemingly “exact” measures such as “statistical complexity” and “effective measure complexity” are not uniquely defined: “For higher dimensional systems, e.g., spins in 2D, there are several ways to define $`E`$ and $`C_\mu `$.” (pg. R1242) Thus, we believe there is a place for simple measures of complexity. The great advantage of $`\mathrm{\Gamma }_{\alpha \beta }`$, as noted in SDL, is that it is available for systems where much less information is available than is necessary to calculate some other measures, such as $`E`$ and $`C_\mu `$.
We do not claim any “universality” for $`\mathrm{\Gamma }_{\alpha \beta }`$ though, and think that one should examine several possible complexity measures to get a handle on the various things which can be meant by saying a system or a process is complex.
This work was supported in part by grant no. 31-42069.94 from the Swiss National Science Foundation. |
warning/0003/nlin0003002.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Long ago, in his famous papers R.Baxter has introduced the object, which is known now as $`Q`$-operator. This operator was used initially for the solution of the eigenvalue problem of $`XYZ`$-spin chain, where usual Bethe ansatz fails. Recently this operator was intensively discussed in the series of papers in the connection with continuous quantum field theory. In it was pointed out the relation of $`Q`$-operator with quantum Bäklund transformations. In we suggested the construction of the one-parametirc family of $`Q`$-operators for the most difficult case of isotropic Heisenberg spin chain. (In spite of the obvious simplicity of this model, the original Baxter construction fails here.)
The existence of the one-parametric family of $`Q`$-operators implies the existence of two basic solutions of Baxter equation, whose linear combinations ( with operator coefficients ) form the one-parametric family.
In the present paper we extend the investigation started in to the periodic Toda chain, the other model with rational $`R`$-matrix. It turns out that apart from the construction of the one-parametric family of $`Q`$-operators (section 2), in the case of Toda chain it is possible to build also two basic $`Q`$-operators separately (section 3). These basic operators satisfy to the set of the functional wronskian relations (section 5), first established for certain field theoretical model in . On the one hand the wronskian relations imply the linear independence of the basic operators, on the other hand they are the origin for numerous fusion relations for the transfer matrix of the model.
In our approach we construct the basic $`Q`$-operators as the trace of the monodromy of certain $`M_n^{(1,2)}(x)`$ operators (section 3). It turns out that these operators also permit us to construct the quantum Bloch functions, the basis of the solutions of the quantum linear problem, which are the eigenvectors of the monodromy matrix (section 6).
The defining relation of the $`Q`$-operator (Baxter equation) for the models with rational $`R`$-matrix looks as follows:
$$t(x)Q(x)=a(x)Q(x+i)+b(x)Q(xi),$$
(1)
where $`t(x)`$ is the corresponding transfer matrix and $`a(x)`$ and $`b(x)`$ are the c-number functions which enter into factorization of quantum determinant of $`t(x)`$. In case of Toda chain the quantum determinant is unity, therefore we can choose the normalization $`a(x)=b(x)=1`$, which we shall use below.
## 2 Toda Chain
The periodic Toda Chain is the quantum system described by the Hamiltonian
$$H=\underset{i=1}{\overset{N}{}}\left(p_i^2/2+\mathrm{exp}(q_{i+1}q_i)\right),$$
(2)
where the canonical variables $`p_i,q_i`$ satisfy commutation relations
$$[p_i,q_j]=i\delta _{ij}$$
(3)
and periodic boundary conditions
$`p_{i+N}`$ $`=`$ $`p_i`$
$`q_{i+N}`$ $`=`$ $`q_i`$ (4)
Following Sklyanin we introduce Lax operator in $`2`$-dimensional auxiliary space as follows:
$`L_n(x)=\left(\begin{array}{cc}xp_n& e^{q_n}\\ e^{q_n}& 0\end{array}\right),`$ (7)
where x is the spectral parameter. The fundamental commutation relations for Lax operator could be written in $`R`$-matrix form:
$$R_{12}(xy)L_n^1(x)L_n^2(y)=L_n^2(y)L_n^1(x)R_{12}(xy),$$
(8)
where indexes $`1,2`$ indicate different auxiliary spaces and $`R`$-matrix is given by
$$R_{12}(x)=x+iP_{12},$$
(9)
where $`P`$-is the operator of permutation of the auxiliary spaces. The same intertwining relation also holds true and for the monodromy matrix corresponding to the $`L`$-operator (5):
$$T_{ij}(x)=\left(\underset{1}{\overset{N}{}}L_n(x)\right)_{ij},$$
(10)
where the multipliers of the product is ordered from the right to the left.
The $`Q(x)`$-operator we are going to construct will be given as the trace of the monodromy $`\widehat{Q}(x)`$ appropriate operators $`M_n(x)`$, which acts in n-th quantum space and its auxiliary space, which we will choose to be the representation space $`\mathrm{\Gamma }`$ of the algebra:
$$[\rho _i,\rho _j^+]=\delta _{ij},i,j=1,2$$
(11)
The operator $`\widehat{Q}(x)`$ will be given by the ordered product:
$$\widehat{Q}(x)=\underset{n=1}{\overset{N}{}}M_n(x),$$
(12)
Further we shall need to consider the product $`\left(L_n(x)\right)_{ij}M_n(x)`$, which acts in the auxiliary space $`\mathrm{\Gamma }\times C^2`$ ($`\mathrm{\Gamma }`$ \- for $`M_n(x)`$ and $`C^2`$ \- is two-dimensional auxiliary space for $`L_n(x)`$). In this space it is convenient to consider a pair of projectors $`\mathrm{\Pi }_{ij}^\pm `$:
$`\mathrm{\Pi }_{ij}^+`$ $`=`$ $`(\rho ^+\rho +1)^1\rho _i\rho _j^+=\rho _i\rho _j^+(\rho ^+\rho +1)^1,`$
$`\mathrm{\Pi }_{ij}^{}`$ $`=`$ $`(\rho ^+\rho +1)^1ϵ_{il}\rho _l^+ϵ_{jm}\rho _m=ϵ_{il}\rho _l^+ϵ_{jm}\rho _m(\rho ^+\rho +1)^1,`$ (13)
where
$`\rho ^+\rho `$ $`=`$ $`\rho _i^+\rho _i`$
$`ϵ_{ij}`$ $`=`$ $`ϵ_{ji},ϵ_{12}=1.`$ (14)
These projectors formally satisfy the following relations:
$`\mathrm{\Pi }_{ik}^\pm \mathrm{\Pi }_{kj}^\pm `$ $`=`$ $`\mathrm{\Pi }_{ij}^\pm ,`$
$`\mathrm{\Pi }_{ik}^+\mathrm{\Pi }_{kj}^{}`$ $`=`$ $`0,`$
$`\mathrm{\Pi }_{ij}^++\mathrm{\Pi }_{ij}^{}`$ $`=`$ $`\delta _{ij}.`$ (15)
Rigorously speaking the r.h.s. of the first equation (13) in the Fock representation has an extra term, proportional to the projector on the vacuum state, but, as we shall see below, this term is irrelevant in the present discussion.
In order to define $`Q`$-operator which satisfies Baxter equation we shall exploit Baxter’s idea , which we reformulate as following: $`M_n(x)`$-operator should satisfies the relation:
$$\mathrm{\Pi }_{ij}^{}\left(L_n(x)\right)_{jl}M_n(x)\mathrm{\Pi }_{lk}^+=0.$$
(16)
If this condition is fulfilled, then
$`\left(L_n(x)\right)_{ij}M_n(x)`$ $`=`$ $`\mathrm{\Pi }_{ik}^+\left(M_n(x)\right)_{kl}M_n(x)\mathrm{\Pi }_{lj}^++`$
$`\mathrm{\Pi }_{ik}^{}\left(L(x)_n\right)_{kl}M_n(x)\mathrm{\Pi }_{lj}^{}`$ $`+`$ $`\mathrm{\Pi }_{ik}^+\left(L_n(x)\right)_{kl}M_n(x)\mathrm{\Pi }_{lj}^{}.`$ (17)
In other words, the condition (14) guaranties that the r.h.s. of (15) in the sense of projectors $`\mathrm{\Pi }^\pm `$ has the triangle form and this form will be conserved for products over $`n`$ due to orthogonality of the projectors.
From (14) we obtain
$$ϵ_{jm}\rho _m\left(L_n(x)\right)_{jk}M_n(x)\rho _k=0.$$
(18)
To satisfy this equation it is sufficient if
$$M_n(x)\rho _k=\left(L_n^1(x)\right)_{kl}\rho _lA_n(x)$$
(19)
or
$$ϵ_{jm}\rho _m\left(L_n(x)\right)_{jk}M_n(x)=B_n(x)ϵ_{kl}\rho _l,$$
(20)
where $`A_n(x)`$ and $`B_n(x)`$ are some operators which we shall find now. Note that the operator $`L_n^1(x)`$ is given by
$`L_n^1(x)=\left(\begin{array}{cc}0& e^{q_n}\\ e^{q_n}& xip_n\end{array}\right),`$ (23)
The equation (18) could be rewritten in the following form:
$$\left(L_n^1(x+i)\right)_{jk}\rho _kM_n(x)=B_n(x)\rho _j$$
(24)
Comparing the equations (17) and (20) we conclude that they both are satisfied provided
$`A_n(x)`$ $`=`$ $`M_n(xi),`$
$`B_n(x)`$ $`=`$ $`M_n(x+i).`$ (25)
In such a way we obtain the following equation for the $`M(x)`$-operator:
$$\left(L_n^1(x+i)\right)_{jk}\rho _kM_n(x)=M_n(x+i)\rho _j.$$
(26)
If the operator $`M_n(x)`$ satisfies this equation, the product $`L_n(x)M_n(x)`$ takes the following form:
$`\left(L_n(x)\right)_{ij}M_n(x)=\rho _iM_n(xi)\rho _j^+(\rho ^+\rho +1)^1`$ (27)
$`+(\rho ^+\rho +1)^1ϵ_{il}\rho _l^+M_n(x+i)ϵ_{jm}\rho _m+\mathrm{\Pi }_{ik}^+\left(L_n(x)\right)_{kl}M_n(x)\mathrm{\Pi }_{lj}^{}`$ .
We do not detail the last term in (23) because, due to triangle structure of it r.h.s. this term will not enter into the trace of $`\widehat{Q}(x)`$.
Now our task is to solve the equation for $`M_n(x)`$-operator. The detailed investigation of the equation (22) shows that the usual Fock representation for (9) does not fit for our purpose, therefore we shall use less restrictive holomorphic representation.
Let the operator $`\rho _i^+`$ be the operator of multiplication by the $`\alpha _i`$, while the operator $`\rho _i`$-the operator of differentiation with respect to $`\alpha _i`$:
$`\rho _i^+\psi (\alpha )`$ $`=`$ $`\alpha _i\psi (\alpha ),`$
$`\rho _i\psi (\alpha )`$ $`=`$ $`{\displaystyle \frac{}{\alpha }}\psi (\alpha ).`$ (28)
The operators $`\rho _i^+,\rho _i`$ are canonically conjugated for the scalar product:
$$(\psi ,\varphi )=\frac{_{i=1,2}d\alpha _id\overline{\alpha }_i}{(2\pi i)^2}e^{\alpha \overline{\alpha }}\overline{\psi }(\alpha )\varphi (\alpha )$$
(29)
The action of an operator in holomorphic representation is defined by its kernel:
$$\left(K\psi \right)(\alpha )=d^2\mu (\beta )K(\alpha ,\overline{\beta })\psi (\beta ),$$
(30)
where we have denoted
$$d^2\mu (\beta )=\frac{_{i=1,2}d\beta _id\overline{\beta }_i}{(2\pi i)^2}.$$
(31)
Now we are ready to make the following
Statement The kernel $`M_n(x,\alpha ,\overline{\beta })`$ of the operator $`M_n(x)`$ in holomorphic representation has the following form:
$$M_n(x,\alpha ,\overline{\beta })=m_n(x)\frac{(\alpha \overline{\beta })^{2l+ix}}{\mathrm{\Gamma }(2l+ix+1)},$$
(32)
where $`l`$ is arbitrary parameter and the operator $`m_n(x)`$ is given by
$`m_n(x)=\mathrm{exp}[\pi /2(\rho _1^+\rho _2e^{q_n}\rho _2^+\rho _1e^{q_n}](1+i\rho _2^+\rho _1e^{q_n})^{i(p_nx)+\rho _1^+\rho _1}`$
$`=(1i\rho _1^+\rho _2e^{q_n})^{i(p_nx)+\rho _1^+\rho _1}\mathrm{exp}[\pi /2(\rho _1^+\rho _2e^{q_n}\rho _2^+\rho _1e^{q_n}]`$ (33)
In (28) the operator $`m_n(x)`$ acts on the argument $`\alpha `$ of the function $`(\alpha \overline{\beta })^{2l+ix}`$ according to (24). The proof of the Statement is straightforward by direct substitution of (28) into equation (22). This calculation give us also the by-product – the meaning of the operator $`m_n(x)`$. Apparently this operator commutes with the operator
$$\widehat{l}=\frac{1}{2}(\rho _1^+\rho _1+\rho _2^+\rho _2).$$
(34)
If we shall fix the subspace of $`\mathrm{\Gamma }`$ corresponding to the definite eigenvalue $`l`$ of the operator $`\widehat{l}`$ then the operator $`m_n(xi(l+1/2))`$ becomes Lax operator of Toda chain with auxiliary space, corresponding to the spin $`l`$. In particular, the operator (5) corresponds to $`l=1/2`$. Generally speaking, the $`m_n(xi(l+1/2))`$ represents Lax operator of Toda chain in the auxiliary space $`\mathrm{\Gamma }`$. This statement could be proved by intertwining of operator (5) with $`m_n(xi(l+1/2))`$.
Now, taking the ordered product of the $`M_n(x)`$ operators we shall obtain the operator $`\widehat{Q}(x,l)`$ whose kernel is given by
$`\widehat{Q}(x,l,\alpha ,\overline{\beta })`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N1}{}}d^2\mu (\gamma _i)M_N(x,l,\alpha ,\overline{\gamma }_{N1})M_{N1}(x,l,\gamma _{N1},\overline{\gamma }_{N2})}`$
$`\mathrm{}`$ $`\times `$ $`M_2(x,l,\gamma _2,\overline{\gamma }_1)M_1(x,l,\gamma _1,\overline{\beta }).`$ (35)
Due to triangle (in the sense of projectors $`\mathrm{\Pi }^\pm `$ ) structure of the r.h.s. of (23) we obtain the following rule of multiplication of the monodromy matrix $`T(x)`$ on operator $`\widehat{Q}(x)`$:
$`\left(T(x)\right)_{ij}\widehat{Q}(x,l,\alpha ,\overline{\beta })=(x+{\displaystyle \frac{i}{2}})^N\rho _i\widehat{Q}(xi,l,\alpha ,\overline{\beta })\rho _j^+(\rho ^+\rho +1)^1`$
$`(x{\displaystyle \frac{i}{2}})^N(\rho ^+\rho +1)^1ϵ_{im}\rho _m^+\widehat{Q}(x+i,l,\alpha ,\overline{\beta })ϵ_{jk}\rho _k+\mathrm{\Pi }_{im}^+\left(\mathrm{}\right)_{mk}\mathrm{\Pi }_{kj}^{},`$ (36)
where we omitted the explicit expression of the last term by obvious reason.
To proceed further we need to remind the definition of trace of an operator in holomorphic representation. If the operator is given by its kernel $`F(\alpha ,\overline{\beta })`$ then, (see e.g. )
$$TrF=d^2\mu (\alpha )F(\alpha ,\overline{\alpha }),$$
(37)
where the measure was defined in (27). Now we can perform the trace operation for both sides of (32 )over the holomorphic variables and over $`i,j`$ indexes, corresponding to the auxiliary $`2`$-dimensional space of $`T(x)`$. The result is the desired Baxter equation:
$$t(x)Q(x,l)=Q(xi,l)+Q(x+i,l),$$
(38)
where, according to (33)
$$Q(x,l)=d^2\mu (\alpha )\widehat{Q}(x,l,\alpha ,\overline{\alpha }).$$
(39)
Note, that the trace of $`\widehat{Q}`$ exists due to the exponential factor in holomorphic measure (27) and has cyclic property, therefore $`Q(x,l)`$ is invariant under cyclic permutation of the quantum variables. Acting as above we can also consider right multiplication $`M_n(x)L_n(x)`$ to obtain
$$Q(x,l)t(x)=Q(xi,l)+Q(x+i,l).$$
(40)
We shall not consider here the derivations of the intertwining relations for $`\widehat{Q}(x,l)`$ for different values of $`x`$ and $`l`$ and for $`\widehat{Q}(x,l)`$ and $`T_{ij}(y)`$. This may be done in the same way as in and these relations imply the following commutation relations:
$`[Q(x,l),Q(y,m)]`$ $`=`$ $`0`$
$`[t(x),Q(y,l)]`$ $`=`$ $`0`$ (41)
In such a way we have constructed the family of solutions of the Baxter equation which are parametrized by the parameter $`l`$. We can prove that this family may be considered as a linear combinations of two basic solutions with operator coefficients. Here arises the question - is it possible to construct these basic operators separately. The answer is positive and now we shall show how our procedure should be modified in this case.
## 3 Basic $`Q`$-operators for Toda Chain.
As above, we shall look for the $`Q`$-operators in the form of the monodromy of appropriate $`M_n^{(i)}(x)`$-operators, which we now supply with the index $`i=1,2`$ and which act in $`n`$-th quantum space. The auxiliary space $`\mathrm{\Gamma }`$ now will be the representation space of one Heisenberg algebra, instead of (9):
$$[\rho ,\rho ^+]=1.$$
(42)
The product $`(L_n(x))_{ij}M_n^{(i)}(x)`$ is an operator in $`n`$-th quantum space and in auxiliary space which is tensor product $`\mathrm{\Gamma }\times C^2`$. In this auxiliary space we shall introduce new projectors :
$`\mathrm{\Pi }_{ij}^+=\left(\begin{array}{c}1\\ \rho \end{array}\right){\displaystyle \frac{1}{\rho ^+\rho +1}}(1,\rho ^+),`$ (45)
$`\mathrm{\Pi }_{ij}^{}=\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right){\displaystyle \frac{1}{\rho ^+\rho +2}}(\rho ,1)`$ (48)
The defining equations for the operators $`M_n^{(i)}`$ ( the analogies of eq. (14) ) are
$`\mathrm{\Pi }_{ik}^{}\left(L_n(x)\right)_{kl}M_n^{(1)}(x)\mathrm{\Pi }_{lj}^+`$ $`=`$ $`0,`$
$`\mathrm{\Pi }_{ik}^+\left(L_n(x)\right)_{kl}M_n^{(2)}(x)\mathrm{\Pi }_{lj}^{}`$ $`=`$ $`0.`$ (49)
The solutions of these equations we again will present as the kernels of the corresponding operators in holomorphic representation of the algebra (38):
$`M_n^{(1)}(x,\alpha ,\overline{\beta })`$ $`=`$ $`\mathrm{exp}(i\overline{\beta }e^{q_n}){\displaystyle \frac{e^{\pi x/2}}{\mathrm{\Gamma }(i(xp_n)+1)}}\mathrm{exp}(i\alpha e^{q_n}),`$
$`M_n^{(2)}(x,\alpha ,\overline{\beta })`$ $`=`$ $`\mathrm{exp}(i\alpha e^{q_n})e^{\pi x/2}e^{(xp_n)}`$ (50)
$`\times \mathrm{\Gamma }(i(xp_n))\mathrm{exp}(i\overline{\beta }e^{q_n}).`$
For the right multiplication by $`L_n(x)`$ these operators automatically satisfy the following equations:
$`\mathrm{\Pi }_{ik}^+M_n^{(1)}(x)\left(L_n(x)\right)_{kl}\mathrm{\Pi }_{lj}^{}`$ $`=`$ $`0,`$
$`\mathrm{\Pi }_{ik}^{}M_n^{(2)}(x)\left(L_n(x)\right)_{kl}\mathrm{\Pi }_{lj}^+`$ $`=`$ $`0.`$ (51)
The full multiplication rules for the operators $`M_n^i(x)`$ and $`L_n(x)`$ have the following form for left multiplication:
$`\left(L_n(x)\right)_{ij}M_n^{(1)}(x)=\left(\begin{array}{c}1\\ \rho \end{array}\right)_iM_n^{(1)}(xi){\displaystyle \frac{1}{\rho ^+\rho +1}}(1,\rho ^+)_j`$ (54)
$`+`$ $`\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right)_i{\displaystyle \frac{1}{\rho ^+\rho +2}}M_n^{(1)}(x+i)(\rho ,1)_j+\mathrm{\Pi }_{ik}^+\left(L_n(x)\right)_{kl}M_n^{(1)}(x)\mathrm{\Pi }_{lj}^{}`$ (60)
$`\left(L_n(x)\right)_{ij}M_n^{(2)}(x)=\left(\begin{array}{c}1\\ \rho \end{array}\right)_i{\displaystyle \frac{1}{\rho ^+\rho +1}}M_n^{(2)}(x+i)(1,\rho ^+)_j`$
$`+`$ $`\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right)_iM_n^{(2)}(xi){\displaystyle \frac{1}{\rho ^+\rho +2}}(\rho ,1)_j+\mathrm{\Pi }_{ik}^{}\left(L_n(x)\right)_{kl}M_n^{(2)}(x)\mathrm{\Pi }_{lj}^+`$ (63)
and for right multiplication:
$`M_n^{(1)}(x)\left(L_n(x)\right)_{ij}=\left(\begin{array}{c}1\\ \rho \end{array}\right)_i{\displaystyle \frac{1}{\rho ^+\rho +1}}M_n^{(1)}(xi)(1,\rho ^+)_j`$ (66)
$`+`$ $`\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right)_iM_n^{(1)}(x+i){\displaystyle \frac{1}{\rho ^+\rho +2}}(\rho ,1)_j+\mathrm{\Pi }_{ik}^{}\left(L_n(x)\right)_{kl}M_n^{(1)}(x)\mathrm{\Pi }_{lj}^+`$ (72)
$`M_n^{(2)}(x)\left(L_n(x)\right)_{ij}=\left(\begin{array}{c}1\\ \rho \end{array}\right)_iM_n^{(2)}(x+i){\displaystyle \frac{1}{\rho ^+\rho +1}}(1,\rho ^+)_j`$
$`+`$ $`\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right)_i{\displaystyle \frac{1}{\rho ^+\rho +2}}M_n^{(2)}(xi)(\rho ,1)_j+\mathrm{\Pi }_{ik}^+\left(L_n(x)\right)_{kl}M_n^{(2)}(x)\mathrm{\Pi }_{lj}^{}`$ (75)
These relations guaranty that the traces of the monodromies, corresponding to both operators $`M_n^{(i)}(x)`$ satisfy Baxter equations:
$`t(x)Q^{(i)}(x)`$ $`=`$ $`Q^{(i)}(x+i)+Q^{(i)}(xi)`$
$`Q^{(i)}(x)t(x)`$ $`=`$ $`Q^{(i)}(x+i)+Q^{(i)}(xi)`$ (76)
We shall conclude this section with the calculation of the operators $`Q^i(x)`$ for the simplest case of one quantum degree of freedom. In this case from (33) we easily obtain
$`Q^{(1)}(x)`$ $`=`$ $`{\displaystyle \frac{d\alpha d\overline{\alpha }}{2\pi i}e^{\alpha \overline{\alpha }}M^1(x,\alpha ,\overline{\alpha })}={\displaystyle \underset{n=0}{}}{\displaystyle \frac{e^{\pi x/2}}{n!}}e^{qn}{\displaystyle \frac{1}{\mathrm{\Gamma }(i(xp)+1)}}e^{qn}`$ (77)
$`=`$ $`{\displaystyle \underset{n=0}{}}{\displaystyle \frac{e^{\pi x/2}}{n!\mathrm{\Gamma }(i(xp)+n+1)}}=e^{\pi x/2}I_{i(xp)}(2),`$
where $`I_\nu (x)`$ is the modified Bessel function. The analogues calculations for the second $`Q`$-operator gives:
$`Q^{(2)}(x)`$ $`=`$ $`e^{\pi x/2}{\displaystyle \frac{\pi e^{\pi (xp)}}{\mathrm{sin}\pi i(xp)}}{\displaystyle \underset{n=0}{}}{\displaystyle \frac{1}{n!\mathrm{\Gamma }(i(xp)+n+1)}}`$ (78)
$`=e^{\pi x/2}{\displaystyle \frac{\pi e^{\pi (xp)}}{\mathrm{sin}\pi i(xp)}}I_{i(xp)}(2).`$
These two expressions could be compared with the results of .
## 4 Intertwining Relations.
In this section we shall consider the set of intertwining relations among $`L_n(x)`$-operator and $`M_n^{(i)}(x)`$-operators which will imply the mutual commutativity of transfer matrix and $`Q^{(i)}(x)`$. Let us start with the simplest relation
$$R_{kl}^{(i)}(xy)\left(L_n(x)\right)_{lm}M_n^{(i)}(y)=M_n^{(i)}(y)\left(L_n(x)\right)_{kl}R_{lm}^{(i)}(xy)$$
(79)
From eq. (40) follows that for $`x=y`$ the $`R^{(i)}`$-matrixes become the corresponding projectors - $`\mathrm{\Pi }^{}`$ for $`i=1`$ and $`\mathrm{\Pi }^+`$ for $`i=2`$. Making use of these properties we easily obtain:
$`R_{kl}^{(1)}(xy)=\left(\begin{array}{cc}xy+i\rho ^+\rho & i\rho ^+\\ i\rho & i\end{array}\right)`$ (82)
$`R_{kl}^{(2)}(xy)=\left(\begin{array}{cc}i& i\rho ^+\\ i\rho & xy+i+i\rho ^+\rho \end{array}\right)`$ (85)
Next relation which we shall consider is
$$M_n^{(1)}(x,\rho )M_n^{(2)}(y,\tau )R^{12}(xy)=R^{12}(xy)M_n^{(2)}(y,\tau )M_n^{(1)}(x,\rho ),$$
(86)
where both $`M`$-operators act in different auxiliary spaces $`\mathrm{\Gamma }^{(i)}`$ and mutual quantum space. The $`R`$-matrix acts in the tensor product of auxiliary spaces $`\mathrm{\Gamma }^{(1)}\times \mathrm{\Gamma }^{(2)}`$. In (52) we have denoted the operators which act in the auxiliary space $`\mathrm{\Gamma }^{(1)}`$ as $`\rho ,\rho ^+`$ and operators in $`\mathrm{\Gamma }^{(2)}`$ as $`\tau ,\tau ^+`$. From explicit expressions for $`M`$-operators (41) follows that
$`(\rho +\tau )M_n^{(1)}(x,\rho )M_n^{(2)}(y,\tau )`$ $`=`$ $`0,`$
$`M_n^{(2)}(y,\tau )M_n^{(1)}(x,\rho )(\rho ^++\tau ^+)`$ $`=`$ $`0.`$ (87)
These relations mean that the products of the $`M`$-operators are triangle operators in the $`\mathrm{\Gamma }^{(1)}\times \mathrm{\Gamma }^{(2)}`$ and , as a result the $`R`$-matrix satisfy the following equations:
$`(\rho +\tau )R^{12}(x)=0`$
$`R^{12}(x)(\rho ^++\tau ^+)=0.`$ (88)
The corollary of (54) is that the kernel of $`R`$-matrix in holomorphic representation depends only on one variable:
$$R^{12}(x,\alpha ,\overline{\beta };\gamma ,\overline{\delta })=f(x,(\alpha \gamma )(\overline{\beta }\overline{\delta })),$$
(89)
where the variables $`\alpha ,\overline{\beta }`$ refer to the operators $`\rho ,\rho ^+`$ and variables $`\gamma ,\overline{\delta }`$ to the operators $`\tau ,\tau ^+`$. Taking (55) into account we can write the intertwining relation (52) in holomorphic representation:
$`{\displaystyle 𝑑\mu (\beta ^{})𝑑\mu (\delta ^{})M_n^{(1)}(x,\alpha ,\overline{\beta }^{})M_n^{(2)}(y,\gamma ,\overline{\delta }^{})f(xy,(\beta ^{}\delta ^{})(\overline{\beta }\overline{\delta }))}=`$
$`{\displaystyle 𝑑\mu (\alpha ^{})𝑑\mu (\gamma ^{})f(xy,(\alpha \gamma )(\overline{\alpha }^{}\overline{\gamma }^{}))M_n^{(2)}(y,\gamma ^{},\overline{\delta })M_n^{(1)}(x,\alpha ^{},\overline{\beta })},`$ (90)
where
$$d\mu (\alpha )=\frac{d\alpha d\overline{\alpha }}{2\pi i}e^{\alpha \overline{\alpha }}.$$
(91)
To simplify this equation let us introduce the new external variables:
$`\xi _1`$ $`={\displaystyle \frac{1}{\sqrt{2}}}(\alpha +\gamma ),\xi _1^{}`$ $`={\displaystyle \frac{1}{\sqrt{2}}}(\beta +\delta ),`$
$`\xi _2`$ $`={\displaystyle \frac{1}{\sqrt{2}}}(\alpha \gamma ),\xi _2^{}`$ $`={\displaystyle \frac{1}{\sqrt{2}}}(\beta \delta )`$ (92)
and new integration variables for l.h.s. (r.h.s.) integral:
$`\xi _1^{\prime \prime }`$ $`={\displaystyle \frac{1}{\sqrt{2}}}(\beta ^{}+\delta ^{})`$ $`\left(\xi _1^{\prime \prime }={\displaystyle \frac{1}{\sqrt{2}}}(\alpha ^{}+\gamma ^{})\right)`$
$`\xi _2^{\prime \prime }`$ $`={\displaystyle \frac{1}{\sqrt{2}}}(\beta ^{}\delta ^{})`$ $`\left(\xi _2^{\prime \prime }={\displaystyle \frac{1}{\sqrt{2}}}(\alpha ^{}\gamma ^{})\right).`$ (93)
Apparently, due to the structure of $`M^{(i)}`$-operators and $`R`$-matrix, both sides of (56) depend only on the variables $`\xi _2,\overline{\xi }_2^{}`$ and integration over $`\xi _1^{\prime \prime }`$ becomes trivial, resulting in elimination of these variables in the integrands. Further, representing the function $`f(x,2\xi ^{\prime \prime }\overline{\xi }^{})`$ as
$$f(x,2\xi ^{\prime \prime }\overline{\xi }^{})=\underset{n=0}{}C_n(x)\frac{(2\xi ^{\prime \prime }\overline{\xi }^{})^n}{n!},$$
(94)
we can perform the integration over $`\xi _2^{\prime \prime }`$ and, comparing similar terms in both sides of (56), conclude that
$$C_n(x)=\frac{1}{\mathrm{\Gamma }(ix+n+1)}.$$
(95)
Therefore $`R`$-matrix in (52) has the following form in holomorphic representation
$$R^{12}(x,\alpha ,\overline{\beta };\gamma ,\overline{\delta })=\underset{n=0}{}\frac{\left((\alpha \gamma )(\overline{\beta }\overline{\delta })\right)^n}{n!\mathrm{\Gamma }(ix+n+1)}.$$
(96)
As the operator in the space $`\mathrm{\Gamma }^{(1)}\times \mathrm{\Gamma }^{(2)}`$ the $`R`$-matrix (62) is pathological because its kernel depends only on part of holomorphic variables. In other words it contains the projector $`\pi `$ on the subspace of $`\mathrm{\Gamma }^{(1)}\times \mathrm{\Gamma }^{(2)}`$ which is formed by the functions depending on the difference of variables. This property may be an obstacle in the derivation of the commutativity of $`Q`$-operators from the intertwining relation (52). The situation is saved due to the same pathological nature of the product of $`M`$-operators. Indeed, let us consider the product
$`Q^{(1)}(x)Q^{(2)}(y)=Tr_1{\displaystyle \underset{k=1}{\overset{N}{}}}M_k^{(1)}(x)Tr_2{\displaystyle \underset{k=1}{\overset{N}{}}}M_k^{(2)}(y)=`$
$`=Tr_{1,2}{\displaystyle \underset{k=1}{\overset{N}{}}}M_k^{(1)}(x)M_k^{(2)}(y),`$ (97)
where the indexes $`1,2`$ mark the corresponding auxiliary space. Due to the property (53) we can supply each term $`M_k^{(1)}(x)M_k^{(2)}(y)`$ in the last product with the projector $`\pi `$. The same holds true also for the product of $`Q`$-operators taken in the inverse order. In such a way for the commutativity of $`Q`$-operators we need to consider only the intertwining relations of $`M`$-operators projected onto the space $`\pi \left(\mathrm{\Gamma }^{(1)}\times \mathrm{\Gamma }^{(2)}\right)`$ , where our $`R`$-matrix is well defined.
Next we shall consider the intertwining relation for the $`M^{(1)}`$-operators with different values of spectral parameter:
$$R^{(11)}(xy)M^{(1)}(x,\rho )M^{(1)}(y,\tau )=M^{(1)}(y,\tau )M^{(1)}(x,\rho )R^{(11)}(xy).$$
(98)
As above, the $`R`$-matrix in (64) acts in the space $`\mathrm{\Gamma }^{(1)}\times \mathrm{\Gamma }^{(2)}`$. From explicit expression for $`M^{(1)}`$-operator (41) we obtain:
$$\rho M^{(1)}(x,\rho )=M^{(1)}(x,\rho )ie^q,ie^qM^{(1)}(x,\rho )=M^{(1)}(x,\rho )\rho ^+$$
(99)
These properties of $`M^{(1)}`$-operator imply the following conditions on the $`R`$-matrix:
$$\tau ^+R^{(11)}(x)=R^{(11)}(x)\rho ^+,\rho R^{(11)}(x)=R^{(11)}(x)\tau ,$$
(100)
which could be satisfied if $`R^{(11)}(x)`$ has the following form:
$$R^{(11)}(x)=P_{\rho \tau }g(x,\rho ^+\tau ),$$
(101)
where $`P_{\rho \tau }`$ denotes the operator of permutation of $`\rho \tau `$ variables. Substituting (67) into relation (64) we get the equation for the function $`g`$:
$$g(xy,\rho ^+\tau )M^{(1)}(x,\rho )M^{(1)}(y,\tau )=M^{(1)}(y,\tau )M^{(1)}(x,\rho )g(xy,\rho ^+\tau )$$
(102)
Making use of the explicit form of the $`M^{(1)}`$-operator and the formal power series expansion for function $`g`$ with respect to it second argument we can solve this equation and find the function $`g`$:
$$g(x,\rho ^+\tau )=(1+\rho ^+\tau )^{ix}$$
(103)
and therefore
$$R^{(11)}(x)=P_{\rho \tau }(1+\rho ^+\tau )^{ix}.$$
(104)
As this $`R`$-matrix intertwines two similar objects, it should satisfies the Yang-Baxer equation (and it really does), but we shall not investigate further this issue.
The last relation which we need to discuss is the intertwining of two $`M^{(2)}`$-operators:
$$R^{(22)}(xy)M^{(2)}(x,\rho )M^{(2)}(y,\tau )=M^{(2)}(y,\tau )M^{(2)}(x,\rho )R^{(22)}(xy).$$
(105)
The $`M^{(2)}`$-operators also satisfy the relations analogues to (65):
$$\rho M^{(2)}(x,\rho )=ie^qM^{(2)}(x,\rho ),M^{(2)}(x,\rho )ie^q=M^{(2)}(x,\rho )\rho ^+,$$
(106)
from where we obtain the analogue of (66):
$$\tau ^+R^{(22)}(x)=R^{(22)}(x)\rho ^,\rho ^+R^{(22)}(x)=R^{(22)}(x)\tau ^+,$$
(107)
and therefore $`R^{(22)}`$ has the following form:
$$R^{(22)}(x)=P_{\rho \tau }h(x,\tau ^+\rho ).$$
(108)
Further , acting as above we find that the unknown function $`h`$ does coincide with the function $`g`$, resulting in the following $`R^{(22)}`$-matrix:
$$R^{(22)}(x)=P_{\rho \tau }(1+\tau ^+\rho )^{ix}.$$
(109)
Now we have completed the derivation of all needed intertwining relation The main corollary of these relations is the mutual commutativity of the transfer matrix and both $`Q`$-operators:
$$[t(x),Q^{(i)}(y)]=0,[Q^{(i)}(x),Q^{(j)}(y)]=0,i(j)=1,2.$$
(110)
## 5 Wronskian-type Functional Relations
It was first pointed out in that Baxter equation (1) which defines the $`Q`$ -operator could be viewed as the finite difference analogue of the second order differential equation which admits two independent solution. The linear independence of the solutions could be established through the calculation of the wronskian corresponding to the equation. In the previous section we have constructed two solution of Baxter equation and now our task is to prove its linear independence i.e. to derive the finite difference analogue of the wronskian. To solve this problem let us consider in the details the representation of the product (63) of two different $`Q`$-operators. In the notations of the previous section the product of two $`M`$-operators which enters into the r.h.s. of (63) has the following form:
$`M_k^{(12)}(x,y,\alpha ,\overline{\beta },\gamma ,\overline{\delta })=M_k^{(1)}(x,\alpha ,\overline{\beta })M_k^{(2)}(y,\gamma ,\overline{\delta })=`$ (111)
$`e^{\pi (x+y)/2}e^{i\overline{\beta }e^{q_k}}{\displaystyle \frac{1}{\mathrm{\Gamma }(i(xp_k)+1)}}e^{i(\alpha \gamma )e^{q_k}}e^{\pi (yp_k)}\mathrm{\Gamma }(i(yp_k))e^{i\overline{\delta }e^{q_k}}`$
Changing the holomorphic variables according to (58) we obtain:
$`M_k^{(12)}(x,y,\xi _1,\xi _2,\overline{\xi }_1^{},\overline{\xi }_2^{})=e^{\pi (x+y)/2}e^{i/\sqrt{2}(\overline{\xi }_1^{}+\overline{\xi }_2^{})e^{q_k}}`$
$`{\displaystyle \frac{1}{\mathrm{\Gamma }(i(xp_k)+1)}}e^{i\sqrt{2}\xi _2e^{q_k}}e^{\pi (yp_k)}\mathrm{\Gamma }(i(yp_k))e^{i/\sqrt{2}(\overline{\xi }_1^{}\overline{\xi }_2^{})e^{q_k}}`$ (112)
This equation demonstrates that the kernel of $`M^{(1)}(x)M^{(2)}(y)`$ does not depends on the variable $`\xi _1`$ and for calculation of the $`Q^{(1)}(x)Q^{(2)}(y)`$ the dependence of (78) on the variable $`\overline{\xi }_1^{}`$ is irrelevant because the integration over $`\xi ^{},\overline{\xi }^{}`$ in (63) results in deleting $`\overline{\xi }_1^{}`$ from (78). In such a way for the calculation of $`Q^{(1)}(x)Q^{(2)}(y)`$ we can use instead of $`M_k^{(12)}(x,y,\xi _1,\xi _2,\overline{\xi }_1^{},\overline{\xi }_2^{})`$ the following reduced object:
$`\stackrel{~}{M}_k^{(12)}(x,y,\xi ,\overline{\xi }^{})=e^{\pi (x+y)/2}e^{i/\sqrt{2}\overline{\xi }^{}e^{q_k}}`$
$`{\displaystyle \frac{1}{\mathrm{\Gamma }(i(xp_k)+1)}}e^{i\sqrt{2}\xi e^{q_k}}e^{\pi (yp_k)}\mathrm{\Gamma }(i(yp_k))e^{i/\sqrt{2}\overline{\xi }^{}e^{q_k}}.`$ (113)
Note that $`\stackrel{~}{M}^{(12)}(x,y)`$ is nothing else but the kernel of $`M^{(1)}(x)M^{(2)}(y)`$ on the space $`\pi \left(\mathrm{\Gamma }^{(1)}\times \mathrm{\Gamma }^{(2)}\right)`$. Now let use expand the exponents which contain $`\xi ,\overline{\xi }`$ in the r.h.s. of (79) and move the all the factors depending on $`p_k`$ to the right:
$`\stackrel{~}{M}_k^{(12)}(x,y,\xi ,\overline{\xi }^{})=e^{\pi (x+y)}{\displaystyle \underset{n,m=0}{}}{\displaystyle \frac{(i\sqrt{2}\xi )^n}{n!}}(i\overline{\xi }^{}/\sqrt{2})^me^{(mn)q_k}`$
$`\times {\displaystyle \underset{l=0}{\overset{m}{}}}{\displaystyle \frac{(1)^{ml}}{l!(ml)!}}{\displaystyle \frac{\mathrm{\Gamma }(i(yp_k)m+l)}{\mathrm{\Gamma }(i(xp_k)+1+nm+l)}}e^{\pi (yp_k)}.`$ (114)
The summation over $`l`$ in (80) gives:
$`{\displaystyle \underset{l=0}{\overset{m}{}}}{\displaystyle \frac{(1)^{ml}}{l!(ml)!}}{\displaystyle \frac{\mathrm{\Gamma }(i(yp_k)m+l)}{\mathrm{\Gamma }(i(xp_k)+1+nm+l)}}=`$
$`{\displaystyle \frac{(1)^m}{m!}}{\displaystyle \frac{\mathrm{\Gamma }(i(yp_k)m)}{\mathrm{\Gamma }(i(xp_k)+n+1)}}{\displaystyle \frac{\mathrm{\Gamma }(i(xy)+m+n+1)}{\mathrm{\Gamma }(i(xy)+n+1)}}`$ (115)
and we arrive at the following expression for the $`\stackrel{~}{M}_k^{(12)}(x,y,\xi ,\overline{\xi }^{})`$:
$`\stackrel{~}{M}_k^{(12)}(x,y,\xi ,\overline{\xi }^{})=e^{\pi (x+y)}{\displaystyle \underset{n,m=0}{}}{\displaystyle \frac{(i\sqrt{2}\xi )^n}{n!}}{\displaystyle \frac{(i\overline{\xi }^{}/\sqrt{2})^m}{m!}}e^{(mn)q_k}`$
$`\times {\displaystyle \frac{\mathrm{\Gamma }(i(yp_k)m)}{\mathrm{\Gamma }(i(xp_k)+n+1)}}{\displaystyle \frac{\mathrm{\Gamma }(i(xy)+m+n+1)}{\mathrm{\Gamma }(i(xy)+n+1)}}e^{\pi (yp_k)}.`$ (116)
Now let $`x`$ and $`y`$ be
$`x=z_+=z+i(l+1/2),y=z_{}=zi(l+1/2),`$ (117)
where $`l`$ is an integer (half-integer). For these values of spectral parameters (82) takes the following form:
$`\stackrel{~}{M}_k^{(12)}(z_+,z_{},\xi ,\overline{\xi }^{})=e^{\pi z}{\displaystyle \underset{n,m=0}{}}{\displaystyle \frac{(i\sqrt{2}\xi )^n}{n!}}{\displaystyle \frac{(i\overline{\xi }^{}/\sqrt{2})^m}{m!}}e^{(mn)q_k}`$
$`\times {\displaystyle \frac{\mathrm{\Gamma }(i(z_{}p_k)m)}{\mathrm{\Gamma }(i(z_+p_k)+n+1)}}{\displaystyle \frac{\mathrm{\Gamma }(2l+m+n+2)}{\mathrm{\Gamma }(2l+n+2)}}e^{\pi (z_{}p_k)}.`$ (118)
Further we need to consider (82) for the opposite shift of spectral parameters
$`x=z_{}iϵ,y=z_++iϵ.`$ (119)
We have introduced infinitesimal $`ϵ`$ in (85) to remove an ambiguity which arises in (82) for these $`x`$ and $`y`$:
$`\stackrel{~}{M}_k^{(12)}(z_{},z_+,\xi ,\overline{\xi }^{})=e^{\pi z}{\displaystyle \underset{n,m=0}{}}{\displaystyle \frac{(i\sqrt{2}\xi )^n}{n!}}{\displaystyle \frac{(i\overline{\xi }^{}/\sqrt{2})^m}{m!}}e^{(mn)q_k}`$
$`{\displaystyle \frac{\mathrm{\Gamma }(i(z_+p_k)m)}{\mathrm{\Gamma }(i(z_{}p_k)+n+l)}}{\displaystyle \frac{\mathrm{\Gamma }(2l2ϵ+m+n)}{\mathrm{\Gamma }(2l2ϵ+n)}}e^{\pi (z_+p_k)}.`$ (120)
For $`ϵ0`$ the fraction of $`\mathrm{\Gamma }`$-functions in (86) takes the following values:
$`\underset{ϵ0}{lim}{\displaystyle \frac{\mathrm{\Gamma }(2l2ϵ+m+n)}{\mathrm{\Gamma }(2l2ϵ+n)}}=\{\begin{array}{cc}{\displaystyle \frac{\mathrm{\Gamma }(2l+m+n)}{\mathrm{\Gamma }(2l+n)}},& n,m2l+1,\\ (1)^m{\displaystyle \frac{(2ln)!}{(2lnm)!}},& 2ln+m0,\\ {\displaystyle \frac{\mathrm{\Gamma }(n+m2l)}{\mathrm{\Gamma }(n2l)}},& n2lm\\ 0,& otherwise\end{array}`$ (125)
Apparently, the vanishing of the (87) in the fourth region manifests the triangularity of the operator the $`\stackrel{~}{M}_k^{(12)}(z_{},z_+)`$, therefore for the calculation of the trace of the product over $`k`$ of these operators we need to consider only the part of (87), corresponding to the first two regions. Thus, the resulting expression for the twice reduced operator has the following form:
$`\underset{k}{\overset{(12)}{\stackrel{}{M}}}(z_{},z_+,\xi ,\overline{\xi }^{})=A_k(z,l,\xi ,\overline{\xi }^{})+B_k(z,l,\xi ,\overline{\xi }^{}),`$ (126)
where $`A`$ contains the part of the r.h.s. of (86) with the summation over $`n,m`$ in the region $`n,m2l+1`$, $`B`$ contains the summation over $`n,m`$ in the region $`2ln+m0`$. In other words, the degrees of $`\xi ,\overline{\xi }^{}`$ in $`A`$ and $`B`$ have no intersection and therefore while the calculation of the product $`Q^{(1)}(z_{})Q^{(2)}(z_+)`$ these two parts will multiply coherently:
$`Q^{(1)}(z_{})Q^{(2)}(z_+)={\displaystyle \underset{k=1}{\overset{N}{}}d\mu (\xi _k)}\underset{N}{\overset{(12)}{\stackrel{}{M}}}(z_{},z_+,\xi _1,\overline{\xi }_N)`$
$`\times \underset{N1}{\overset{(12)}{\stackrel{}{M}}}(z_{},z_+,\xi _N,\overline{\xi }_{N1})\mathrm{}\underset{1}{\overset{(12)}{\stackrel{}{M}}}(z_{},z_+,\xi _2,\overline{\xi }_1^{})`$
$`={\displaystyle \underset{k=1}{\overset{N}{}}d\mu (\xi _k)A_N(z,l,\xi _1,\overline{\xi }_N)A_{N1}(z,l,\xi _N,\overline{\xi }_{N1})\mathrm{}A_1(z,l,\xi _2,\overline{\xi }_1)}`$
$`+{\displaystyle \underset{k=1}{\overset{N}{}}d\mu (\xi _k)B_N(z,l,\xi _1,\overline{\xi }_N)B_{N1}(z,l,\xi _N,\overline{\xi }_{N1})\mathrm{}B_1(z,l,\xi _2,\overline{\xi }_1)}`$ (127)
Let us consider first $`A`$. For the convenience we will shift the values of $`n,m`$ by $`2l+1`$, then
$`A_k(z,l,\xi ,\overline{\xi })=(\xi \overline{\xi }^{})^{2l+1}e^{\pi z}{\displaystyle \underset{n,m=0}{}}{\displaystyle \frac{(i\sqrt{2}\xi )^n}{n!}}{\displaystyle \frac{(i\overline{\xi }^{}/\sqrt{2})^m}{(m+2l+1)!}}e^{(mn)q_k}`$
$`\times {\displaystyle \frac{\mathrm{\Gamma }(i(z_{}p_k)m)}{\mathrm{\Gamma }(i(z_+p_k)+n+l)}}{\displaystyle \frac{\mathrm{\Gamma }(2l+m+n+2)}{\mathrm{\Gamma }(2l+n+2)}}e^{\pi (z_{}p_k)}.`$ (128)
Comparing (90) with (84), we see that they differ from each other by the factor $`(\xi \overline{\xi }^{})^{2l+1}`$ and shift of the factorial $`m!`$. This difference may be presented as appropriate transformation of $`\stackrel{~}{M}_k^{(12)}(z_+,z_{},\xi ,\overline{\xi }^{})`$ :
$$A_k(z,l,\xi ,\overline{\xi }^{})=𝑑\mu (\zeta )𝑑\mu (\zeta ^{})g_l(\xi ,\overline{\zeta })\stackrel{~}{M}_k^{(12)}(z_+,z_{},\zeta ,\overline{\zeta }^{})f_l(\zeta ^{},\overline{\xi }^{}),$$
(129)
where
$$g_l(\xi ,\overline{\zeta })=(\xi )^{2l+1}e^{\xi \overline{\zeta }},f_l(\zeta ,\overline{\xi })=(\overline{\xi })^{2l+1}\underset{n=0}{}\frac{(\zeta \overline{\xi })^n}{(n+2l+1)!}.$$
(130)
These two functions possess the following property:
$$𝑑\mu (\xi )f_l(\zeta ,\overline{\xi })g_l(\xi ,\overline{\zeta }^{})=e^{\zeta \overline{\zeta }^{}}$$
(131)
The r.h.s. of (93) is the $`\delta `$-function in holomorphic representation. But note that
$$𝑑\mu (\xi )g_l(\zeta ,\overline{\xi })f_l(\xi ,\overline{\zeta }^{})=\underset{n=2l+1}{}\frac{(\zeta \overline{\zeta }^{})^n}{n!}.$$
(132)
Taking into account (93), we immediately obtain:
$`{\displaystyle \underset{k=1}{\overset{N}{}}d\mu (\xi _k)A_N(z,l,\xi _1,\overline{\xi }_N)A_{N1}(z,l,\xi _N,\overline{\xi }_{N1})\mathrm{}A_1(z,l,\xi _2,\overline{\xi }_1)}`$
$`=Q^{(1)}(z_+)Q^{(2)}(z_{})`$ (133)
Our next step is the consideration of $`B`$ part of the $`M^{(12)}(z_{},z_+)`$. First of all we shall remove the $`\sqrt{2}`$ from its holomorphic arguments, because in the integral (89) these factors will cancelled out. Therefore we need to consider the following expression for $`B`$:
$`B_k(z,l,\xi ,\overline{\xi }^{})=e^{\pi z}{\displaystyle \underset{t=0}{\overset{2l}{}}}{\displaystyle \underset{m=0}{\overset{t}{}}}{\displaystyle \frac{\xi ^{tm}}{(tm)!}}{\displaystyle \frac{\overline{\xi }^m}{m!}}(1)^mi^{t+2l+1}e^{(2mt)q_k}`$
$`\times {\displaystyle \frac{(2l+mt)!}{(2lt)!}}{\displaystyle \frac{\mathrm{\Gamma }(i(zp_k)+lm+1/2)}{\mathrm{\Gamma }(i(zp_k)l+tm+l/2)}}e^{\pi (zp_k)}`$ (134)
We intend to compare this operator with Lax operator $`L_k^l(x)`$ of Toda chain with auxiliary space corresponding to the spin $`l`$. As it follows from the results of the 2-nd section, $`L_k^l(x)`$ could be obtain by the reduction of the operator $`m_k(x)`$ defined in (29) to the subspace corresponding to spin $`l`$. In the holomorphic representation the kernel of $`L_k^l(x)`$ could be easily found using the projection:
$$L_k^l(x,\alpha ,\overline{\beta })=m_k(xi(l+1/2))\frac{(\alpha \overline{\beta })^{2l}}{\mathrm{\Gamma }(2l+1)}.$$
(135)
(Note that here we again use two-component variables $`\alpha _i,\beta _i,i=1,2`$). In (97) the operator $`m_k(x)`$ should be understood as the differential operator, acting on the projection kernel $`\frac{(\alpha \overline{\beta })^{2l}}{\mathrm{\Gamma }(2l+1)}`$. For the calculation of the r.h.s. of the (97) recall that the operator exponential function in (29) is well defined because
$$[i(px)+l_3,\rho _1^+\rho _2e^q]=[i(px)+l_3,\rho _2^+\rho _1e^q]=0,$$
(136)
therefore we can expand the exponential function into formal series and find the action of each term on the projection kernel:
$`m_k(xi(l+1/2)){\displaystyle \frac{(\alpha \overline{\beta })^{2l}}{\mathrm{\Gamma }(2l+1)}}=`$ (137)
$`{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \frac{(1)^n\mathrm{\Gamma }(i(p_kx)+\rho _1^+\rho _1l+\frac{1}{2})}{\mathrm{\Gamma }(i(p_kx)+\rho _1^+\rho _1ln+\frac{1}{2})}}{\displaystyle \frac{(i\rho _1^+\rho _2e^{q_k})^n}{n!}}{\displaystyle \frac{(\alpha _1\overline{\beta }_2e^{q_k}\alpha _2\overline{\beta }_1e^{q_k})^{2l}}{\mathrm{\Gamma }(2l+1)}}.`$
Apparently, only $`2l`$ terms in (99) will survive because the differential operator $`(\rho _2)^n`$ acts on the polynomial. The result has the following form:
$`L_k^l(x,\alpha ,\overline{\beta })={\displaystyle \underset{t=0}{\overset{2l}{}}}{\displaystyle \underset{m=0}{\overset{t}{}}}e^{(2mt)q_k}{\displaystyle \frac{\mathrm{\Gamma }(i(xp_k)m+l+1/2)}{\mathrm{\Gamma }(i(xp_k)m+tl+1/2)}}`$
$`(1)^mi^{2l+t}{\displaystyle \frac{\alpha _1^{2lt+m}\alpha _2^{tm}\overline{\beta }_1^{2lm}\overline{\beta }_2^m}{(2lt)!(tm)!m!}}`$ (138)
This $`L`$-operator defines the transfer matrix of Toda chain with auxiliary space, corresponding spin $`l`$:
$`t^l(x)={\displaystyle \underset{k=1}{\overset{N}{}}d^2\mu (\alpha _k)L_N^l(x,\alpha _1,\overline{\alpha }_N)L_{N1}^l(x,\alpha _N,\overline{\alpha }_{N1})\mathrm{}L_1^l(x,\alpha _2,\overline{\alpha }_1)}`$ (139)
If in this formula we will perform the integration over one pair of the holomorphic variables, corresponding for example $`\alpha _1,\overline{\beta }_1`$ in (100), the integrand still will be presented in the factorized form, but with new, reduced kernel of $`L`$-operator:
$`\stackrel{~}{L}_k^l(x,\alpha ,\overline{\beta })={\displaystyle \underset{t=0}{\overset{2l}{}}}{\displaystyle \underset{m=0}{\overset{t}{}}}e^{(2mt)q_k}{\displaystyle \frac{\mathrm{\Gamma }(i(xp_k)m+l+1/2)}{\mathrm{\Gamma }(i(xp_k)m+tl+1/2)}}`$
$`(1)^mi^{2l+t}{\displaystyle \frac{\alpha _2^{tm}\overline{\beta }_2^m}{(2lt)!(tm)!m!}}(2lt+m)!`$ (140)
Comparing (102) with (96) we find that
$$B_k(z,l,\xi ,\overline{\xi }^{})=\stackrel{~}{L}_k^l(z,\xi ,\overline{\xi })ie^{\pi p_k}.$$
(141)
Therefore
$`{\displaystyle \underset{k=1}{\overset{N}{}}d\mu (\xi _k)B_N(z,l,\xi _1,\overline{\xi }_N)B_{N1}(z,l,\xi _N,\overline{\xi }_{N1})\mathrm{}B_1(z,l,\xi _2,\overline{\xi }_1)}`$
$`=i^Nt^l(x)e^{\pi P},`$ (142)
where
$$P=\underset{k=0}{\overset{N}{}}p_k$$
(143)
is the integral of motion, which commutes with $`t^l(x)`$. In the derivation of (104) we have moved all the factors $`e^{\pi p_k}`$ to the right to form $`e^{\pi P}`$. Gathering together (89), (95) and (104) we obtain the following functional relations:
$`Q^{(1)}(zi(l+{\displaystyle \frac{1}{2}}))Q^{(2)}(z+i(l+{\displaystyle \frac{1}{2}}))Q^{(1)}(z+i(l+{\displaystyle \frac{1}{2}}))Q^{(2)}(zi(l+{\displaystyle \frac{1}{2}}))`$
$`=i^Nt^l(x)e^{\pi P}`$ (144)
For $`l=0`$ the transfer matrix turns into $`1`$ and we have the simplest wronskian relation:
$`Q^{(1)}(zi/2)Q^{(2)}(z+i/2)Q^{(1)}(z+i/2)Q^{(2)}(zi/2)=i^Ne^{\pi P}`$ (145)
For the illustration of this identity the reader can use the $`Q^{(i)}`$-operators for one degree of freedom (47), (48). In this simplest case (107) reduces to the well-known identity for Bessel functions:
$$I_\nu (z)I_{\nu +1}(z)I_\nu (z)I_{\nu 1}(z)=\frac{2\mathrm{sin}(\pi \nu )}{\pi z}$$
(146)
The general case (106) for one degree on freedom is related to Lommel polynomials .
The functional relations of the type (106) was first established for certain field theoretical model in . In the recent paper of the author with Yu.Stroganov we have discussed the analogues relation for the eigenvalues of $`Q`$-operators in the case of isotropic Heisenberg spin chain. Originally, since the Baxter paper the existence of one $`Q`$-operator was considered as important alternative for Bethe ansatz. The relations (106) show the importance of the second $`Q`$-operator which together with the first one give rise to the numerous fusion relations (see e.g.,).
## 6 Discussion
The approach we have considered in the present paper could be applied also to the other with rational $`R`$-matrix – the discrete self-trapping (DST) model, considered in . The quantum determinant of Lax operator for this model is not unity and Baxter equation has the following form:
$$t(x)Q(x)=(xi/2)^NQ(xi)+Q(x+i).$$
(147)
The general properties of the $`Q`$-operators for DST-model are similar to that of Toda system. The eigenvalues of one $`Q`$-operator are polynomial in spectrum parameter , while the eigenvalues of the second are meromorphic functions. In the case of Toda system the eigenvalues of $`Q^{(1)}(x)`$ are entire functions, the eigenvalues of $`Q^{(2)}(x)`$ are meromorphic. For the DST -model there also exist the functional relations similar to (106).
The most interesting would be the application of the formalism to the case of XXX-spin chain. The situation here is the following. In we have constructed the family of $`Q(x,l)`$-operators similar to (31). Moreover, from the results of follows that for XXX-spin chain there exist the basic $`Q`$-operators. Making use of the formalism of the section 3, it is possible to the find the $`M_k^{(i)}(x)`$-operators for this case, but the trace of monodromies corresponding to $`M_k^{(i)}(x)`$ diverges. This puzzle deserves further investigation.
Another interesting point we want to discuss is the relation of our $`M_k^{(i)}(x)`$-operators with quantum linear problem for Lax operator (5). In classical case the linear problem is the main ingredient of inverse scattering method, in the same time for the quantum theory it seems to be unnecessary (see for example excellent review on the subject ). However, let us consider the following problem:
$$\psi _{n+1}(x)=L_n(x)\psi _n(x),$$
(148)
where $`L_n(x)`$ is given in (5) and $`\psi _n`$ is two component quantum operator. From the multiplication rules (43) we obtain:
$`\left(L_n(x)\right)_{ij}M_n^{(1)}(x)\left(\begin{array}{c}1\\ \rho \end{array}\right)_j=\left(\begin{array}{c}1\\ \rho \end{array}\right)_iM_n^{(1)}(x.i.).`$ (153)
Now let us define the operator
$`\left(\psi _n^{(1)}\right)_i(x)=Tr\left({\displaystyle \underset{k=n}{\overset{N}{}}}M_k^{(1)}(x)\left(\begin{array}{c}1\\ \rho \end{array}\right)_i{\displaystyle \underset{k=1}{\overset{n1}{}}}M_k^{(1)}(xi)\right),`$ (156)
where the trace is taken over auxiliary space. Apparently, due to (111) the operator (112) does satisfy the equation (110). For $`n=1`$, the solution has the following form:
$`\left(\psi _1^{(1)}\right)_i(x)=Tr\left({\displaystyle \underset{k=1}{\overset{N}{}}}M_k^{(1)}(x)\left(\begin{array}{c}1\\ \rho \end{array}\right)_i\right)=Q^{(1)}(x)\left(\begin{array}{c}1\\ ie^{q_N}\end{array}\right)_i,`$ (161)
where on the last step we have used the explicit form of $`Q^{(1)}(x)`$-operator for the calculation of the trace. On the other hand the solution (113) translated to the period $`N`$ by the monodromy (8), due to (111) is given by
$`\left(\psi _{N+1}^{(1)}\right)_i(x)=Tr\left(\left(\begin{array}{c}1\\ \rho \end{array}\right)_i{\displaystyle \underset{k=1}{\overset{N}{}}}M_k^{(1)}(xi)\right)=Q^{(1)}(xi)\left(\begin{array}{c}1\\ ie^{q_N}\end{array}\right)_i.`$ (166)
In other words we obtain
$$T_{ij}(x)\left(\psi _1^{(1)}\right)_j(x)=\frac{Q^{(1)}(xi)}{Q^{(1)}(x)}\left(\psi _1^{(1)}\right)_i(x).$$
(167)
This equation may be understood as quantum analogue of the property of Bloch solutions, which are the eigenvectors of the translation to the period.
Similarly we can consider the second solution. Indeed, from the multiplications rules (43) for the $`M_n^{(2)}(x)`$ we obtain:
$`\left(L_n(x)\right)_{ij}M_n^{(2)}(x)\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right)_j=\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right)_iM_n^{(2)}(xi).`$ (172)
Therefore the operator
$`\left(\psi _n^{(2)}\right)_i(x)=Tr\left({\displaystyle \underset{k=n}{\overset{N}{}}}M_k^{(2)}(x)\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right)_i{\displaystyle \underset{k=1}{\overset{n1}{}}}M_k^{(2)}(xi)\right),`$ (175)
possesses the same properties as (112). The initial value of (117) is given by
$`\left(\psi _1^{(2)}\right)_i(x)=Tr\left({\displaystyle \underset{k=1}{\overset{N}{}}}M_k^{(2)}(x)\left(\begin{array}{c}\rho ^+\\ 1\end{array}\right)_i\right)=Q^{(2)}(x)\left(\begin{array}{c}ie^{q_1}\\ 1\end{array}\right)_i,`$ (180)
where we again on the last step have used the explicit form of $`M^{(2)}(x)`$. As above we obtain
$$T_{ij}(x)\left(\psi _1^{(2)}\right)_j(x)=\frac{Q^{(2)}(xi)}{Q^{(2)}(x)}\left(\psi _1^{(2)}\right)_i(x).$$
(181)
In such a way using $`M_n^{(i)}(x)`$-operators we succeeded in the construction of the operators which may be interpreted as the quantum analogues of the Bloch functions of the corresponding linear problem. In the classical theory of finite-zone ”potentials”, two Bloch solutions of the linear problem, as the functions of spectral parameter are actually the projections of the Backer-Akhiezer function, which is the single-valued meromorphic functions on an hyper-elliptic surface. In quantum case the Bloch functions (112) and (117) do not possess the branching points (in weak sense) which is the trace of the projection in the classical case, therefore their intimate relation is somehow hidden and it will be very interesting to uncover this relationship.
## 7 Acknowledgments
The author is grateful to E.Skyanin, S.Sergeev, for their interest and discussions, This work was supported in part by ESPIRIT project NTCONGS and RFFI Grant 98-01-0070. |
warning/0003/physics0003062.html | ar5iv | text | # Limits on the Applicability of Classical Electromagnetic Fields as Inferred from the Radiation Reaction
## I Introduction
The ultimate test of the applicability of a physical theory is the accuracy with which it describes natural phenomena. Yet on occasion the difficulty of a theory in dealing with a “thought experiment” provides a clue as to limitations of that theory.
It has long since been recognized that classical electrodynamics has been surplanted by quantum electrodynamics in some respects. But one doubts that quantum electrodynamics, or even its generalization, the Standard Model of elementary particles, is valid in all domains. To aid in the search for new physics, it is helpful to review the warning signs of the past transitions from one theoretical description to another.
The debates as to the meaning of the classical radiation reaction for pointlike particles provide examples of such warning signs. One case is the “4/3 problem” of electromagnetic mass, where covariance does not imply uniqueness . Such difficulties have often been interpreted as suggesting that classical electrodynamics cannot be a complete description of matter on the scale of the classical electron radius, $`r_0=e^2/mc^2`$ in Gaussian units.
It seems less appreciated that the part of the classical radiation reaction that is independent of particle size provides clues as to the limits of applicability of classical electromagnetic fields. For example, a recent article ends with the sentence, “Only when all distances involved are in the classical domain is classical dynamics acceptable for electrons”. While this condition is necessary, it is not sufficient. For a classical description to be accurate, an electron can only be subject to fields that are not too strong. This paper seeks to illustrate what “not too strong” means.
Considerations of strong fields have been very influential in the development of other modern theories besides quantum electrodynamics. In classical gravity, i.e., general relativity, the strong-field problem is identified with black holes. One of the best known intersections between gravity and quantum electrodynamics is the Hawking radiation of black holes. In the case of the strong (nuclear) interaction, the fields associated with nuclear matter all appear to be strong, and weak fields are thought to exist only in the high-energy limit (asymptotic freedom). Such considerations led to the introduction of non-Abelian gauge theories. These constructs, when applied to the weak interaction, led to the concept of a background (Higgs) field that is strong in the sense of having a large vacuum expectation value, which in turn has the effect of generating the masses of the $`W`$ and $`Z`$ gauge bosons. Most recently, considerations of the strong-field (strong-coupling) limit of string theories have led to the notion of “duality”, i.e., the various string theories of the 1980’s are actually different weak-field limits of a single strong-field theory. These string theories are noteworthy for suggesting that particles are to be considered as excitations of small, but extended quantum strings, thereby avoiding the infinite self energies that have appeared in theories of point particles since J.J. Thomson introduced the concept of electromagnetic mass in 1881 .
The main argument concerning classical electromagnetic fields is given in sec. 2, and is brief. This argument could have been given around 1900 by Lorentz or by Planck . who made remarks of a related nature. But the argument seems to have been first made in 1935 by Oppenheimer , and more explicitly by Landau and Lifshitz . Additional historical commentary is given in sec. 1.1. Sections 2.1-2.5 comment on various aspects of the main argument, still from a classical perspective. A quantum view in introduced in sec. 3, and the important examples of Compton scattering and the QED critical field are discussed in secs. 3.1-2. The paper concludes in sec. 4 with remarks on the role of strong fields on the development of quantum electrodynamics, and presents two examples (secs. 4.1-2) of speculative features of strong-field QED and one of very short distance QED (sec. 4.3).
### A Historical Introduction
The relation between Newton’s third law and electromagnetism has been of concern at least since the investigations of Ampère, who insisted that the force of one current element on another be along their line of centers. See Part IV, Chap. II, especially sec. 527, of Maxwell’s Treatise for a review . However, the presently used differential version of the Biot-Savart law does not satisfy Newton’s third law for pairs of current elements unless they are parallel.
Perhaps discomfort with this fact contributed to the delay in acceptance of the concept of isolated electrical charges, in contrast to complete loops of current, until the late nineteenth century.
A way out of this dilemma became possible after 1884 when Poynting and Heaviside argued that electromagnetic fields (in suitable configurations) can be thought of as transmitting energy. The transmission of energy was then extended by Thomson , Poincaré and Abraham to include transmission (and storage) of momentum by an electromagnetic field.
That a moving charge interacting with thermal radiation should feel a radiation pressure was anticipated by Stewart in 1871 , who inferred that both the energy and the momentum of the charge would be affected.
In 1873, Maxwell discussed the pressure of light on conducting media at rest, and on “the medium in which waves are propagated” (, secs. 792-793). In the former case, the radiation of a reflected wave by a (perfectly conducting) medium in response to an incident wave results in momentum, but not energy, being transferred to the medium. The energy for the reflected wave comes from the incident wave.
The present formulation of the radiation reaction is due to Lorentz’ investigations of the self force of an extended electron, beginning in 1892 and continuing through 1903 . The example of dipole radiation of a single charge contrasts strikingly with Maxwell’s discussion of reaction forces during specular reflection. There is no net momentum radiated by an oscillating charge with zero average velocity, but energy is radiated. The external force alone can not account for the energy balance. An additional force is needed, and was identified by Lorentz as the net electromagnetic force of one part of an extended, accelerated charge distribution on another. See eq. (1) below.
In 1897, Planck applied the radiation reaction force of Lorentz to a model of charged oscillators and noted the existence of what are sometimes called “runaway” solutions, which he dismissed as having no physical meaning (keine physikalishe Bedeutung).
The basic concepts of the radiation reaction were brought essentially to their final form by Abraham , who emphasized the balance of both energy and momentum in the motion of extended electrons moving with arbitrary velocity.
Important contributions to the subject in the early twentieth century include those of Sommerfeld , Poincaré , Larmor , Lindemann , Von Laue , Born , Schott , Page , Nordström , Milner , Fermi , Wenzel , Wesel and Wilson . The main theme of these works was, however, models of classical charges and the related topic of electromagnetic mass.
The struggle to understand the physics of atoms led to diminished attention to classical models of charged particles in favor of quantum mechanics and quantum electrodynamics (QED). In 1935, there was apparent disagreement between QED and reported observations at energy scales of 10-100 MeV. Oppenheimer then conjectured whether QED might fail at high energies and, in partial support of his view, invoked a classical argument concerning difficulties of interpretation of the radiation reaction at short distances. The present article illustrates an aspect of Oppenheimer’s argument that was developed further by Landau .
Another response to Oppenheimer’s conjecture was by Dirac in 1938, when he deduced a covariant expression for the radiation reaction force (previously given by Abraham, Lorentz and von Laue in noncovariant notation) by an argument not based on a model of an extended electron. Dirac also gave considerable discussion of the paradoxes of runaway solutions and pre-acceleration. This work of Dirac, and most subsequent work on the classical radiation reaction, emphasized the internal consistency of classical electromagnetism as a mathematical theory, rather than as a description of nature. But, as has been remarked by Schott , “there is considerable danger, in a purely mathematical investigation, of losing touch with reality”. Quantum mechanics had triumphed.
Research articles on the classical radiation reaction are still being produced; see, for example, Refs. -. Sarachik and Schappert (, sec. IIID) present a brief version of the argument given below in sec. 2.
Reviews of the subject include Refs. -. Most noteworthy in relation to the present article are the reviews by Lorentz , Erber and Klepikov , which are the only ones that indicate an awareness of the problem of strong fields. The texts of Landau and Lifshitz , Jackson and Milonni briefly mention that issue.
The radiation reaction has been a frequent topic of articles in the American Journal of Physics, including Refs. and -. The article of Page and Adams is noteworthy for illustrating how the concept of electromagnetic field momentum restores the full validity of Newton’s third law in an interesting example of the interaction of a pair of moving charges.
## II A General Result for the Radiation Reaction
Consider an electron of charge $`e`$ and mass $`m`$ moving in electric and magnetic fields E and B. The mass $`m`$ is the “effective mass” in the language of Lorentz , now called the “renormalized” mass, for which the divergent electromagnetic self energy of a small electron is cancelled in a manner beyond the scope of this article. Then the remaining leading effect of the radiation reaction is the “radiation resistance” which is independent of hypotheses as to the structure of the electron. Our argument emphasizes the effect of radiation resistance, since any deductions about properties of electromagnetic fields will then be as free as possible from controversy as to the nature of matter.
The (nonrelativistic) equation of motion including radiation resistance is (in Gaussian units)
$$m\dot{𝐯}=𝐅_{\mathrm{ext}}+𝐅_{\mathrm{resist}},$$
(1)
where
$$𝐅_{\mathrm{ext}}=e𝐄+e\frac{𝐯}{c}\times 𝐁$$
(2)
is the Lorentz force on the electron due to the external field,
$$𝐅_{\mathrm{resist}}=\frac{2e^2}{3c^3}\ddot{𝐯}+𝒪(𝐯/c)$$
(3)
is the force of radiation resistance, v is the velocity of the electron, $`c`$ is the speed of light and the dot indicates differentiation with respect to time. Equation (3) is the form of the radiation reaction given in the original derivations of Lorentz and Planck , which is sufficient for the main argument of this paper. Some discussion of the larger context of the classical radiation reaction is given in secs. 2.1-5.
If the second time derivative of the velocity is small we estimate it by taking the derivative of (1):
$$\ddot{𝐯}\frac{e\dot{𝐄}}{m}+\frac{e}{m}\frac{\dot{𝐯}}{c}\times 𝐁+\frac{e}{m}\frac{𝐯}{c}\times \dot{𝐁}.$$
(4)
We further suppose that the velocity is small (without loss of generality according to the principle of relativity; see sec. 2.4 for a relativistic discussion). so it suffices to approximate $`\dot{𝐯}`$ as $`e𝐄/m`$ in (4). Hence,
$$\ddot{𝐯}\frac{e\dot{𝐄}}{m}+\frac{e^2}{m^2c}𝐄\times 𝐁.$$
(5)
The radiation resistance is now
$$𝐅_{\mathrm{resist}}\frac{2e^2}{3c^3}\left(\frac{e\dot{𝐄}}{m}+\frac{e^2}{m^2c}𝐄\times 𝐁\right).$$
(6)
The first term in (6) contributes only for time-varying fields, which I take to have frequency $`\omega `$ and reduced wavelength $`\mathrm{\lambda ̄}`$; hence, $`\dot{𝐄}\omega 𝐄`$. The second term contributes only when $`𝐄\times 𝐁0`$, which is most likely to be in a wave (with $`E=B`$) if the fields are large. So, for an electron in an external wave field, the magnitude of the radiation-resistance force is
$`F_{\mathrm{resist}}`$ $``$ $`{\displaystyle \frac{2}{3}}eE\sqrt{\left({\displaystyle \frac{e^2}{mc^2}}{\displaystyle \frac{\omega }{c}}\right)^2+\left({\displaystyle \frac{e^3E}{m^2c^4}}\right)^2}`$ (7)
$``$ $`F_{\mathrm{ext}}\sqrt{\left({\displaystyle \frac{r_0}{\mathrm{\lambda ̄}}}\right)^2+\left({\displaystyle \frac{E}{e/r_0^2}}\right)^2},`$ (8)
where $`r_0=e^2/mc^2=2.8\times 10^{13}`$ cm is the classical electron radius.
Equation (8) makes physical sense only when the radiation reaction force is smaller than the external force. Here we don’t explore whether the length $`r_0`$ describes a physical electron; we simply consider it to be a length that arises from the charge and mass of an electron. Rather, we concentrate on the implication of eq. (8) for the electromagnetic field. Then we infer that a classical description becomes implausible for fields whose wavelength is small compared to length $`r_0`$, or whose strength is large compared to $`e/r_0^2`$.
### A Commentary
The argument related to eq. (8) is that there are classical electromagnetic fields that lead to physically implausible behavior when radiation-reaction effects are included. This does not necessarily imply any mathematical inconsistency in the theory. Indeed, various authors have displayed solutions for electron motion coupled to an oscillator of very high natural frequency . Such solutions are well-defined mathematically but appear “physically implausible”. Of course, the mathematics might be correct in predicting the physical behavior in an unfamiliar situation. So it becomes a matter of experiment to decide whether the characterization “implausible” corresponds to physical reality or not. The experiments that produce the most influential results are typically those that reveal new phenomena in realms where prevailing theories are “implausible”.
Thus far, there is no evidence for the behavior predicted by the classical equations for electrons interacting with waves of frequencies greater than $`c/r_0`$. Rather, quantum mechanics is needed for a good description of the phenomena observed in that case, Compton scattering being an early example (sec. 3.1). Laboratory studies of strong-field electrodynamics have been undertaking only recently (sec. 3.2), and deal primarily with effects not anticipated in a classical description.
The argument of sec. 2 can also be considered as a model-independent version of a restriction that Lorentz placed on his derivation of eqs. (1-3) (, sec. 37, eq. (73)). Namely, the derivation makes physical sense only if
$$\frac{l}{ct}1,$$
(9)
where $`l`$ is a characteristic length of the problem, and $`t`$ is a characteristic time “during which the state of motion is sensibly altered”.
Lorentz would certainly have considered the classical electron radius, $`r_0`$, as an example of a relevant characteristic length. Hence, for an electron in an electromagnetic wave of (reduced) wavelength $`\mathrm{\lambda ̄}`$, the characteristic time of the resulting motion is $`\mathrm{\lambda ̄}/c`$, and Lorentz’ condition (9) becomes
$$\frac{r_0}{\mathrm{\lambda ̄}}1,$$
(10)
A close variant of the above argument was also given by Planck .
In case of a strong field with a long (possibly infinite) wavelength, Lorentz’ condition (9) can be interpreted as requiring the change in the electron’s velocity to be small compared to the speed of light during the time it takes light to travel one classical electron radius. That is, we require
$$\mathrm{\Delta }v=a\mathrm{\Delta }t=\frac{eE}{m}\frac{r_0}{c}c,$$
(11)
and hence,
$$E\frac{mc^2}{er_0}=\frac{e}{r_0^2}.$$
(12)
Thus, we arrive by another (although closely related) path to the conclusion drawn previously from eq. (8). Perhaps because the limiting field strength implied by (12) is extraordinarily large by practical standards, neither Lorentz nor Planck mentioned it explicitly.
In the first sentence of his 1938 article, Dirac stated that “the Lorentz model of the electron…has proved very valuable…in a certain domain of problems, in which the electromagnetic field does not vary too rapidly and the accelerations of the electrons are not too great”. However, he does not elaborate on the meaning of “not too great”.
Dirac’s derivation of the radiation-reaction 4-force was not based on a model of an extended electron, and so the derivation was not subject to Lorentz’ restriction (9). But as a co-inventor of quantum mechanics, Dirac cannot have expected his classical results to have unrestricted validity in the physical world.
In the decade after Dirac’s 1938 paper, a few works appeared that commented on the concept of a limiting field strength, typically in classical discussions of electron-positron pair creation. In sec. 3.2 we return to the issue of pair creation, but in a quantum context.
After the discovery of pulsars in 1967 there was a burst of interest in the behavior of electrons in very strong magnetic fields. Several papers appeared in which classical electrodynamics was applied , often with the intent of clarifying the boundary between the classical and quantum domains. For very large fields, classical solutions to the motion were obtained in which the electron has a damping time constant that is small compared to the period of cyclotron motion. Whether or not such highly damped solutions are “implausible”, they are outside ordinary experience. Again, one must perform experiments to decide whether the classical theory is valid in this domain. If such experiments had been possible prior to the development of quantum mechanics, they would have revealed deviations from the classical theory that would have encouraged development of a new theory. Arguments such as those leading to eqs. (8), (10) and (12) would have motivated the experiments.
### B Another Strong-Field Regime
Are there any other domains in which classical electrodynamics might be called into question?
Another interpretation of Lorentz’ criterion (9) is that the amplitude of the oscillatory motion of an electron in a wave of frequency $`\omega `$ should be small compared to the wavelength. As is well known (see prob. 2, sec. 47 of Ref. ), this leads to the condition that the dimensionless, Lorentz-invariant quantity,
$$\eta =\frac{eE_{\mathrm{rms}}}{m\omega c},$$
(13)
should be small compared to one. Parameter $`\eta `$ can exceed unity for waves of very low field strength if the frequency is low enough. An interesting result is that the electron can be said to have an effective mass,
$$m_{\mathrm{eff}}=m\sqrt{1+\eta ^2},$$
(14)
when inside a wave field . An electron in a spatially varying wave experiences a force $`𝐅=m_{\mathrm{eff}}c^2`$ which is often called the “ponderomotive force”, but which can be regarded as a kind of radiation pressure for a case where the “reflected” wave cannot be distinguished from the incident wave, and hence as a kind of radiation reaction force in its broadest meaning.
Debates continue regarding energy-transfer mechanisms between electrons and strong classical waves (as represented by a laser beam with η
>
1
>
𝜂1\eta\mathrel{\vbox{\kern 0.0pt\hbox{$>$} \kern 0.0pt\hbox{$\sim$} }}1). To what extent can net energy be exchanged between a free electron and a laser pulse in vacuum? Does a classical discussion suffice? Our understanding suggests that quantum aspects should be unimportant even for $`\eta 1`$ so long as condition (12) is satisfied, but full understanding has been elusive. Detailed discussion of this matter is deferred to a future article.
### C Utility of the Classical Radiation Reaction
Besides provoking extensive discussion on the validity of classical electrodynamics, the radiation reaction has enjoyed some well-known success in classical phenomenology. In particular, the topics of linewidth of radiation by a classical oscillator and resonance width in scattering of waves off such an oscillator show how partial understanding of atomic systems can be obtained in a classical context. Also, the radiation reaction is very important in antenna engineering where the power source must provide for the energy (and momentum, if any) radiated as well as that consumed in Joule losses. It is worth noting that these successes hold where the electron is part of an extended system.
In contrast, the radiation reaction has been almost completely negligible in descriptions of the radiation of free electrons for practical parameters in the classical domain (i.e., outside the domain of quantum mechanics). That this might be the case is the main argument of sec. 2. Section 3 discusses effects of the radiation reaction in the quantum domain.
### D Relativistic Radiation Reaction
For purposes of additional commentary, it is useful to record relativistic expressions for the radiation reaction.
The relativistic version of (1) in 4-vector notation is
$$mc^2\frac{du^\mu }{ds}=F_{\mathrm{ext}}^\mu +F_{\mathrm{resist}}^\mu ,$$
(15)
with external 4-force $`F_{\mathrm{ext}}^\mu =\gamma (𝐅_{\mathrm{ext}}𝐯/c,𝐅_{\mathrm{ext}})`$, and radiation-reaction 4-force given by
$$F_{\mathrm{resist}}^\mu =\frac{2e^2}{3}\frac{d^2u^\mu }{ds^2}\frac{Ru^\mu }{c},$$
(16)
where
$$R=\frac{2e^2c}{3}\frac{du_\nu }{ds}\frac{du^\nu }{ds}=\frac{2e^2\gamma ^6}{3c^3}\left[\dot{𝐯}^2\frac{(𝐯\times \dot{𝐯})^2}{c^2}\right]0$$
(17)
is the invariant rate of radiation of energy of an accelerated charge, $`u^\mu =\gamma (1,𝐯/c)`$ is the 4-velocity, $`\gamma =1/\sqrt{1v^2/c^2}`$, $`ds=cd\tau `$ is the invariant interval and the metric is $`(1,1,1,1)`$.
The time component of eq. (15) can be written
$$\frac{d\gamma mc^2}{dt}=𝐅_{\mathrm{ext}}𝐯+\frac{d}{dt}\left(\frac{2e^2\gamma ^4𝐯\dot{𝐯}}{3c^2}\right)R,$$
(18)
and the space components as
$`{\displaystyle \frac{d\gamma m𝐯}{dt}}=𝐅_{\mathrm{ext}}`$ (19)
$`+`$ $`{\displaystyle \frac{2e^2\gamma ^2}{3c^3}}\left[\ddot{𝐯}+{\displaystyle \frac{3\gamma ^2}{c^2}}(𝐯\dot{𝐯})\dot{𝐯}+{\displaystyle \frac{\gamma ^2}{c^2}}(𝐯\ddot{𝐯})𝐯+{\displaystyle \frac{3\gamma ^4}{c^4}}(𝐯\dot{𝐯})^2𝐯\right].`$ (20)
Keeping terms only to first order in velocity, eqs. (18-19) become
$$\frac{dmv^2/2}{dt}=𝐅_{\mathrm{ext}}𝐯+\frac{2e^2𝐯\ddot{𝐯}}{3c^3},$$
(21)
and
$$\frac{dm𝐯}{dt}=𝐅_{\mathrm{ext}}+\frac{2e^2\ddot{𝐯}}{3c^3}+\frac{2e^2(𝐯\dot{𝐯})\dot{𝐯}}{c^3}.$$
(22)
Equations (18-19) were first given by Abraham . Von Laue was the first to show that these equations can be obtained by a Lorentz transformation of the nonrelativistic results (21-22). The covariant notation of eqs. (15-17) was first applied to the radiation reaction by Dirac . An interesting discussion of the development of eqs. (18-19) has been given recently by Yaghjian .
### E Terminology
During a century of discussion of the radiation reaction a variety of terminology has been employed. In this article I use the phrase “radiation reaction” to cover all aspects of the physics of “Rückwirkung der Strahlung” as introduced by Lorentz and Abraham. This usage contrasts with a proposed narrow interpretation discussed at the end of this section.
“Æthereal friction” was the first description by Stewart in 1871, which he used in only a qualitative manner.
In 1873, Maxwell wrote on the “pressure exerted by light” in secs. 792-793 of his Treatise .
Lorentz used the French word “résistance” in describing eq. (3) when he presented it in 1892, and used the English equivalent “resistance” in his 1906 Columbia lectures .
Planck also discussed eq. (3), which he described as “Dämpfung” (damping) and “Dämpfung durch Strahlung” (literally, “damping by radiation” but translated more smoothly as “radiation damping”). The term “Strahlungsdämpfung” (radiation damping) does not, however, appear in the German literature until 1933 .
Around 1900, Larmor used the terms “frictional resistance” and “ray pressure” to describe a result meant to quantify Stewart’s insight, but which analysis has not stood the test of time.
The massive analyses of Abraham were accompanied by the introduction of several new terms. The title of Abraham’s 1904 article included the term “Strahlungsdruck” (radiation pressure). This use of the phrase “radiation pressure” can, however, be confused with the simpler concept of the pressure that results when a wave is reflected from a conducting surface . Perhaps for this reason, Abraham also introduced the phrase “Reaktionskraft der Strahlung”, which I translate as “radiation reaction force”. This appears to be the origin of the phrase “radiation reaction”, although in German that phrase remained a qualifier to “Kraft” (force) for many years. The variant “Strahlungsreaktion” (radiation reaction) appeared for the first time in 1933 .
Lorentz’ 1903 Encyklopädie article introduced the topic of the radiation reaction with the phrase “Rückwirkung des Äthers” (back interaction of the æther). In his 1905 monograph , Abraham used the variant “Rückwirkung der Strahlung” (back interaction of radiation, which could also be translated agreeably as “radiation reaction”).
In England in 1908, the Adams Prize examiners chose the topic of the radiation reaction, suggesting the cumbersome title “The Radiation from Electric Systems or Ions in Accelerated Motion and the Mechanical Reactions on their Motion which arise from it”. The winning essay by Schott adopted much of this title, but in the text Schott refers to “radiation pressure” and indicates that he follows Abraham in this. In his 1915 article, Schott also used the phrase “reaction due to radiation” and indicated that it was equivalent to his use of the phrase “radiation pressure”.
Schott also introduced other terms that seem less than ideal descriptions of the phenomena associated with the radiation reaction. His argument of 1912 is less crisp than one he gave in 1915 , so I follow the latter here. Schott considered the rate at which a radiating charge loses energy, and deduced eq. (18) in essentially that form. Schott noted that term $`R`$ is just the rate of radiation of energy by an accelerated charge, which he described as an “irreversible” process. He then interpreted the term
$$Q=\frac{2\gamma ^4e^2𝐯\dot{𝐯}}{3c^3},$$
(23)
as an energy stored “in the electron in virtue of its acceleration” and gave it the name “acceleration energy”. Schott considered the term $`\dot{Q}`$ in eq. (18) to be a “reversible” loss of energy.
Insights related to the concept of the “acceleration energy” have been useful in resolving the paradox of whether a charge radiates if its acceleration is uniform, i.e., if $`\ddot{𝐯}=0`$. In this case the radiation reaction force (3) vanishes and many people have argued that this means there is no radiation . But as first argued by Schott , in the case of uniform acceleration “the energy radiated by the electron is derived entirely from its acceleration energy; there is as it were internal compensation amongst the different parts of its radiation pressure, which causes its resultant effect to vanish”. This view is somewhat easier to follow if “acceleration energy” means energy stored in the near and induction zones of the electromagnetic field .
Schott’s use of the word “irreversible” to describe the process of radiation seems inapt. He may have meant that in a classical universe containing only one electron and an external force field, the radiated energy can never return to the electron. But as noted by Planck , “the fundamental equations of mechanics as well as those of electrodynamics allow the direct reversal of every process as regards time”. For example, “if we now consider any radiation processes whatsoever, taking place in a perfect vacuum enclosed by reflecting walls, it is found that, since they are completely determined by the principles of classical electrodynamics, there can be in their case no question of irreversibility of any kind”. However, “an irreversible element is introduced by the addition of emitting and absorbing substance”. Thus, consistent use of the word “irreversible” goes beyond classical electron theory. These views of Planck were seconded by Einstein and elaborated upon in the absorber theory of radiation of Wheeler and Feynman .
As another counterexample to the view that radiation is irreversible, a theme of contemporary accelerator physics is that every radiation process can be inverted to produce energy gain, not loss. Hence, there are now devices that accelerate electrons based on inverse Čerenkov radiation, inverse free-electron radiation, inverse Smith-Purcell radiation, inverse transition radiation, etc. Uniform acceleration is the inverse of uniform deceleration, and the inverse transformation is especially simple here: since $`𝐅_{\mathrm{resist}}`$ vanishes, it suffices to reverse the sign of the external force. These inverse radiation processes will be the subject of a future paper.
Schott’s use of “irreversible” as applied to the term $`Ru^\mu /c`$ of the radiation reaction has not been followed in the German literature. See Ref. for an interesting contrast.
The English phrase “radiation reaction” appears to have been first used by Page in 1918 .
In his 1938 paper, Dirac used the phrase “the effect of radiation damping on the motion of the electron”. As a consequence, most subsequent papers use “radiation damping” interchangeably with “radiation reaction” as a general description of the subject. Thus, in German there appeared the use of “Strahlungsdämfung” (already in 1933), in French, “force de freinage” (braking force, compare “rayonnement freinage” = Bremsstrahlung), and in Russian the equivalent of “radiation damping” must have been used as well . Dirac seconded Schott’s use of the terms “irreversible” and “acceleration energy”, and these become fairly common in the English literature thereafter. Indeed, “acceleration energy” becomes “Schott acceleration energy”, or just “Schott energy”.
The terminology of Schott and Dirac was taken a step further by Rohrlich in 1961 and 1965 , who proposed that only the second term in the covariant expression (16) is entitled to be called “the radiation reaction”. The first term of (16) is to be called the “Schott term”. A motivation for this terminology appears to be that in the case of uniform acceleration, expression (16) vanishes by virtue of cancellation of its two nonzero terms. Then the broadly defined “radiation reaction” (i.e., eq. (16)) vanishes, but the radiation does not (although it takes considerable effort to demonstrate this ). The terminology of Rohrlich has the merit that the paradox “how can there be radiation if there is no radiation reaction” is avoided in this case since only the (nonvanishing) term $`Ru^\mu /c`$ is called the “radiation reaction”.
However, this terminology is at odds with the origins of the subject, which emphasize the low-velocity limit, eqs. (21-22). Here, the radiated momentum enters only in terms of order $`v^2/c^2`$, so the direct back reaction of the radiated momentum (i.e., $`Ru^\mu /c`$) plays no role in the nonrelativistic limit. Thus, according to Rorhlich’s terminology there is no “radiation reaction” in the nonrelativistic limit.
But the original, and continuing, purpose of the concepts of the radiation reaction is to describe how a charge reacts to the radiation of energy when it does not radiate net momentum. To define the “radiation reaction” to be zero in this circumstance is counterproductive.
It appears that the terminology of Rohrlich has been adopted only by three subsequent workers .
## III A Quantum Interpretation
To go further, we pass beyond the realm of classical electromagnetism. The remainder of this paper is not a direct consequence of that theory, but considers how only a modest admixture of quantum concepts greatly clarifies the hints deduced by classical argument.
A simple device is to multiply and divide eq. (8) by Planck’s constant $`\mathrm{}`$, which was introduced by him shortly after his work on eq. (1) . Then we can write
$`F_{\mathrm{resist}}`$ $``$ $`F_{\mathrm{ext}}\sqrt{\left({\displaystyle \frac{e^2}{\mathrm{}c}}{\displaystyle \frac{\mathrm{}}{mc}}{\displaystyle \frac{\omega }{c}}\right)^2+\left({\displaystyle \frac{e^2}{\mathrm{}c}}{\displaystyle \frac{e\mathrm{}}{m^2c^3}}E\right)^2}`$ (24)
$``$ $`\alpha F_{\mathrm{ext}}\sqrt{\left({\displaystyle \frac{\mathrm{\lambda ̄}_C}{\mathrm{\lambda ̄}}}\right)^2+\left({\displaystyle \frac{E}{E_{\mathrm{crit}}}}\right)^2},`$ (25)
where $`\alpha =e^2/\mathrm{}c`$ is the QED fine structure constant, $`\mathrm{\lambda ̄}_C=\mathrm{}/mc`$ is the reduced Compton wavelength of an electron and
$$E_{\mathrm{crit}}=\frac{m^2c^3}{e\mathrm{}}=1.6\times 10^{16}\text{V/cm}=3.3\times 10^{13}\text{Gauss}$$
(26)
is the QED critical field strength, discussed in sec. 3.2 below.
Thus, our naïve quantum theory (classical electromagnetism plus $`\mathrm{}`$) leads us to expect important departures from classical electromagnetism for waves of wavelength much shorter than the Compton wavelength of the electron, and for fields of strength larger than the QED critical field strength.
### A The Radiation Reaction and Compton Scattering
Compton scattering was one of the earlier predictions of quantum theory and its observation had an important historical role in widespread acceptance of photons as quanta of light. Compton scattering is distinguished from Thomson scattering of classical electromagnetism in that wavelengths of the photons involved in Compton scattering are not small compared to the Compton wavelength of the electron, when measured in the frame in which the electron is initially at rest. Hence Compton scattering appears to be exactly the kind of example discussed above in which the radiation reaction should be important.
A description of a quantum scattering experiment relates the energy and momentum (plus relevant internal quantum numbers) of the initial state to those of the final state without discussion of forces. Yet, we can identify various correspondences between the quantum and classical descriptions.
In the case of Compton scattering, the initial photon corresponds to the external force field on the electron, while the final photon corresponds to the radiated wave. The quantum changes in momentum (and energy) of the electron in the scattering process can be said to correspond to classical time integrals of force (and of $`𝐅𝐯`$). Conservation of momentum (and energy) is described in the scattering process by including the back reaction of the final photon on the electron as well as the direct reaction of the initial photon. Thus, the quantum description, which incorporates conservation of momentum (and energy), can be said to include automatically the (time-integrated) effects of the radiation reaction.
Compton scattering is an electromagnetic scattering process in which large changes in momentum (and energy) of the electron are observed (in the frame in which the electron is initially at rest). It can therefore be said to correspond to a situation in which the radiation reaction is large, in agreement with the semiclassical inferences of secs. 2 and 3.
The correspondence between quantum conservation of energy and the classical radiation reaction appears to involve only the second term, $`Ru^\mu /c`$, in expression (16) for the radiation-reaction force. Since the electron has constant (though different) initial and final velocities in a scattering experiment, the “acceleration energy” $`Q`$ of eq. (23) is zero both before and after the scattering, and the equivalent of $`\dot{Q}`$ cannot be expected to appear in the quantum description (at “tree level”, in the technical jargon) of Compton scattering.
Effects corresponding to the near-field “acceleration energy” can be said to occur in quantum electrodynamics in the case of so-called vertex corrections and propagator (mass) corrections, in which a virtual photon is emitted and absorbed by the same electron. These “loop corrections” to the behavior of a quantum point charge implement the equivalent of the self interaction of an extended charge, but diverge when the emission and absorption occur at the same spacetime point. They are the source of the famous infinities of QED that are dealt with by “renormalization”. See also sec. 4.1 below.
In the early 1940’s, Heitler and coworkers formulated a version of QED in which radiation damping played a prominent role. Following the suggestion of Oppenheimer , they hoped that this theory would provide a general method of dealing with the divergences of QED. By selecting a subset of “loop corrections”, they deduced an expression for Compton scattering that corresponds to classical Thomson scattering plus classical radiation damping. While this result is suggestive, it does not appear to be endorsed in detail by subsequent treatments of “renormalization” in QED.
### B The Critical Field
The second term under the radical in eq. (25) may be less familiar. The concept of a critical field in quantum mechanics began with Klein’s paradox : an electron that encounters an (electric) potential step appears to be reflected with greater than unit probability in Dirac’s theory.
Sauter noted that this effect arises only when the potential gradient is larger than the critical field, $`m^2c^3/e\mathrm{}`$. The resolution of the paradox is due to Heisenberg and Euler , who noted that electrons and positrons can be spontaneously produced in critical fields – a very extreme form of the radiation reaction. The critical field has been discussed at a sophisticated level by Schwinger and by Brezin and Itzykson , among many others.
An electron that encounters an electromagnetic wave of critical strength produces not only Compton scattering of the wave photons but also electron-positron pairs. These effects have recently been observed in experiments in which the author participated .
There is speculation that critical magnetic fields exist at the surface of neutron stars , and may be responsible for some aspects of pulsar radiation.
Pomeranchuk noted that the Earth’s magnetic field appears to have critical strength from the point of view of an electron of energy $`10^{19}`$ eV, which energy is at the upper limit of observation of cosmic rays.
The critical field arises in discussion of the radiation, commonly called synchrotron radiation, of electrons moving in circular orbits under the influence of a magnetic field $`B`$. If an electron of laboratory energy $`mc^2`$ moves in an orbit with angular velocity $`\omega _0`$, then the characteristic frequency of the synchrotron radiation is
$$\omega \gamma ^3\omega _0,$$
(27)
where $`\gamma =/mc^2`$ is the Lorentz boost to the rest frame of the electron. For motion in a magnetic field $`B`$, the cyclotron frequency $`\omega _0`$ can be written
$$\mathrm{}\omega _0=\frac{mc^2B}{\gamma B_{\mathrm{crit}}},$$
(28)
where $`B_{\mathrm{crit}}=m^2c^3/e\mathrm{}`$. Thus, the characteristic energy of synchrotron-radiation photons (often called the critical energy) is
$$\mathrm{}\omega \frac{\gamma B}{B_{\mathrm{crit}}}.$$
(29)
Hence an electron radiates away roughly 100% of its energy in a single synchrotron-radiation photon if the magnetic field in the electron’s rest frame, $`B^{}=\gamma B`$, has critical field strength. In this regime a classical theory of synchrotron radiation is inadequate .
Critical electric fields can be created for short times in the collision of nonrelativistic heavy ions, resulting in positron production .
As a final example of the inapplicability of classical electromagnetism in strong fields, the performance of future high-energy electron-positron colliders will be limited by the disruptive (quantum) effect of the critical fields experienced by one bunch of charge as it passes through the oncoming bunch .
## IV Discussion
In this paper I have followed the example of Landau in using the argument of sec. 2 to suggest limitations to the concepts of classical electrodynamics. However, this line of argument appears to have played no role in the early development of quantum mechanics. Rather, the argument was used in the 1930’s to suggest that quantum electrodynamics might have conceptual limitation when carried beyond the leading order of approximation . The history of this era has been well reviewed in the recent book by Schweber .
While the program of renormalization, of which Lorentz was an early advocate in the classical context , appears to have been successful in eliminating the formal divergences that were so troublesome in the 1930’s, quantum electrodynamics is still essentially untested for fields in excess of the critical field strength (26) . It still may be the case that this realm contains new physical phenomena that will validate the cautionary argument of Oppenheimer .
We close with three examples to stimulate additional discussion. Two are from strong-field electrodynamics; while not necessarily suggesting defects in the theory, they indicate that not all aspects of QED are integrated in the most familiar presentations. The third example considers the case of extraordinarily short wavelengths.
### A The Mass Shift of an Accelerated Charge
We can rewrite the nonrelativistic expressions (1-3) for the radiation reaction as
$$\frac{d}{dt}\left(m𝐯\frac{2e^2\dot{𝐯}}{3c^3}\right)=𝐅_{\mathrm{ext}},$$
(30)
and the relativistic expressions (15-17) as
$$\frac{d}{ds}\left(mc^2u^\mu \frac{2e^2}{3}\frac{du^\mu }{ds}\right)=F_{\mathrm{ext}}^\mu \frac{Ru^\mu }{c}.$$
(31)
These forms suggest the interpretation that the momentum, $`m𝐯`$, of a moving charge is decreased by amount $`2e^2\dot{𝐯}/3c^2`$ if that charge is accelerating as well .
If we take $`mc`$ as the scale of the ordinary momentum, then the effect of acceleration, $`eE/m`$, due to an electric field $`E`$ becomes large in eq. (30) only when E
>
m2c4/e3=e/r02
>
𝐸superscript𝑚2superscript𝑐4superscript𝑒3𝑒superscriptsubscript𝑟02E\mathrel{\vbox{\kern 0.0pt\hbox{$>$} \kern 0.0pt\hbox{$\sim$} }}m^{2}c^{4}/e^{3}=e/r_{0}^{2}, i.e., when the electric field is large compared to the classical critical field found in sec. 2.
This interpretation has been seconded by Ritus based on a semiclassical analysis (classical electromagnetic field, quantum electron) of the behavior of electrons in a strong, uniform electric field. He finds that the mass of an electron (= eigenvalue of the mass operator) obeys
$$m=m_0\left(1\frac{\alpha E}{2E_{\mathrm{crit}}}+𝒪(E^2/E_{\mathrm{crit}}^2)\right),$$
(32)
and remarks on the relation between this result and the classical interpretations (30-31). The mass shift of an accelerated charge becomes large when E
>
Ecrit/α=e/r02
>
𝐸subscript𝐸crit𝛼𝑒superscriptsubscript𝑟02E\mathrel{\vbox{\kern 0.0pt\hbox{$>$} \kern 0.0pt\hbox{$\sim$} }}E_{\rm crit}/\alpha=e/r_{0}^{2}, as found above.
The physical meaning of Ritus’ result remains somewhat unclear. For example, a mass shift of the form (32) does not appear in Ritus’ treatment of Compton scattering in intense wave fields (which treatment agrees with other works), although the effective mass (14) does appear.
### B Hawking-Unruh Radiation
According to Hawking , an observer outside a black hole experiences a bath of thermal radiation of temperature
$$T=\frac{\mathrm{}g}{2\pi ck},$$
(33)
where $`g`$ is the local acceleration due to gravity and $`k`$ is Boltzmann’s constant. In some manner, the background gravitational field interacts with the quantum fluctuations of the electromagnetic field with the result that energy can be transferred to the observer as if he(she) were in an oven filled with black-body radiation. Of course, the effect is strong only if the background field is strong.
An extreme example is that if the temperature is equivalent to 1 MeV or more, virtual electron-positron pairs emerge from the vacuum into real particles.
As remarked by Unruh , this phenomenon can be demonstrated in the laboratory according to the principle of equivalence: an accelerated observer in a gravity-free environment experiences the same physics (locally) as an observer at rest in a gravitational field. Therefore, an accelerated observer (in zero gravity) should find him(her)self in a thermal bath of radiation characterized by temperature
$$T=\frac{\mathrm{}a^{}}{2\pi ck},$$
(34)
where $`a^{}`$ is the acceleration as measured in the observer’s instantaneous rest frame.
The Hawking-Unruh temperature finds application in accelerator physics as the reason that electrons in a storage ring do not reach 100% polarization despite emitting polarized synchrotron radiation . Indeed, the various limiting features of performance of a storage ring that arise due to quantum fluctuations of the synchrotron radiation can be understood quickly in terms of eq. (34) .
Here we consider a more speculative example. Suppose the observer is an electron accelerated by an electromagnetic field $`E`$. Then, scattering of the electron off photons in the apparent thermal bath would be interpreted by a laboratory observer as an extra contribution to the radiation rate of the accelerated charge . The power of the extra radiation, which I call Unruh radiation, is given by
$`{\displaystyle \frac{dU_{\mathrm{Unruh}}}{dt}}`$ $`=`$ (energy flux of thermal radiation) (36)
$`\times \text{(scattering cross section)}.`$
For the scattering cross section, we use the well-known result for Thomson scattering, $`\sigma _{\mathrm{Thomson}}=8\pi r_0^2/3`$. The energy density of thermal radiation is given by the usual expression of Planck:
$$\frac{dU}{d\nu }=\frac{8\pi }{c^3}\frac{h\nu ^3}{e^{h\nu /kT}1},$$
(37)
where $`\nu `$ is the frequency. The flux of the isotropic radiation on the electron is just $`c`$ times the energy density. Note that these relations hold in the instantaneous rest frame of the electron. Then
$$\frac{dU_{\mathrm{Unruh}}}{dtd\nu }=\frac{8\pi }{c^2}\frac{h\nu ^3}{e^{h\nu /kT}1}\frac{8\pi }{3}r_0^2.$$
(38)
On integrating over $`\nu `$ we find
$$\frac{dU_{\mathrm{Unruh}}}{dt}=\frac{8\pi ^3\mathrm{}r_0^2}{45c^2}\left(\frac{kT}{\mathrm{}}\right)^4=\frac{\mathrm{}r_0^2a^4}{90\pi c^6},$$
(39)
using the Hawking-Unruh relation (34). The presence of $`\mathrm{}`$ in eq. (39) reminds us that Unruh radiation is a quantum effect.
This equals the classical Larmor radiation rate, $`dU/dt=2e^2a^2/3c^3`$, when
$$E^{}=\sqrt{\frac{60\pi }{\alpha }}E_{\mathrm{crit}}\frac{E_{\mathrm{crit}}}{\alpha },$$
(40)
where $`E_{\mathrm{crit}}`$ is the QED critical field strength introduced in eq. (26). In this case, the acceleration $`a^{}=eE^{}/m`$ is about $`10^{31}`$ Earth $`g`$’s.
The physical significance of Unruh radiation remains unclear. Sciama has emphasized how the apparent temperature of an accelerated observed should be interpreted in view of quantum fluctuations. Unruh radiation is a quantum correction to the classical radiation rate that grows large only in situations where quantum fluctuations in the radiation rate become very significant. This phenomenon should be contained in the standard theory of QED, but a direct demonstration of this is not yet available. Likewise, the relation between Unruh radiation and the mass shift of an accelerated charge, both of which become prominent at fields of strength $`E_{\mathrm{crit}}/\alpha `$, is not yet evident.
The existence of Unruh radiation provides an interesting comment on the “perpetual problem” of whether a uniformly accelerated charge emits electromagnetic radiation ; this issue has been discussed briefly in sec. 2.5. The interpretation of Unruh radiation as a measure of the quantum fluctuations in the classical radiation implies that the classical radiation exists. It is noteworthy that while discussion of radiation by an accelerated charge is perhaps most intricate classically in case of uniform acceleration, the discussion of quantum fluctuations is the most straightforward for uniform acceleration.
In addition, Hawking-Unruh radiation helps clarify a residual puzzle in the discussion of the equivalence between accelerated charges and charges in a gravitational field. Because of the difficulty in identifying an unambiguous wave zone for uniformly accelerated motion of a charge (in a gravity-free region) and also in the case of a charge in a uniform gravitational field, there remains some doubt as to whether the ‘radiation’ deduced by classical arguments contains photons. Thus, on p. 573 of the article by Ginzburg we read: “neither a homogeneous gravitational field nor a uniformly accelerated reference frame can actually “generate” free particles, expecially photons”. We now see that the quantum view is richer than anticipated, and that Hawking-Unruh radiation provides at least a partial understanding of particle emission in uniform acceleration or gravitation. Hence, we can regard the concerns of Bondi and Gold , Fulton and Rohrlich , the DeWitt’s and Ginzburg on radiation and the equivalence principle as precursors to the concept of Hawking radiation.
### C Can a Photon Be a Black Hole?
While quantum electrodynamics appears valid in all laboratory studies so far, which have explored photons energies up to the TeV energy scale, will this success continue at arbitrarily high energies (i.e., arbitrarily short wavelengths)?
Consider a photon whose (reduced) wavelength $`\mathrm{\lambda ̄}`$ is the so-called Planck length ,
$$L_P=\sqrt{\frac{\mathrm{}G}{c^3}}10^{33}\text{cm},$$
(41)
where $`G`$ is Newton’s gravitational constant. The gravitational effect of such a photon is quite large. A measure of this is the “equivalent mass”,
$$m_{\mathrm{equiv}}=\frac{\mathrm{}\omega }{c^2}=\frac{\mathrm{}}{c\mathrm{\lambda ̄}}=\frac{\mathrm{}}{cL_P}.$$
(42)
The Schwarzschild radius corresponding to this equivalent mass is
$$R=\frac{2Gm_{\mathrm{equiv}}}{c^2}=\frac{2\mathrm{}G\omega }{c^3L_P}=2L_P=2\mathrm{\lambda ̄}.$$
(43)
A naïve interpretation of this result is that a photon is a black hole if its wavelength is less than the Planck length. Among the scattering processes involving such a photon and a charged particle would be the case in which the charged particle is devoured by the photon, which would increase the energy of the latter, making its wavelength shorter still.
At very short wavelengths, electromagnetism and gravitation become intertwined in a manner that requires new understanding. The current best candidate for the eventual theory that unifies the fundamental interactions at short wavelengths is string theory. Variants of the preceding argument are often used to motivate the need for a new theory.
## V Acknowledgements
The author wishes to thank John Wheeler and Arthur Wightman for discussions of the history of the radiation reaction, Bill Unruh for discussions of the Hawking-Unruh effect, and Igor Klebanov, Larus Thorlacius and Ed Witten for discussions of string theory. |
warning/0003/astro-ph0003127.html | ar5iv | text | # Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field
## 1 INTRODUCTION
The luminosity function of galaxies (LF) plays a crucial role for extragalactic astronomy and observational cosmology. It is one of the basic descriptions of the galaxy population itself, and sometimes treated as a function of color (e.g. Efstathiou, Ellis, & Peterson 1988, hereafter EEP; Metcalfe et al. 1998; Lin et al. 1999) or morphology (e.g. Bingelli, Sandage, & Tammann 1988; Marzke et al. 1998), or other additional parameters of galaxies. It is also essential for interpreting galaxy number counts (e.g. Koo & Kron 1992; Ellis 1997) and for analyzing galaxy clustering (e.g. Strauss & Willick 1995; Efstathiou 1996). Furthermore, the LF is a fundamental test for the theory of galaxy formation (e.g. Baugh, Cole, & Frenk 1996). Recently, the exact shape of the LF has been of particular interest, because it is one of the key issues to the “faint blue galaxy problem” of galaxy number counts (Koo & Kron 1992; Ellis 1997), and may be related to dwarf galaxy formation (e.g. Babul & Rees 1992; Babul & Ferguson 1996; Hogg & Phinney 1997). The evolution of the LF is also important to derive the cosmic luminosity density, in the context of the ‘Madau plot’, i.e., cosmic star formation density as a function of redshift (e.g. Madau et al. 1996; Cowie et al. 1996; Sawicki, Lin, & Yee 1997, hereafter SLY97; Pascarelle, Lanzetta, & Fernández-Soto 1998).
Estimating galaxy luminosity function from an observational galaxy catalog is a fundamental work, but it is not a trivial task. Because of the flux-limited nature of the redshift survey data, the catalogs are inevitably censored, and suitable statistical technique is required. In the early stage of the extragalactic astronomy, the classical estimator, the number of galaxies in a given volume, $`\mathrm{\Phi }=N/V`$, was used to estimate the LF (Hubble 1936). Of course this is not sufficient for detailed studies, and many experts have proposed ingenious methods. Schmidt (1968) invented the famous $`1/V_{\mathrm{max}}`$ estimator in the studies of quasar population. Felten (1977) introduced the direction dependence of the magnitude limit. The extension for combining some different catalogs coherently was discussed in Avni & Bahcall (1980). Further extension to examine the evolution of the LF with redshift was proposed by Eales (1993), and the integrated variant of the Eales’ estimator was used in the survey of the Hawaii Deep Fields by Cowie et al. (1996). Qin & Xie (1999) also developed this estimator with a similar line of study. The fundamental assumption of this estimator is that the distribution of the objects is spatially uniform. Nowadays this is regarded as a drawback, because we know that the galaxies have strong clustering properties in the large-scale structure. In spite of the drawback, $`1/V_{\mathrm{max}}`$ estimator has been frequently used for extragalactic studies (e.g. Lilly et al. 1995; Ellis et al. 1996), probably because of its simplicity in calculation.
In order to overcome the difficulty in treating inhomogeneous galaxy distribution, some density-insensitive methods have been invented. Lynden-Bell (1971) proposed the $`C^{}`$ method, and applied it to the quasar sample of Schmidt (1968). This method is based on a quite sophisticated statistical idea as we discuss in subsequent sections. Carswell (1973) reported numerical experiments in its use. Jackson (1974) improved the method to combine several different catalogs, and studied the error estimation when the LF is expressed as an analytical form. The original method could derive only the shape of the probability density function, but Chołoniewski (1987)(hereafter C87) improved the method to obtain the density normalization and to trace the density evolution simultaneously. Lynden-Bell himself, and later Felten (1976) and Nicoll & Segal (1983) pointed out the drawback of this method that it cannot work in the faintest regime where the data points are too sparse. This drawback was basically overcome the introduction of smoothing method by Caditz & Petrosian (1993) (CP93). Subba Rao et al. (1996) and Szokoly et al. (1998) used the method in the recent studies of distant galaxies. We note that this method was further generalized by Maloney & Petrosian (1999) to treat the doubly truncated data, but here we do not discuss it further.
The method proposed by Turner (1979) and Kirshner, Oemler, & Schechter (1979) used the ratio of the number of objects between the absolute magnitude interval $`[M,M+\mathrm{d}M]`$ and the number of objects brighter than $`M`$, which canceled out the density inhomogeneity. Marinoni et al. (1999) used this method in their analysis of the effect of the Local infall motion on the estimation of the LF. Similar estimators were used by Davis & Huchra (1982) and later de Lapparent, Geller, & Huchra (1989). However, as mentioned in Efstathiou (1996), this estimator does not use the whole sample.
In contrast, some estimation methods using analytical LF models, which are often called parametric estimation methods, have been developed. Sandage, Tammann, & Yahil (1979)(STY) introduced the maximum likelihood method, which was free of the effects induced by density inhomogeneity, in this field by using parametric Schechter form for the LF. This parametric form was extended for evolutionary studies of galaxies by Lin et al. (1999). Marshall et al. (1983) presented another parametric estimator which can treat both the LF and the evolution parameter simultaneously, assuming the Poisson distribution of the objects on the magnitude–redshift space.
The maximum likelihood approach was widely used and extended to the methods which did not use analytical forms, often referred to as nonparametric methods. Nicoll & Segal (1983) proposed such a type of estimator and used it for the study of their chronometric cosmology. The estimator which can be regarded as an advanced version of Nicoll & Segal’s method was invented by Chołoniewski (1986)(C86). This method adopts the same assumption as Marshall et al. (1983), and is regarded as a binned nonparametric version of it. Another stepwise estimator, which was a binned analog of STY’s estimator, was introduced by EEP. Now this method seems to be most commonly used, and is called ‘the stepwise maximum likelihood method’. But note that not only EEP’s but also most of the other estimators are based on the maximum likelihood principle. The EEP’s method was extended to treat density evolution (Heyl et al. 1997; Springel & White 1998).
In spite of the variety of the methods, as we see above, there had been only the comparisons of some methods in the literature (e.g. Felten 1976; C86; EEP; Heyl et al. 1997) before the elaborate intercomparison by Willmer (1997)(W97). Statistically detailed discussions are not so frequently seen, either, except the rigorous works of Petrosian (1992). In W97, each method was examined by Monte Carlo simulations and CfA1 (e.g. de Lapparent et al. 1989) data. The obtained results were fitted by Schechter form, and W97 discussed the distribution of the estimates by each method after 1000 simulations. Based on the fitting parameter distributions, W97 reported the bias trends for some estimators. Furthermore W97 studied the normalization estimates, and concluded that the serious discrepancies between the LFs of local and distant galaxies is not attributed to the difference of the estimators used in the analyses.
Now further questions arise after W97. The tests of W97 were restricted to the Schechter form LF. They considered, for example, the bias in the faint-end slope estimation, and concluded that even for the spatially homogeneous samples, $`1/V_{\mathrm{max}}`$ estimator gives biased results. It is often claimed that the faint-end overestimation of the $`1/V_{\mathrm{max}}`$ estimator is caused by the density inhomogeneity of the Local Supercluster (e.g. Efstathiou 1996). Thus if galaxies are homogeneously distributed, the estimator is expected to give the correct value. If any subtler problem dwells in the slope estimation, further extensive experiments are required. They also mentioned the binning size selection. For the analysis of the recent very high-redshift data, data sparseness should be considered properly.
Recently the LF of galaxies at extremely high redshift has become available with the aid of large telescope facilities and improved detectors. Added to this, redshift surveys have entered upon a new phase by development of the photometric redshift technique. The technique requires much lower observational cost than the spectroscopic survey, and is suitable for the analyses of the deep photometric data like the Hubble Deep Field (HDF; Williams et al. 1996). Though some problems are inherent in the technique and in the faint source finding itself (Ferguson 1998), vast advances have been produced by the method (e.g. Furusawa et al. 2000). The intermediate–high redshift results are, however, still controversial with each other.
To settle down these problems, reliable and robust analyses of the LF are required. In this paper, we examined and made practical improvements for these estimation methods<sup>1</sup><sup>1</sup>1Numerical calculations in this paper are based on the public software package for cosmological study written by one of the authors (KY). The C library can be downloaded from http://www.kusastro.kyoto-u.ac.jp/$`\stackrel{~}{}`$kohji/research/libcosm/.. Considering the complicated understanding of the evolution of galaxy population, we concentrated our discussions on the nonparametric methods without any assumed functional forms for the LF. Besides we restricted our concerns only to the methods which use the whole sample. We used the mock catalog generated from various shapes of the probability density function (namely the LF). As we noted above, the density inhomogeneity is a basic property of the galaxy distribution. First we tested how accurately these methods reproduce the input density function, by using spatially homogeneous mock catalogs with varying sample size. Next, we examined the estimators by using mock catalogs with a dense cluster and with a large void. We also used the mock 2dF catalog prepared by Cole et al. (1998) in this study. After checking the reliability of each method, we finally applied the methods to the photometric redshift catalog prepared by Fernández-Soto et al. (1999) (FLY99) and studied the evolution of the LF at the very large redshift.
This paper is organized as follows: in Section 2 we review and discuss the methods and our extensions. Section 3 is devoted to the tests for the performance of these methods by mock catalogs. We apply the methods to the photometric redshift catalog and discuss the LF evolution in section 4. Our summary and conclusions are presented in section 5. We briefly introduce the statistical model selection criterion which we used in our discussions in Appendix A.
## 2 NONPARAMETRIC METHODS FOR ESTIMATING LUMINOSITY FUNCTION
Before we discuss each method, we define some fundamental quantities. Let $`M`$ : absolute magnitude, $`m`$ : apparent magnitude, and $`d_\mathrm{L}(z)`$ : luminosity distance in unit of \[Mpc\] corresponding to redshift $`z`$. Then
$`M=m5\mathrm{log}d_\mathrm{L}(z)25K(z),`$ (1)
where $`K(z)`$ is the $`K`$-correction. Here $`\mathrm{log}\mathrm{log}_{10}`$. We use the following notation unless otherwise stated: $`\varphi (M)`$ : the luminosity function \[$`\mathrm{Mpc}^3\mathrm{mag}^1`$\], $`N_{\mathrm{obs}}`$ : number of detected galaxies in the survey. When we use stepwise estimators, we must select the optimal binning size to suppress the statistical fluctuation (Sturges 1926; Beers 1992 and references therein; Sakamoto, Ishiguro, & Kitagawa 1986; Heyl et al. 1997). We used Akaike’s information criteria (AIC: Akaike 1974) in order to select the optimal binning size with the least loss of information (Takeuchi 1999; for general discussion, see Sakamoto et al. 1986).
### 2.1 Schmidt–Eales ($`1/V_{\mathrm{max}}`$) Method
The method to construct the LF we will discuss here was originally proposed by Schmidt (1968) and well known as the $`1/V_{\mathrm{max}}`$ method. Eales (1993) developed it further to trace the evolution with redshift. Cowie et al. (1996) used this estimator in an integral form.
We consider the absolute magnitude and redshift range
$`\{\begin{array}{cc}M_\mathrm{l}MM_\mathrm{u}\hfill & \\ z_\mathrm{l}zz_\mathrm{u}\hfill & \end{array}`$ (4)
with a survey solid angle $`\mathrm{\Omega }`$ and upper and lower limiting apparent magnitude, $`m_\mathrm{u}`$ and $`m_\mathrm{l}`$. Then we have
$`{\displaystyle _{M_\mathrm{l}}^{M_\mathrm{u}}}\varphi (M)dM`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}{\displaystyle \frac{1}{V_{\mathrm{max}}(i)}},`$ (5)
$`V_{\mathrm{max}}(i)`$ $``$ $`{\displaystyle _\mathrm{\Omega }}{\displaystyle _{z_{\mathrm{min},i}}^{z_{\mathrm{max},i}}}{\displaystyle \frac{\mathrm{d}^2V}{\mathrm{d}\mathrm{\Omega }\mathrm{d}z}}dzd\mathrm{\Omega },`$ (6)
where $`z_{\mathrm{max},i}`$ and $`z_{\mathrm{min},i}`$ are the upper and lower redshift limits that a galaxy with the absolute magnitude $`M_i`$ can be detected in the survey. We note that
$`z_\mathrm{l}z_{\mathrm{min},i}<z_{\mathrm{max},i}z_\mathrm{u}.`$
Defining $`z(M,m)`$ to be the redshift that a galaxy with the absolute magnitude $`M`$ is observed as an object with the apparent magnitude $`m`$, we get
$`z_{\mathrm{max},i}`$ $`=`$ $`\mathrm{min}\{z_\mathrm{u},z(M_i,m_\mathrm{u})\},`$ (7)
$`z_{\mathrm{min},i}`$ $`=`$ $`\mathrm{max}\{z_\mathrm{l},z(M_i,m_\mathrm{l})\}.`$ (8)
Actually, both $`z_{\mathrm{max},i}`$ and $`z_{\mathrm{min},i}`$ depend on the galaxy spectral energy distributions (SEDs). Thus we must account for the $`K`$-correction when calculating the $`1/V_{\mathrm{max}}(i)`$.
Felten (1976) proved that the Schmidt $`1/V_{\mathrm{max}}`$ estimator is unbiased, but does not yield a minimum variance. He also proved that the “classical estimator $`N/V`$”, which is different from the $`1/V_{\mathrm{max}}`$ estimator, is biased. Willmer (1997) gave a comment that Felten (1976) had shown this estimator to be biased, but it is not exact. A complication of the terminology may have led to such a comment.
### 2.2 Efstathiou–Ellis–Peterson (EEP) Method
In this subsection, we consider the stepwise maximum likelihood method introduced by EEP. Since the estimator of EEP method completely cancels the density information, this method requires an independent estimation of the galaxy density. The EEP method uses the form of the LF
$`\varphi (M)={\displaystyle \underset{k=1}{\overset{K}{}}}\varphi _kW(M_kM).`$ (9)
The window function $`W(M_{\mathrm{}}M)`$ is defined by
$`W(M_{\mathrm{}}M)\{\begin{array}{cc}1\hfill & \mathrm{for}M_{\mathrm{}}{\displaystyle \frac{\mathrm{\Delta }M}{2}}MM_{\mathrm{}}+{\displaystyle \frac{\mathrm{\Delta }M}{2}},\hfill \\ 0\hfill & \mathrm{otherwise}.\hfill \end{array}`$ (12)
According to EEP <sup>2</sup><sup>2</sup>2 Koranyi & Strauss (1997) have pointed out that the discreteness of the assumed LF causes a systematic error in the estimation. In order to avoid this effect, we can use the linear extrapolated form. We do not discuss it further in this paper (see the appendix of Koranyi & Strauss 1997)., the likelihood function is
$`(\{\varphi _k\}_{k=1,\mathrm{},K}|\{M_i\}_{i=1,\mathrm{},N_{\mathrm{obs}}})={\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}{\displaystyle \frac{_{\mathrm{}=1}^KW(M_{\mathrm{}}M_i)\varphi _{\mathrm{}}}{_{\mathrm{}=1}^K\varphi _{\mathrm{}}H(M_{\mathrm{lim}}(z_i)M_{\mathrm{}})\mathrm{\Delta }M}},`$ (13)
$`H(M_{\mathrm{lim}}(z_i)M)\{\begin{array}{cc}1\hfill & M_{\mathrm{lim}}(z_i)\mathrm{\Delta }M/2>M\hfill \\ {\displaystyle \frac{M_{\mathrm{lim}}(z_i)M}{\mathrm{\Delta }M}}+{\displaystyle \frac{1}{2}}\hfill & M_{\mathrm{lim}}(z_i)\mathrm{\Delta }M/2M<M_{\mathrm{lim}}(z_i)+\mathrm{\Delta }M/2\hfill \\ 0\hfill & M_{\mathrm{lim}}(z_i)+\mathrm{\Delta }M/2M\hfill \end{array}`$ (17)
where $`M_{\mathrm{lim}}(z_i)`$ is the absolute magnitude corresponding to the survey limit $`m_{\mathrm{lim}}`$ at redshift $`z_i`$.
The logarithmic likelihood is expressed as
$`\mathrm{ln}={\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}\left[{\displaystyle \underset{\mathrm{}=1}{\overset{K}{}}}W(M_{\mathrm{}}M_i)\mathrm{ln}\varphi _{\mathrm{}}\mathrm{ln}\left\{{\displaystyle \underset{\mathrm{}=1}{\overset{K}{}}}\varphi _{\mathrm{}}H(M_{\mathrm{lim}}(z_i)M_{\mathrm{}})\mathrm{\Delta }M\right\}\right].`$ (18)
Hence, the likelihood equation becomes
$`{\displaystyle \frac{\mathrm{ln}}{\varphi _k}}={\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}{\displaystyle \frac{W(M_kM_i)}{\varphi _k}}{\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}{\displaystyle \frac{H(M_{\mathrm{lim}}(z_i)M_k)\mathrm{\Delta }M}{_{\mathrm{}=1}^K\varphi _{\mathrm{}}H(M_{\mathrm{lim}}(z_i)M_{\mathrm{}})\mathrm{\Delta }M}}=0`$ (19)
and it reduces to
$`\varphi _k\mathrm{\Delta }M={\displaystyle \frac{_{i=1}^{N_{\mathrm{obs}}}W(M_kM_i)}{_{i=1}^{N_{\mathrm{obs}}}{\displaystyle \frac{H(M_{\mathrm{lim}}(z_i)M_k)}{_{\mathrm{}=1}^K\varphi _{\mathrm{}}H(M_{\mathrm{lim}}(z_i)M_{\mathrm{}})\mathrm{\Delta }M}}}}.`$ (20)
This equation can be solved by iteration, and we obtain the maximum likelihood estimator $`\widehat{\varphi }=\{\widehat{\varphi }_k\}_{k=1,\mathrm{},K}`$.
As for the normalization of the LF, some estimators have been proposed. We use the following estimator of the mean galaxy density, $`n`$, which was used by EEP:
$`n={\displaystyle \frac{1}{V}}{\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}{\displaystyle \frac{1}{\mathrm{\Psi }(z_i)}},`$ (21)
where $`V`$ is the maximum volume defined by the largest redshift in the sample, and $`\mathrm{\Psi }(z)`$ is the selection function, defined by
$`\mathrm{\Psi }(z){\displaystyle \frac{{\displaystyle _{\mathrm{}}^{M_{\mathrm{lim}}(z)}}\varphi (M)dM}{{\displaystyle _{\mathrm{}}^{\mathrm{}}}\varphi (M)dM}}.`$ (22)
For further discussions about other normalization estimators, see the appendix of Davis & Huchra (1982). By combining eqs. (20) and (21), we get the final results. As Strauss & Willick (1995) pointed out, at large redshift where the selection function eq. (22) is small, the estimator eq. (21) becomes noisy. Therefore in practice a certain cutoff should be introduced in redshift.
### 2.3 Chołoniewski Method
Here we discuss the method for estimating the LF developed by C86. The advantage of the method is that we can obtain the density and the shape of the LF simultaneously, and can easily examine the galaxy density evolution with redshift. The method explained here is an extended version applicable for the sample with a cosmological scale. In this subsection, we use indices $`i,j`$ as the labels of the cells.
We again consider the absolute magnitude and redshift range
$`\{\begin{array}{cc}M_\mathrm{l}MM_\mathrm{u}\hfill & \\ z_\mathrm{l}zz_\mathrm{u}\hfill & \end{array}`$
with a survey solid angle $`\mathrm{\Omega }`$ and survey limiting magnitude $`m_{\mathrm{lim}}`$. And let $`n(𝒓)`$ \[$`\mathrm{Mpc}^3`$\], the number density of galaxies in the neighborhood of the position $`𝒓`$. If we define $`V`$ as the total comoving volume under consideration, i.e.
$`V={\displaystyle _\mathrm{\Omega }}{\displaystyle _{z_\mathrm{l}}^{z_\mathrm{u}}}{\displaystyle \frac{\mathrm{d}^2V}{\mathrm{d}\mathrm{\Omega }\mathrm{d}z}}dzd\mathrm{\Omega },`$ (24)
then it leads to the following expression for the mean number density as
$`\overline{n}={\displaystyle \frac{N}{V}},`$ (25)
where $`N`$ is the total number of galaxies within the redshift range $`z_\mathrm{l}zz_\mathrm{u}`$. Here we adopt a statistical model: on the absolute magnitude–position space ($`M`$$`𝒓`$ space) the galaxy distribution is $`f(M,𝒓)`$, and the probability that we find $`k`$ galaxies in a volume element $`\mathrm{d}M\mathrm{d}V`$ at $`(M,𝒓)`$, $`P_k`$, is described as a Poisson distribution:
$`P_k`$ $`=`$ $`{\displaystyle \frac{e^\lambda \lambda ^k}{k!}},`$ (26)
$`\lambda `$ $`=`$ $`{\displaystyle \frac{1}{\overline{n}}}f(M,𝒓)\mathrm{d}M\mathrm{d}V.`$ (27)
Here
$`\varphi (M)`$ $`=`$ $`{\displaystyle \frac{1}{V}}{\displaystyle _\mathrm{\Omega }}{\displaystyle _{z_\mathrm{l}}^{z_\mathrm{u}}}f(M,𝒓){\displaystyle \frac{\mathrm{d}^2V}{\mathrm{d}\mathrm{\Omega }\mathrm{d}z}}dzd\mathrm{\Omega },`$ (28)
$`n(𝒓)`$ $`=`$ $`{\displaystyle \frac{1}{V}}{\displaystyle _{M_\mathrm{l}}^{M_\mathrm{u}}}f(M,𝒓)dM.`$ (29)
If we apply an assumption that the random variables $`M`$ and $`𝒓`$ are independent, i.e. $`f(M,𝒓)=\psi (M)\nu (𝒓)`$, then we obtain
$`\lambda `$ $`=`$ $`{\displaystyle \frac{1}{\overline{n}}}\psi (M)\mathrm{d}M\nu (𝒓)\mathrm{d}V.`$ (30)
We integrate $`\lambda `$ over the spherical shell at redshift $`z`$ and divide the $`M`$$`z`$ plane into small rectangular cells such that
$`\begin{array}{cc}M_iMM_{i+1}=M_i+\mathrm{\Delta }M\hfill & (i=1,\mathrm{},I),\hfill \\ z_jzz_{j+1}=z_j+\mathrm{\Delta }z\hfill & (j=1,\mathrm{},J).\hfill \end{array}`$ (33)
Now we see that the problem is to estimate the intensity parameter $`\lambda _{ij}`$ inhomogeneously defined on the $`M`$$`z`$ plane (see Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field). The probability of finding $`k_{ij}`$ galaxies in the cell $`(i,j)`$, $`P_{k_{ij}}`$, is
$`P_{k_{ij}}={\displaystyle \frac{e^{\lambda _{ij}}\lambda _{ij}^{k_{ij}}}{k_{ij}!}},`$ (34)
which is characterized by the parameter
$`\lambda _{ij}={\displaystyle _{M_i}^{M_{i+1}}}{\displaystyle _{V[z_j,z_{j+1}]}}\lambda {\displaystyle \frac{1}{\overline{n}}}\psi _i\mathrm{\Delta }M\nu _jV_j`$ (35)
where $`V[z_j,z_{j+1}]`$ is the comoving volume between redshifts $`z_j`$ and $`z_{j+1}`$. Here
$`\psi _i`$ $``$ $`{\displaystyle \frac{1}{\mathrm{\Delta }M}}{\displaystyle _{M_i}^{M_{i+1}}}\psi (M)dM,`$ (36)
$`\nu _j`$ $``$ $`{\displaystyle \frac{\mathrm{\Omega }}{V_j}}{\displaystyle _{z_j}^{z_{j+1}}}\nu (𝒓){\displaystyle \frac{\mathrm{d}V}{\mathrm{d}z}}dz,`$ (37)
and
$`V_j`$ $``$ $`\mathrm{\Omega }{\displaystyle _{z_j}^{z_{j+1}}}{\displaystyle \frac{\mathrm{d}V}{\mathrm{d}z}}dz.`$ (38)
The likelihood is given by
$`={\displaystyle \underset{(M_i,z_j)S}{}}{\displaystyle \frac{e^{\lambda _{ij}}\lambda _{ij}^{k_{ij}}}{k_{ij}!}},`$ (39)
and we obtain the log likelihood
$`\mathrm{ln}={\displaystyle \underset{(M_i,z_j)S}{}}\left\{k_{ij}\mathrm{ln}\lambda _{ij}\lambda _{ij}\mathrm{ln}k_{ij}!\right\},`$ (40)
where $`S`$ stands for the subset of the $`M`$$`z`$ plane surrounded with $`M_\mathrm{u}`$, $`M_\mathrm{l}`$, $`z_\mathrm{u}`$, $`z_\mathrm{l}`$, and the curve $`𝒞`$ defined by the selection line
$`M+5\mathrm{log}d_\mathrm{L}(z)+K(z)+25=m_{\mathrm{lim}}.`$ (41)
We define the following quantities:
$`i_{\mathrm{max}}(j)\mathrm{min}\{I,i_S(j)\},j_{\mathrm{max}}(i)\mathrm{min}\{J,j_S(i)\},`$ (42)
where $`M_{i_S(j)}\{M:𝒞\{(M,z):z=z_j\}\}`$, and $`z_{j_S(i)}\{z:𝒞\{(M,z):M=M_i\}\}`$ (see Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field). Using these notations, we can reduce eq. (40) as
$`\mathrm{ln}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{I}{}}}{\displaystyle \underset{j=1}{\overset{j_{\mathrm{max}}(i)}{}}}\left\{k_{ij}\mathrm{ln}\lambda _{ij}\lambda _{ij}\mathrm{ln}k_{ij}!\right\},`$ (43)
$`=`$ $`{\displaystyle \underset{j=1}{\overset{J}{}}}{\displaystyle \underset{i=1}{\overset{i_{\mathrm{max}}(j)}{}}}\left\{k_{ij}\mathrm{ln}\lambda _{ij}\lambda _{ij}\mathrm{ln}k_{ij}!\right\}.`$
The maximum likelihood estimates (MLEs) $`(\widehat{\psi }_0,\mathrm{},\widehat{\psi }_i,\widehat{\nu }_0,\mathrm{},\widehat{\nu }_J,\widehat{\overline{n}})`$ are the set of solutions which maximizes $``$. They can be obtained, in practice, by setting the following equations to zero:
$`{\displaystyle \frac{\mathrm{ln}}{\psi _i}}`$ $`=`$ $`{\displaystyle \underset{s=1}{\overset{I}{}}}{\displaystyle \underset{t=1}{\overset{t_{\mathrm{max}}(s)}{}}}\left(k_{st}{\displaystyle \frac{1}{\lambda _{st}}}{\displaystyle \frac{\lambda _{st}}{\psi _i}}{\displaystyle \frac{\lambda _{st}}{\psi _i}}\right)`$ (44)
$`=`$ $`{\displaystyle \underset{t=1}{\overset{t_{\mathrm{max}}(i)}{}}}\left({\displaystyle \frac{k_{it}}{\psi _i}}{\displaystyle \frac{1}{\overline{n}}}\mathrm{\Delta }M\nu _tV_t\right)=0,`$
$`{\displaystyle \frac{\mathrm{ln}}{\nu _j}}`$ $`=`$ $`{\displaystyle \underset{t=1}{\overset{J}{}}}{\displaystyle \underset{s=1}{\overset{s_{\mathrm{max}}(t)}{}}}\left(k_{st}{\displaystyle \frac{1}{\lambda _{st}}}{\displaystyle \frac{\lambda _{st}}{\nu _j}}{\displaystyle \frac{\lambda _{st}}{\nu _j}}\right)`$ (45)
$`=`$ $`{\displaystyle \underset{s=1}{\overset{s_{\mathrm{max}}(j)}{}}}\left({\displaystyle \frac{k_{sj}}{\nu _j}}{\displaystyle \frac{1}{\overline{n}}}\psi _s\mathrm{\Delta }MV_j\right)=0.`$
Thus we have a set of equations which are referred to as likelihood equations.
$`{\displaystyle \frac{\mathrm{\Delta }M}{\overline{n}}}\psi _i{\displaystyle \underset{t=1}{\overset{t_{\mathrm{max}}(i)}{}}}\nu _tV_t`$ $`=`$ $`{\displaystyle \underset{t=1}{\overset{t_{\mathrm{max}}(i)}{}}}k_{it},`$ (46)
$`{\displaystyle \frac{\mathrm{\Delta }M}{\overline{n}}}\nu _jV_j{\displaystyle \underset{s=1}{\overset{s_{\mathrm{max}}(j)}{}}}\psi _s`$ $`=`$ $`{\displaystyle \underset{s=1}{\overset{s_{\mathrm{max}}(j)}{}}}k_{sj}.`$ (47)
These equations are solved by iterative procedure. At this stage, these solutions obtained here are not exactly the MLEs themselves, but relative values. We need one more step to obtain absolute values. We denote the relative solutions by ‘$`\stackrel{~}{}`$’ and exact MLEs by ‘$`\widehat{}`$’. From eqs. (46) and (47), we have
$`\stackrel{~}{\psi }_i`$ $`=`$ $`{\displaystyle \frac{\overline{n}}{\mathrm{\Delta }M}}{\displaystyle \frac{_{t=1}^{t_{\mathrm{max}}(i)}k_{it}}{_{t=1}^{t_{\mathrm{max}}(i)}\stackrel{~}{\nu }_tV_t}}={\displaystyle \frac{_{t=1}^{t_{\mathrm{max}}(i)}k_{it}}{_{t=1}^{t_{\mathrm{max}}(i)}{\displaystyle \frac{_{s=1}^{s_{\mathrm{max}}(t)}k_{st}}{_{s=1}^{s_{\mathrm{max}}(t)}\stackrel{~}{\psi }_s}}}},`$ (48)
$`\stackrel{~}{\nu }_jV_j`$ $`=`$ $`{\displaystyle \frac{\overline{n}}{\mathrm{\Delta }M}}{\displaystyle \frac{_{s=1}^{s_{\mathrm{max}}(j)}k_{sj}}{_{s=1}^{s_{\mathrm{max}}(j)}\stackrel{~}{\psi }_s}}={\displaystyle \frac{_{s=1}^{s_{\mathrm{max}}(j)}k_{sj}}{_{s=1}^{s_{\mathrm{max}}(j)}{\displaystyle \frac{_{t=1}^{t_{\mathrm{max}}(s)}k_{st}}{_{t=1}^{t_{\mathrm{max}}(s)}\stackrel{~}{\nu }_tV_t}}}}.`$ (49)
Then we properly normalize these solutions. Clearly it follows that
$`{\displaystyle \underset{(M_i,z_j)S}{}}\lambda _{ij}=\mathrm{\Delta }M{\displaystyle \underset{(M_i,z_j)S}{}}\widehat{\psi }_i\widehat{\nu }_jV_j=N_{\mathrm{obs}}.`$ (50)
If we set $`\stackrel{~}{\psi }_i\stackrel{~}{\nu }_j=w\widehat{\psi }_i\widehat{\nu }_j`$, then we straightforwardly obtain the numerical factor $`w`$ by eq. (50):
$`w={\displaystyle \frac{N_{\mathrm{obs}}}{\mathrm{\Delta }M_{(M_i,z_j)S}(\stackrel{~}{\psi }_i\stackrel{~}{\nu }_jV_j)}}.`$ (51)
Now we obtain the LF $`\varphi (M)`$ and density $`n(z)`$ as
$`\varphi (M_i)`$ $`=`$ $`{\displaystyle \frac{1}{V}}\widehat{\psi }_i{\displaystyle \underset{j=1}{\overset{J}{}}}\widehat{\nu }_jV_j,`$ (52)
$`n(z_i)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Delta }M}{V}}\widehat{\nu }_jV_j{\displaystyle \underset{i=1}{\overset{I}{}}}\widehat{\psi }_i.`$ (53)
### 2.4 Lynden-Bell–Chołoniewski–Caditz–Petrosian (LCCP) Method
In this section, we discuss the method originally introduced by Lynden-Bell (1971) as the ‘$`C^{}`$ method’. The estimator of this method is an analog of the Kaplan–Meier estimator used for censored data analyses, like survival analysis (e.g. Feigelson & Nelson 1985; Feigelson 1992; Babu & Feigelson 1996; for reference of survival analysis itself, see e.g. Kleinbaum 1996). The method may be the most natural application of the nonparametric statistics to the problem (e.g. Petrosian 1992). The rederived version of the method by C87 was improved so that it could estimate the LF and density evolution of galaxies at the same time. In addition, the derivation of the estimator was much simplified. The original method was invented to estimate the cumulative LF as a step function, thus the differential LF was described as a weighted sum of Dirac’s $`\delta `$-function. But obviously this form is not practical, and C87 suggested to smooth the LF. In modern statistics, the kernel estimator is used in the problem of nonparametric density estimation (Silverman 1986; Lehmann 1999). The kernel is a smooth function which is used as a substitute of the delta function, in order to keep the estimated density function smooth. This improvement was introduced to the LF estimation problem by CP93, and used for a photometric redshift catalog by Subba Rao et al. (1996)<sup>3</sup><sup>3</sup>3But we note that Subba Rao et al.’s eq.(6) erroneously includes an extra exponential..
We unify these improvements, and show the practically convenient calculation here, which we call the ‘LCCP method’ after the names of the above contributors. We use the same notations for luminosity function, galaxy number density, distribution of galaxies, etc., and we consider the same absolute magnitude and redshift ranges as in section 2.3. But we must note that, in this subsection, indices represent the labels of galaxies. This method is completely free of binning procedure.
For the later discussion, we suppose that the galaxies are ordered as $`M_kM_{k+1}`$. In the LCCP method, the independence assumption is also adopted for $`M`$ and $`z`$, which leads to the expression
$`f(M,z)=\psi (M)\nu (z).`$
The empirical distribution (distribution of observational data) is expressed as
$`f_{\mathrm{obs}}(M,z)={\displaystyle \underset{k=1}{\overset{N_{\mathrm{obs}}}{}}}\delta (MM_k,zz_k),`$ (54)
again $`N_{\mathrm{obs}}`$ is the observed sample size, and let
$`\psi (M)`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}\psi _i\delta (MM_i),`$ (55)
$`\nu (z)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{N_{\mathrm{obs}}}{}}}\nu _j\delta (zz_j).`$ (56)
Then the empirical distribution is
$`f_{\mathrm{obs}}(M,z)`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}\psi _i\delta (MM_i){\displaystyle \underset{j=1}{\overset{N_{\mathrm{obs}}}{}}}\nu _j\delta (zz_j)\chi _S(M,z)`$ (57)
$`=`$ $`{\displaystyle \underset{(i,j)S}{}}\psi _i\nu _j\delta (MM_i)\delta (zz_j).`$
Here, $`\chi _S(M,z)`$ is the characteristic function of the set $`S`$ defined as
$`\chi _S(M,z)\{\begin{array}{cc}0\hfill & (M,z)S,\hfill \\ 1\hfill & (M,z)S.\hfill \end{array}`$ (60)
In the following discussions, the quantities $`M_{\mathrm{max}}(j)`$ and $`z_{\mathrm{max}}(i)`$ are defined as
$`M_{\mathrm{max}}(j)\mathrm{min}\{M_\mathrm{u},M_{S(j)}\},z_{\mathrm{max}}(i)\mathrm{min}\{z_\mathrm{u},z_{S(i)}\},`$ (61)
where $`M_{S(j)}\{M:𝒞\{(M,z):z=z_j\}\}`$, and $`z_{S(i)}\{z:𝒞\{(M,z):M=M_i\}\}`$. Though they look like those used in the subsection 2.3, we note again that the indices are of galaxies. These are schematically described in Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field. Integration of eq. (57) over the interval $`[M_\mathrm{l},M_\mathrm{u}],[z_k\epsilon ,z_k+\epsilon ]`$ ($`\epsilon >0`$) gives
$`{\displaystyle _{M_\mathrm{l}}^{M_\mathrm{u}}}{\displaystyle _{z_k\epsilon }^{z_k+\epsilon }}{\displaystyle \underset{\mathrm{}=1}{\overset{N_{\mathrm{obs}}}{}}}\delta (MM_{\mathrm{}},zz_{\mathrm{}})\mathrm{d}z\mathrm{d}M=1`$
$`=`$ $`{\displaystyle _{M_\mathrm{l}}^{M_\mathrm{u}}}{\displaystyle _{z_k\epsilon }^{z_k+\epsilon }}{\displaystyle \underset{(i,j)S}{}}\psi _i\nu _j\delta (MM_i)\delta (zz_j)\mathrm{d}z\mathrm{d}M`$
$`=`$ $`\nu _k{\displaystyle \underset{i=1}{\overset{M_i<M_{\mathrm{max}(k)}}{}}}\psi _i`$
therefore
$`\nu _j{\displaystyle \underset{i=1}{\overset{M_i<M_{\mathrm{max}(j)}}{}}}\psi _i=1,(j=1,\mathrm{},N_{\mathrm{obs}}).`$ (62)
Similarly, integration over the $`[M_k\epsilon ,M_k+\epsilon ],[z_\mathrm{l},z_\mathrm{u}]`$ gives
$`\psi _i{\displaystyle \underset{j=1}{\overset{z_j<z_{\mathrm{max}(i)}}{}}}\nu _j=1,(i=1,\mathrm{},N_{\mathrm{obs}}).`$ (63)
Formally, we can obtain $`\{\psi _i\}_{i=1,\mathrm{},N_{\mathrm{obs}}}`$ and $`\{\nu _j\}_{j=1,\mathrm{},N_{\mathrm{obs}}}`$ by solving the eqs. (62) and (63), and the estimates of the real galaxy distribution, $`f(M,z)`$, as
$`f(M,z)={\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}{\displaystyle \underset{j=1}{\overset{N_{\mathrm{obs}}}{}}}\psi _i\delta (MM_i)\nu _j\delta (zz_j).`$ (64)
Thus
$`\varphi (M)`$ $`=`$ $`{\displaystyle \frac{1}{V}}{\displaystyle _{z_\mathrm{l}}^{z_\mathrm{u}}}f(M,z)dz={\displaystyle \frac{1}{V}}{\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}\psi _i\delta (MM_i){\displaystyle \underset{j=1}{\overset{N_{\mathrm{obs}}}{}}}\nu _j,`$ (65)
$`n(z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{d}z}{\mathrm{d}V}}{\displaystyle _{M_\mathrm{l}}^{M_\mathrm{u}}}f(M,z)dM={\displaystyle \frac{\mathrm{d}z}{\mathrm{d}V}}{\displaystyle \underset{j=1}{\overset{N_{\mathrm{obs}}}{}}}\nu _j\delta (zz_j){\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}\psi _i,`$ (66)
where $`V`$ is the volume considered. The total number of galaxies $`N`$ is
$`N={\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}\psi _i{\displaystyle \underset{j=1}{\overset{N_{\mathrm{obs}}}{}}}\nu _j.`$ (67)
These solutions are MLEs as discussed in C87.
In spite of the clarity of the derivation, it is, actually, not an easy task to solve the eqs. (62) and (63) numerically if the data size $`N_{\mathrm{obs}}`$ is large. Thus we use the usual $`C`$ estimator together, in order to calculate the Chołoniewski’s coefficients more easily. The Lynden-Bell’s $`C^{}`$-function, $`C^{}(M_k)`$, is the number of galaxies in the region
$`\{\begin{array}{c}M_\mathrm{l}M<M_k,\hfill \\ z_\mathrm{l}zz_{\mathrm{max}(k)}.\hfill \end{array}`$ (70)
Let
$`C_kC^{}(M_k),k=1,\mathrm{},N_{\mathrm{obs}}.`$ (71)
Then, using eqs. (62) and (63), we have
$`C_k+1={\displaystyle \underset{i=1}{\overset{k}{}}}\psi _i{\displaystyle \underset{j=1}{\overset{z_j<z_{\mathrm{max}(k)}}{}}}\nu _j={\displaystyle \underset{i=1}{\overset{k}{}}}{\displaystyle \frac{\psi _i}{\psi _k}}.`$ (72)
and we obtain the following recursion relation:
$`\psi _{k+1}={\displaystyle \frac{C_k+1}{C_{k+1}}}\psi _k.`$ (73)
Thus, the distribution function of $`M`$ (cumulative LF), $`\mathrm{\Phi }(M)`$, is
$`\mathrm{\Phi }(M){\displaystyle \underset{k=1}{\overset{M_k<M}{}}}\psi _k=\psi _1{\displaystyle \underset{k=1}{\overset{M_k<M}{}}}{\displaystyle \frac{C_k+1}{C_k}}.`$ (74)
In the real procedure, we set $`(C_1+1)/C_1=1`$, so the product in the above equation begins with $`k=2`$. We can prove the second step of the above equation by mathematical induction. This is equivalent to the Lynden-Bell’s solution (C87). We obtain the weight $`\{\psi _i\}_{i=1,\mathrm{},N_{\mathrm{obs}}}`$ by eq. (73), and we are able to calculate the density weight $`\{\nu _j\}_{j=1,\mathrm{},N_{\mathrm{obs}}}`$ by eq. (63).
As we mentioned above, the weighted sum of the $`\delta `$-function is not a practically useful form, and random fluctuation would be serious in the region where the data points are sparse. Therefore, the kernel estimator, which is often used in modern nonparametric density estimation, was introduced by CP93. This estimator is simply obtained by replacing the $`\delta `$-function with a smooth kernel function $`\kappa `$ as
$`f(M,z)={\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}{\displaystyle \underset{j=1}{\overset{N_{\mathrm{obs}}}{}}}\psi _i\nu _j{\displaystyle \frac{1}{h_Mh_z}}\kappa \left({\displaystyle \frac{MM_i}{h_M}}\right)\kappa \left({\displaystyle \frac{zz_j}{h_z}}\right).`$ (75)
The minimum value of the ‘smoothing scale’ $`h`$ is restricted by the observational uncertainty, which was used by Subba Rao et al. (1996), but it does not provide sufficient smoothing in general (CP93). The optimal value of $`h_M`$ or $`h_z`$ may be estimated as
$`h_M\mathrm{max}\{M_{i+1}M_i\}_{i=1,\mathrm{},N_{\mathrm{obs}}},h_z\mathrm{max}\{z_{j+1}z_j\}_{j=1,\mathrm{},N_{\mathrm{obs}}}.`$ (76)
It is obvious that the larger the data size $`N_{\mathrm{obs}}`$ is, the smaller the smoothing scale becomes. Furthermore, CP93 discussed the effect of the kernel shape on the estimates. Now it is known that the best shape of the kernel is parabolic, so-called the Epanechnikov kernel (Epanechnikov 1969), because it gives the minimum variance (Lehmann 1999; van Es 1991):
$`\kappa (x)={\displaystyle \frac{3}{4}}(1x^2).`$ (77)
It should be noted that, in principle, the kernel estimator is asymptotically biased, i.e. the expectation value is slightly different from the true value even if the sample size is large.
## 3 TEST OF THE METHODS BY SIMULATION
### 3.1 Numerical Examination with Mock Catalogs
The validity of the estimation methods of the LF is often examined by mathematical statistics. For example, their statistical unbiasedness and statistical convergence were discussed in many early works (e.g. Felten 1976). However, quantitative evaluation frequently appears to be difficult by such approach, and numerical examination is quite important. Jackson (1974) used numerical experiments, as well as the analytical error estimation by Fisher’s information matrix (see Stuart, Ord, & Arnold 1999), in the study of quasar LF, and EEP also checked the errors of their method by Monte Carlo simulations as well as traditional information matrix approach. Mobasher, Sharples, & Ellis (1993) performed Monte Carlo error estimation to test the special method developed to construct the LF at a certain waveband from the data selected at another wavelength. Heyl et al. (1997) examined the effect of galaxy clustering to the LF estimation by their extended EEP method. But computer-aided extensive intercomparison between the estimators had not been performed until the work of W97. They discussed the performance of several estimators when the LF is represented by the Schechter form
$`\varphi (M)\mathrm{d}M=0.4\mathrm{ln}10\varphi _{}10^{0.4(\alpha +1)(MM_{})}\mathrm{exp}\left(10^{0.4(MM_{})}\right)\mathrm{d}M,`$ (78)
and tested the results in some cases with different Schechter parameters. Their main conclusions are as follows:
1. The STY and $`C^{}`$ methods are the best.
2. The $`1/V_{\mathrm{max}}`$ method gives biased results and tends to give higher values for the faint-end slope even for spatially homogeneous samples.
3. The STY fit tends to underestimate the faint-end slope.
4. The mean densities (normalization of the LF) recovered by most estimators are lower than the input values by factors (up to 20 %).
Among these, the second one looks most strange, because as we mentioned in section 2.1, Felten has proved mathematically that the $`1/V_{\mathrm{max}}`$ estimator is unbiased when the homogeneous assumption holds. The $`1/V_{\mathrm{max}}`$ method is quite frequently used in the estimation for the LFs of quasars, clusters of galaxies, etc., and if W97’s claim is true, some widely accepted conclusions must be significantly changed. Thus it is necessary to examine the estimators further, not only for the Schechter form but for various shapes of the LF, in order to clarify the trends of the results.
In this section, we test the four estimators discussed in the previous section by using simulated mock galaxy samples with a variety of the LFs which have the following functional forms:
1. Uniform distribution,
2. Power-law form which increases toward fainter magnitude,
3. Power-law form which decreases toward fainter magnitude,
4. Gaussian distribution (with standard deviation 1.67 mag),
5. Schechter form (steep faint-end slope: $`\alpha =1.6`$),
6. Schechter form (flat faint-end slope: $`\alpha =1.1`$),
with magnitude range $`M=[24,14]`$ (Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field). The first three forms are designed to examine the effect of LF slope for estimation, and the form D, Gaussian, is to check the effect of curvature of the function. Power-law LF of the form B appears ubiquitously in various types of objects. The form C looks apparently unrealistic, but we added this for making thorough investigation. The form D is interesting because approximate Gaussian form is often found in the LFs of individual galaxy types. We applied Box–Muller method (Box & Muller 1958) to generate Gaussian distribution from uniform random number, and von Neumann’s acception–rejection method to obtain other distributions (see Knuth 1998 for details). We set the sample sizes $`100`$ and $`1000`$, to study the behavior of the statistical estimators with galaxy number. Here ‘sample size’ means the detected number of galaxies after magnitude selection (observation) procedures. Therefore the underlying population density for each LF form is different from each other. The estimation of galaxy spatial density is an important part of the derivation of the LF. What to be estimated is the total galaxy number including the galaxies too faint to be observed. In our simulations, we stochastically produced galaxies according to the assumed LF, distribute them in space, calculate their observed flux, and judge that they could be observed or not. Therefore, the total number corresponds to the number of Monte Carlo trials. We fixed the number of trials through one sequence of simulations with a certain LF shape and spatial density.
#### 3.1.1 Mock Catalog with Spatially Homogeneous Distribution
First we construct a set of mock galaxy samples with spatially homogeneous distribution in order to investigate the bias trend of the estimators, especially for the Schmidt and Eales’ $`1/V_{\mathrm{max}}`$. We set the redshift range up to 0.1, and we adopted the Hubble parameter $`H_0=75\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$, $`\mathrm{\Omega }_0=0.2(q_0=0.1)`$, $`\lambda _0=0`$, and limiting magnitude $`m_{\mathrm{lim}}=13`$ mag in the series of simulations. No $`K`$-correction is considered here. We constructed 100 representations for each LF form and sample size, and applied the four estimators to each sample.
#### 3.1.2 Mock Catalog with a Dense Cluster and with a Void
We, next, investigate the response against density inhomogeneity of galaxies. We consider some extreme cases for clear understanding. For the case with density enhancement, we constructed a series of mock catalogs with a dense spherical clump, to which half of the galaxies belong. The clump lies at a distance of 0.8 Mpc, and its radius is 0.8 Mpc. We call the mock catalog the “cluster sample”. An example of the spatial configuration of galaxies of a cluster sample is described in Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field. Then we also constructed a set of the mock catalogs with a large spherical void without galaxies. The void lies at a distance of 0.8 Mpc and its radius is 1.6 Mpc. We call this mock catalog the “void sample”. The overall underlying density of cluster and void samples defined in a considered volume is the same as the homogeneous samples for each LF shape, i.e. we set the number of Monte Carlo trials the same as that of the homogeneous sample for each LF shape. Therefore, the observed sample size of the cluster sample is larger than that of the homogeneous sample, because we put the dense clump in the considered volume. In the case of the void sample, the observed galaxy number is smaller than that of the homogeneous one.
#### 3.1.3 Results
The results for the 1000-galaxy samples are shown in Figure 5 – 10. The solid lines represent the input distributions, and the symbols are the averages of the estimates. The error bars depict the standard deviations of the mean of the estimates for 100 representations. Figures 5a, 5b, and 5c are the results from the spatially homogeneous sample, from the cluster sample, and from the void sample, respectively. This is also the same for Figures 6 – 10.
At a glance, we see that all estimators give consistent results with each other, and we do not find any bias trends in our numerical experiments for any LF forms in the case of homogeneous samples. For cluster samples, the $`1/V_{\mathrm{max}}`$ method yields strongly distorted estimations, as widely recognized. The overestimation corresponding to the dense clump clearly appeared in the $`1/V_{\mathrm{max}}`$ results. In contrast, the other three estimators were not affected by the dense cluster at all. The estimates appeared to be consistent with each other, and showed perfect agreement with the input LFs. The $`1/V_{\mathrm{max}}`$ method was also affected by the large void, and gave underestimated results.
Large fluctuations appear at the faint end of the LF, because the number of available data points is small, especially in the case of the LF form C and D. We can obtain statistically stable estimates if the slope is properly steep, and the more shallow the slope is, the larger the fluctuation becomes. This is clearly shown in Figure 5 – 10.
In principle, the error bar of the Chołoniewski method is larger than those of the other methods, because the method subdivides the $`Mz`$ plane both in $`M`$ and $`z`$. This procedure enables us to estimate the shape, the normalization, and the evolution of the LF at the same time. On the other hand, this becomes a drawback when the data size is small, because the shot noise dominates. Therefore we cannot expect a firm estimation with the Chołoniewski method when the sample size is smaller than 100.
Here we mention the calculation time that each method needed for the same sample size. Because of its algorithmic simplicity, Chołoniewski method is the fastest among the four methods. When we analyze the 1000-mock data, the relative calculation times of the $`1/V_{\mathrm{max}}`$, EEP, and LCCP methods normalized with that of Chołoniewski method are 2.76, 2.73, and 1.87, respectively. This advantage is quite significant when we treat a large sample of $`10^{45}`$ galaxies. We estimate the LFs from large datasets of sample size 250,000 in Section 3.2 by the $`1/V_{\mathrm{max}}`$, EEP, and Chołoniewski methods. The relative calculation times of the $`1/V_{\mathrm{max}}`$ and EEP methods normalized with that of Chołoniewski method are, in this case, 15.01 and 131.74, respectively. The EEP method takes longer calculation time because it needs more iterations in the procedure than others do. The $`1/V_{\mathrm{max}}`$ method derives the maximum volume $`V_{\mathrm{max}}`$ for each galaxy, and also needs some calculation time. The LCCP method requires a large stack for data sorting procedure, which is a requirement of this method. Thus we stress that the Chołoniewski method is most economic from the standpoint of practical computing.
Figures 11 – 16 are the same as Figures 5 – 10, except that the data size is 100. We see it is often not possible to determine the faint end of the LF accurately for such small datasets. The fluctuation became larger than the result of the 1000-sample, but we did not find the systematic bias trend from our results. Thus we conclude that when the galaxy distribution is homogeneous, all four estimators provide the consistent and correct results, even the $`1/V_{\mathrm{max}}`$ estimator.
### 3.2 Mock 2dF Redshift Catalog
The Anglo–Australian 2-degree field (2dF) galaxy redshift survey is now underway<sup>4</sup><sup>4</sup>4See http://msowww.anu.edu.au/~colless/2dF for recent status.. This survey will measure 250,000 redshifts, up to $`z0.2`$, and be complete to an extinction corrected apparent magnitude of $`b_\mathrm{J}<19.45`$ mag. In order to develop statistical methods and faster algorithms for the analyses of such large upcoming redshift surveys, Cole et al. (1998) prepared an extensive set of mock 2dF catalogs constructed from a series of large cosmological $`N`$-body simulations. The simulations span a wide range of cosmological models, with various values of the density parameters, $`\mathrm{\Omega }_0`$, the cosmological constant, $`\lambda _0`$, and the shape parameter $`\mathrm{\Gamma }`$ and amplitude of the density fluctuation $`\sigma _8`$. The LF is assumed to be a Schechter form with the parameters reported by APM-Stromlo bright galaxy survey (Loveday et al. 1992), $`M_{b_\mathrm{J}}5\mathrm{log}h=19.5`$ mag, $`\alpha =0.97`$, and $`\varphi _{}=1.4\times 10^2h^3\mathrm{Mpc}^3`$. The $`K`$-correction is assumed to be canceled by evolutionary correction.
We applied the three methods to the mock 2dF catalog in order to see how accurately they can reproduce the true LF when they are used in the analysis of realistic large redshift surveys. We did not use the LCCP method for this sample. When we treat such a large catalog, the advantage of the Chołoniewski methods is extremely significant. We also focused on the difference between the real-space data and the redshift-space data which is affected by the redshift distortion. The redshift distortion causes a scatter in the estimated luminosities of galaxies. In this study, we used three mock catalogs, named E1 (Einstein–de Sitter: $`\mathrm{\Omega }_0=1,\lambda _0=0,\mathrm{\Gamma }=0.5,\sigma _8=0.55`$), L3S ($`\mathrm{\Omega }_0=0.3,\lambda _0=0.7,\mathrm{\Gamma }=0.25,\sigma _8=1.13`$), and O3S ($`\mathrm{\Omega }_0=0.3,\lambda _0=0,\mathrm{\Gamma }=0.25,\sigma _8=1.13`$). The catalogs we selected are all cluster-normalized, i.e. the amplitude of the initial power spectrum is set to reproduce present abundance of rich galaxy clusters in the local Universe (e.g. Viana & Liddle 1996; Kitayama & Suto 1997) and $`h=\mathrm{\Gamma }/\mathrm{\Omega }_0`$.
We compare the input LF and the estimated LF in Figures Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field, Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field, and Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field. Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field shows the LF derived from the Einstein–de Sitter (EdS) Universe, Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field is the LF derived from L3S data, and Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field is the LF derived from O3S data. The left panels in these Figures show the LFs derived from the redshift-space data, and the right panels, those from the real-space data. First, we see that all the estimators provided perfectly consistent results, and they show an excellent agreement with the input LF. There are no significant difference between the real- and redshift-space datasets. The slight deviations of $`1/V_{\mathrm{max}}`$ estimates are caused by the clustering in the 2dF mock catalog. Thus we do not have to consider the redshift distortion effect seriously when we derive the galaxy LF from such large-volume redshift surveys. When we use such a large survey, we should rather mention the photometric calibration as a more important error source.
## 4 APPLICATION TO THE HUBBLE DEEP FIELD
Recently some authors claim that the faint-end slope of the LF becomes steeper with redshift at $`z<1`$ (e.g. Ellis et al. 1996; Heyl et al. 1997; but see Lin et al. 1999). The LFs for some special classes of galaxies such as Lyman-break objects (Steidel et al. 1998) or Ly-$`\alpha `$ emitters (Pascarelle et al. 1999) are now also available. We, however, do not have a coherent understanding of the evolution of the LF and the evolution of the luminosity density, $`\rho _\mathrm{L}`$. At low redshift, Zucca et al. (1997) reported a high normalization LF with $`\varphi _{}=0.020h^3\mathrm{Mpc}^3`$, and Ellis et al. (1996) obtained $`\varphi _{}=0.026h^3\mathrm{Mpc}^3`$, while Loveday et al. (1992) derived $`\varphi _{}=0.014h^3\mathrm{Mpc}^3`$, and Las Campanas Redshift Survey result (Lin et al. 1996) is similar to the value of Loveday et al. (1992). The local value of the LF parameters plays a crucial role in the study of galaxy evolution, since it controls the redshift dependence of $`\rho _\mathrm{L}`$. Cowie et al. (1999) showed a rather mild evolution of the UV luminosity density at $`z<1`$ from their surveys. On the other hand, high redshift LF estimations are also controversial with each other. Gwyn & Hartwick (1996) claimed dramatic changes in the LF from $`z=0`$ to $`z5`$, becoming flat between $`24M_B15`$ for $`3<z<5`$. On the contrary, SLY97 reported more familiar Schechter form with $`\alpha =1.3`$ for the LF at $`3<z<4`$. Mobasher et al. (1996) suggested a stronger evolution of the LF. From a deep multiband photometric survey, Bershady et al. (1997) gave a constraint which ruled out Gwyn & Hartwick’s result.
Thus in this section, we apply the four estimators to the photometric redshift catalog of the HDF to study the evolution of the LF shape. For the observational data, the error estimation is complicated, because the estimation procedure of the LF involves the magnitude selection, weighting, etc. In such cases, bootstrap resampling analysis is known to be often superior to classical analytic methods in order to estimate statistical properties (e.g. Efron & Tibshirani 1993; Babu & Feigelson 1996; Davison & Hinkley 1997). Thus we used the bootstrap method for the estimation of the statistical uncertainties. When we perform the bootstrapping, how to generate good random numbers is important. We generated the uniform random number by Mersenne Twister method<sup>5</sup><sup>5</sup>5 For recent development, see http://www.math.keio.ac.jp/matumoto/mt.html. developed by Matsumoto & Nishimura (1998).
### 4.1 Sample
We used the photometry and photometric redshift catalog of the HDF prepared by FLY99. Their catalog contains 1067 galaxies, with $`\mathrm{AB}(8140)<26.0`$. The photometric redshifts are derived based on both UBVI (F300W, F450W, F606W, and F814W, respectively; Williams et al. 1996) obtained by WFPC2, and JHK obtained by the IRIM camera on the Kitt Peak National Observatory 4-m telescope. The object detection and photometry are performed using SExtractor (Bertin & Arnouts 1996). Details of the procedures are found in FLY99. In the peripheral region of the WFPC2 image (referred to as zone 2), the detection limit is $`\mathrm{AB}(8140)=26`$ mag, and in the inner region (zone 1), $`\mathrm{AB}(8140)=28`$ mag. We restricted our analysis to the inner zone 1 sample. The solid angle of zone 1 is $`3.92\mathrm{arcmin}^2=3.32\times 10^7\mathrm{sr}`$. The sample size is then 946 galaxies.
FLY99 used four spectral templates given by Coleman, Wu, & Weedman (1980) to determine the photometric redshifts. For ultraviolet wavelengths, the templates are extrapolated by using the results of Kinney et al. (1993), and for infrared, by the models of Bruzual & Charlot (1993). Evolutionary corrections are not included in the model spectra to avoid additional parameter dependence. According to Coleman et al. (1980), they classified the galaxy spectra into four categories: 1. Elliptical, 2. Sbc, 3. Scd, and 4. Irr. We used these labels to set the $`K`$-corrections.
In principle, the SED must be the same as the templates used in FLY99, but for simplicity and comparison with other studies, we used the galaxy SED sample compiled by Kinney et al. (1996). The data of Kinney et al. (1996) have almost the same properties as the SED templates of FLY99, thus we can use them comfortably. To construct the $`K`$-correction function, we first selected the sample galaxy SEDs corresponding to the labels of FLY99, and fitted polynomial functions from 1st order to 6th order. The order of polynomial fitting was decided by referring to AIC, and we chose the 5th order.
### 4.2 Results and Discussions
We show the redshift-dependent LF at $`I`$-band and $`B`$-band in Figures Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field and Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field, respectively. The symbols represent the estimated LFs by the four methods. We show the LFs of the HDF at $`0<z<0.5`$ (106 galaxies), $`0.5<z<1.0`$ (193), $`1.0<z<1.5`$ (204), $`1.5<z<2.0`$ (193), $`2.0<z<3.0`$ (117), and $`3.0<z<6.0`$ (109). The sample is $`I`$-band selected, and we derived the $`B`$-band LF by following the discussion of Lilly et al. (1995). We stress that the four different LF estimators give consistent results for the HDF sample, same as the results for the mock catalogs.
We clearly see the evolutionary trend of the LF with redshift. But we note that, though we can fit Schechter function, it is not so easy to derive the parameters $`\alpha `$, $`M_{}`$, or $`\varphi _{}`$ precisely, because the Schechter function is rather smooth and the errors of these characteristic parameters are strongly correlated. These parameters can be easily affected by statistical fluctuations. We will discuss more details of the $`I`$\- and $`B`$-band results at each redshift range in the following.
#### 4.2.1 $`I`$-band LF at $`0<z<0.5`$
In Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field, the dotted line represents the local $`I`$-band LF obtained by Metcalfe et al. (1998). Metcalfe et al. (1998) pointed out a possible upturn of the faint end of their multiband LFs, though they took a prudent attitude in concluding firmly. The upturn magnitude $`M_I15+\mathrm{log}h`$ mag (in Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field, $`h=0.75`$) is in good agreement with that of our lowest redshift LF except the normalization.
#### 4.2.2 Evolution at $`B`$-band: $`0<z<0.5`$
We compare the normalization of the LF with other previous results. Our $`B`$-band LF shows roughly good agreement with other local LFs. In Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field we put our LF, SLY97 Schechter fit, a nd Schechter functions reported by Metcalfe et al. (1998) and Ellis et al. (1996) (Autofib Redshift Survey) The dotted line depicts SLY97 LF, dot-dashed line represents the Metcalfe et al. (1998) $`B`$-band LF, and long dashed line is the Autofib LF at $`z<0.1`$. Our LF and that of SLY97 agree with higher-normalization LF reported by Autofib Survey, but are significantly higher than that of Metcalfe et al. (1998) Autofib LF is also consistent with the LF of ESP Survey (Zucca et al. 1997), while Metcalfe et al.’s LF is consistent with EEP LF and Stromlo–APM LF (Loveday et al. 1992). But since the solid angle covered by HDF is extremely small and thus the normalization can be strongly affected by cosmic variance, we should not go into further discussion.
We note that the SLY97 $`M_{}`$ value is significantly higher than those of other surveys. This is because the exponential decline at bright end is not observed in the HDF LF at $`0<z<0.5`$, and a bump exists at $`M_B20`$ mag. We should also mention that the error bar of the $`M_{}`$ of SLY97 is very large (1.6 mag). Considering the large error bars and the uncertainty of the photometric redshift, we conclude that the bright $`M_{}`$ is not a real feature.
At this lowest redshift, the rise of the faint end is prominent. The problem of the faint-end slope of galaxy LF has long been a matter of debate, and we do not have a widely accepted consensus yet. As we already pointed out in the above, even in $`I`$-band we find a steepening of the faint end. If this steep faint end is the artifact of the clustering, the LF derived from $`1/V_{\mathrm{max}}`$ and those derived from other estimators should have been different (Section 3.1). But in fact, they are consistent with each other. Thus we conclude that, at least in the HDF, the faintest end of the LF has a steep slope in the Local Universe.
#### 4.2.3 Evolution at $`B`$-band: $`0.5<z<1.0`$
It seems that the brighter galaxies are more numerous than the local value at $`0.5<z<1.0`$. Here we should remember the fact that the “fuzzy” redshift determination is known to affect the shape estimation (Liu et al. 1998). Liu et al. (1998) showed by numerical experiments that the faint-end slope is underestimated and $`M_{}`$ is overestimated by the photometric redshift blurring. The uncertainty of the photometric redshift is rather independent of the object redshift, so the effect will be severer at the low-$`z`$, and the $`M_{}`$ can be overestimated. Thus the increase of the bright galaxies is partially due to this effect. But we can discuss the trend of the LF evolution by comparison of the LF derived from photometric redshifts consistently (Liu et al. 1998).
#### 4.2.4 Evolution at $`B`$-band: $`1.0<z<2.0`$
The LFs of the redshift range $`1.0<z<1.5`$ and $`1.5<z<2.0`$ are the most reliable ones among the LFs in Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field, since the sample size is twice larger than those of the other redshift ranges, and in addition, the photometric redshift error becomes worse again at $`z>2`$. SLY97 suggested the steepening of the faint-end slope at this redshift. Our LF of $`1.5<z<2.0`$ presents a similar feature, though the slope becomes flatter at the faintest regime. The deformation of the LF from $`z0`$ to $`z2`$ supports that the steepening of the fainter side of the LF, which is confirmed at $`z<1`$, is continued up to $`z2`$. We do not find a significant shift of $`M_{}`$ at this redshift range.
#### 4.2.5 Evolution at $`B`$-band: $`2.0<z`$
The normalization of the furthest redshift LFs settles down to the local value, while we also find a brightening of $`M_{}`$ at $`z>3.0`$. We must be careful that in such high redshift, cosmological surface brightness dimming is quite severe, and selection effect becomes significant (Ferguson 1998; Weedman et al. 1998). Other kinds of selection effects are discussed in Pascarelle et al. (1998). Therefore, there can exist more numerous galaxies than estimated. Further discussions require delicate treatment of such effects.
#### 4.2.6 Luminosity density evolution
In order to explore the cosmic star formation history, we derived the luminosity density at $`B`$\- and $`I`$-band based on our LFs. We fit the Schechter function and extrapolate the faint end below the detection limit. As we mentioned above, the Schechter parameters are poor indicators of the galaxy evolution, but the integrated luminosity density $`\rho _\mathrm{L}`$ is regarded as an indicator of the evolution of galaxies, because in case the Schechter parameters are significantly affected by the fluctuations, $`\rho _\mathrm{L}`$ is robust against the effect. We showed the derived $`\rho _\mathrm{L}`$ in Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field. The upper panel shows the evolution of the $`B`$-band luminosity density, $`\rho _\mathrm{L}(B)`$, and the lower, the $`I`$-band luminosity density, $`\rho _\mathrm{L}(I)`$. Open squares are $`\rho _\mathrm{L}(B)`$ derived from CFRS (Lilly et al. 1995), open circles, $`\rho _\mathrm{L}(B)`$ from Autofib (Ellis et al. 1996), open triangle represents the value from Stromlo-APM (Loveday et al. 1992) and open diamond, ESP value (Zucca et al. 1997). Crosses are the estimates of SLY97. In this paper we did not try to correct for the reddening effect of dust.
We see the local diversity of the $`\rho _\mathrm{L}(B)`$, corresponding to the normalization discrepancy in Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field. Despite the fact that the local LF is hard to derive from the HDF data, our low-$`z`$ value is consistent with other previous results. Added to this, our $`\rho _\mathrm{L}(B)`$ at $`0.5<z<1.0`$ significantly suffers from the redshift blurring effect, but it is also consistent with CFRS highest redshift point within the errors. As a whole, our result is consistent with that of SLY97, except for $`2.0<z<3.0`$. In this redshift range, $`\rho _\mathrm{L}(B)`$ of SLY97 is several times larger than our estimate. This difference may be because SLY97 obtained a steeper $`\alpha `$ and brighter $`M_{}`$ than ours. We find a flatter LF slope, and the estimates fainter than $`20`$ mag are not reliable in our result since the fluctuation is horribly large at this redshift. If we choose steeper slope, our $`\rho _\mathrm{L}`$ will be higher. We have to wait for larger datasets to address this problem. At very high-$`z`$, we derived a moderately high $`\rho _\mathrm{L}`$, implying significant numbers of stars have already formed at such a high redshift. The evolution of $`\rho _\mathrm{L}(I)`$ appears to be flat. At the longer wavelength, the observed light is dominated by the contribution from lower-mass stars, and the temporal change of the SFR is less prominent. We, in addition, should note that the $`I`$-band results are subject to larger $`K`$-correction extrapolation uncertainties, compared with the $`B`$-band results.
At last, we must notice that the above discussions do not account for the fact that the sample is selected at $`I`$-band, and the selection criterion is different for each redshift range. At $`z<0.5`$, the sample is safely regarded as $`I`$-selected, while at $`z>3.0`$, they are in fact rest UV-selected. Thus the ideal discussion on the evolution of the galaxy LF should be based on the suitably designed survey as performed by Cowie et al. (1999). We will consider this point, and make more sophisticated discussions elsewhere (Takeuchi 2000, in preparation).
## 5 SUMMARY AND CONCLUSION
The estimation of the LF from observational data is not a trivial task, because of the flux-limited nature of the astronomical data. We focused on the following four estimators: 1) Schmidt–Eales ($`1/V_{\mathrm{max}}`$) method , 2) Efstathiou–Ellis–Peterson (EEP) method, 3) Chołoniewski method, and 4) Lynden-Bell–Chołoniewski–Caditz–Petrosian (LCCP) method. We improved some of the estimators for studying the very distant universe, and examined their performances for much wider class of functional forms by Monte Carlo simulation. We tested these four estimators by the numerical experiments with mock catalogs. We also used the mock 2dF catalogs prepared by Cole et al. (1998). Then we applied these estimators to the HDF photometric redshift catalog of Fernández-Soto et al. (1999). Our conclusions are as follows:
1. If the sample is spatially homogeneous, all estimators give consistent results with each other, and we did not find any bias for any LF shapes. Thus, when we have a sufficiently large galaxy sample, we can use any of the estimators examined in this paper. Even when the sample size is smaller, the mean values remain unbiased, though the standard deviations become larger.
2. Large fluctuation appears at the faint end of the LF, because the amount of available data is small. Therefore, the flatter the LF slope is, the larger the fluctuations become. When the sample size is small, fluctuations in the Chołoniewski method become seriously large due to shot noise, and thus we recommend this method for the analysis of large samples.
3. When a large cluster or void exists, $`1/V_{\mathrm{max}}`$ estimator is severely affected in its LF shape estimation. The other three estimators are not affected by a cluster or void at all. They gave consistent results with each other, and the estimates showed perfect agreement with the input LFs.
4. We examined the calculation time of each method. Because of its algorithmic simplicity, Chołoniewski method is the fastest among the four methods. The EEP method needs more iterations in the procedure than others do, and longer calculation time. The $`1/V_{\mathrm{max}}`$ method calculates the maximum volume $`V_{\mathrm{max}}`$ for each galaxy, and also needs significant calculation time. The LCCP method requires a large stack for data sorting procedure, which is a requirement of this method. Thus we stress that the Chołoniewski method is the most economic from the standpoint of practical computing.
5. We examined more realistic large mock samples, specifically mock 2dF catalogs prepared by Cole et al. (1998). We found that the redshift distortion does not affect the LF estimates. When we treat such a large catalog, the advantage of the Chołoniewski method is extremely significant in terms of the computation time.
6. We derived the $`I`$\- and $`B`$-band luminosity function of the HDF. The four different LF estimators gave consistent results for the HDF sample. We found the overall brightening of the LF. It seems that the faint-end steepens toward $`z=23`$, and settles down to the local value at $`z3`$. We note that the “fuzzy” redshift determination is known to affect the shape estimation (Liu et al. 1998).
7. We found a rather mild evolution of the LF. Despite the fact that the local LF is hard to derive from the HDF data, our low-$`z`$ value is consistent with other previous results. Our $`\rho _\mathrm{L}(B)`$ at $`0.5<z<1.0`$ is also consistent with CFRS highest redshift point within the errors. As a whole, our result is roughly consistent with that of SLY97, but lower at $`2.0<z<3.0`$. At very high-$`z`$, we derived a moderately high $`\rho _\mathrm{L}`$, implying that a significant numbers of stars have already formed at such a high redshift. We found that the evolution of $`\rho _\mathrm{L}(I)`$ is flat.
First we would like to thank the anonymous referee for his useful suggestions and comments, which improved our paper very much in its clarity and English presentation. We offer our gratitude to Hiroyuki Hirashita and Fumiko Eizawa who gave a lot of useful suggestions. We also thank Kouji Ohta, Kouichiro Nakanishi, Toru Yamada, Takashi Ichikawa, Kazuhiro Shimasaku for their fruitful discussions and useful comments. Mamoru Saitō, Hiroki Kurokawa and Yasushi Suto are thanked for their continuous encouragements. This work owes a great debt to the photometric redshift catalog prepared by Fernández-Soto et al. TTT and KY acknowledge the Research Fellowships of the Japan Society for the Promotion of Science for Young Scientists. We carried out the numerical computations and extensively used the databases at the Astronomical Data Analysis Center of the National Astronomical Observatory, Japan, which is an inter-university research institute of astronomy operated by Ministry of Education, Science, Culture, and Sports.
## Appendix A AKAIKE’S INFORMATION CRITERION
In this appendix, we make an informal introduction of Akaike’s theory. The meaning of the maximum likelihood method is clearly understood by using the concepts of information theory. Since the middle of 1970’s, vast advances have been made in the field of the statistical inference by the discovery of Akaike’s Information Criterion (AIC: Akaike 1974). The AIC is closely related to the information entropy, especially to the ‘relative entropy’ of two probability distributions. The relative entropy has a property just like a distance in differential geometry, i.e. it is a distance between the two probability distributions. Using AIC enables us to compare the goodness of a certain model with that of another type directly. For this fascinating property, AIC is applied to various fields of studies. The AIC is expressed as
$`\mathrm{AIC}=2(\mathrm{ln}(\widehat{\theta })K),`$
where $``$ is a likelihood function, $`\widehat{\theta }`$ is a set of maximum likelihood estimators, and $`K`$ is the number of free parameters of the assumed model. The “most preferred” model is the one which minimizes the AIC.
Here we present the problem of polynomial regression model selection by using AIC as an example. As we mentioned in Section 4, we adopted this procedure to determine the order of $`K`$-correction as a function of redshift. Given a set of $`n`$ pairs of observations $`(x_1,y_1),\mathrm{},(x_n,y_n)`$, we fit the $`m`$-th order polynomial model
$`y_i={\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}a_{\mathrm{}}x_i^{\mathrm{}}+\epsilon _i,`$ (A1)
where $`\epsilon _i`$ is an independent random variable which follows the normal distribution with mean $`0`$ and dispersion $`\sigma ^2`$. The variables $`x_i`$ and $`y_i`$ are called the explanatory variable and the objective variable, respectively. This model is a conditional distribution of which the distribution of the objective variable $`y`$ is a normal distribution $`f(y_i)`$ with the mean $`a_0+a_1x_i+\mathrm{}+a_mx_i^m`$ and the variance $`\sigma ^2`$, i.e.
$`f(y_i|a_0,\mathrm{},a_m,\sigma ^2)={\displaystyle \frac{1}{\sqrt{2\pi \sigma ^2}}}\mathrm{exp}\left\{{\displaystyle \frac{1}{2\sigma ^2}}\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}a_{\mathrm{}}x_i^{\mathrm{}}\right)^2\right\}.`$ (A2)
Therefore, when a set of data is $`(x_1,y_1),\mathrm{},(x_n,y_n)`$, the likelihood is given by
$`(y_1,\mathrm{},y_n|a_0,\mathrm{},a_m,\sigma ^2)`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}f(y_i|a_0,\mathrm{},a_m,\sigma ^2)`$ (A3)
$`=`$ $`\left({\displaystyle \frac{1}{\sqrt{2\pi \sigma ^2}}}\right)^{\frac{n}{2}}{\displaystyle \underset{i=1}{\overset{n}{}}}\mathrm{exp}\left\{{\displaystyle \frac{1}{2\sigma ^2}}\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}a_{\mathrm{}}x_i^{\mathrm{}}\right)^2\right\}.`$
The log likelihood is then expressed as
$`\mathrm{ln}(y|a_0,\mathrm{},a_m,\sigma ^2)`$ $`=`$ $`{\displaystyle \frac{n}{2}}\mathrm{ln}2\pi {\displaystyle \frac{n}{2}}\mathrm{ln}\sigma ^2{\displaystyle \frac{1}{2\sigma ^2}}{\displaystyle \underset{i=1}{\overset{n}{}}}\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}a_{\mathrm{}}x_i^{\mathrm{}}\right)^2.`$ (A4)
The log likelihood eq. (A4) is maximized with respect to $`a_0,\mathrm{},a_m`$ when
$`S{\displaystyle \underset{i=1}{\overset{n}{}}}\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}a_{\mathrm{}}x_i^{\mathrm{}}\right)^2.`$ (A5)
is minimized. Thus, in the case of polynomial model fitting, the maximum likelihood procedure is equivalent to the least square method. The necessary conditions that $`a_0,\mathrm{},a_m`$ maximize $`S`$ are the normal equations of the least square,
$`{\displaystyle \frac{S}{a_0}}`$ $`=`$ $`2{\displaystyle \underset{i=1}{\overset{n}{}}}\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}a_{\mathrm{}}x_i^{\mathrm{}}\right)=0`$
$`{\displaystyle \frac{S}{a_1}}`$ $`=`$ $`2{\displaystyle \underset{i=1}{\overset{n}{}}}x_i\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}a_{\mathrm{}}x_i^{\mathrm{}}\right)=0`$
$`\mathrm{}`$
$`{\displaystyle \frac{S}{a_m}}`$ $`=`$ $`2{\displaystyle \underset{i=1}{\overset{n}{}}}x_i^m\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}a_{\mathrm{}}x_i^{\mathrm{}}\right)=0,`$ (A6)
and the maximum likelihood estimates $`\widehat{a_0},\mathrm{},\widehat{a_m}`$ are obtained by solving these linear equations. Besides, the necessary condition that $`\sigma ^2`$ maximizes eq. (A4) is
$`{\displaystyle \frac{\mathrm{ln}}{\sigma ^2}}={\displaystyle \frac{n}{2\sigma ^2}}+{\displaystyle \frac{1}{2(\sigma ^2)^2}}{\displaystyle \underset{i=1}{\overset{n}{}}}\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}\widehat{a_{\mathrm{}}}x_i^{\mathrm{}}\right)^2=0.`$ (A7)
The maximum likelihood estimate of the residual variance $`\sigma ^2`$ is
$`\widehat{\sigma ^2}={\displaystyle \frac{1}{n}}{\displaystyle \underset{i=1}{\overset{n}{}}}\left(y_i{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}\widehat{a_{\mathrm{}}}x_i^{\mathrm{}}\right)^2={\displaystyle \frac{1}{n}}\left({\displaystyle \underset{i=1}{\overset{n}{}}}y_i^2{\displaystyle \underset{\mathrm{}=0}{\overset{m}{}}}\widehat{a_{\mathrm{}}}{\displaystyle \underset{i=1}{\overset{n}{}}}x_i^{\mathrm{}}y_i\right).`$ (A8)
Hereafter we denote the residual variance $`\sigma ^2`$ for a model with $`m`$-th order as $`\sigma ^2(m)`$. Then, from eqs. (A4) and (A8), the maximum log likelihood becomes
$`\mathrm{ln}(y|\widehat{a_0},\mathrm{},\widehat{a_m},\widehat{\sigma ^2})={\displaystyle \frac{n}{2}}\mathrm{ln}2\pi {\displaystyle \frac{n}{2}}\mathrm{ln}\widehat{\sigma ^2}(m){\displaystyle \frac{n}{2}}.`$ (A9)
The $`m`$-th order polynomial model has $`m+2`$ parameters ($`a_0,\mathrm{},a_m,\sigma ^2(m)`$). Substituting $`K=m+2`$ and eq. (A9) into eq. (A) gives the AIC of the $`m`$-th order model,
$`\mathrm{AIC}(m)=n(\mathrm{ln}2\pi +1)+n\mathrm{ln}\widehat{\sigma ^2}(m)+2(m+2).`$ (A10)
The result of our polynomial fitting to the $`K`$-correction is presented in Figure Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field. We also summarize the AIC value for each polynomial order in Table Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field. The AIC values of elliptical, Sbc, and Irr in table Tests of Statistical Methods for Estimating Galaxy Luminosity Function and Applications to the Hubble Deep Field really took their minima in the case that the fitting polynomials were those with 5th order, and only Scd data preferred the 6th order. Putting all accounts together, we chose 5th order polynomial model.
Figure Captions<sup>1</sup><sup>1</sup>1All figures listed below are available from ftp://ftp.kusastro.kyoto-u.ac.jp/pub/kohji/lf/ . |
warning/0003/cond-mat0003109.html | ar5iv | text | # Double-Layer Systems at Zero Magnetic Field
## I Introduction
In the last several years, double-layer electron and hole systems have provided an exceptionally useful tool for investigating the effects of interparticle Coulomb interactions in two dimensions, particularly at low particle densities where exchange and correlation effects are significant. This has been especially true in the quantum Hall regime, where the combination of a strong perpendicular magnetic field (which quenches the kinetic energy) and very small layer separation (which enhances interlayer exchange) stabilizes remarkable interlayer-coherent quantum Hall states. Even in the absence of any magnetic field, interlayer capacitance and drag measurements in double-layer electron systems (2LES’s) have provided quantitative measures of the effects of electronic interactions on both the thermodynamics and transport (Coulomb drag) of two-dimensional electron and hole systems. Unless otherwise specified, we shall take the notation 2LES to also include double-layer hole systems.
Our work is motivated by the rapid pace of advancements in the engineering of double-layer semiconductor devices. We expect that high-mobility double-layer devices will eventually be built with both (1) separately contactable layers, and (2) layer separations and carrier densities so small that the interlayer correlations between the carriers are substantial, perhaps even without the aid of a strong quantizing magnetic field. Such devices will allow direct measurements of the effects interlayer many-body effects in double-layer systems. As a starting point to analyze the zero magnetic field situation, we have developed a simple mean-field model that incorporates both intralayer and interlayer exchange in biased double-layer electron and hole systems in the absence of a magnetic field. We use the model to calculate theoretically the effect of interlayer exchange on the layer densities and interlayer capacitance as a function of layer spacing, particle density, and applied gate voltage.
Gated double-layer systems have the great advantage of allowing the layer densities of electrons or holes to be varied by the application of a bias (gate voltage). At high densities, the kinetic energy per unit area dominates the exchange and correlation energies, and in a translationally invariant system, the symmetry and properties of the ground state of a many-body system are in one-to-one correspondence with those of a free-electron gas. At lower electron densities, the Coulombic exchange and correlation energies can produce qualitative changes in the nature of the many-particle ground state: numerical work on two- and three-dimensional electron gases show that a spin ferromagnetic state is obtained at low densities, which is eventually supplanted by a Wigner crystal state at the lowest densities. The low-density ferromagnetic state of the interacting electron gas was anticipated some 70 years ago by Bloch; ferromagnetism can be found even within the Hartree-Fock approximation when the exchange interaction energy, which favors occupation of single-particle states of the same spin, dominates the kinetic energy, which favors reducing the Fermi energy by equal occupation of both spin states.
Multilayer semiconductor devices enhance the effects of interparticle interactions through the combination of reduced dimensionality and low particle density, and by the presence of an additional electronic degree of freedom, the layer index. In double-layer systems, layer occupancy can be specified by a introducing a pseudospin variable that points up for one layer and down for the other layer. Extending the notion of an exchange-driven ferromagnetic transition to double-layer systems suggests that at low enough densities, the electronic ground state should be both spin and pseudospin polarized. Such reasoning, supported by Hartree-Fock calculations, led Ruden and Wu to propose that, at low enough electron densities and for small enough layer separations, the electrons of a 2LES would minimize their ground-state energy by having all electrons occupy a single layer. The Ruden-Wu scenario implies that as the total density of a balanced 2LES is lowered, the electrons should eventually experience an interlayer charge transfer instability that spontaneously empties out one of the two layers. Ruden and Wu also suggested that the inherent bistability of the resulting low-density 2LES (either layer could be the one to lose or gain particles) would constitute an exchange-driven logic gate. The possibility of an exchange-driven interlayer charge-transfer instability has been a subject of both theoretical and experimental interest.
Although pseudospin polarization at sufficiently low densities is likely, it does not require unequal layer densities. This has been demonstrated theoretically in great detail for closely spaced double-layer systems in a strong magnetic field at unit filling factor and appears to give a good explanation of experimental results. The key point is that electrons in double-layer structures are not restricted to occupying only one of two layer eigenstates: quantum mechanics allows states that are superpositions of the two layers. The layer pseudospin must therefore be treated as a Heisenberg variable, as was done in Refs. and in the quantum Hall regime, and by Zheng and co-workers in zero magnetic field, rather than as an Ising variable (where only “up” and “down” allowed) as was done by Ruden and Wu. For example, when interlayer tunneling is present, the single-particle eigenstates are symmetric and antisymmetric combinations of layer states. A major insight of Refs. was the concept of “spontaneous interlayer coherence” (SILC): at sufficiently small layer separations, electrons can spontaneously create and occupy linear combinations of layer states in which each layer has the same average number of electrons, even without any interlayer tunneling. The spontaneous formation of superposed layer states in double-layer quantum Hall (2LQH) sytems can be accomplished by the interlayer exchange interaction alone. SILC in balanced 2LQH systems corresponds to $`XY`$ pseudospin ferromagnetism in which the pseudospins spontaneously magnetize, but do not point either up or down, since neither layer has (on average) more particles than the other.
The application of SILC to the zero magnetic field case was first made by Zheng and co-workers, who considered the same model system as Ruden and Wu – electrostatically balanced zero-thickness layers of interacting electrons without interlayer tunneling – but came to a very different conclusion. They proved within the Hartree-Fock approximation (HFA) that (1) the 2LES becomes spin ferromagnetic before it becomes pseudospin ferromagnetic for any finite layer separation (this possibility was not considered by Ruden and Wu), and (2) at low enough densities and small enough layer separations, the pseudospin ferromagnetic state possesses SILC, with all electrons occupying one subband composed of a superposition of layer states with equal average density in each layer. Conti and Senatore have also argued that for electrostatically balanced layers, the 2LES ground state cannot be one in which all electrons are eigenstates of the same single layer. SILC at sufficiently small layer separations should also follow from earlier calculations presented in Ref. , although no explicit inference of SILC was made in that work. Recent work that goes beyond the HFA and includes correlation effects within the STLS approximation also finds that SILC is favored over single-layer occupancy for balanced layers that are sufficiently close together.
When studying a 2LES using a density-functional approach, it is important to note that SILC is a nonlocal effect. Calculations based on local-density approximations will not find SILC if they treat the pseudospin as an Ising-like variable. The same caveat applies to the work of Ruden and Wu, who used a restricted HFA that excluded the possibility of SILC. Once the intralayer separation between electrons becomes comparable to the interlayer separation between layers, the possibility of interlayer correlations (such as SILC) must be considered. The spontaneous charge-transfer state predicted by Ruden and Wu follows quite generally (even beyond the HFA) from the fact that when interlayer correlations are ignored, the negative compressibility of the electron gas guarantees an interlayer charge-transfer instability at sufficiently small layer spacing. But it is precisely at small layer spacings that interlayer correlations become important, and so their effects must be included to obtain physically meaningful results.
We have extended the study the effects of Coulombic exchange in double-layer electron and hole systems to include an applied bias due to front and back gate voltages, while allowing for the possibility interlayer exchange. In the balanced case, we have found that the four- to two-component transition is always interrupted by the presence of a three-component phase with slightly unequal layer densities. There are therefore four possible noncrystalline phases for a 2LES with balanced gates. Under bias or tunneling a second (pseudospin-polarized) two-component state is also possible. We have enumerated the transitions between the five allowed states in the presence of bias, and explored the effects of bias on the one-component state.
The rest of this paper is organized as follows: In Sec. II, we introduce a simplified model for double-layer systems, review the concept of interlayer capacitance, and give a general criterion for stability against spontaneous interlayer charge transfer. In Sec. III, we develop a mean-field approximation for biased double-layer systems that allows for the possibility of interlayer coherence. In Sec. IV, we examine the special case of electrostatically balanced layers, enumerate the resulting four possible noncrystalline phases, and explore the onset of interlayer coherence and its effect on the size of the subband splitting. We also obtain a phase diagram for the balanced case, and perform an alternate calculation for the onset of the one-component phase. In Sec. V, we explore the effect of bias on the layer occupancies and interlayer capacitance for large, intermediate, and small layer separations. We develop simple models capable of closely fitting experimental layer-occupancy data, and expore the transitions between different phases induced by layer imbalance. In Sec. VI, we analyze the onset and properties of the one-component state under bias. We summarize our findings and speculate on the possible relevance of these results to the strong magnetic-field regime in Sec. VII.
## II Double-Layer Model and Interlayer Capacitance
In this section, we introduce an idealized model of double-layer systems. We review the condition for thermodynamic equilibrium between the inner layers, obtain a necessary condition for stability against interlayer charge transfer, and review an experimentally useful measure of the interlayer capacitance, the Eisenstein ratio.
Figure 1 illustrates schematically the geometry of the 2LES device. We treat the quantum wells as zero-thickness layers sandwiched between two plates of neutralizing charge, which represent the effects of the front and back gates. The distance between the front gate (at far left) and the first layer (layer 1) is $`D_F`$; that between the back gate (at far right) and the second layer (layer 2) is $`D_B`$; the interlayer separation is denoted by $`d`$. Typically, $`d10`$ nm, $`D_F1\mu `$m, and $`D_B1`$ mm, so that $`dD_FD_B`$; thus, Fig. 1 is not at all to scale.
The two inner layers are assumed to be in thermodynamic equilibrium with each other, and the voltage of the front (back) gate relative to the common chemical potential of the inner layers is denoted by $`V_F`$ ($`V_B`$). It is also assumed that small changes $`\delta V_\alpha `$ in the gate voltages $`V_\alpha `$ ($`\alpha =F,B`$) produce small changes in surface charge densities only at the gates ($`e\delta p_\alpha `$) and in the layers ($`e\delta n_i`$, $`i=1,2`$). Overall charge neutrality requires that the total charge density vanish:
$$ep_F+ep_Ben_1en_2=0.$$
(1)
(Strictly speaking, we only require that the change in the total charge density vanish: $`\delta ep_F+\delta ep_B\delta en_1\delta en_2=0`$.) In writing Eq. (1), we have assumed that any stray charge not included in Eq. (1) is unchanged when the gate voltage are varied. This implies that the only significant effect of the stray charges is to shift the gate voltages by constant (empirically determined) amounts. Sheet charge densities on the gates and inner layers produce electric fields between the double-layer and the gates ($`E_\alpha `$) and between the two layers ($`E_{12}`$) according to Gauss’s law,
$`E_\alpha `$ $`=`$ $`(e/ϵ)p_\alpha ,`$ (2)
$`E_{12}`$ $`=`$ $`(e/ϵ)(p_Fn_1)=(e/ϵ)(n_2p_B).`$ (3)
We now obtain the conditions for thermodynamic equilibrium and stability between the layers. Regarding the gate charge densities $`p_\alpha `$ as fixed quantities, we seek the values of the particle density in the inner layers that minimize the total energy per unit area $`_0/L_xL_y`$. Following Ref. , we separate the electrostatic part of the total energy per unit area from the rest:
$$\frac{_0}{L_xL_y}=\frac{ϵ}{2}E_{12}^2d+\epsilon (n_1,n_2),$$
(4)
up to an irrelevant constant, where the first term is the interlayer electrostatic energy density, with $`E_{12}`$ given by Eq. (2). The quantity $`\epsilon (n_1,n_2)`$ represents the total energy per unit area for a fully interacting double-layer system in which each layer contains a uniform neutralizing charge density, for particle densities $`n_1`$ and $`n_2`$ in layers 1 and 2, respectively. The Fermi, exchange, and correlation energies, both intralayer and interlayer, are contained in $`\epsilon (n_1,n_2)`$, and it is this quantity that is calculated using many-body techniques. In the next section, we make an approximate calculation of $`\epsilon (n_1,n_2)`$ which includes the effects of interlayer and intralayer exchange.
To obtain the condition for thermodynamic equilibrium between the inner layers, we note that for fixed external charge densities $`p_\alpha `$, the constraint of overall charge neutrality implies that the particle density in one layer (e.g, $`n_1`$) is determined by that in the other layer (e.g., $`n_2`$): $`n_1=p_F+p_Bn_2`$. We may thus regard the total energy per unit area $`_0/L_xL_y`$ as a function of the layer density $`n_2`$, and extremize $`_0/L_xL_y`$ with respect to $`n_2`$ at fixed $`p_F`$ and $`p_B`$ to obtain
$$\mu _1\mu _2=eE_{12}d,$$
(5)
where
$$\mu _i\epsilon (n_1,n_2)/n_i$$
(6)
is the chemical potential measured relative to the energy minimum of layer $`i`$. Equation (5) states that the difference in the layer values of the chemical potential is equal to the electrostatic potential energy difference between the layers. If the equation of state determining $`\mu _i(n_1,n_2)`$ were known, then Eqs. (2) and (5) would together determine the values of layer densities $`n_1`$ and $`n_2`$ for which the total energy $`_0/L_xL_y`$ is an extremum.
We now examine a necessary condition for interlayer thermodynamic stability (i.e., for the local extremum to be a local minimum of the energy). First we follow Ref. and introduce a set of lengths that describe the dependence of the layer chemical potentials $`\mu _i`$ on the layer densities $`n_j`$,
$$s_{ij}\frac{ϵ}{e^2}\frac{\mu _i}{n_j}=\frac{ϵ}{e^2}\frac{^2\epsilon }{n_jn_i}.$$
(7)
For the extremum condition in Eq. (5) to represent a local minimum of the total energy per unit area, we require that the second derivative of $`_0/L_xL_y`$ with respect to $`n_2`$ be positive. (We again regard $`n_1`$ as being determined by $`n_2`$ for fixed $`p_\alpha `$ by the requirement for overall charge neutrality.) This gives a necessary condition for stability:
$$d+s_1+s_2>0,$$
(8)
where
$$s_1s_{11}s_{12},s_2s_{22}s_{21}.$$
(9)
The above inequality guarantees that the 2LES is at least metastable. If Eq. (8) is violated, then the layer densities constitute a local energetic maximum rather than a local minimum, and there will be an interlayer charge instability that causes charge to flow between the layers until a new energetic minimum is reached.
In the absence of interlayer correlations (the case considered in Refs. and ), $`s_{12}=s_{21}=0`$, and the length $`s_{ii}`$ is directly related to the electronic compressibility $`\kappa _i`$ in layer $`i`$ according to
$$s_{ii}=\frac{ϵ}{e^2n_i^2\kappa _i}.$$
(10)
Equations (8) and (10) together with the experimentally measured negative compressibility ($`\kappa <0`$) of the electrons imply that when interlayer correlations are ignored ($`s_{12}=0`$), sufficiently small interlayer separation $`d`$ will always lead to a charge-transfer instability when the density of either (or both) layer is sufficiently small. This result is true even when intralayer correlations are included beyond the HFA calculation of Ruden and Wu; only interlayer correlations that produce sufficiently negative values of $`s_{12}`$ to satisfy Eq. (8) can suppress interlayer charge-transfer instabilities at very small interlayer separations.
If the two layers of the 2LES sample can be contacted separately, then the Eisenstein ratio $`R_E`$ provides a sensitive measure of the interlayer capacitance that avoids the large gate-distance factors that dominate the gate capacitances per unit area. The Eisenstein ratio is defined as the ratio of the differential change in the electric field $`E_{12}`$ between the inner layers to that of the electric field $`E_F`$ between the front gate and the inner layers:
$$R_E\frac{\delta E_{12}}{\delta E_F}=1\delta n_1/\delta p_F,$$
(11)
where we have made use of Gauss’s law, Eq. (2). In the following sections, we calculate the layer occupancies $`n_i`$ as a function of the gate charges $`p_\alpha `$; we then use Eq. (11) to obtain $`R_E`$ by computing the derivative of $`n_1`$ with respect to $`p_F`$. In the classical limit (corresponding to large enough particle densities and layer separations so that only the electrostatic energies are relevant), $`n_1=p_F`$ (for $`p_F>0`$), so that by Eq. (11), $`R_E=0`$. Note also that if $`n_1=0`$ (e.g., due to $`p_F<0`$), then $`R_E=1`$. By using Eqs. (5) and (7) to express differential changes in $`E_{12}`$ in terms of the electronic lengths $`s_{ij}`$ and using Gauss’s law, the Eisenstein ratio may be expressed as
$$R_E=\frac{s_1s_2\delta E_B/\delta E_F}{d+s_1+s_2}.$$
(12)
The Eisenstein ratio has an especially simple form for fixed total density since then $`\delta E_B=\delta E_F`$; from Eq. (12), $`R_E=s/(d+s)`$, where $`ss_1+s_2`$. However, most experiments fix the back-gate voltage $`V_B`$ rather than the total density, and sweep the front-gate voltage $`V_F`$. Because of the relatively large size of the back-gate distance $`D_B`$, the constraint of fixed back-gate voltage ($`\delta V_B=0`$) is very nearly equivalent to that of fixed back-gate sheet charge density ($`\delta p_B=0`$ or, by Gauss’s law, $`\delta E_B=0`$). For $`\delta E_B=0`$, Eq. (12) shows that the Eisenstein ratio is very nearly
$$R_E\left(\delta E_{12}/\delta E_F\right)_{p_B}=\frac{s_1}{d+s_1+s_2}.$$
(13)
The advantage of measuring $`R_E`$ rather than the usual gate capacitances per unit area (such as $`e\delta p_F/\delta V_F`$) is that $`R_E`$ depends only on the electronic lengths $`s_i`$ and the interlayer distance $`d`$, not on the much larger gate distances $`D_\alpha `$. The difficulty in measuring $`R_E`$ is that (at least one of) layers 1 and 2 must be separately contactable, which becomes increasingly difficult as the layer separation $`d`$ becomes very small. Nonetheless, measurements of the Eisenstein ratio have been used to demonstrate the negative compressibility of the electron gas, and it is to be expected that devices with separately contactable layers will be built with increasingly narrow layer separations. Note also that $`R_E`$ is very sensitive to charge-transfer instabilities. In fact, from Eq. (8), the condition for the onset of an instability to interlayer charge transfer, $`d+s_1+s_2=0`$, shows that $`R_E`$ formally diverges at the instability. This can also be seen from the relation $`R_E=1\delta n_1/\delta p_F`$, since a charge transfer instability would produce an abrupt change in the layer density $`n_1`$ in response to a small change in the gate density $`p_F`$. Although evidence for abrupt interlayer charge transfers has been reported based on Shubnikov-de Haas (SdH) measurements, the Eisenstein ratio would be a far more sensitive measure of abrupt interlayer charge transfers.
In subsequent sections, we calculate the layer densities $`n_i`$ as a function of the front-gate particle density $`p_F`$, for fixed back-gate particle density $`p_B`$. Fortunately, $`p_B`$ may be found experimentally from SdH measurements as the value of the layer densities when the system is balanced: i.e., for equal layer densities ($`n_1=n_2`$) and minimum subband separation $`(n_an_b)`$. Once $`p_B`$ (which we assume is very nearly constant) is known, $`p_F`$ may be determined from charge conservation by measuring the total layer density, either by SdH (as the sum of the subband densities) or by Hall effect measurements. It is therefore possible to determine $`p_F`$ experimentally, without recourse to the gate voltages. Of course, experimentally, it is $`V_F`$ that is varied directly while $`V_B`$ is kept fixed, and $`p_F`$ changes in response to $`V_F`$ (while $`p_B`$ changes very little for large $`D_B`$). We describe how the gate voltages may be determined from a knowledge of $`p_F,p_B`$, and $`\mu _i`$ in Sec. 1 of the Appendix.
## III Mean-Field Approximation
In this section, we introduce the microscopic Hamiltonian for the gated double-layer system and make a variational approximation for the ground-state wave function that allows for interlayer coherence. The resulting approximate ground-state energy per unit area depends on both intralayer and interlayer exchange.
We now consider the microscopic Hamiltonian for the double-layer system illustrated in Fig. 1. We idealize the inner layers as being two-dimensional, and treat interlayer tunneling in the tight-binding approximation. The Hamiltonian for the interacting system is then
$``$ $`=`$ $`{\displaystyle \underset{j𝐤s}{}}\epsilon _kc_{j𝐤s}^{}c_{j𝐤s}t{\displaystyle \underset{𝐤s}{}}(c_{1𝐤s}^{}c_{2𝐤s}+c_{2𝐤s}^{}c_{1𝐤s})`$ (14)
$`+`$ $`{\displaystyle \frac{1}{2L_xL_y}}{\displaystyle \underset{𝐪}{}}{\displaystyle \underset{j_1𝐤_1s_1}{}}{\displaystyle \underset{j_2𝐤_2s_2}{}}V_{j_1j_2}(q)`$ (16)
$`\times c_{j_1𝐤_\mathrm{𝟏}+𝐪s_1}^{}c_{j_2𝐤_\mathrm{𝟐}𝐪s_2}^{}c_{j_2𝐤_\mathrm{𝟐}s_2}c_{j_1𝐤_\mathrm{𝟏}s_1}`$
$``$ $`{\displaystyle \underset{j𝐤s}{}}{\displaystyle \underset{\alpha }{}}V_{j\alpha }(q=0)p_\alpha c_{j𝐤s}^{}c_{j𝐤s}`$ (17)
$`+`$ $`{\displaystyle \frac{L_xL_y}{2}}{\displaystyle \underset{\alpha \beta }{}}V_{\alpha \beta }(q=0)p_\alpha p_\beta ,`$ (18)
where $`c_{j𝐤s}`$ ($`c_{j𝐤s}^{})`$ denotes the second-quantized destruction (creation) operator for an electron or hole in layer $`j`$ with momentum $`\mathrm{}𝐤`$ and spin $`s`$. Here $`\epsilon _k=\mathrm{}^2k^2/2m^{}`$ is the kinetic energy in the effective mass ($`m^{}`$) approximation, $`t`$ is the interlayer tunneling amplitude, and the Fourier-transformed Coulomb potential is given by
$$V_{ij}(q)=\frac{e^2}{2ϵq}e^{qd_{ij}},$$
(19)
where $`d_{ij}`$ is the distance between layer $`i`$ and layer $`j`$. The indices $`\alpha ,\beta `$ in Eq. (18) take the values $`F`$ (front gate) or $`B`$ (back gate), $`V_{\alpha \beta }`$ is the direct Coulomb interaction between gates $`\alpha `$ and $`\beta `$, and $`V_{\alpha j}`$ is the direct Coulomb interaction between gate $`\alpha `$ and layer $`j`$. The last two terms in Eq. (18) represent the direct interaction between the layers and gates, and between front and back gates, respectively.
We use a mean-field approximation (MFA) that is variationally based and that reduces to the Hartree-Fock approximation for the balanced case of equal layer densities. This approximation includes the effects of interlayer correlations in the simplest possible way. The variational ground-state wave function is composed of two subbands, $`a`$ and $`b`$, containing spin up ($``$) and spin down ($``$) electrons:
$$|\mathrm{\Psi }_0=\underset{𝐤_\mathrm{𝟒}}{\overset{k_4k_b}{}}b_{𝐤_\mathrm{𝟒}}^{}\underset{𝐤_\mathrm{𝟑}}{\overset{k_3k_b}{}}b_{𝐤_\mathrm{𝟑}}^{}\underset{𝐤_\mathrm{𝟐}}{\overset{k_2k_a}{}}a_{𝐤_\mathrm{𝟐}}^{}\underset{𝐤_\mathrm{𝟏}}{\overset{k_1k_a}{}}a_{𝐤_\mathrm{𝟏}}^{}|0,$$
(20)
where $`k_{as}`$ and $`k_{bs}`$ denote the Fermi wave vectors for electrons or holes of spin $`s`$ in subbands $`a`$ and $`b`$. The creation operators for the subbands are related to the layer creation operators by a canonical transformation that we take to be of the form
$`a_{𝐤s}^{}`$ $`=`$ $`\mathrm{cos}(\theta /2)c_{1𝐤s}^{}+\mathrm{sin}(\theta /2)e^{i\varphi }c_{2𝐤s}^{}`$ (21)
$`b_{𝐤s}^{}`$ $`=`$ $`\mathrm{sin}(\theta /2)e^{i\varphi }c_{1𝐤s}^{}+\mathrm{cos}(\theta /2)c_{2𝐤s}^{}`$ (22)
When $`\theta =\pi /2`$ and $`\varphi =0`$, subband $`a`$ is the symmetric subband and subband $`b`$ is the antisymmetric subband. In the language of pseudospin, the superposition of layer states in Eq. (22) corresponds to treating the layer pseudospin as a Heisenberg, rather than an Ising, spin variable. The form of the canonical transformation in Eq. (22) is not completely equivalent to a fully self-consistent Hartree-Fock calculation because we have taken $`\theta `$ and $`\varphi `$ to be independent of the wave vector $`𝐤`$ and spin $`s`$. It would be interesting to explore the effects of including the $`𝐤`$ and $`s`$ dependence of $`\theta `$ and $`\varphi `$ in a future calculation. Our simpler variational calculation, which is equivalent to Ref. for the special case of balanced layers, offers a reasonable starting point, which is probably qualitatively correct over a large range of layer densities. It certainly gives layer densities that are in close agreement with experimental values obtained from SdH measurements, as we shall show.
The layer occupation numbers may be expressed in terms of the subband occupation numbers by using Eq. (22):
$`c_{1𝐤s}^{}c_{1𝐤s}`$ $`=`$ $`\mathrm{cos}^2(\theta /2)a_{𝐤s}^{}a_{𝐤s}+\mathrm{sin}^2(\theta /2)b_{𝐤s}^{}b_{𝐤s}`$ (23)
$`c_{2𝐤s}^{}c_{2𝐤s}`$ $`=`$ $`\mathrm{sin}^2(\theta /2)a_{𝐤s}^{}a_{𝐤s}+\mathrm{cos}^2(\theta /2)b_{𝐤s}^{}b_{𝐤s}`$ (24)
$`c_{1𝐤s}^{}c_{2𝐤s}`$ $`=`$ $`\mathrm{sin}(\theta /2)\mathrm{cos}(\theta /2)e^{i\varphi }(a_{𝐤s}^{}a_{𝐤s}b_{𝐤s}^{}b_{𝐤s})`$ (25)
$`c_{2𝐤s}^{}c_{1𝐤s}`$ $`=`$ $`c_{1𝐤s}^{}c_{2𝐤s}^{},`$ (26)
where the asterisk denotes complex conjugation, and we have used Eq. (20) to eliminate cross terms such as $`a_{𝐤s}^{}b_{𝐤s}`$. SILC occurs when
$$c_{1𝐤s}^{}c_{2𝐤s}0$$
(27)
in the absence of interlayer tunneling. According to Eqs. (26) and (27), SILC requires that the following occurs: (1) $`\theta 0,\pi `$ so that the subband densities are different from the layer densities. SILC is therefore excluded when the pseudospin is treated as an Ising variable. (2) $`n_an_b`$ so that the subband densities are not equal (nonzero pseudospin polarization). When $`n_{as}=n_{bs}`$ (completely unpolarized pseudospin), the MFA ground state can be expressed as the product of two uncorrelated single-layer wavefunctions by performing a global pseudospin rotation.
Although it is straightforward to generalize our approach to finite temperature, we calculate numerical results in the limit of zero temperature ($`T=0`$), both for the sake of simplicity and because measurements can (and have) been made on double-layer systems at low temperatures, even down to millikelvin temperatures in the quantum Hall regime. For the zero-magnetic-field case treated here, we expect that finite temperature will not produce signifcant qualitative changes in the layer densities if the temperature $`T`$ is below a fraction of the Fermi temperature $`T_FE_F/k_B`$, where $`E_F`$ is Fermi energy and $`k_B`$ is Boltzmann’s constant. For a layer density of $`n=10^{10}`$cm<sup>-2</sup>, $`T_F`$ is roughly 4 K for $`n`$-type GaAs, and 1K for $`p`$-type GaAs. The scale of the Hartree charge transfer energy, $`e^2dn/2ϵ`$, is larger than the Fermi energy except for ultrasmall layer separations ($`d<5`$ nm for $`n`$-type GaAs and $`d<1`$ nm for $`p`$-type GaAs.)
The other reason we work at zero temperature is to address matters of principle, such as whether an interlayer charge transfer instability can occur when the layers are very close together; finite temperatures would presumably smear out such a transfer, if it could occur. In the limit of zero temperature, Eq. (20) implies that
$$a_{𝐤s}^{}a_{𝐤s}=\mathrm{\Theta }(k_{as}k),b_{𝐤s}^{}b_{𝐤s}=\mathrm{\Theta }(k_{bs}k),$$
(28)
where the subband Fermi wave vectors and number densities are related through
$$k_{as}=\sqrt{4\pi n_{as}},k_{bs}=\sqrt{4\pi n_{bs}}.$$
(29)
Summing Equations (26) over wave vector $`𝐤`$ relates the number densities of the layers to those of the subbands:
$`n_{1s}+n_{2s}`$ $`=`$ $`n_{as}+n_{bs},`$ (30)
$`n_{1s}n_{2s}`$ $`=`$ $`(n_{as}n_{bs})\mathrm{cos}\theta .`$ (31)
We may use the preceding equations to express the ground-state energy per unit area in terms of the subband occupancies $`n_{\alpha s}`$ and the angle $`\theta `$:
$`{\displaystyle \frac{_0}{L_xL_y}}`$ $`=`$ $`{\displaystyle \frac{1}{\nu _0}}{\displaystyle \underset{\alpha s}{}}n_{\alpha s}^2t(n_an_b)\mathrm{sin}\theta \mathrm{cos}\varphi `$ (32)
$`+`$ $`{\displaystyle \frac{e^2d}{8ϵ}}\left[(n_an_b)\mathrm{cos}\theta (p_Fp_B)\right]^2`$ (33)
$``$ $`{\displaystyle \frac{1}{2L_xL_y}}{\displaystyle \underset{𝐪\alpha s}{}}V_{11}(q)I_{\alpha \alpha s}(q)`$ (34)
$`+`$ $`{\displaystyle \frac{\mathrm{sin}^2\theta }{4L_xL_y}}{\displaystyle \underset{𝐪s}{}}\left[V_{11}(q)V_{12}(q)\right]`$ (36)
$`\times \left[I_{aas}(q)+I_{bbs}(q)2I_{abs}(q)\right],`$
where $`\nu _0=m^{}/(\pi \mathrm{}^2)`$ is the density of states per unit area for noninteracting spin-1/2 particles in two dimensions, $`𝐪=𝐤_1𝐤_2`$, and
$`I_{\alpha \beta s}(q){\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐊}{}}`$ $`\mathrm{\Theta }(k_{\alpha s}|𝐊+𝐪/2|)`$ (38)
$`\times \mathrm{\Theta }(k_{\beta s}|𝐊𝐪/2|).`$
Here $`𝐊=(𝐤_1+𝐤_2)/2`$, and the subband indices $`\alpha `$ and $`\beta `$ can be either $`a`$ or $`b`$. Equation (38) says that $`I_{\alpha \beta s}(q)`$ is $`1/(2\pi )^2`$ times the common area of two circles of radii $`k_{\alpha s}`$ and $`k_{\beta s}`$ whose centers are separated by $`q`$. When $`\beta =\alpha `$, then $`k_{\beta s}=k_{\alpha s}`$, and the first exchange integral in Eq. (36) may be carried out explicitly:
$$\frac{1}{2L_xL_y}\underset{𝐪}{}V_{11}(q)I_{\alpha \alpha s}(q)=\frac{8}{3\sqrt{\pi }}\frac{e^2}{4\pi ϵ}n_{\alpha s}^{3/2}.$$
(39)
Equation (39) is just the exchange energy per unit area for a uniform single-layer spin-polarized two-dimensional electron gas of areal density $`n_{\alpha s}`$.
The last term in Eq. (36), which contains the interlayer exchange contribution, may be conveniently expressed as
$$\frac{\mathrm{sin}^2\theta }{4}\frac{e^2d}{2ϵ}(n_an_b)^2\mathrm{\Gamma },$$
(40)
where $`\mathrm{\Gamma }`$ is the interlayer exchange parameter, given by
$`\mathrm{\Gamma }`$ $``$ $`{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐪s}{}}\left[{\displaystyle \frac{V_{11}(q)V_{12}(q)}{e^2d/2ϵ}}\right]`$ (42)
$`\times \left[{\displaystyle \frac{I_{aas}(q)+I_{bbs}(q)2I_{abs}(q)}{(n_an_b)^2}}\right].`$
The properties of $`\mathrm{\Gamma }`$ are described in Sec. 4 of the Appendix. The last term in Eq. (36) must in general be evaluated numerically, although it vanishes at $`d=0`$ or when $`n_{as}=n_{bs}`$. It also vanishes if $`\theta `$ is $`0`$ or $`\pi `$, in which case subband $`a`$ is the layer (1 or 2) with most particles, while subband $`b`$ is the layer with the fewest particles. Ruden and Wu implicitly treated the layer pseudospin as an Ising-like variable with $`0`$ and $`\pi `$ as the only allowed values for $`\theta `$; in their approximation, the last term in Eq. (36) vanishes, and the interlayer effects we shall discuss here do not appear.
For definiteness, we take $`n_an_b`$, $`n_\alpha n_\alpha `$, and $`0\theta \pi `$. Our procedure consists of finding the values of $`n_{\alpha s}`$ and $`\theta `$ that minimize the expected energy per unit area, Eq. (36). Within our variational approximation, we find (as in the HFA) that the spins in a given subband are always either completely polarized (ferromagnetic at sufficiently low subband densities) or completely unpolarized (paramagnetic at higher densities). Real systems probably possess intermediate polarization for a range of low densities. For finite $`t>0`$, the ground-state energy per unit area, Eq. (36), is minimized for $`\varphi =0`$. In the absence of interlayer tunneling, the ground-state energy per unit area is independent of $`\varphi `$, provided that $`\varphi `$ is constant; for convenience we set $`\varphi =0`$. The layer densities $`n_{1s}`$ and $`n_{2s}`$ may be obtained from $`n_{\alpha s}`$ and $`\theta `$ via Eq. (31). We begin our calculations in the next section, by considering the case of electrostatically balanced gates.
## IV Balanced Gates
In this section, we consider the case of electrostatically balanced gates ($`p_F=p_B`$), beginning with zero interlayer tunneling. This was the situation originally considered by Ruden and Wu, and more recently in Refs. , and . For balanced gates, our approximation is equivalent to the unrestricted HFA of Zheng and co-workers, and except for our analysis of the three-component phase, most of our results agree with theirs.
### A Zero tunneling
The balanced case raises an important question of principle: can exchange and correlation effects alone, unaided by applied gate biases and unhindered by interlayer tunneling, ever produce a ground state in which the densities of the inner layers are not equal? Based on a restricted HFA (which did not allow for SILC) Ruden and Wu proposed that for small enough layer densities and layer separations, the answer is yes. Zheng and co-workers argued recently that an unrestricted HFA (which allows for, but does not mandate, SILC) gives the opposite answer. We find that, except for a small region in density which supports a three-component phase that has a slight layer imbalance, the layer densities are equal when the gates are balanced.
#### 1 Zero layer separation
In order to classify the four noncrystalline phases that we find for gate-balanced double-layer systems, it is useful to begin with the idealized case of zero layer separation ($`d=0`$). For $`d=0`$, the Hamiltonian is invariant under spin rotation, pseudospin rotation, and the interchange of spin and pseudospin. It is the same as the Hamiltonian for a four-layer system of spinless fermions with zero separation between all the layers: layers, subbands, and spins become interchangeable labels for the four components. The $`d=0`$ double-layer system is therefore equivalent to a single-layer two-dimensional system of fermions with $`CP(3)`$ symmetry. At $`d=0`$, the interlayer Hartree energy is zero and the interlayer ($`V_{12}`$) and intralayer ($`V_{11}`$) Coulomb interactions are equal. As a consequence, the variational energy in Eq. (36) is independent of $`\theta `$ and (for $`t=0`$) of $`\varphi `$ when $`d=0`$,
$$\frac{_0}{L_xL_y}=\underset{\alpha s}{}\left[\frac{n_{\alpha s}^2}{\nu _0}\frac{8}{3\sqrt{\pi }}\frac{e^2}{4\pi ϵ}n_{\alpha s}^{3/2}\right]\underset{\alpha s}{}\epsilon (n_{\alpha s}),$$
(43)
and the MFA is equivalent to the HFA.
For generality, we first consider an $`N`$-component system, where $`N`$ is twice the number of layers: $`N=4`$ for double-layer spin-1/2 systems. At $`d=0`$, the MFA lacks intercomponent correlations; thus, the total energy of the system is just the sum of the individual energies $`\epsilon (n_{\alpha s})`$ of each component. We can investigate the distribution of component densities in the MFA ground state in an $`N`$-component system by taking all but two of component densities to be fixed, and minimizing the energy of the remaining two-component system. If we label the two components we seek to minimize as $`1`$ and $`2`$, then according to Sec. II, the condition for stable equilibrium (local minimum of the total energy) is
$$\mu (n_1)=\mu (n_2),s(n_1)+s(n_2)>0,$$
(44)
where $`\mu (n_j)=\epsilon (n_j)/n_j`$ is the chemical potential relative to the minimum energy of component $`j`$ and $`s(n_j)=\mu (n_j)/n_j`$ is inversely proportional to the compressibility of component $`j`$. When $`\epsilon (n_j)`$ is the sum of the kinetic and exchange energies as in Eq. (43), then the MFA energy is minimized only if (1) both layer densities are equal ($`n_1=n_2`$), or (2) one or both layers are empty. There are no intermediate possibilities in the MFA. This $`d=0`$ result for $`N=2`$ gives the results found by Ruden and Wu: when $`n_1+n_2`$ is sufficiently large, the component densities are equal; when $`n_1+n_2`$ is sufficiently small, there is an exchange-driven intercomponent charge instability that empties out one of the components. In the absence of intercomponent correlations, we expect that at low enough densities the component compressibilities will be negative and that therefore one of the components will empty out, even if intracomponent correlations are included. In the $`N`$-component $`d=0`$ MFA ground state, any pair of layers either has equal density or has at least one of the layers empty.
There are therefore $`N`$ possible MFA ground states in an $`N`$-component system at $`d=0`$, characterized by the number of components $`p`$ that have nonzero and equal densities. The remaining $`Np`$ components have zero density. Defining the dimensionless average interparticle spacing per component by $`r_s=1/\sqrt{\pi (n_T/N)a_0^2}`$ for an $`N`$-component system, where $`a_0=4\pi ϵ\mathrm{}^2/m^{}e^2`$ is the effective Bohr radius, we may write the energy per unit area for the “$`p`$-component” MFA ground state as
$$\frac{_p}{L_xL_y}=\frac{e^2}{4\pi ϵa_0^3}p\left[\pi \left(\frac{N}{\pi r_s^2p}\right)^2\frac{8}{3\sqrt{\pi }}\left(\frac{N}{\pi r_s^2p}\right)^{3/2}\right].$$
(45)
In the limit $`N\mathrm{}`$, $`p`$ becomes a continuous variable that minimizes the energy (45) for $`p=p_{\mathrm{}}`$, where
$$p_{\mathrm{}}=N\left(\frac{3\pi }{4r_s}\right)^2=\left(\frac{3\pi }{4}\right)^2\pi a_0^2n_T,$$
(46)
where we have assumed that $`r_s(3\pi /4)`$; otherwise, $`p_{\mathrm{}}=N`$. Equation (46) shows that in the limit $`N\mathrm{}`$, $`p`$ is proportional to $`n_T`$ for $`r_s(3\pi /4)`$, and equal to $`N`$ otherwise. As expected, the number $`p`$ of equally occupied components drops as $`n_T`$ is reduced. Equation (46) is equivalent to saying the dimensionless interparticle separation per component for which $`p=p_{\mathrm{}}`$ is
$$r_s(p_{\mathrm{}})=\frac{3\pi }{4}\sqrt{\frac{N}{p_{\mathrm{}}}}$$
(47)
when $`p_{\mathrm{}}/N1`$. The equilibrium value of the $`N\mathrm{}`$ MFA energy corresponding to Eq. (46) is
$$\epsilon _{\mathrm{}}=\left(\frac{4}{3\pi }\right)^2\frac{e^2}{4\pi ϵa_0}n_T.$$
(48)
For arbitrary finite $`N`$, Eq. (45) can be used to find the interparticle spacing per component $`r_s^{(0)}(p,p+1)`$ at which the $`p`$\- and the $`(p+1)`$-component phases have the same energies at $`d=0`$:
$$r_s^{(0)}(p,p+1)=\frac{3\pi }{8}\left(\sqrt{\frac{N}{p}}+\sqrt{\frac{N}{p+1}}\right).$$
(49)
This is the interparticle spacing per component for the MFA transition between the $`p`$\- and $`(p+1)`$-component phases. It is interesting to compare this result to Eq. (47) and to note that
$$r_s^{(0)}(p,p+1)<r_s(p_{\mathrm{}})<r_s^{(0)}(p1,p),$$
(50)
so that even for finite $`N`$, $`r_s(p_{\mathrm{}})`$ always gives a value for the interparticle spacing that is in the $`p`$-component MFA phase at $`d=0`$.
For systems of physical interest containing $`N/2`$ layers of spin-1/2 particles, it is convenient to work with the interparticle spacing per layer (rather than the spacing per component). This is accomplished by dividing Eq. (49) by $`\sqrt{2}`$. For double-layer systems of spin-1/2 particles ($`N=4`$),
$`r_s^{(0)}(1,2)`$ $`=`$ $`{\displaystyle \frac{3\pi }{8}}(\sqrt{2}+1)2.844`$ (51)
$`,r_s^{(0)}(2,3)`$ $`=`$ $`{\displaystyle \frac{3\pi }{8}}(1+\sqrt{2/3})2.140`$ (52)
$`,r_s^{(0)}(3,4)`$ $`=`$ $`{\displaystyle \frac{3\pi }{8}}(\sqrt{2/3}+\sqrt{1/2})1.795,`$ (53)
where the superscript $`(0)`$ denotes zero layer separation. Note that the direct four- to two-component MFA transition predicted to occur at
$$r_s(2,4)=\frac{3\pi }{8}(1+\sqrt{1/2})2.011$$
(54)
does not exist. In fact, we find that when $`r_s=r_s(2,4)`$, the gate-balanced ($`p_F=p_B`$) double-layer system is always in the three-component MFA phase, regardless of the layer separation. The three-component system, which at $`d=0`$ has $`n_a=n_a=n_b=n_T/3`$, has a spin-unpolarized subband ($`n_a=n_a`$) with greater density than the other subband ($`n_a>n_b`$), which is spin-polarized. These features of the three-component phase persist at finite layer separations, although the layer imbalance is greatly reduced.
#### 2 Finite layer separation
Classically (when only the electrostatic energies are considered), balanced layers ($`n_1=n_2`$) are obtained when $`p_F=p_B`$, in order to make the electric field $`E_{12}`$ between the inner layers vanish. This result gives the asymptotically correct behavior for high layer densities and large layer separations. At sufficiently low densities and layer separations, the exchange energy can dominate the kinetic and electrostatic energies, so that the possibility of strong intralayer exchange leading to an interlayer charge-transfer instability must be considered. However, within the MFA, it can be proved that the inner layer densities are always equal, except in the three-component phase.
If the subband densities are equal ($`n_a=n_b`$, the case of “pseudospin paramagnetism”), then Eq. (31) shows that $`n_1=n_2`$. Thus the four-component ($`n_{\alpha s}=n_T/4`$) and two-component (with $`n_a=n_b=n_T/2`$) phases have balanced layers. This is because the MFA state constructed by occupying equally the single-particle subband states $`a`$ and $`b`$ is equivalent (up to a global pseudospin rotation) to the MFA state constructed by occupying equally the single-particle layer states $`1`$ and $`2`$. The fact that the ground-state energy in Eq. (36) is independent of $`\theta `$ and $`\varphi `$ when $`n_a=n_b`$ is due to the invariance of the ground-state energy under global pseudospin rotation.
If the subband densities are not equal ($`n_a>n_b`$), extremizing the total energy per unit area in Eq. (36) with respect to $`\theta `$ for $`p_F=p_B`$ and $`t=0`$ gives the condition $`\mathrm{sin}(2\theta )=0`$. The requirement that the extremum be a minimum (i.e., that the second derivative of the total energy per unit area with respect to $`\theta `$ be positive) gives
$$\mathrm{sin}\theta =\{\begin{array}{cc}1\hfill & \text{if }\mathrm{\Gamma }<1\hfill \\ 0\hfill & \text{if }\mathrm{\Gamma }>1\hfill \end{array}$$
(55)
where the interlayer exchange parameter $`\mathrm{\Gamma }`$ is defined in Eq. (42). The properties of $`\mathrm{\Gamma }`$ are described in Sec. 4 of the Appendix. Using the inequality
$$e^2d/2ϵ>V_{11}(q)V_{12}(q),$$
(56)
which is true for $`d>0`$, it follows from Eq. (42) that for $`d>0`$,
$`\mathrm{\Gamma }`$ $`<`$ $`{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐪s}{}}\left[{\displaystyle \frac{I_{aas}(q)+I_{bbs}(q)2I_{abs}(q)}{(n_an_b)^2}}\right]`$ (58)
$`={\displaystyle \underset{s}{}}{\displaystyle \frac{(n_{as}n_{bs})^2}{(n_an_b)^2}}.`$
Thus the condition
$$(n_an_b)(n_an_b)0$$
(59)
is sufficient to guarantee that $`\mathrm{\Gamma }<1`$ for $`d>0`$, so that $`\theta =\pi /2`$, which balances the layers. The one-component phase ($`n_a=n_T`$) satisfies Eq. (59), so $`\theta =\pi /2`$ and the layers are balanced. No interlayer charge transfer is obtained in the one-component phase because the combined effects of electrostatics and interlayer exchange, which favor balanced layers, always dominate the unbalancing influence of intralayer exchange: $`e^2d/2ϵ+V_{12}(q)>V_{11}(q)`$.
Within the MFA, the spins in subbands $`a`$ and $`b`$ are either completely unpolarized (at higher densities) or fully polarized (at sufficiently low densities). Therefore the only possible MFA configurations of spin and pseudospin which could have unbalanced layers ($`n_a>n_b`$, $`\theta \pi /2`$) when $`p_F=p_B`$ would be three-component states with
$$n_a=n_a<n_b,n_b=0.$$
(60)
We find numerically that such states always have $`\mathrm{\Gamma }>1`$ when $`t=0`$. so that $`\mathrm{sin}\theta =0`$. Hence, there is no interlayer phase coherence and the pseudospins are Ising-like. States with $`n_a>n_b`$ and $`\mathrm{sin}\theta =0`$ have partially unbalanced layers, with the lower-density layer being spin-polarized and the higher-density layer being spin-unpolarized, even for balanced gates ($`p_F=p_B`$). This is the behavior we find for the three-component MFA phase. For infinitesimal $`d`$, one (spin-unpolarized) layer has twice the density of the other (spin-polarized) layer, and the phase exists in the range $`r_s^{(0)}(3,4)<r_s<r_s^{(0)}(2,3)`$. At finite $`d`$, the three-component phase has only a slight layer imbalance and exists only in a narrow region of average interparticle spacing around $`r_sr_s(2,4)`$.
The equality of the inner layer densities in the balanced case (except for the three-component phase) has also been shown to be true for the one-component phase when intralayer and interlayer correlations are included within the STLS approximation. We note that if interlayer exchange were omitted from the total energy per unit area by setting $`V_{12}(q)=0`$ in Eq. (36), then four MFA ground states would still be obtained, and the four-, three-, and two-component states would be unchanged. However, Eq. (56) would not be satisfied at small $`q`$, and at large interparticle distances (at small values of $`k_F`$, or low densities) the vanishing of the second derivative with respect to $`\theta `$ would give the condition $`\mathrm{cos}(2\theta )>0`$, so that $`\mathrm{cos}(\theta )=\pm 1`$. This would produce the interlayer charge instability of Ruden and Wu for the one-component phase. The fact that the one-component phase has equal densities is due to the effects of interlayer exchange.
Before obtaining the MFA phase diagram for double-layer systems, it is convenient to express lengths and energies as dimensionless quantities. We therefore express the layer separation $`d`$ and the average density per layer $`n_T/2`$ in terms of the effective Bohr radius of the sample, $`a_0=4\pi ϵ\mathrm{}^2/m^{}e^2`$, where $`ϵ`$ is the dielectric constant in the material, and $`m^{}`$ is the effective mass. For GaAs, the dielectric constant is $`ϵ13ϵ_0`$. For $`n`$-type GaAs, $`m^{}0.07m_e`$ so that $`a_09.8`$ nm, while for $`p`$-type GaAs, $`m^{}0.3m_e`$ so that $`a_02.3`$ nm. The average density per layer can be expressed in terms of dimensionless ratio $`r_s=r_0/a_0`$ of the interparticle spacing $`r_0`$ for a single layer of averaged density $`n_T/2`$ (defined through $`\pi r_0^2n_T/2=1`$) to the effective Bohr radius $`a_0`$. Thus, for the same total density, $`p`$-type GaAs will have a value for $`r_s`$ that is about $`9.8/2.34.3`$ times larger than for $`n`$-type GaAs. We define the Fermi wave vector $`k_F`$ in terms of the total density $`n_T`$ as
$$k_F=\sqrt{4\pi n_T/p}=\sqrt{\frac{2}{p}}\frac{2}{r_sa_0}$$
(61)
for $`p=1,2,4`$. For the spin-polarized $`p=2a`$ state and and for the completely unpolirized $`p=4`$, $`k_F`$ is equal to the Fermi wave vector corresponding to the average density per layer ($`n_T/2`$). The energy scale associated with the effective Bohr radius is $`v_0=e^2/4\pi ϵa_0=\mathrm{}^2/m^{}a_0^2`$, which gives $`v_011`$ meV for $`n`$-type GaAs, and $`v_048`$ meV for $`p`$-type GaAs.
The HFA phase diagram for balanced double-layers without tunneling was obtained by Zheng and co-workers, except for the three-component phase. Like them, we find that within the MFA, three of the stable phases have equal average inner layer densities. Only the three-component phase has unequal layer densities. To understand the origin of the MFA phases, it useful to consider the five terms that contribute to the total energy per unit area in Eq. (36). The first term of Eq. (36) is the kinetic (Fermi) energy, which favors distributing the particles equally among the subbands and spins. At the highest densities, the kinetic energy term dominates, and the double-layer system is a four-component spin and pseudospin paramagnet: $`n_a=n_a=n_b=n_b=n_T/4`$. The second term of Eq. (36) is the tunneling energy, which we take to be zero for now. In general, it favors $`n_a>n_b`$ (pseudospin polarization), without regard to the real spin. The third term of Eq. (36) is the electrostatic energy, which vanishes when the gates and inner layers are balanced. In general, the electrostatic term favors complete screening, which would make $`n_1=p_F`$ and $`n_2=p_B`$, without regard to the real spin.
The fourth term in Eq. (36) is an intrasubband exchange term that dominates at the smallest densities and layer separations. It has the opposite effect of the kinetic energy, eventually producing a one-component spin and pseudospin ferromagnet at very low densities and small layer separations. The last term (containing the interlayer exchange) favors pseudospin paramagnetism ($`n_{as}=n_{bs}`$), but is indifferent to the polarization of the real spin, so long as it is the same in both subbands. Thus when $`p_F=p_B`$ and $`d>0`$, the last term and the electrostatic terms are responsible for producing a two-component phase that is ferromagnetic in real spin rather than in pseudospin: $`n_a=n_b=n_T/2`$. (At $`d=0`$, two-component states that are ferromagnetic in either the spin or pseudospin are degenerate. For $`d>0`$, the pseudospin ferromagnetic $`p=2b`$ state is favored only for substantial tunneling $`t`$ and/or layer imbalance $`|p_Fp_B|`$.) In the absence of the kinetic energy term (e.g., in the limit $`m^{}\mathrm{}`$), the real spin is always polarized due to intralayer exchange. However, because of the interlayer exchange, the two-component state is always obtained at high densities or large layer separations, while the one-component state is favored only for low densities and small layer separations.
Finite layer separation ($`d>0`$) differentiates between spin and pseudospin, so that the symmetry of the problem becomes $`SU(2)\times U(1)`$ rather than the $`CP(3)`$ symmetry at $`d=0`$. At finite layer separation, $`V_{12}(q)<V_{11}(q)`$, and the last term in the energy per unit area in Eq. (36) is minimized by equal subband densities, rather than by equal real spin densities. This does not effect the $`p=1`$ (fully spin- and pseudospin-polarized) or $`p=4`$ (completely spin and pseudospin unpolarized) phases. However, when $`t=0`$ and $`p_F=p_B`$, the two-component ($`p=2`$) MFA phase has its real spin fully polarized and its pseudospin completely unpolarized.
Finite layer separation also changes the densities at which the transitions between neighboring $`p`$-component phases occur. The densities at which the transitions between the phases occur in the MFA can be determined by comparing MFA energies. This task is made easier by the fact that the MFA always makes the real spins in a given subband either completely unpolarized (paramagnetic) of fully polarized (maximally ferromagnetic). This would probably not be true if correlation-energy effects were properly included, and it is likely that in real double-layer systems, states with partial polarization may be stable in some regions of density. The same MFA behavior (the restriction to the two extremes of either zero or full polarization) is also found for the pseudospin when the layers are balanced and the interlayer tunneling is zero, except for the three-component phase, which has partial pseudospin polarization.
The four-component and two-component phases both have equal subband densities, so their MFA energies are independent of the layer separation $`d`$. If there were a direct transition between these two phases, it would be simply a spin paramagnetic to ferromagnetic transition in each subband or layer. Therefore, within the MFA, such a four-component to two-component transition would occur at the same layer density as the spin-polarization transition for a single-layer system with a layer density equal to the subband densities: i.e., for $`r_s(2,4)=3\pi (1+\sqrt{1/2})/82.011`$, independent of the layer separation. As we discuss below, the direct four- to two-component transition is interrupted by a three-component phase, which has one subband (layer) spin-polarized and the other spin-unpolarized. So it is still true that MFA spin polarization transitions occur near $`r_sr_s(2,4)`$. However, the actual value of $`r_s`$ needed to obtain spin-polarization in a real sample is likely to be significantly higher. For single-layer systems, diffusion Monte Carlo simulations show that the low-density ferromagnetic state predicted by the HFA does occur; however, correlation-energy effects move the transition to densities that are probably 100 times lower, to $`r_s20`$. Such high values of $`r_s`$ have been achieved in low-density $`p`$-type GaAs samples, which possess a larger effective mass (and therefore larger $`r_s`$) than $`n`$-type samples. Large values of the effective mass will favor the existence of the lower-component ($`p<4`$) described here, in that they increase $`r_s`$.
We find empirically that the three-component MFA phase has $`\mathrm{sin}\theta =0`$. In order to analyze this phase, consider a MFA ground state with
$`n_b`$ $`=`$ $`(n_T/3)(1+x/2),n_b=0,`$ (62)
$`n_a`$ $`=`$ $`n_a=(n_T/3)(1x/4),`$ (63)
and $`\mathrm{sin}\theta =0`$, for $`0x1`$. Note that the layer imbalance is given by
$$\mathrm{\Delta }nn_an_b=(n_T/3)(1x).$$
(64)
When $`d=0`$, the three-component phase distributes the densities equally between the three components (but not between the two layers), so that $`x=0`$. As $`d\mathrm{}`$, the Coulombic cost of layer imbalance becomes prohibitive, and $`x1`$ so that $`\mathrm{\Delta }n0`$. We plot the layer imbalance ratio $`\mathrm{\Delta }n/(n_T/3)=(1x)`$ for $`r_s=r_s(2,4)2.011`$ in Fig. 2.
As $`d0`$,
$$x\frac{(4/3)(d/a_0)}{1r_s/(\pi \sqrt{2/3})}6.174\frac{d}{a_0},$$
(65)
to linear order in $`d/a_0`$, where the last expression in the equation above is for $`r_s=r_s(2,4)`$, which is the only value of $`r_s`$ for which the double-layer system is in the three-component phase for arbitrary layer separation. As $`d\mathrm{}`$,
$$\frac{\mathrm{\Delta }n}{n_T/3}=1x\frac{a_0}{d}\frac{3}{4}\left[\frac{r_s}{(1+1/\sqrt{2})\pi /2}\right]\frac{3}{16}\frac{a_0}{d},$$
(66)
to linear order in $`a_0/d`$, where the last expression in the equation above is for $`r_s=r_s(2,4)`$. Equation (66) says that the layer imbalance $`\mathrm{\Delta }n`$ in the three-component phase is inversely proportional to $`d/a_0`$ for large values of $`d/a_0`$.
For infinitesimal $`d`$, the energy per unit area of three-component phase increases by the amount $`(e^2d/8ϵ)(n_T/3)^2`$ to linear order in $`d`$, while that of the four- and two-component phases are unchanged. Equating the three-component energy to the four- and to the two-component energies gives
$`r_s(2,3)`$ $``$ $`r_s^{(0)}(2,3)\left(1{\displaystyle \frac{1}{3}}{\displaystyle \frac{d}{a_0}}\right),`$ (67)
$`,r_s(3,4)`$ $``$ $`r_s^{(0)}(3,4)\left(1+{\displaystyle \frac{2}{3}}{\displaystyle \frac{d}{a_0}}\right),`$ (68)
to first order in $`d/a_0`$ as $`d/a_00`$, and
$`r_s(2,3)`$ $``$ $`r_s^{(0)}(2,4)\left(1+{\displaystyle \frac{1}{16}}{\displaystyle \frac{a_0}{d}}\right),`$ (69)
$`r_s(3,4)`$ $``$ $`r_s^{(0)}(2,4)\left(1{\displaystyle \frac{1}{16}}{\displaystyle \frac{a_0}{d}}\right),`$ (70)
to first order in $`a_0/d`$ as $`d/a_0\mathrm{}`$. Note that both $`r_s(2,3)`$ and $`r_s(3,4)`$ approach $`r_s(2,4)`$ in the limit $`d\mathrm{}`$. This is because as $`d\mathrm{}`$, the double-layer system consists of two independent layers, and $`r_s(2,4)`$ is the interparticle spacing at which the spin-polarized and spin-unpolarized energies are equal in a single-layer system. Thus, as $`d\mathrm{}`$, the energies of the four-, three-, and two-component phases are all equal at $`r_s=r_s(2,4)`$.
We now consider the two-component to one-component transition in the MFA. For $`d=0`$, the MFA transition to pseudospin ferromagnetism is equivalent to a real-spin paramagnetic to ferromagnetic transition in a single layer having total density $`n_T`$ (rather than $`n_T/2`$). Thus for $`d=0`$, the MFA critical density per layer for the two- to one-component transition is exactly half the critical density for the four- to two-component transition, so that $`r_s^{(0)}(1,2)=3\pi (\sqrt{2}+1)/82.844`$. By equating the one- and two-component phase energies per area, the critical density for the one- to two-component transition may be obtained:
$`{\displaystyle \frac{r_s^{(0)}(1,2)}{r_s(1,2)}}`$ $`=`$ $`1(1+1/\sqrt{2}){\displaystyle \frac{3\pi z}{32}}\mathrm{\Gamma }_1(z)`$ (71)
$`=`$ $`1(1+1/\sqrt{2})\left[1S(z)\right],`$ (72)
where
$`z`$ $`=`$ $`2k_Fd=2d\sqrt{4\pi n_T}=4\sqrt{2}d/(r_sa_0)`$ (73)
$`=`$ $`(2\sqrt{2}){\displaystyle \frac{32}{3\pi }}{\displaystyle \frac{d}{a_0}}{\displaystyle \frac{r_s^{(0)}(1,2)}{r_s}},`$ (74)
$`\mathrm{\Gamma }_1=\mathrm{\Gamma }(n_a=n_T)`$, and
$`S(z)`$ $`=`$ $`{\displaystyle \frac{3}{2}}{\displaystyle _0^1}𝑑xe^{zx}\left[\mathrm{arccos}(x)x\sqrt{1x^2}\right]`$ (75)
$`=`$ $`{\displaystyle \frac{3\pi }{4z}}\left\{1{\displaystyle \frac{2}{z}}\left[I_1(z)L_1(z)\right]\right\}`$ (76)
$``$ $`\{\begin{array}{cc}1(3\pi /32)z+(1/5)z^2,\hfill & z0\hfill \\ 3\pi /4z,\hfill & z\mathrm{}\hfill \end{array}`$ (79)
Here $`I_1`$ and $`L_1`$ are modified Bessel and modified Struve functions of the first kind, respectively. The derivation of the above formula is discussed in Sec. 4 c of the Appendix. Equations (71) and (73) determine $`r_s(1,2)`$, which we have plotted as the upper solid line in Fig. 3.
It follows from Eq. (75) that $`r_s(1,2)/r_s^{(0)}(1,2)(1+d/a_0)`$ in the limit $`d/a_00`$. The value of $`r_s(1,2)`$ at very large layer separations ($`d/a_0\mathrm{}`$) can be obtained by setting $`r_s(1,2)/r_s^{(0)}(1,2)=0`$ in the left-hand side of Eq. (71) and then solving the resulting equation numerically for $`z`$. The result is $`z4.015`$, or $`r_s(1,2)/r_s^{(0)}(1,2)0.495d/a_0`$, so that $`r_s(1,2)1.409d/a_0`$. We note that if interlayer exchange were ignored and the pseudospins treated as Ising variables ($`\mathrm{sin}\theta =0`$), then the two- to one-component MFA transition would occur at $`r_s^{(0)}(1,2)(1+d/a_0)`$ (for all values of $`d/a_0`$) and would put all (spin-polarized) particles in a single layer. Interlayer exchange makes the layer densities equal in the one-component MFA state, and causes the two- to one-component MFA transition to occur at a somewhat lower value of $`r_s`$ than would be predicted using Ising pseudospins.
### B Infinitesimal tunneling
In this subsection, we discuss the effects of very small interlayer tunneling. The contribution of interlayer tunneling to the MFA energy per unit area is given by $`2t(n_an_b)\mathrm{sin}\theta `$. Finite interlayer tunneling thus has two important effects. First, it removes the Ising character of the spins by making $`\mathrm{sin}\theta >0`$. We discuss this in more detail in the next subsection. Second, it always produces some degree of pseudospin polarization ($`n_an_b>0`$). We have parametrized the dependence of the subband splitting $`(n_an_b)`$ on $`t`$ in terms of the pseudospin Stoner interaction parameter, which we calculate in the second subsection below.
#### 1 Effect of tunneling on phase angle
We first consider the effect of interlayer tunneling on the interlayer phase angle $`\theta `$. Extremizing the energy per unit area \[Eq. (36)\] with respect to $`\theta `$ for $`t>0`$ gives two possible solutions. The solution
$$\mathrm{cos}\theta =0$$
(80)
is always an extremum, and is a minimum whenever $`\mathrm{\Gamma }<1`$, which is true for all phases except $`p=3`$ near balance ($`p_Fp_B`$). Of course, if $`t`$ were large enough to cause full pseudospin polarization ($`n_a=n_T`$), then $`\mathrm{cos}\theta =0`$.
If $`\mathrm{\Gamma }>1`$ (three-component phase) then the solution
$$\mathrm{sin}\theta =\frac{t}{(\mathrm{\Gamma }1)e^2d(n_an_b)/2ϵ}$$
(81)
gives the correct minimum, provided that the right-hand side is positive and does not exceed unity. Even in the three-component phase, $`t`$ will increase $`(n_an_b)`$, but since $`(n_an_b)`$ is already finite at $`t=0`$, $`\mathrm{sin}\theta `$ will be proportional to $`t`$, to first order in $`t`$.
The effect of including a small amount of interlayer tunneling can often be described perturbatively, and produces a smooth increase in the pseudospin polarization $`(n_an_b)`$ that is proportional to $`t`$. However, the results obtained in the combined limits $`t0`$ and $`d0`$ can depend on the order of these limits. For example, at $`d=0`$, the lowest-energy two-component phase has $`n_a=n_a=n_T/2`$ (pseudospin-polarized but spin-unpolarized) for arbitrarily small but finite $`t`$. However, at $`t=0`$, the lowest-energy two-component phase has $`n_a=n_b=n_T/2`$ (spin-polarized but pseudospin-unpolarized) for arbitrarily small but finite $`d`$. By comparing the MFA energies per unit area of two competing $`p=2`$ ground states with $`\mathrm{cos}\theta =0`$ (spin-polarized $`p=2a`$ versus pseudospin-polarized $`p=2b`$), it can be shown that the pseudospin-polarized two-component ($`p=2b`$) ground state requires
$`t`$ $`>`$ $`{\displaystyle \frac{e^2dn_T}{8ϵ}}\mathrm{\Gamma }(n_a=n_a=n_T/2)={\displaystyle \frac{e^2dn_T}{8ϵ}}\mathrm{\Gamma }_1(p=2)`$ (82)
$`=`$ $`{\displaystyle \frac{e^2dn_T}{8ϵ}}{\displaystyle \frac{16}{3\pi z}}\left[1S(z)\right]`$ (86)
$`\{\begin{array}{cc}e^2dn_T/16ϵ,\hfill & k_Fd0\hfill \\ (4/3\sqrt{2\pi })e^2\sqrt{n_T}/4\pi ϵ,\hfill & k_Fd\mathrm{}.\hfill \end{array}`$
Here $`z=2k_Fd=2d\sqrt{2\pi n_T}`$, and $`S(z)`$ is defined in Eq. (75). Therefore it is the size of $`t`$ relative to $`e^2dn_T/16ϵ`$ that must be considered as both $`t`$ and $`d`$ approach zero in the two-component phase. A rough estimate of the minimum tunneling energy $`t_c`$ necessary to obtain the pseudospin-polarized two-component phase may be made by calculating $`e^2dn_T/16ϵ`$ for the smallest value of $`n_T`$ that still gives the two-component phase: i.e., for $`r_sr_s^{(0)}(1,2)2.844`$. For $`d=10`$ nm, this gives $`t_c0.7`$ meV for $`n`$-type GaAs and $`t_c13`$ meV for $`p`$-type GaAs. The differences between these two values of $`t_c`$ arise from the fact that $`n`$-type GaAs spin-polarizes at a much lower density than $`p`$-GaAs, due to the differences in the effective mass (and therefore in $`r_s`$.) We stress that there exist two $`p=2`$ two-component states, which can be either spin- or pseudospin-polarized. The $`p=2`$ state that we focus on most will be the spin-polarized state ($`p=2a`$).
#### 2 Pseudsopin Stoner parameter
The difference $`\mathrm{\Delta }nn_an_b`$ between the subband densities (obtained from SdH measurements) of a balanced double-layer system is often used as a measure of the size of the interlayer tuneling matrix element $`t`$ by applying the formula $`n_an_b=(p/2)t\nu _0`$, which is valid for noninteracting particles. Here $`p=4`$ for spin-unpolarized particles, $`p=2`$ for spin-polarized particles, and $`\nu _0=m^{}/(\pi /\mathrm{}^2)`$ is the two-dimensional density of states per unit area. For interacting electrons or holes, the pseudospin Stoner interaction parameter $`I`$ is defined through
$$n_an_b=\frac{(p/2)t\nu _0}{1I}.$$
(87)
For noninteracting particles, $`I=0`$. For interacting particles, $`I`$ is a function of $`r_s`$ and $`d/a_0`$; in general, it also a function of $`t`$. We do not consider $`p=3`$ or $`p=1`$, since both the three- and the one-component phases have $`n_a>n_b`$ at $`t=0`$, corresponding to $`I=1`$. The onset of SILC ($`n_a>n_b`$ and $`\mathrm{sin}\theta >0`$ even when $`t=0`$) occurs at the two- to one-component transition, and corresponds to $`I1`$ for $`p=2`$.
The Stoner interaction parameter $`I`$ can be calculated analytically in the limit of vanishing interlayer tunneling ($`t0`$), as we show in Sec. 5 of the Appendix. (For finite $`t`$, we calculate $`I`$ numerically.) The basic idea is to start with equal subband densities ($`n_a=n_b`$) and infinitesimally small $`t`$, and then calculate the change in energy due to moving an infinitesimal amount of charge from subband $`b`$ to subband $`a`$. Minimizing the change in energy with respect to the amount of charge transferred between the subbands gives the linear response to small interlayer tunneling and yields the following expression for the $`t0`$ limit of the Stoner interaction parameter:
$`I`$ $`=`$ $`{\displaystyle \frac{\nu _0e^2}{2\pi ϵk_F}}\left(1p{\displaystyle \frac{\pi }{4}}k_Fd\mathrm{\Gamma }_0\right)`$ (88)
$`=`$ $`{\displaystyle \frac{\nu _0}{\pi }}\left\{2V_{11}(2k_F){\displaystyle _0^1}𝑑x{\displaystyle \frac{[V_{11}(2k_Fx)V_{12}(2k_Fx)]}{\sqrt{1x^2}}}\right\}`$ (89)
$`=`$ $`{\displaystyle \frac{r_s}{\pi }}\sqrt{{\displaystyle \frac{p}{2}}}\left[1{\displaystyle \frac{1}{2}}{\displaystyle _0^{\pi /2}}𝑑\theta {\displaystyle \frac{\left(1e^{2k_Fd\mathrm{sin}\theta }\right)}{\mathrm{sin}\theta }}\right],`$ (90)
where $`k_F=\sqrt{4\pi n_T/p}`$ for $`p=2,4`$, and $`\mathrm{\Gamma }_0\mathrm{\Gamma }(n_an_b)`$. Equation (88) is equivalent to the generalized random-phase approximation (GRPA) result in Eq. (14) of Ref. .
Note that $`I=1`$ (when $`p=1`$) corresponds to SILC, and that the GRPA result for the phase boundary, which is shown as the dashed sloped line in Fig. 3, is different (has larger $`r_s`$) than the MFA result. That the GRPA gives a higher value for $`r_s(1,2)`$ is not surprising, given that the GRPA goes beyond MFA and contains correlation effects in an approximate way. To lowest order in $`d`$, $`I(d)/I(0)=(1d/a_0)`$, so that for $`d0`$, the GRPA gives a higher critical value of the interparticle spacing for SILC than the MFA: $`r_s(1,2)\pi (1+d/a_0)`$. As expected, a similar calculation of the linear response of the real spins to a weak Zeeman field shows that a hypothetical four- to two-component GRPA transition would occur at twice the density of the $`d=0`$ two- to one-component transition, i.e., at $`r_s(1,2)=\pi /\sqrt{2}`$. However, as with the MFA, we expect that a three-component GRPA phase preempts any direct four- to two-component GRPA transition, and that a three-component is always obtained within the GRPA at $`r_s=\pi /\sqrt{2}`$.
As pointed out in Ref. , interactions enhance the subband splitting ($`I>0`$) for $`k_Fd<1.13`$, but reduce the splitting ($`I<0`$) for $`k_Fd>1.13`$. The critical value for $`r_s`$ at large $`d`$ that separates the one- and two-component phases is determined by solving Eq. (88) for $`I=1`$ and $`r_s\mathrm{}`$. Asymptotically (for $`r_s,d/a_0\mathrm{}`$), it occurs at the same value of $`r_s`$ that has $`I=0`$ i.e., $`k_Fd1.134`$ or $`r_s(1,2)/r_s^{(0)}(1,2)0.620d/a_0`$, which gives $`r_s(1,2)1.764d/a_0`$. Within the GRPA, SILC occurs once the interparticle spacing is roughly twice the interlayer spacing. In the limit $`k_Fd\mathrm{}`$, $`I(d)/I(0)=1(1/2)[\mathrm{ln}(4k_Fd)+\gamma ]`$, where $`\gamma 0.5772`$ is Euler’s constant. In Ref. it is argued that although $`I(d)`$ is large and negative as $`d\mathrm{}`$, its apparently divergent behavior is an unphysical artifact of the GRPA.
We have calculated the Stoner interaction parameter $`I`$ for a few values of the interlayer tunneling $`t`$ in Figs. 4 ($`n`$-type GaAs) and 5 ($`p`$-type GaAs), for a hypothetical sample with total density $`n_T=10^{11}`$cm<sup>-2</sup>. The complete polarization of the pseudospin ($`n_a=n_T`$) is indicated by the mesa-like regions where $`I`$ becomes flat: $`n_a=n_a=n_T/2`$ for $`n`$-type GaAs and $`n_a_{}=n_T`$ for $`p`$-type GaAs. Increasing the size of $`t`$ favors pseudospin polarization and allows it to persist to larger values of $`k_Fd`$. Note that at fixed $`t`$, the Stoner interaction is equal to $`I_{\mathrm{max}}=12t\nu _0/n_T`$ as $`k_Fd0`$, so that $`I_{\mathrm{max}}`$ decreases with increasing $`t`$. The fact that $`I_{\mathrm{max}}(t=0.1)>I_{\mathrm{max}}(t=0)`$ is an artifact of the order in which the limits $`t0`$ and $`d0`$ are taken: for any finite $`t`$, the two-component phase will have $`n_a=n_a=n_T/2`$ as $`d0`$, but for any finite $`d`$, the two-component phase will have $`n_a=n_b=n_T/2`$ as $`t0`$. For $`n`$-type GaAs with $`t=1`$ meV (dotted line in Fig. 4), the system goes through three phases as a function of $`k_Fd`$: (1) a pseudospin-polarized two-component phase ($`n_a=n_a=n_T/2`$) for $`k_Fd0`$, (2) a three-component phase ($`n_a=n_a<n_b`$) for intermediate values of $`k_Fd`$, indicated in by the “missing piece” on the right side of the mesa in Fig. 4, (3) a real-spin-polarized two-component phase ($`n_a=n_b=n_T/2`$) for larger $`k_Fd\mathrm{}`$.
SILC is indicated by having $`I(t=0)=1`$ and occurs only for $`p`$-type GaAs (Fig. 5, for $`k_Fd<0.7`$): $`I(t=0)`$ is directly proportional to $`\nu _0`$, and therefore to the effective mass of the particles, so that SILC is more likely to be observed in $`p`$-type ($`m^{}/m_e0.3`$) rather than $`n`$-type ($`m^{}/m_e0.07`$) GaAs.
## V Effect of Bias
In this section, we study the effect of bias ($`p_Fp_B`$, due to applied gate voltages) on the subband and layer densities. The classical results
$$n_1=p_F\mathrm{\Theta }(p_F),n_2=p_B\mathrm{\Theta }(p_B),$$
(91)
and
$$V_F=eD_FE_F+edE_{12}+eV_F^{(0)}$$
(92)
give the asymptotically correct behavior for high layer densities and large layer separations. Double-layer systems at low densities and small layer separations show measureable deviations from the classical behavior, most notably because of quantum-mechanical exchange. We shall find it convenient to study the effects of layer imbalance by fixing the total density ($`n_T=p_F+p_B`$) and then varying the gate-imbalance parameter $`\zeta `$, defined by
$$\zeta \frac{p_Fp_B}{p_F+p_B}=(p_Fp_B)/n_T.$$
(93)
The case of balanced gates ($`p_F=p_B`$) corresponds to $`\zeta =0`$.
In the presence of bias and/or tunneling, there are five possible noncrystalline MFA ground states, which we write in order of increasing $`r_s`$ in Table I. Only the last ($`p=1`$) phase can exhibit SILC. The last two phases are pseudospin-polarized and can have $`0\mathrm{sin}\theta 1`$, with the value of $`\theta `$ being determined by the layer separation, density, bias, and tunneling. The first three phases are Ising-like ($`\mathrm{sin}\theta =0`$) when $`t=0`$, and therefore do not involve interlayer exchange, in the absence of interlayer tunneling. Note that there are two $`p=2`$ phases, one that is pseudospin-polarized but spin-unpolarized ($`p=2b`$), and another that is spin-polarized but pseudospin unpolarized ($`p=2a`$). The pseudospin-polarized $`p=2b`$ requires bias and/or tunneling. For all MFA phases, the real spin is either fully polarized or completely unpolarized.
We begin this section by studying the case of vanishing interlayer exchange, which is the relevant situation for the majority of double-layer samples that have been studied experimentally, except in the quantum Hall regime. This is the case when layer separations are sufficiently far apart that interlayer exchange is negligible in zero or weak magnetic fields. A model without interlayer exchange is able to fit existing double-layer data well, and obtains the correct four-, three-, and two-component phases, although it fails to properly describe the one-component phase. We then include the effects of interlayer exchange and solve for the subband and layer densities using our MFA, which allows for the possibility of interlayer exchange, even in the absence of interlayer tunneling. Finally, we give a full treatment of the one-component phase within the MFA, including bias and tunneling.
### A No interlayer exchange
It is simplest to begin our study of the effects of layer imbalance ($`p_Fp_B`$) by first considering the limit of vanishing interlayer exchange. This limit is relevant to most of double-layer samples that have been studied experimentally, except in the quantum Hall regime. It corresponds to layer separations that are sufficiently far apart that interlayer exchange is negligible in zero or weak magnetic fields. We shall also demonstrate that a very simple model which assumes that the particles are always spin-unpolarized gives a good fit to existing data on the subband occupancies of double-layer systems, except at low densities.
#### 1 No tunneling
Interlayer exchange is negligible when the layer separation is large compared to the interparticle spacing. This condition may be expressed in various ways, e.g., $`k_Fd1`$ or $`r_sd/a_0`$, and is satisfied for most samples. In this limit, we ignore interlayer tunneling and correlations and write the exchange-correlation energy in Eq. (4) as
$$\epsilon (n_1,n_2)\epsilon (n_1)+\epsilon (n_2),$$
(94)
where $`\epsilon (n)`$ is the sum of the kinetic (Fermi), exchange, and correlation (but not the electrostatic) energies, for a single-layer two-dimensional electron gas (2DEG) of density $`n`$. In the absence of interlayer exchange and tunneling, we can work directly with the layer densities rather than the subband densities because the pseudospins are Ising variables:
$`n_a`$ $`=`$ $`\mathrm{max}(n_1,n_2),n_b=\mathrm{min}(n_1,n_2),`$ (95)
$`\theta `$ $`=`$ $`\pi \mathrm{\Theta }(p_Bp_F)=\{\begin{array}{cc}\pi ,\hfill & p_F<p_B\hfill \\ 0,\hfill & p_F>p_B\hfill \end{array}`$ (98)
so that $`\mathrm{sin}\theta =0`$.
No interlayer exchange is required to correctly describe the four-, three-, and two-component MFA phases at $`t=0`$, since all have $`\mathrm{sin}\theta =0`$. Our analysis of these phases therefore proceeds as before. It is nonetheless useful to look at the phases of system in terms of the equilibrium and stability conditions of Sec. II, which we do below.
Although it would be straightforward to include intralayer correlation-energy effects in $`\epsilon (n)`$, we do not do so here for the sake of simplicity. Even so, the resulting approximate model does a very good job of fitting SdH data until the layer densities get so low that they violate our initial assumption that $`k_Fd`$ is large. We therefore begin with
$`\epsilon (n)`$ $``$ $`{\displaystyle \underset{s}{}}\left[{\displaystyle \frac{n_s^2}{\nu _0}}{\displaystyle \frac{8}{3\sqrt{\pi }}}{\displaystyle \frac{e^2}{4\pi ϵ}}n_s^{3/2}\right]`$ (99)
$`=`$ $`\{\begin{array}{cc}n^2/2\nu _0(8/3\sqrt{2\pi })(e^2/4\pi ϵ)n^{3/2},\hfill & n>n_c\hfill \\ n^2/\nu _0(8/(3\sqrt{\pi })(e^2/4\pi ϵ)n^{3/2},\hfill & n<n_c\hfill \end{array},`$ (102)
where $`n_c`$ is the critical density for the MFA spin-polarization transition in a single-layer system. The condition $`n>n_c`$ corresponds to spin-unpolarized electrons and occurs at higher densities, whereas $`n<n_c`$ corresponds to spin-polarized electrons and occurs at lower densities. When $`\mathrm{sin}\theta =0`$ (which is the case we are considering in this section), $`n_c`$ is the critical density of the MFA (or the HFA) spin-polarization transition for a single layer, which occurs when the single-layer $`r_s`$ has the value $`r_s(2,4)=(3\pi /8)(1+\sqrt{1/2})2.011`$, shown as the dashed line in Fig. 3. Note that $`r_s1/\sqrt{\pi na_0^2}`$ for a single-layer system of number density $`n`$, which gives $`n_ca_0^2=(2/3)(4/\pi )^3(12\sqrt{2}/3)0.07870`$. In the MFA, which is equivalent to the HFA for balanced layers, the spin polarization is either completely unpolarized (at higher densities) or completely polarized (at lower densities). Correlation-energy effects probably produce a range of intermediate spin polarizations.
The chemical potential measured relative to layer $`i`$ is $`\mu _i=\mu (n_i)`$, where
$`\mu (n)`$ $`=`$ $`\epsilon (n)/n`$ (103)
$`=`$ $`\{\begin{array}{cc}n/\nu _0(4/\sqrt{2\pi })(e^2/4\pi ϵ)\sqrt{n},\hfill & n>n_c\hfill \\ 2n/\nu _0(4/\sqrt{\pi })(e^2/4\pi ϵ)\sqrt{n},\hfill & n<n_c\hfill \end{array}`$ (106)
where we have used Eqs. (6) and (99). The values of the layer densities can be determined by using Eq. (103) in the equilibrium condition of Eq. (5), $`\mu _1\mu _2=eE_{12}d`$.
The electronic lengths $`s_{ij}`$ that determine the Eisenstein ratio $`R_E`$ \[Eq. (12)\] and the condition for stability against interlayer charge transfer \[Eq. (8)\] can be calculated from Eqs. (7) and (103). Ignoring interlayer correlations as in Eq. (94), gives $`s_{ij}=0`$ for $`ij`$, and $`s_i=s_{ii}=s(n_i)`$, where
$`{\displaystyle \frac{s(n)}{a_0}}={\displaystyle \frac{ϵ}{e^2a_0}}{\displaystyle \frac{\mu (n)}{n}}`$ $`=`$ $`\{\begin{array}{cc}1/4(\sqrt{2/\pi })/(4\pi a_0\sqrt{n}),\hfill & n>n_c\hfill \\ 1/2(2/\sqrt{\pi })/(4\pi a_0\sqrt{n}),\hfill & n<n_c\hfill \end{array}`$ (109)
$`=`$ $`\{\begin{array}{cc}1/4(\sqrt{2}/4\pi )r_s,\hfill & n>n_c\hfill \\ 1/2(1/2\pi )r_s,\hfill & n<n_c\hfill \end{array}.`$ (112)
Note that the MFA compressibility $`\kappa `$ can be calculated from the above result using Eq. (10), and that $`s(n)`$ and therefore $`\kappa `$ are negative at sufficiently low densities. The length $`s(n)`$ and the compressibility $`\kappa `$ jump discontinuously at $`n=n_c`$, where the ground state changes abruptly from spin-unpolarized to spin-polarized. For densities just above $`n_c`$, $`s(n_c^+)/a_00.0237`$, while for densities just below $`n_c`$, $`s(n_c^{})/a_00.1799`$. Sensitive measurements of the interlayer capacitance (e.g., the Eisenstein ratio $`R_E`$) could detect the exchange-driven spin polarization of a 2DEG through its effect on the compressibility, especially in $`p`$-type GaAs samples when the density in a layer could be made small enough to polarize the holes in that layer.
It is straightforward to calculate the effect of spin polarization in a low-density layer on $`R_E`$ using Eqs. (13) and (109). For the usual case in which the interlayer separation $`d`$ is substantially larger than the electronic lengths $`s(n_i)`$ (i.e., for $`d/a_00.2`$), the MFA gives an abrupt jump in $`R_E`$ by almost a factor of 8, from approximately $`0.0237a_0/d`$ for densities just above $`n_c`$, to approximately $`0.1799a_0/d`$ for densities just below $`n_c`$. Of course, the MFA overestimates the size of the jump, but it is nonetheless plausible that measurements of $`R_E`$ could detect changes in the compressibility of a low-density layer due to the exchange-driven polarization of the spins.
#### 2 Unpolarized spins
Ruden and Wu assumed not only that the pseudospins were Ising-like ($`\mathrm{sin}\theta =0`$), but also that the real spins were always unpolarized. This limited the phases they found at $`p_F=p_B`$ to two: pseudospin-unpolarized ($`p=4`$) at high density and pseudospin-polarized ($`p=2b`$) for low density. It is straightforward to compare the energy of the four-component phase with that of the hypothetical pseudospin-polarized ($`p=2b`$) phase of Ruden and Wu and show that they are equal when $`r_s/r_s^{(0)}(1,2)=1+2d/a_0`$. Although neither assumption was, strictly speaking, correct, it is an interesting and useful fact that making such assumptions can yield a simple model that fits experimental data for layer densities versus gate bias quite well, except at the lowest densities. Figure 6 shows experimental SdH data and a theoretical fit from a simple theory that ignores interlayer exchange and takes the spins to be unpolarized. The value of the interlayer separation $`d`$ used in the model is taken to be a fitting parameter. The values of $`d`$ that we obtain with this simplified model always locate the idealized two-dimensional electrons layers inside the confining quantum wells, although $`d`$ always seems to be somewhat larger than the midwell to midwell distance.
Note that the experimental data fits very well almost everywhere, except where the density $`n_1`$ is very small when layer 1 is near depletion. Here the simplest theory (which omits the possibilities of spin and pseudospin polarization) erroneously predicts an interlayer charge transfer instability. As we discussed in Sec. II, such an instability is unavoidable when interlayer correlations ($`s_{12}`$) are neglected. According to the stability criterion of Eq. (8), the interlayer charge-transfer instability should occur in the simplest theory (spin-unpolarized, no interlayer exchange) when
$$r_{s1}\pi \sqrt{2}(1+2d/a_0)r_{s2},$$
(113)
where $`r_{si}=1/\sqrt{\pi n_ia_0^2}`$ is the dimensionless interparticle separation in layer $`i`$, and we have taken $`n_1n_2`$ (or $`r_{s1}r_{s2}`$). At balance, setting $`r_{s1}=r_{s2}`$ in Eq. (113) gives $`r_s=(\pi /\sqrt{2})(1+2d/a_0)`$ as the GRPA version of the critical particle separation for Ruden and Wu’s hypothetical pseudospin-polarization ($`p=4p=2b`$) transition. \[In the MFA, $`r_s(2b,4)=r_s^{(0)}(2,4)(1+2d/a_0)`$ gives the Ruden-Wu hypothetical pseudospin-polarization transition at balance.\]
Note that even in the limit of equal layer densities and zero layer separation, $`r_{s1}\sqrt{\pi }/2`$, which is the GRPA value of the particle separation for a single layer to spin polarize. Thus even in a theory that neglects interlayer correlations, the particles in the lower-density layer (or both layers, if they are balanced) spin-polarize before the layer empties out. As noted in Ref. , the spin polarization of the electrons predicted by the HFA was ignored by Ruden and Wu. However, including the spin polarization does not eliminate the interlayer charge transfer instability, which according to Eqs. (8) and (109) would occur at
$$r_{s1}>2\pi (3/4+d/a_0)r_{s2}/\sqrt{2},$$
(114)
when layer 1 is spin-polarized but layer 2 is not, and at
$$r_{s1}2\pi (1+d/a_0)r_{s2},$$
(115)
when both layers are spin polarized, if interlayer correlations could be ignored.
#### 3 LDF model
We now introduce a tight-binding local density functional (LDF) model, which includes the effects of interlayer tunneling in a simple way. We shall follow the previous section and ignore for now the effects of interlayer exchange, and we shall even take the electron densities to be always spin-unpolarized. Such an elementary model is capable of fitting experimental data quite well, despite its simplicity.
The Kohn-Sham single-particle equations for our tight-binding LDF model is conveniently expressed as a $`2\times 2`$ matrix equation:
$$\left(\begin{array}{cc}ϵ_1& t\\ t& ϵ_2\end{array}\right)\left(\begin{array}{c}z_1^{(\lambda )}\\ z_2^{(\lambda )}\end{array}\right)=E_\lambda \left(\begin{array}{c}z_1^{(\lambda )}\\ z_2^{(\lambda )}\end{array}\right),$$
(116)
where
$$ϵ_j=(1)^j\frac{1}{2}eE_{12}d+\mu _{xc}(n_j)$$
(117)
represents the “on-site” energy of layer $`j`$, and the tunneling matrix element $`t`$ is off-diagonal. The amplitude of the wave function for subband $`\lambda `$ ($`\lambda =a,b`$) in layer $`j`$ is $`z_j^{(\lambda )}`$, and the subband energy is $`E_\lambda `$. The Hartree contribution to the on-site energy enters via the interlayer electric field $`E_{12}`$, as shown in Eq. (2). The intralayer exchange and correlation contributions to the on-site energy are given by the exchange-correlation potential $`\mu _{xc}`$. In LDF theory, $`\mu _{xc}(n)`$ is equal to the derivative, with respect to density, of the exchange plus correlation energies per unit area of a two-dimensional single-layer system of uniform areal density $`n`$. Equivalently, $`\mu _{xc}(n_j)`$ is equal to $`\mu _j`$ \[see Eq. (6)\] minus the kinetic energy contribution to $`\mu _j`$. For simplicity, we do not include intralayer correlation energy contributions to $`\mu _{xc}`$, so we write
$$\mu _{xc}(n)\frac{4}{\sqrt{\pi }}\frac{e^2}{4\pi ϵ}\sqrt{n}.$$
(118)
The density in layer $`j`$ is given by
$$n_j=\underset{\lambda =1}{\overset{2}{}}N_\lambda |z_j^{(\lambda )}|^2,$$
(119)
where
$$N_\lambda (E_FE_\lambda )\nu _0\mathrm{\Theta }(E_FE_\lambda )$$
(120)
is the areal-density contribution from subband $`\lambda `$, and $`\nu _0=m^{}/\pi \mathrm{}^2`$ is the two-dimensional density of states for noninteracting particles. The self-consistency of the Kohn-Sham equations enters via Eqs. (119) and (120), since the layer densities $`n_j`$, together with the gate densities $`p_\alpha `$, determine the interlayer electric field $`E_{12}`$ appearing in Eq. (117). The Fermi energy $`E_F`$ is chosen so that the sum of the subband densities $`N_\lambda `$ is equal to the total density $`p_F+p_B`$.
This simple LDF model, which takes the layers to have zero thickness and assumes that the real spins are unpolarized, is capable of fitting the experimental layer density data closely. This is illustrated in Fig. 7, which shows SdH data taken from sample A of Ref. . The front-gate voltages used in Fig. 7 were calculated using Eq. (92).
### B Interlayer exchange
We now allow for the possibility of interlayer exchange in biased systems. We found that for balanced systems, interlayer exchange becomes important only at low densities and small layer separations, occuring only in the one-component phase. In this section, we explore the effect of interlayer exchange on biased double layers, and find that it can reduce or suppress interlayer (although not intersubband) charge transfers. We find that bias always increases pseudospin polarization and sometimes reduces the total density required to achieve SILC.
#### 1 Interlayer phase angle
When the interlayer tunneling $`t`$ is zero, it is possible to determine the equilibrium value of $`\theta `$ that minimizes the total energy per unit area, in terms of the equilibrium subband densities. Extremizing Eq. (36) with respect to $`\theta `$ for $`n_a>n_b`$ gives
$`\mathrm{cos}\theta =\{\begin{array}{cc}1,& X1\\ X,& 1X1\\ 1,& X1\end{array}`$ (124)
$`X={\displaystyle \frac{(p_Fp_B)/(n_an_b)}{1\mathrm{\Gamma }}},`$ (125)
where $`\mathrm{\Gamma }`$ is defined in Eq. (42). In the four-component phase, the kinetic energy dominates over the exchange energy, and $`(n_an_b)<|p_Fp_B|`$, so that $`\mathrm{sin}\theta =0`$. In practice, the only time that we find $`|\mathrm{cos}\theta |<1`$ (for $`t=0`$) is in the one-component phase. For Eq. (124) to minimize the energy with respect to $`\theta `$, the second derivative of the energy with respect to $`\theta `$ must be positive, which is equivalent to requiring that $`\mathrm{\Gamma }<1`$. If $`\mathrm{\Gamma }>1`$ (which is true for the $`p=3`$ phase near balance), then $`\mathrm{cos}\theta =\pm 1`$.
It follows from Eqs. (31) and (124) that
$`(n_1n_2)`$ $`=`$ $`\{\begin{array}{cc}(n_an_b),& X1\\ (p_Fp_B)/(1\mathrm{\Gamma }),& 1X1\\ (n_an_b),& X1\end{array}`$ (129)
In the special case of interest where $`|X|<1`$ (which requires $`p=1`$), we can calculate the Eisenstein ratio $`R_E`$ at the point where the layers are balanced:
$$R_E(p_F=p_B)=\frac{1}{2}(1\delta p_B/\delta p_F)\frac{\mathrm{\Gamma }}{1\mathrm{\Gamma }},$$
(130)
where we have used Eqs. (11) and (129). (Recall that $`\delta p_B/\delta p_F0`$ if the back-gate voltage $`V_B`$ is kept fixed, but that $`\delta p_B/\delta p_F=1`$ if the total density is kept fixed.)
Within the tight-binding model of tunneling, the MFA model shows that any finite value of the tunneling matrix element $`t`$ prohibits either layer from completely emptying out, regardless of the gate charges $`(p_F,p_B)`$. Extremizing the energy per unit area \[Eq. (36)\] with respect to $`\theta `$ for $`t>0`$ and $`p_Fp_B`$ shows that neither $`\mathrm{sin}\theta =0`$ nor $`\mathrm{cos}\theta =0`$ are ever local extrema. When $`t`$ is small and $`p_F<<p_B`$, including any negative values of $`p_F`$ (i.e., for $`\zeta 1`$, roughly speaking), then extremizing the energy per unit area yields
$$\pi \theta \frac{t/(e^2dn_T/2ϵ)}{\left[|\zeta |(1\mathrm{\Gamma })\right]},$$
(131)
to lowest order in $`t/(e^2dn_T/2ϵ)`$. This result is a local minimum provided that it is positive, which is true in the limit we are considering here. It follows from Eq. (31) that $`n_1/n_T(\varphi /2)^2`$, so that $`n_1t^2`$. Hence $`n_1`$, although small, is always nonzero for $`t>0`$. In actual samples, large bias changes the tunneling matrix element $`t`$. Sufficiently large bias shifts the bottom (minimum energy) of the wells relative to one another so greatly that $`t`$ can be driven (for all practical purposes) to zero.
Interlayer exchange is significant only when the layer densities and their separation are sufficiently small. In order for interlayer exchange to contribute, there must be (1) more particles in one subband than another ($`n_a>n_b`$), and (2) nonzero $`\mathrm{sin}\theta `$ ($`\theta 0,\pi `$). Thus, for example, the case of balanced layers with $`\theta =\pi /2`$ at very low densities ($`n_a=n_T`$) is a situation in which interlayer exchange contributes strongly. We discuss this case in Sec. VI, and we find there that interlayer exchange does indeed suppress interlayer charge transfer. For the case of unbalanced layers at high total density, generally $`n_a>n_b`$; but when $`t=0`$, $`\theta `$ is usually equal to $`0`$ or $`\pi `$, so that in the MFA, interlayer exchange does not contribute. Near depletion, where one of the layers empties out, the situation is not as clear, so we now analyze that situation in some detail later below.
Even with interlayer exchange, it turns out that the MFA model is always unstable with respect to an abrupt exchange-driven intersubband charge transfer from (low-density) subband $`b`$ to (higher-density) subband $`a`$ when the particle density in subband $`b`$ gets small enough. The abrupt intersubband charge transfer is probably an unphysical feature of the MFA model that is not observed in real experiments. We believe that a proper treatment of the correlation energies (which have been entirely omitted here) would help fix this shortcoming. Nevertheless, we can still investigate what the MFA has to say about interlayer charge transfer, since the subbands are in general different from the layers, when we include interlayer exchange.
#### 2 Subbands densities nearly equal
We now consider the limit in which $`\mathrm{\Delta }n(n_an_b)n_T`$, so that the double-layer system is only slightly pseudospin-polarized. This will be the case for the two- ($`p=2a`$) and four- ($`p=4`$) component ground states for small $`t`$ and $`|p_Fp_B|`$. We begin by computing the change in the ground-state energy per unit area due to changing $`\mathrm{\Delta }n`$ from zero to a small but finite value. Expanding Eq. (36) to second order in $`\mathrm{\Delta }n`$ gives
$`{\displaystyle \frac{\mathrm{\Delta }_0}{L_xL_y}}={\displaystyle \frac{(\mathrm{\Delta }n)^2}{p\nu _0}}t\mathrm{\Delta }n\mathrm{sin}\theta {\displaystyle \frac{e^2}{4\pi ϵ}}{\displaystyle \frac{(\mathrm{\Delta }n)^2}{\sqrt{p\pi n_T}}}`$ (132)
$`+{\displaystyle \frac{e^2d}{8ϵ}}\left[\mathrm{\Delta }n\mathrm{cos}\theta (p_Fp_B)\right]^2+\mathrm{\Gamma }_0{\displaystyle \frac{e^2d}{2ϵ}}\left({\displaystyle \frac{\mathrm{\Delta }n}{2}}\right)^2\mathrm{sin}^2\theta ,`$ (133)
where
$$\mathrm{\Gamma }_0\underset{n_an_b}{lim}\mathrm{\Gamma }.$$
(134)
The quantity $`\mathrm{\Gamma }_0`$ is calculated in Sec. 4 b of the Appendix. Extremizing Eq. (132) with respect to $`\theta `$ gives
$`2t\mathrm{cos}\theta +{\displaystyle \frac{e^2d}{2ϵ}}(p_Fp_B)\mathrm{sin}\theta `$ (136)
$`=(1\mathrm{\Gamma }_0){\displaystyle \frac{e^2d}{2ϵ}}\mathrm{\Delta }n\mathrm{sin}\theta \mathrm{cos}\theta .`$
By taking the second derivative of Eq. (132) with respect to $`\theta `$, we find that Eq. (136) is a local minimum provided that
$`2t\mathrm{sin}\theta +{\displaystyle \frac{e^2d}{2ϵ}}(p_Fp_B)\mathrm{cos}\theta `$ (137)
$`(1\mathrm{\Gamma }_0){\displaystyle \frac{e^2d}{2ϵ}}\mathrm{\Delta }n\mathrm{cos}(2\theta )>0.`$ (138)
Except for $`p=3`$ near balance, $`0<\mathrm{\Gamma }_0<1`$, so that
$$0<1\mathrm{\Gamma }_0<1.$$
(139)
Hence Eqs. (136) and (137) imply that if $`t>0`$ but $`|p_Fp_B|=0`$, then $`\mathrm{cos}\theta =0`$ so that $`\theta =\pi /2`$ (except for $`p=3`$.) If, however, $`|p_Fp_B|>0`$ but $`t=0`$, then $`\mathrm{sin}\theta =0`$ so that $`\theta =0,\pi `$ (except for $`p=1`$.) There is thus a competition between the effects of $`t`$ and $`|p_Fp_B|`$. If neither $`t`$ nor $`|p_Fp_B|`$ is zero, then $`\mathrm{sin}\theta 0`$ and $`\mathrm{cos}\theta 0`$. In the limit that $`(e^2d/2ϵ)|p_Fp_B|t`$, then
$$\mathrm{cos}\theta \frac{(e^2d/2ϵ)(p_Fp_B)}{2t}1$$
(140)
for $`\mathrm{\Delta }n0`$. On the other hand, in the limit that $`t(e^2d/2ϵ)|p_Fp_B|`$, then
$$\mathrm{sin}\theta \frac{2t}{(e^2d/2ϵ)[|p_Fp_B|(1\mathrm{\Gamma }_0)\mathrm{\Delta }n]}1.$$
(141)
In general, $`\theta `$ must be solved numerically.
Extremizing Eq. (132) with respect to $`\mathrm{\Delta }n`$ gives
$$\mathrm{\Delta }n=\frac{2t\mathrm{sin}\theta +(e^2d/ϵ)(p_Fp_B)\mathrm{cos}\theta }{\frac{4}{p\nu _0}+\frac{e^2d}{2ϵ}(\mathrm{cos}^2\theta +\mathrm{\Gamma }_0\mathrm{sin}^2\theta )\frac{e^2}{4\pi ϵ}\frac{4}{\sqrt{p\pi n_T}}},$$
(142)
which is a local minimum provided that its denominator is positive, as may be seen by computing the second derivative of Eq. (132) with respect to $`\mathrm{\Delta }n`$. Note that both interlayer tunneling ($`t`$) and gate bias ($`|p_Fp_B|`$) produce pseudospin polarization (increase $`\mathrm{\Delta }n`$.)
When $`p_F=p_B`$ and $`t`$ is (arbitrarily) small but nonzero so that $`\theta =\pi /2`$, Eq. (142) yields the pseudospin Stoner enhancement factor in Eq. (88). This is discussed in more detail in Sec. IV B 2.
When $`\mathrm{sin}\theta =0`$ (which requires $`t=0`$), then Eq. (142) gives
$$\frac{\mathrm{\Delta }n}{|p_Fp_B|}=\frac{d/a_0}{d/a_0+(2/p)[1r_s/(\pi \sqrt{2/p})]}.$$
(143)
This is the case for the $`p=2a`$ and $`p=4`$ states for $`\mathrm{\Delta }n/n_T1`$. It follows from this that near balance ($`p_Fp_B`$), the Eisenstein ratio for fixed $`p_B`$ is given by
$$R_E=\frac{1}{2}\left[1+\frac{(p/2)d/a_0}{1r_s/(\pi \sqrt{2/p})}\right]^1,$$
(144)
where we have use of Eq. (11). Equation (143) says that, when $`\mathrm{sin}\theta =0`$, then $`\mathrm{\Delta }n<|p_Fp_B|`$ for small $`\mathrm{\Delta }n/n_T`$, provided that
$$r_s<\pi \sqrt{2/p}.$$
(145)
Now, the GRPA estimate of the interparticle separation required for spin polarization in a single two-dimensional layer is $`r_s=\pi /\sqrt{2}`$, which is just the right-hand side of Eq. (145) for $`p=4`$. We therefore expect that, as long as the ground state has four components, it will be true that $`\mathrm{\Delta }n<|p_Fp_B|`$, and this is indeed what our MFA calculations find for small $`r_s`$.
According to Eq. (143), $`\mathrm{\Delta }n<|p_Fp_B|`$ for small $`\mathrm{\Delta }n/n_T`$ for $`p=2a`$ until $`r_s>\pi `$. The interparticle separation $`r_s=\pi `$ is also the GRPA estimate of $`r_s^{(0)}(1,2)`$ required for pseudospin polarization. Thus for $`d>0`$ we expect that at higher densities (small $`r_s`$), $`\mathrm{\Delta }n<|p_Fp_B|`$ throughout the $`p=4`$ state and in the low-$`r_s`$ region of the $`p=2a`$ state, but that $`\mathrm{\Delta }n>|p_Fp_B|`$ for the high-$`r_s`$ region of the $`p=2a`$ state, at least for small $`\mathrm{\Delta }n/n_T`$. This is in fact what our MFA calculations show. It is also true that in the $`p=3`$ phase (which has $`\mathrm{\Delta }n>0`$ even when $`p_F=p_B`$), $`\mathrm{\Delta }n>|p_Fp_B|`$ for $`\zeta 1`$, although not for $`\zeta `$ on the order of one. Of course, for $`r_s`$ sufficiently small, pseudospin polarization occurs so that $`\mathrm{\Delta }n=n_T>|p_Fp_B|`$.
Figure 8 shows a plot of the subband densities $`n_{\alpha s}`$ and the ratio $`(p_Fp_B/(n_an_b)`$ versus $`r_s`$ for fixed layer separation $`d/a_0=5`$ and fixed layer imbalance $`\zeta =0.2`$. For small $`r_s`$, the $`p=4`$ phase is obtained, and $`(p_Fp_B)/(n_an_B)>1`$. For $`r_s2.011`$, the $`p=4`$ phase is obtained, and $`(p_Fp_B)/(n_an_B)<1`$. For larger $`r_s`$, the spin-polarized $`p=2a`$ phase is obtained. Note that for $`p=2a`$ the ratio $`(p_Fp_B)/(n_an_B)`$ is larger than one for smaller $`r_s`$, but smaller than one for larger $`r_s`$. For even larger $`r_s`$ (not shown), the $`p=1`$ phase would be obtained with $`(p_Fp_B)/(n_an_B)=\zeta `$, which in this case has $`\zeta =0.2`$.
If the denominator in Eq. (142) is not positive, then the global minimum for the energy per unit area occurs for $`\mathrm{\Delta }n=n_T`$, corresponding to pseudospin polarization. Thus the GRPA condition for stability against abrupt intersubband charge transfer is just the condition that denominator in Eq. (142) be positive, or equivalently
$$\frac{d}{a_0}+\frac{(2/p)[1r_s/(\pi \sqrt{2/p})]}{(\mathrm{cos}^2\theta +\mathrm{\Gamma }_0\mathrm{sin}^2\theta )}>0.$$
(146)
When $`\mathrm{sin}\theta =0`$ (which requires $`t=0`$) so that the pseudospin is Ising-like and interlayer exchange does not contribute, then Eq. (146) is equivalent to the stability condition against abrupt interlayer charge transfer given in Eq. (8), when the electron lengths $`s(n)`$ are approximated by Eq. (109). When $`\mathrm{cos}\theta =0`$ (e.g., when $`p_F=p_B`$ and $`t`$ has any finite positive value), then the violation of the inequality in Eq. (146) is equivalent to the condition that the pseudospin Stoner ehancement factor $`I`$, given in Eq. (88), is equal to one, which signals the transition to pseudospin ferromagnetism and SILC. Because $`0<\mathrm{\Gamma }_0<1`$, gate imbalance ($`|p_Fp_B|>0`$, so that $`\mathrm{sin}^2\theta `$ decreases and $`\mathrm{cos}^2\theta `$ increases) makes the double-layer system more unstable towards pseudospin polarization.
#### 3 Subband densities versus bias
In this section, we show some illustrative calculations of the effect of layer imbalance. We plot the subband densities $`n_{\alpha s}`$ and the value of $`\mathrm{\Gamma }`$ versus the gate imbalance parameter $`\zeta =(p_Fp_B)/n_T`$ at fixed layer separation $`d/a_0=5`$ assuming zero interlayer tunneling ($`t=0`$), for different values of $`r_s`$. We find that if we fix layer density (or equivalently, $`r_s`$) and vary the gate imbalance parameter $`\zeta `$, then (for $`t=0`$) there are six distinct patterns of transitions between noncrystalline MFA ground states. We list the six possibilities below, in order of increasing layer imbalance (beginning at $`\zeta =0`$) from left to right, and in order of increasing average interparticle separation per layer $`r_s`$ from top to bottom:
1. $`r_s<r_s(3,4)`$:
$`(p=4)(p=3)(p=2b)`$,
all with $`\mathrm{sin}\theta =0`$
2. $`r_s(3,4)<r_s<r_s(2,3)`$:
$`(p=3)(p=2b)`$,
all with $`\mathrm{sin}\theta =0`$
3. $`r_s(2,3)<r_s<\sqrt{2}r_s(2,4)`$:
$`(p=2a)(p=3)(p=2b)`$,
all with $`\mathrm{sin}\theta =0`$
4. $`\sqrt{2}r_s(2,4)<r_s<r_s(1,2)`$ and $`\zeta _c>(1\mathrm{\Gamma }_1)`$:
$`(p=2a)(p=1`$),
all with $`\mathrm{sin}\theta =0`$
5. $`\sqrt{2}r_s(2,4)<r_s<r_s(1,2)`$ and $`\zeta _c<(1\mathrm{\Gamma }_1)`$:
$`(p=2a)(p=1)`$,
with $`\mathrm{sin}\theta >0`$ for $`\zeta _c<\zeta <(1\mathrm{\Gamma }_1)`$
6. $`r_s>r_s(1,2)`$:
$`(p=1)`$ only,
with $`\mathrm{sin}\theta >0`$ for $`0\zeta <(1\mathrm{\Gamma }_1)`$
Here $`(p=2a)`$ denotes the spin-polarized two-component state, $`(p=2b)`$ denotes the pseudospin-polarized two-component state, and $`\zeta _c`$ is the value of the gate imbalance parameter $`\zeta `$ at which the $`(p=2a)(p=1)`$ transition occurs. The quantity $`\sqrt{2}r_s(2,4)`$ appearing in cases (4) and (5) above correspond to the critical density $`n_c`$ for the MFA spin polarization transition for a single layer, expressed in terms of the average interparticle spacing per layer: $`1/\sqrt{\pi (n_c/2)a_0^2}`$. (Note that for the same total density $`n_T`$, a double-layer system has an average layer $`r_s=1/\sqrt{\pi (n_T/2)a_0^2}`$ that is $`\sqrt{2}`$ larger than the single-layer $`r_s=1/\sqrt{\pi n_Ta_0^2}`$.) The quantity $`n_c`$ is discussed below Eq. (99). We shall illustrate the first four of these possibilities in the remainder of this section, and discuss the last two possibilities (which exhibit SILC) in Sec. VI. It is evident from the above list that SILC, which requires $`\mathrm{sin}\theta 0`$, occurs only for $`p=1`$.
Figure 9 is an example of case (1), with $`r_s=1`$. This gives a four-component ($`p=4`$) phase when the gates are balanced ($`\zeta =0`$), and maintains a $`p=4`$ state for most of the range of $`\zeta `$, followed by a $`p=3`$ state for $`\zeta `$ near one. If we were to increase $`\zeta `$ beyond one (not shown), corresponding to $`p_B<0`$, then a pseudospin-polarized $`p=2b`$ state would be obtained.
Figure 10 is an example of case (2), with $`r_s=r_s(2,4)2.011`$. This gives a three-component ($`p=3`$) state when the gates are balanced ($`\zeta =0`$), and maintains a $`p=3`$ state for most of the range of $`\zeta `$, followed by a pseudospin-polarized $`p=2b`$ state for $`\zeta `$ near one and beyond. Note that for small $`\zeta `$, $`\mathrm{\Gamma }(p=3)>1`$.
Figure 11 is an example of case (3), with $`r_s=2.5`$. This gives a spin-polarized two-component ($`p=2a`$) state when the gates are balanced ($`\zeta =0`$), and maintains a $`p=2a`$ state for $`\zeta <0.55`$, followed by a $`p=3`$ state for $`0.55<\zeta <0.95`$, and then a pseudospin-polarized $`p=2b`$ state for $`\zeta >0.95`$.
Figure 12 is an example of case (4), with $`r_s=9`$. This gives a spin-polarized two-component ($`p=2a`$) state when the gates are balanced ($`\zeta =0`$), and maintains a $`p=2a`$ state for $`\zeta <\zeta _c0.45`$, followed by a one-component ($`p=1`$) state for $`\zeta >\zeta _c`$. Because $`\zeta _c>1\mathrm{\Gamma }_1`$, $`\mathrm{sin}\theta =0`$ throughout, and thus no SILC is found. In the next section, we consider values of $`r_s`$ large enough that the $`p=1`$ state is achieved for $`\zeta _c<1\mathrm{\Gamma }_1`$, thereby producing SILC.
## VI Pseudospin-Polarized States
In this section, we consider the case in which all the particles are in the lowest-energy subband ($`n_a=n_T`$), corresponding to full pseudospin polarization. There are two types of pseudospin-polarized MFA ground states: spin-unpolarized ($`p=2b`$), and spin-polarized ($`p=1`$). The spin-unpolarized case requires either interlayer tunneling ($`t>0`$) or gate imbalance $`|p_Fp_B|>0`$, or both. In the absence of tunneling, the pseudospin-polarized $`p=2b`$ state has $`\mathrm{sin}\theta =0`$ (Ising-like pseudospin), and occurs whenever the total density and the layer imbalance are sufficiently large. If the tunneling $`t`$ is sufficiently large and the total density $`n_T`$ is not too small, then it is possible in principle to obtain a $`p=2b`$ state with $`\mathrm{cos}\theta =0`$, for $`p_F=p_B`$. Tunneling also reduces the value of $`r_s`$ required to achieve the $`p=1`$ state. Because the pseudospin-polarized $`p=2b`$ MFA state does not occur without bias or tunneling, we shall focus mainly on the spin-polarized one-component ($`p=1`$) phase, which can in principle arise without bias or tunneling. The one-component phase is especially interesting because it can occur as a broken-symmetry ground state of a double-layer system in the absence of tunneling or layer imbalance, at very small particle densities and layer separations.
When $`n_a=n_T`$, the ground-state energy (36) becomes
$`{\displaystyle \frac{_0}{L_xL_y}}={\displaystyle \frac{n_T^2}{p\nu _0}}tn_T\mathrm{sin}\theta {\displaystyle \frac{8}{3\sqrt{p\pi }}}{\displaystyle \frac{e^2}{4\pi ϵ}}n_T^{3/2}`$ (147)
$`+`$ $`{\displaystyle \frac{e^2dn_T^2}{8ϵ}}(\mathrm{cos}\theta \zeta )^2+{\displaystyle \frac{\mathrm{sin}^2\theta }{4}}{\displaystyle \frac{e^2dn_T^2}{2ϵ}}\mathrm{\Gamma }_1,`$ (148)
where $`\mathrm{\Gamma }_1=\mathrm{\Gamma }(n_a=n_T)`$ and $`\zeta =(p_Fp_B)/n_T`$. The properties of $`\mathrm{\Gamma }_1`$ are described in Sec. 4 of the Appendix. Equation (147) includes the two cases $`p=1`$ ($`n_a=n_T`$) and $`p=2b`$ ($`n_a=n_a=n_T/2`$). For $`t=0`$, the $`p=2b`$ case has $`\mathrm{cos}\theta =\pm 1`$, so we will focus on the spin- and pseudospin-polarized one-component ($`p=1`$) ground state.
### A Spontaneous interlayer coherence
The pseudospin-polarized ground state offers the possibility of SILC. Recall that SILC occurs when the off-diagonal (or interlayer) density matrix $`\rho _{12}`$ is nonzero in the absence of interlayer tunneling:
$$\rho _{12}\underset{𝐤s}{}c_{1𝐤s}^{}c_{2𝐤s}=\frac{1}{2}(n_an_b)\mathrm{sin}\theta 0,$$
(149)
which requires both finite pseudospin polarization ($`n_a>n_b`$) and $`\mathrm{sin}\theta 0`$. In the pseudospin-polarized ground state ($`n_a=n_T`$), $`\rho _{12}`$ is just the geometric mean of the layer densities:
$$\rho _{12}=\frac{1}{2}n_T\mathrm{sin}\theta =\sqrt{n_1n_2},$$
(150)
where we have made use of Eq. (26). So if the pseudospin-polarized ground state has some density of particles in each layer, it has interlayer phase coherence, $`\rho _{12}0`$. Note that in the one-component phase,
$$\mathrm{sin}\theta =(\rho _{12}+\rho _{21})/n_T$$
(151)
measures the interlayer density matrix, normalized by the total density.
For $`t=0`$ and $`p=1`$, Eq. (124) gives
$`{\displaystyle \frac{n_1n_2}{n_T}}=\mathrm{cos}\theta =\{\begin{array}{cc}1,& X1\\ X,& 1X1\\ 1,& X1\end{array}`$ (155)
$`X={\displaystyle \frac{\zeta }{1\mathrm{\Gamma }_1}}={\displaystyle \frac{\zeta }{1(32/3\pi p)[1S(z)]/z}}`$ (156)
$`{\displaystyle \frac{\zeta }{(32/45\pi )z(1/24)z^2}},`$ (157)
where $`z=2k_Fd`$, and we have made use of Eq. (31). The last line of Eq. (155) holds in the limit that $`z0`$. The layer densities are equal ($`n_1=n_2`$) only at exacly $`p_F=p_B`$; when $`p_F>p_B`$, layer 1 tends to be occupied, and when $`p_F<p_B`$, layer 2 tends to be occupied. Thus, the hypothetical bistability of the one-component phase proposed by Ruden and Wu does not exist, due to SILC. Equation (155) gives
$$\mathrm{sin}\theta =\sqrt{1[\zeta /(1\mathrm{\Gamma }_1)]^2}\mathrm{\Theta }(1\mathrm{\Gamma }_1|\zeta |)\delta _{n_a,n_T},$$
(158)
so that $`\mathrm{sin}\theta `$ in nonzero only for $`|\zeta |<1\mathrm{\Gamma }_1`$ and for $`n_a=n_T`$. It is interesting to note that when the ground state is pseuodspin-polarized, the dependence of the layer densities on external parameters (e.g., layer imbalance $`\zeta `$) does not involve the effective mass $`m^{}`$ of the electrons or holes.
We have found that layer imbalance ($`p_Fp_B`$) can induce SILC at higher total densities than in the balanced case. This is illustrated for $`r_s=11`$ in Fig. 13, which is an example of case (5) introduced in Sec. V B 3. Figure 13 shows a spin-polarized two-component ($`p=2a`$) state when the gates are balanced ($`\zeta =0`$), and maintains a $`p=2a`$ state for $`\zeta <\zeta _c0.28`$, followed by a one-component ($`p=1`$) state for $`\zeta >\zeta _c`$. Because $`\zeta _c<1\mathrm{\Gamma }_1`$, $`\mathrm{sin}\theta >0`$ for $`\zeta _c<\zeta <1\mathrm{\Gamma }_1`$, producing SILC in a finite region of layer imbalance away from $`\zeta =0`$, at a smaller value of $`r_s`$ than is required to achieve SILC for balanced layers.
When the toal density (and layer separation) are sufficiently small, the $`p=1`$ phase with SILC is obtained even in the balanced case. This is illustrated for $`r_s=15`$ in Fig. 14, which is an example of case (6) introduced in Sec. V B 3. Figure 14 shows a one-component ($`p=1`$) state throughout the range of $`\zeta `$, with $`\mathrm{sin}\theta >0`$ for $`\zeta <1\mathrm{\Gamma }_1`$, producing SILC in that region. Note that as $`r_s\mathrm{}`$, then $`z=2k_Fd0`$, so that $`\mathrm{\Gamma }_1(z)1`$, and thus $`1\mathrm{\Gamma }_10`$. Therefore the maximum amount of imbalance $`\zeta `$ which allows SILC decreases with the density, at very low densities.
Figure 15 shows $`\mathrm{sin}\theta `$ versus $`\zeta `$ at $`d/a_04.356`$ for the three values of $`r_s`$, where $`\mathrm{sin}\theta `$ obeys Eq. (158). At the lowest density shown ($`r_s=11`$), the system exhibits SILC when balanced ($`\zeta =0`$) and under bias, until $`\zeta =1\mathrm{\Gamma }_1`$. As the density is raised, SILC is lost for the balanced system but appears suddenly around $`\zeta 0.2`$ for $`r_s=10.5`$ when an abrupt intersubband charge transfer produces pseudospin polarization ($`p=1`$). For $`r_s=9`$ there is only a very small region of layer imbalance $`\zeta `$ that exhibits SILC, and for $`r_s`$ slightly smaller than this value, SILC disappears completely.
Using Eqs. (147), (150), and (155), we may calculate the energy per unit area $`\epsilon (n_1,n_2)`$ defined in Eq. (4), for the one-component phase:
$$\epsilon (n_1,n_2)=\frac{n_T^2}{\nu _0}\frac{8}{3\sqrt{\pi }}\frac{e^2}{4\pi ϵ}n_T^{3/2}+\frac{e^2dn_1n_2}{2ϵ}\mathrm{\Gamma }_1,$$
(159)
where $`n_T=n_1+n_2`$. Recall that $`\epsilon `$ does not include the electrostatic contribution to the energy per unit area. We may use Eqs. (6) and (159) to calculate $`\mu _1`$, the chemical potential measure relative to the energy minimum of layer 1,
$`\mu _1`$ $`=`$ $`{\displaystyle \frac{2n_T}{\nu _0}}{\displaystyle \frac{4}{\sqrt{\pi }}}{\displaystyle \frac{e^2}{4\pi ϵ}}\sqrt{n_T}`$ (161)
$`+{\displaystyle \frac{e^2dn_2}{2ϵ}}\left(\mathrm{\Gamma }_1+n_1{\displaystyle \frac{d\mathrm{\Gamma }_1}{dn_T}}\right),`$
where we have used the fact that in the one-component phase, $`\mathrm{\Gamma }_1`$ depends on $`n_1`$ and $`n_2`$ only through $`n_T=n_1+n_2`$. The quantity $`\mu _2`$ may be obtained by interchanging $`n_1`$ and $`n_2`$ in Eq. (161) and can be used to compute the front-gate voltage $`V_F`$ using Eq. (182) in the Appendix. It is straightforward to check that the difference between $`\mu _1`$ and $`\mu _2`$ satisfies the equilibrium condition in Eq. (5). Equations (7) and (161) may be used to calculate the electronic length $`s_{ij}`$ defined in Sec. II:
$`{\displaystyle \frac{s_{11}}{a_0}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{r_s}{2\pi \sqrt{2}}}+{\displaystyle \frac{1}{2}}{\displaystyle \frac{d}{a_0}}n_2\left(2{\displaystyle \frac{\mathrm{\Gamma }_1}{n_T}}+n_1{\displaystyle \frac{^2\mathrm{\Gamma }_1}{n_T^2}}\right).`$ (162)
The quantity $`s_{22}`$ may be obtained by interchanging $`n_1`$ and $`n_2`$ in Eq. (162).
Interlayer correlations produce a nonzero value of the electron length $`s_{12}`$. From Eqs. (7), (9), and (161) we have
$$s_{12}=s_{11}+\frac{d}{2}\left[\mathrm{\Gamma }_1+(n_1n_2)\frac{\mathrm{\Gamma }_1}{n_T}\right].$$
(163)
Note that $`s_{21}=s_{12}`$, and that
$$s_1s_{11}s_{12}=\frac{d}{2}\left[\mathrm{\Gamma }_1+(n_1n_2)\frac{\mathrm{\Gamma }_1}{n_T}\right].$$
(164)
The length $`s_2`$ can be obtained from $`s_1`$ by interchanging $`n_1`$ and $`n_2`$. Note that
$$ss_1+s_2=\mathrm{\Gamma }_1d,$$
(165)
so that, from Eq. (12), the Eisenstein ratio for fixed total density (constant $`n_T`$) is
$$R_E=\frac{s}{d+s}=\frac{\mathrm{\Gamma }_1}{1\mathrm{\Gamma }_1}\frac{1}{(32/45\pi )z},$$
(166)
where $`z=2k_Fd`$, and the right-hand side holds in the limit $`z0`$. It is interesting to note that $`R_E1/d`$ as $`d0`$, just as was found in Ref. for the $`\nu _T=1`$ 2LQH state. For fixed $`p_B`$ (nearly equivalent to keeping the back-gate voltage $`V_B`$ constant),
$$R_E=\frac{s_1}{d+s}=\frac{\left[\mathrm{\Gamma }_1+(n_1n_2)\mathrm{\Gamma }_1/n_T\right]}{2(1\mathrm{\Gamma }_1)},$$
(167)
which in the balanced case ($`p_F=p_B`$ so that $`n_1=n_2`$) gives Eq. (130), which is exactly half of Eq. (166).
Figure 16 shows an example of the one-component phase under bias for fixed back-gate density $`p_B`$ (essentially fixed back-gate voltage $`V_B`$.) The normalized layer densities $`n_1`$ and $`n_2`$ are shown, together with the interlayer density matrix $`\rho _{12}/p_B`$ and $`\mathrm{\Gamma }`$, and also the Eisenstein ratio $`R_E`$. For $`p_F/p_B<0.5`$, layer 2 contains all the charge ($`n_2=n_T`$) and layer 1 is empty, so that Eq. (11) gives $`R_E=1`$. For $`0.5<p_F/p_B<2.25`$, both $`n_1`$ and $`n_2`$ are partially occupied, $`\rho _{12}=\sqrt{n_1n_2}`$ is nonzero, and $`R_E`$ has dropped abruptly and become negative, reflecting the presence of SILC, with its value in this region given by Eq. (167). For $`p_F/p_B>2.25`$, layer 2 is empty, $`n_1=n_T`$, and $`R_E=0`$. Figure 16 illustrates an interesting hypothetical situation in which bias and the exchange interaction have completely emptied out layer 2, despite the fact that $`p_B`$ is nonzero. It turns out that within the MFA, sufficently large $`p_F`$ would eventually repopulate layer 2. We analyze this issue further below, when we calculate the energy gap $`\mathrm{\Delta }_{ab}`$ in a pseudospin-polarized state ($`n_a=n_T`$) to an otherwise empty subband ($`n_b=0`$).
Although Zheng and co-workers showed (within an unrestricted HFA) that an abrupt interlayer charge transfer does not occur when the gates are electrostatically balanced ($`p_F=p_B`$), one may ask if an abrupt interlayer charge transfer can occur if the system is biased. According to the MFA developed here, the answer is yes, except at sufficiently small densities. (We are speaking here of zero tunneling; otherwise, neither layer is strictly empty, according to the MFA.) In the MFA, sufficiently strong bias or tunneling or sufficient lowering of the densities eventually produces an abrupt intersubband charge transfer, i.e., at some point, $`n_a`$ suddenly jumps. Whether this translates into abrupt interlayer charge transfers depends on what happens with the phase angle $`\theta `$. If $`\mathrm{sin}\theta 0`$ (SILC), then the interlayer charge transfer is suppressed, or at least somewhat reduced. If the density is so high that strong bias produces the $`p=2b`$ pseudospin-polarized phase with $`\mathrm{sin}\theta =0`$, then the MFA does give an abrupt interlayer charge transfer, because then $`\mathrm{cos}\theta =\pm 1`$, and there is no difference between subband densities and layer densities. So, for example, a system with a density that would correspond to $`p=4`$ when balanced will not exhibit SILC. It is likely that including correlation-energy effects eliminates the abruptness of the transition, but these effects have not been included here.
### B Intersubband gap
The intersubband energy gap $`\mathrm{\Delta }_{ab}`$ for the pseudospin-polarized ($`p=1`$ or $`p=2b`$) phase is defined as the energy required to move a particle from the occupied $`a`$ subband with $`n_a=n_T`$ to the (otherwise empty) $`b`$ subband:
$$\mathrm{\Delta }_{ab}=\frac{\delta }{\delta n}\left(\frac{_0}{L_xL_y}\right),$$
(168)
for
$`\begin{array}{cc}n_a(1/p)(n_T\delta n),\hfill & n_b\delta n,\hfill \\ n_a(11/p)(n_T\delta n),\hfill & n_b=0.\hfill \end{array}`$ (171)
An outline of the MFA calculation of $`\mathrm{\Delta }_{ab}`$ is given in Sec. 6 of the Appendix. In units of the energy scale $`v_0=e^2/4\pi ϵa_0`$, the MFA intersubband gap is
$`{\displaystyle \frac{\mathrm{\Delta }_{ab}}{v_0}}={\displaystyle \frac{2t}{v_0}}\mathrm{sin}\theta {\displaystyle \frac{4}{r_s^2}}\left[{\displaystyle \frac{1}{p}}+{\displaystyle \frac{d}{a_0}}\mathrm{cos}\theta (\mathrm{cos}\theta \zeta )\right]`$ (172)
$`+\sqrt{{\displaystyle \frac{2}{p}}}{\displaystyle \frac{4}{\pi }}{\displaystyle \frac{1}{r_s}}\mathrm{sin}^2\theta \sqrt{{\displaystyle \frac{2}{p}}}{\displaystyle \frac{1}{r_s}}\{{\displaystyle \frac{\left[e^{z/2}(1z/2)\right]}{z/2}}`$ (173)
$`+{\displaystyle \frac{2}{\pi }}{\displaystyle _0^1}dx(1e^{zx})\mathrm{arccos}(x)\}`$ (174)
where $`\zeta =(p_Fp_B)/n_T`$, and $`z=2k_Fd`$ is the layer imbalance.
The intersubband gap $`\mathrm{\Delta }_{ab}`$ is useful for at least two purposes. First, it provides an estimate for the location of the pseudospin-polarization transition. The condition $`\mathrm{\Delta }_{ab}>0`$ means the pseudospin-polarized ground state is stable against intersubband charge transfers, whereas $`\mathrm{\Delta }_{ab}<0`$ implies the opposite. Thus, solving the equation $`\mathrm{\Delta }_{ab}=0`$ in the MFA yields an estimate of the location of the pseudospin-polarization transition. It turns out that this procedure gives a lower value of $`r_s`$ for the $`p=1`$ transition than the GRPA estimate (obtained using the pseudospin Stoner enhancement factor $`I`$): at $`d=0`$, the MFA $`\mathrm{\Delta }_{ab}`$ calculation (for $`t=\zeta =0=\mathrm{cos}\theta =0`$) gives $`r_s^{(0)}(1,2)=\pi /\sqrt{2}`$, compared to $`r_s^{(0)}(1,2)=\pi `$ from the GRPA. Figure 17 shows $`\mathrm{\Delta }_{ab}`$ versus $`d/a_0`$ when $`t=\zeta =\mathrm{cos}\theta =0`$ for $`r_s=1,2,3,6`$. It is evident that $`\mathrm{\Delta }_{ab}>0`$ only for sufficiently large $`r_s`$, and that it decreases with layer separation $`d/a_0`$. Negative values of $`\mathrm{\Delta }_{ab}`$ indicate regions where the pseudospin-polarized state is not stable.
The intersubband gap has a simple form when $`t=\mathrm{sin}\theta =0`$:
$$\frac{\mathrm{\Delta }_{ab}}{v_0}=\frac{4}{r_s^2}\left[\frac{1}{p}+\frac{d}{a_0}(1|\zeta |)\right]+\sqrt{\frac{2}{p}}\frac{4}{\pi }\frac{1}{r_s}.$$
(175)
Equation (175) provides an estimate of when one of the layers cotains all the particles. It tells us that for sufficently high total density (small $`r_s`$), $`\mathrm{\Delta }_{ab}<0`$ and both layers must be occupied, provided that $`1/2+(d/a_0)(1|\zeta |)>0`$ (which includes $`\zeta =1`$, corresponding to $`p_B=0`$.) This is because the kinetic energy (and the Coulomb energy, for $`|\zeta |<1`$), which favors occupying both layers, dominates over the exchange energy, which favors occupying a single layer, at higher densities. For example, for $`p=2`$ and $`\zeta =1`$, Eq. (175) shows that both layers will be occupied even when $`p_B=0`$, provided that $`r_s<\pi /2`$. On the other hand, for $`p=1`$ and $`\mathrm{sin}\theta =0`$ (i.e., $`|\zeta |>1\mathrm{\Gamma }_1`$), only one layer will be occupied, provided that $`r_s>(\pi /\sqrt{2})[1+(d/a_0)(1|\zeta |)]`$. This is the situation shown in Fig. 16, which shows that layer 2 has completely emptied for $`p_F/p_B>2.25`$. At higher values of $`p_F/p_B`$ (e.g, $`r_s<\pi /2`$, not shown), layer 2 would no longer be empty.
The second use of the MFA calculation of $`\mathrm{\Delta }_{ab}`$ is as a rough estimate of the minimum energy (thermal, or from photons) required to excite particles from the occupied to the unoccupied subband in the one-component state. Perhaps this could be detected with sensitive heat-capacity measurements or by measuring microwave absorption.
### C Coulomb drag
One very interesting feature of the one-component state with SILC is that it is expected to exhibit interlayer drag (finite dc transresistance), even at zero temperature. Ordinarily, if the layers are not correlated in the ground state, current in one layer can drag along particles in the other layer (due to the Coulomb interaction between the layers) only at finite temperature. But if $`\rho _{12}0`$, due to either tunneling or, more interestingly, to SILC, then interlayer correlations present in the 2LES ground state will produce interlayer drag even at zero temperature, as has been predicted for interlayer-correlated 2LQH states. Based on the Kubo formula with the ground state of Eq. (20), we expect that a calculation of the zero-temperature dc transconductivity $`\sigma _d`$ will give
$$\sigma _d\frac{e^2\rho _{12}}{m^{}}\frac{e^2(n_an_b)\mathrm{sin}\theta }{m^{}}.$$
(176)
Calculations of the drag conductivity for a pseudospin-polarized ground state are currently being carried out by other researchers. According to Eqs. (150) and (176), we expect that in the $`p=1`$ phase ($`n_a=n_T`$), $`\sigma _d\sqrt{n_1n_2}`$, approximately (i.e., within an MFA calculation of $`\sigma _d`$.)
It would be interesting to clarify the relationship between $`s_{12}`$, $`\rho _{12}`$, and $`\sigma _d`$. We conjecture that finite interlayer drag at zero temperature requires $`\rho _{12}0`$ at zero temperature (although we have not proved this) and that $`\rho _{12}0`$ (or at least $`\rho _{12}\rho _{21}0`$) at zero implies finite interlayer drag at zero temperature. It is certainly true that $`\rho _{12}0`$ and $`\sigma _d0`$ occur together at zero temperature in the interlayer correlated 2LQH effect. We also think it likely that a similar relation holds between $`s_{12}`$ and $`\rho _{12}`$, and hence between $`s_{12}`$ and $`\sigma _d`$, although we have not proved this either.
## VII Conclusions
We investigated the effects of intralayer and interlayer exchange in biased double-layer systems, in the absence of a magnetic field. This was accomplished using a mean-field approximation (MFA) which, in the limit of balanced layers (no bias), is equivalent to the unrestricted HFA of Zheng and co-workers
### A Findings
We found that a balanced 2LES possesses four possible noncrystalline MFA ground states. The spin- and pseudospin-unpolarized four-component ($`p=4`$) state is obtained at the highest densities. In contrast to earlier work, we found that as the density is lowered, a three-component ($`p=3`$) state with slightly unequal layer densities is obtained. Thus, we find that there is no direct four- to two-component transition. At finite layer separation ($`d>0`$) and zero interlayer tunneling ($`t=0`$), the $`(p=4)(p=3)`$ MFA transition involves a small but abrupt interlayer charge transfer. Any such abrupt interlayer charge transfer will in principle result in a large (formally infinite) value of the Eisenstein ratio $`R_E`$ at the transition. The Eisenstein ratio is a sensitive measure of the interlayer capacitance discussed in Sec. II.
Like Zheng and co-workers, we found that as the total density was lowered, a spin-polarized two-component ($`p=2a`$) state preceded a low-density one-component ($`p=1`$) state possessing SILC, provided that the gates were balanced ($`p_F=p_B`$). This $`p=1`$ state is different from that of Ruden and Wu, whose proposed one-component state occupied a single layer, rather than a single subband consisting of a linear combination of both layers. We obtained a MFA phase diagram for the noncrystalline phases of the 2LES, shown in Fig. 3. This phase diagram is similar to that of Ref. , except for the presence of the $`p=3`$ phase between the $`p=4`$ and $`p=2`$ phases. Only the $`p=1`$ phase was found to possess SILC — i.e., a nonzero interlayer density matrix ($`\rho _{12}0`$) even with zero interlayer tunneling ($`t=0`$). We also defined the pseudospin Stoner interaction parameter $`I`$, and considered the linear response of the MFA ground state to interlayer tunneling, equivalent to a GRPA calculation. We used $`I`$ to obtain an alternate (GRPA) estimate of the location of the $`(p=2)(p=1)`$ transition, shown as the dotted line at the top of Fig. 3. Of course, in the limit of vanishing total density ($`r_s\mathrm{}`$), we expect that a Wigner crystal state is obtained (in the absence of disorder). We did not consider the effects of disorder here, except to note that it limits the maximum $`r_s`$ for a state with mobile particles.
Under bias ($`|p_Fp_B|>0`$), we found that there are five possible noncrystalline ground states. In every case, the MFA gave subbands, which, when occupied, were either completely spin-unpolarized or fully spin-polarized. Including correlation-energy effects would likely produce ground states with intermediate spin polarizations, as is apparently the case in three dimensions. The additional state that can appear under sufficiently large bias and/or interlayer tunneling is a pseudospin-polarized two-component ($`n_a=n_a=n_T/2`$) state, which we labeled $`p=2b`$. The $`p=2b`$ state requires bias and/or tunneling, and has $`\mathrm{sin}\theta =0`$ (and thus no SILC) for $`t=0`$. In Sec. V we studied the effect of bias by considering sytems at fixed total density $`n_T`$, for a range of values of the layer-imbalance parameter $`\zeta =(p_Fp_B)/n_T`$. We enumerated the six possible scenarios for bias-driven transitions between the (five possible) noncrystalline MFA ground states for $`t=0`$. We also showed that a very simple model that assumes no interlayer exchange and no spin polarization is capable of fitting experimental SdH data quite well (see Fig. 6), and that a simple LDF model can do the same in the presence of interlayer tunneling (see Fig. 7).
We studied the one-component phase under applied bias, finding that bias lowers the $`r_s`$ required for SILC (see Fig. 15). Within the MFA, SILC occurs only in the one-component phase: when $`t=0`$, $`\mathrm{sin}\theta `$ is nonzero only when $`n_a=n_T`$. Perhaps including correlation-energy effects would allow SILC for states with partial pseudospin polarization. But $`p=1`$ is only a necessary condition for SILC, not a sufficient one. We found that if the layer imbalance parameter was too large ($`\zeta >1\mathrm{\Gamma }_1`$), then SILC was lost. When SILC occurs, the MFA gave a value of the interlayer density matrix equal to the geometric mean of the layer densities: $`\rho _{12}=\sqrt{n_1n_2}`$. For the case that SILC is present, we calculated the layer densities ($`n_i`$), local values of the chemical potential ($`\mu _i`$), electronic lengths ($`s_{ij}`$), and Eisenstein ratio ($`R_E`$) (see Fig. 16).
Ruden and Wu originally predicted an abrupt interlayer charge tranfer for $`t=0`$ at sufficiently low densities and layer separations, in the balanced case ($`p_F=p_B`$). Like Zheng and co-workers, we found that an abrupt interlayer charge transfer does not occur in the balanced case, due to SILC. However, the MFA (and the unrestricted HFA of Ref. ) does produce abrupt intersubband charge transfers, even for the balanced case. This a feature of the HFA that requires correlation-energy effects to remedy. In the case of nonzero bias, intersubband transfers are equivalent to interlayer transfers if $`\mathrm{sin}\theta =0`$, which is the usual case, except possibly for $`p=1`$. Interlayer subband transfers at $`t=0`$ are reduced or suppressed in the MFA only for the $`(p=2a)(p=1)`$ transition, and only if $`\zeta <(1\mathrm{\Gamma }_1)`$ in the $`p=1`$ phase. So SILC, when present, does reduce or eliminate abrupt interlayer subband transfers, but the MFA does not always eliminate them under bias ($`p_Fp_B`$). If the system is at sufficiently low density and layer separation that it stays in the $`p=1`$ state, then there are no abrupt interlayer charge transfers under bias in the MFA, despite the fact that the layers can empty out as $`p_F`$ is changed (see Fig. 16).
We also calculated the intersubband gap $`\mathrm{\Delta }_{ab}`$ for the pseudospin-polarized ($`p=1`$ or $`p=2b`$) phases within the MFA, defined as the energy to move a particle from the lower energy $`a`$ subband to the higher energy (empty) $`b`$ subband. This energy provides an estimate of the single-particle intersubband gap in the pseudospin-polarized $`p=1`$ and $`p=2b`$ phases, and can be used to estimate the stability of those phases. If the $`p=1`$ phase can be obtained experimentally, $`\mathrm{\Delta }_{ab}`$ might be measured using heat-capacity or microwave/optical techniques. A very interesting feature of the one-component phase with SILC is that it should have nonzero interlayer drag, even at zero temperature, with the size of the interlayer drag conductivity being proportional to the interlayer density matrix $`\rho _{12}`$.
Pseudospin polarization can be detected by SdH measurements, which exhibit oscillations that are periodic in $`1/H`$ (where $`H`$ is the applied magnetic field). The periods of the SdH oscillations are given by
$$\mathrm{\Delta }_{\alpha s}(1/H)=\frac{2\pi e}{\mathrm{}}\frac{1}{A_{\alpha s}}=\frac{2\pi e}{\mathrm{}}\frac{1}{\pi k_{\alpha s}^2}=\frac{e}{h}\frac{1}{n_{\alpha s}},$$
(177)
where $`A_{\alpha s}`$ is the cross-sectional area of the Fermi surface perpendicular to the applied magnetic field for electrons in subband $`\alpha `$ with spin $`s`$. Knowing the total density $`n_T`$ (e.g., from Hall measurements), SdH measurements of the subband densities $`n_{\alpha s}`$ could allow a determination of the degree of spin and pseudospin polarization. For example, in the case of equally balanced layers ($`n_1=n_2=n_T/2`$) having a $`p`$-component ground state ($`p=1,2,4`$) in the absence of tunneling ($`t=0`$), there is a single ($`p`$-fold degenerate) SdH oscillation period,
$$\mathrm{\Delta }_p(1/H)=\frac{2\pi e}{\mathrm{}}\frac{1}{A_p}=\frac{2\pi e}{\mathrm{}}\frac{1}{\pi k_F^2}=\frac{e}{h}\frac{p}{n_T},$$
(178)
which allows $`p`$ to be determined directly from SdH measurements.
### B Can it exist?
Can the one-component state be realized in the balanced case? The $`p=1`$ state is a legitimate solution of in the HFA, but we have not examined its stability here. It is hypothetically possible that the $`p=1`$ state might always be preempted by a Wigner crystal state. Conti and Senatore carried out diffusion Monte Carlo (DMC) simulations in the $`d=0`$ limit, calculating the $`p=4`$ ground-state energy as a function of $`r_s`$ and using previous single-layer DMC results to estimate the $`p=2`$ and $`p=1`$ energies. They also estimated the ground-state energy for a Wigner crystal state, and found that the $`p=4`$ state is obtained for $`r_s<42`$, and that the Wigner crystal state is obtained for larger values of $`r_s`$. In their calculation, neither the $`p=2`$ nor the $`p=1`$ states are ever favored energetically.
Although the DMC results in Ref. show the need to examine the existence and stability of the $`p=1`$ state beyond the HFA, they do not rule out its existence. This is because at $`d=0`$ the fermions possess $`CP(3)`$ symmetry (spin and pseudospin fully rotateable and interchangeable), and estimating the $`p=2`$ and $`p=1`$ energies using single-layer results misses part of the correlation energy, which lowers the $`p=1`$ and $`p=2`$ ground-state energies. Ideally, a DMC simulation of $`CP(3)`$ fermions would be most useful to determine theoretically if the $`p=1`$ state can be obtained, but such calculations might be prohibitively difficult to carry out. As a start, allowing for the possibility of Wigner crystallization (broken translational symmetry) within the MFA calculation would be helpful. Alternatively, a time-dependent MFA calculation of the collective mode would indicate where (for what density and layer separation) the collective mode of the 2LES goes soft ($`\omega 0`$), signaling the onset of Wigner crystallization. We are currently developing such a calculation of the collective mode. Better yet would be a double-layer STLS calculation allowing for the possibility of Wigner crystallization, or at least a determination of when the STLS collective mode goes soft in a double-layer system.
More important is the question of whether SILC can be achieved experimentally in the absence of a strong magnetic field, which serves to quench the kinetic energy of the particles. The MFA and HFA underestimate the value of $`r_s`$ required for the transitions, perhaps by a factor of 10. For example, the spin-polarization transition for a single layer has been estimate to occur for $`r_s20`$. Although such high values of $`r_s`$ have been achieved in $`p`$-type GaAs samples, even higher values of $`r_s`$ will be required to achieve spontaneous pseudospin polarization. Disorder imposes further constraints, because it limits the maximum $`r_s`$ for which the particles are still mobile. However, given the impressive progress in producing double-layer systems with ever-lower particle densities and ever-higher mobilities, it does not seem prudent to rule out the possibility that such a state might someday be realized.
As pointed out by Zheng and co-workers and by Conti and Senatore, most of the considerations presented in Ref. for SILC in the quantum Hall regime should be relevant to the $`p=1`$ phase (with SILC) in the absence of a magnetic field. In both cases, there is a ground state with broken $`U(1)`$ symmetry (when $`t=0`$), due to interlayer exchange at small layer separations and particle densities. It is therefore expected that the $`p=1`$ state will exhibit many of the novel features of the 2LQH state with SILC, including zero-temperature interlayer drag, vortex excitations \[of the angle $`\varphi `$ in Eq. (36)\] and an associated a finite-temperature Kosterlitz-Thouless transition (for $`t=0`$), and interesting many-body effects in tilted magnetic fields for finite interlayer tunneling.
### C Speculations regarding the 2LQH regime
It is interesting to speculate about the applicability of these ideas to the 2LQH effect at total filling factor unity. For sufficiently small distances ($`d<d_c1.2\mathrm{}`$, where $`\mathrm{}=\sqrt{\mathrm{}/eB}`$ is the magnetic length), the 2LQH system exhibits a quantum Hall effect. Theoretically, the small-$`d`$ 2LQH state has nonzero $`\rho _{12}`$ even for $`t=0`$ (SILC), and the 2LQH system exhibits strong Coulomb drag. At sufficiently large layer separations, it is found experimentally that the quantum Hall effect disappears, and it is has been proposed that there is a quantum phase transition to a state without SILC. The nature of the ground state for $`d>d_c`$ is a topic of active investigation. It has been analyzed as a system of two weakly coupled layers of $`\nu =1/2`$ composite fermions (CF’s). Theoretical calculations of the drag at low temperatures predict that the drag resistivity should scale with temperature as $`T^{4/3}`$, based on calculating the effects of gauge fluctuations on two CF layers in the metallic state.. It has also been proposed that the weak coupling between the CF’s in different layers produces BCS pairing between them at sufficiently low temperatures, and that this paired state leads to a finite drag resistivity at zero temperature.
We point out here that besides the apparent BCS instability between CF’s in different layers, double-layer CF systems might be unstable to pseudospin polarization. In the limit $`d\mathrm{}`$, the double-layer $`\nu _T=1`$ system may be regarded as a $`p=2a`$ phase (i.e., spin-polarized but pseudospin-unpolarized) of CF’s in zero effective magnetic field ($`\nu =1/2`$ per layer). Naively, the presumably large effective mass of the CF’s would correspond to a much larger effective value of $`r_s`$ than for the zero-field case, perhaps producing a value of $`r_s`$ sufficently large to obtain a pseudospin-polarized $`p=1`$ phase. Another way of saying this is that the large magnetic field experienced in each $`\nu =1/2`$ layer quenches the kinetic energy of the particles, and that this quenching might strongly enhance exchange instabilities – in this case towards pseudospin polarization ($`n_a>n_b`$), presumably with $`\mathrm{sin}\theta =1`$ (when $`p_F=p_B`$), since no spontaneous interlayer transfer has been found to occur for $`p_Fp_B`$. One appealing feature of a hypothetical $`p=1`$ CF state is that it would have small or zero resistivity in the pseudospin channel (i.e., for oppositely directed currents), the same channel in which the 2LQH state exhibits superfluidity at sufficiently small layer separations. Such a $`p=1`$ state of CF’s would exhibit SILC ($`\rho _{12}0`$ even when $`t=0`$) and therefore possess zero-temperature Coulomb drag. Recent experiments with double-layer systems corresponding to filling factor $`\nu =1/2`$ in each layer provide evidence for the possibility of zero-temperature drag. We are currently investigating the possibility and consequences of pseudospin polarization in double-layer CF systems.
We also note that a perpendicular magnetic field $`𝐁`$ will generally enhance spin and pseudospin polarization because it tends to quench the kinetic energy. This effect enhances the exchange and correlation effects that lead to polarization. We therefore expect that SILC can in principle be found at any filling factor $`\nu _T=(h/e)n_T/B`$, provided that the total density $`n_T`$ and the layer separation $`d`$ are sufficiently small. In particular, if it exists in the one-component phase for zero magnetic field, SILC will probably persist, and even grow stronger, when a perpendicular magnetic field is applied. Finally, we remark that it may prove instructive to view the $`\nu _T=1`$ 2LQH state as a Chern-Simons bosonic condensate of spin- and pseudospin-polarized ($`p=1`$) electrons bound to unit flux quanta.
It has recently been found that double-layer systems in strong magnetic fields near total filling factor unity exhibit a huge resonant enhancement of the interlayer tunneling conductivity when SILC is present. It would be interesting to measure the tunneling conductivity for a tilted sample, since many-body effects in a state with SILC strongly suppress the interlayer tunneling amplitude when the parallel component of the magnetic field exceeds a critical value. This suppression is much stronger than for a system without SILC. We expect a similar strong enhancement of the tunneling conductivity at zero magnetic field, provided that the system possesses SILC. Such tunneling measurements could prove very useful for measuring the strength of SILC in double-layer systems at zero (or higher) magnetic field.
## VIII Acknowledgments
We thank F. David Núñez in deep appreciation of his patience and enthusiasm, and for his invaluable assistance. We thank A. R. Hamilton for providing us with SdH data for a double-layer hole system, and for patiently answering several questions regarding experimental measurements on double-layer systems. Special thanks are also owed to S. Das Sarma and A.H. MacDonald for useful discussions. C.B.H. thanks the Institute for Theoretical Physics (University of California, Santa Barbara), where part of this work was carried out, and for their support through the ITP Scholars Program. This work was supported by a grant from the Research Corporation, and by the National Science Foundation under grant 9972332.
##
### 1 Gate voltages
The layer densities $`(n_1,n_2)`$ are determined theoretically by minimizing the total ground-state energy per unit area \[Eq. (36)\] for fixed gate densities $`(p_F,p_B)`$. Experimentally however, it is the gate voltages which are tuned. If needed, the gate voltages for a given value of $`(p_F,p_B)`$ can be calculated using
$`eV_F`$ $`=`$ $`eD_FE_F+\mu _1+eV_F^{(0)},`$ (179)
$`eV_B`$ $`=`$ $`eD_BE_B+\mu _2+eV_B^{(0)},`$ (180)
and Eq. (6). Here the gate electric fields depend on the gate sheet densities through Gauss’s law, $`E_\alpha =(e/ϵ)p_\alpha `$, and $`eV_\alpha ^{(0)}`$ are sample-dependent constant gate-voltage shifts. Although $`eV_\alpha `$ is approximately equal to $`eD_\alpha E_\alpha `$, Eq. (179) shows that the layer values $`\mu _i`$ of the chemical potential also contribute to the gate voltages.
The layer values of the chemical potential $`\mu _i`$ can be computed numerically from the variation of the equilibrium value of the total energy per unit area (regarded as a function of the front- and back-gate densities $`p_F`$ and $`p_B`$) with respect to infinitesimal changes in gate densities:
$$\delta \overline{}_0/L_xL_y=\mu _1\delta p_F+\mu _2\delta p_B.$$
(181)
For the typical case in which the back-gate voltage $`V_B`$ is kept constant and the back-gate distance $`D_B`$ is much larger than the interparticle and interlayer separation so that $`p_B`$ is nearly constant, it is convenient to use Eq. (5) and write
$$eV_F=eD_FE_F+edE_{12}+\mu _2+eV_F^{(0)},$$
(182)
where Eq. (181) gives
$$\mu _2\frac{}{p_B}\left(\frac{\overline{}_0}{L_xL_y}\right)_{p_F}.$$
(183)
Equation (182) has the advantage of being applicable even when layer 1 empties out. Equations (182) and (183) can be used to calculate theoretically the front-gate voltage. In the limit where the interlayer separation is larger than the intralayer particle separation and (which usually amounts to the same thing) interlayer correlations can be neglected, then variations of $`\mu _2`$ with $`p_F`$ are small in comparison with $`edE_{12}`$, so that the effects of $`\mu _2`$ can be absorbed into the voltage shift $`eV_F^{(0)}`$, thus giving
$$V_FE_FD_F+E_{12}d+V_F^{(0)}.$$
(184)
### 2 Hartree-Fock approximation
In the Hartree-Fock Approximation (HFA) the two-body interaction is factored so that the ground-state energy per unit area is
$`{\displaystyle \frac{_{HF}}{L_xL_y}}`$ $`=`$ $`{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐤s}{}}\epsilon _kc_{1𝐤s}^{}c_{1𝐤s}+c_{2𝐤s}^{}c_{2𝐤s}`$ (185)
$``$ $`{\displaystyle \frac{t}{L_xL_y}}{\displaystyle \underset{𝐤s}{}}c_{1𝐤s}^{}c_{2𝐤s}+c_{2𝐤s}^{}c_{1𝐤s}`$ (186)
$``$ $`{\displaystyle \frac{1}{2(L_xL_y)^2}}{\displaystyle \underset{j_1𝐤_1s_1}{}}{\displaystyle \underset{j_2𝐤_2s_2}{}}V_{j_1j_2}(|𝐤_\mathrm{𝟐}𝐤_\mathrm{𝟏}|)`$ (188)
$`\times c_{j_1𝐤_\mathrm{𝟐}s_1}^{}c_{j_2𝐤_\mathrm{𝟐}s_2}c_{j_2𝐤_\mathrm{𝟏}s_2}^{}c_{j_1𝐤_\mathrm{𝟏}s_1}`$
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{j_1}{}}{\displaystyle \underset{j_2}{}}V_{j_1j_2}(q=0)n_{j_2}n_{j_1}`$ (189)
$``$ $`{\displaystyle \underset{j}{}}{\displaystyle \underset{\alpha }{}}V_{j\alpha }(q=0)p_\alpha n_j`$ (190)
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\alpha \beta }{}}V_{\alpha \beta }(q=0)p_\alpha p_\beta .`$ (191)
The effect last three (the Hartree) terms of Eq. (185) may be calculated by noting that
$`\underset{q0}{lim}V_{ij}(q)=\underset{q0}{lim}{\displaystyle \frac{e^2}{2ϵq}}\left[1\left(1e^{qd_{ij}}\right)\right]`$ (192)
$`=`$ $`\left(\underset{q0}{lim}{\displaystyle \frac{e^2}{2ϵq}}\right){\displaystyle \frac{e^2}{2ϵ}}d_{ij}(\mathrm{}){\displaystyle \frac{e^2}{2ϵ}}d_{ij},`$ (193)
where $`(\mathrm{})`$ denotes the formally divergent part in the last line of Eq. (192). The last three (the Hartree) terms of Eq. (185) become
$`(\mathrm{})`$ $`{\displaystyle \frac{1}{2}}(n_1+n_2p_Fp_B)^2+{\displaystyle \frac{e^2d}{2ϵ}}(p_Fn_1)(n_2p_B)`$ (194)
$`+`$ $`{\displaystyle \frac{e^2D_F}{2ϵ}}p_F(n_1+n_2p_B)+{\displaystyle \frac{e^2D_B}{2ϵ}}p_B(n_1+n_2p_F).`$ (195)
Requiring the first term of Eq. (194) to not diverge imposes charge neutrality: $`n_1+n_2=p_F+p_B`$. From Gauss’s law, the Hartree energy \[Eq. (194)\] may therefore be written, up to an overall constant, as
$$\frac{ϵ}{2}\left[E_{12}^2d+E_F^2D_F+E_B^2D_B\right],$$
(196)
where $`E_{12}`$ is the electric field between layers 1 and 2, $`E_F`$ is the electric field between the front gate and layer 1, and $`E_B`$ is the electric field between the back gate and layer 2.
Equation (196) is just the electric field energy per unit area for the sample; we drop the last two terms since they may be regarded as constants for fixed $`p_F`$ and $`p_B`$. The ground-state energy per unit area may thus be written as
$`{\displaystyle \frac{_0}{L_xL_y}}`$ $`=`$ $`{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐤s}{}}\epsilon _kc_{1𝐤s}^{}c_{1𝐤s}+c_{2𝐤s}^{}c_{2𝐤s}`$ (197)
$``$ $`{\displaystyle \frac{t}{L_xL_y}}{\displaystyle \underset{𝐤s}{}}c_{1𝐤s}^{}c_{2𝐤s}+c_{2𝐤s}^{}c_{1𝐤s}+{\displaystyle \frac{e^2d}{2ϵ}}(n_1p_F)^2`$ (198)
$``$ $`{\displaystyle \frac{1}{2(L_xL_y)^2}}{\displaystyle \underset{j_1𝐤_1s_1}{}}{\displaystyle \underset{j_2𝐤_2s_2}{}}V_{j_1j_2}(|𝐤_2𝐤_1|)`$ (200)
$`\times c_{j_1𝐤_\mathrm{𝟐}s_1}^{}c_{j_2𝐤_\mathrm{𝟐}s_2}c_{j_2𝐤_\mathrm{𝟏}s_2}^{}c_{j_1𝐤_\mathrm{𝟏}s_1}.`$
### 3 Exchange integrals
The exchange integral $`I_{\alpha \beta s}(q)`$ is defined as
$`I_{\alpha \beta s}(q)={\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐊}{}}`$ $`\mathrm{\Theta }(k_{\alpha s}|𝐊+𝐪/2|)`$ (202)
$`\times \mathrm{\Theta }(k_{\beta s}|𝐊𝐪/2|),`$
where $`\alpha `$ and $`\beta `$ can be either $`a`$ or $`b`$, and where $`k_{\alpha s},k_{\beta s}`$ denote the Fermi wave vectors for particles of spin $`s=,`$ in subbands $`a`$ or $`b`$. Note that Eq. (202) implies that $`I_{\alpha \beta s}`$ is $`1/(2\pi )^2`$ times the shaded area shown in Fig. 18, where for concreteness $`𝐪`$ is taken be in the $`\widehat{𝐱}`$ direction.
Let the quantity $`K_0/2`$ equal the value of $`K_x`$ at which the Fermi circles of radius $`k_{\alpha s}`$ and $`k_{\beta s}`$ intersect:
$$K_0=(k_{\alpha s}^2k_{\beta s}^2)/q=4\pi (n_{\alpha s}n_{\beta s})/q.$$
(203)
Then
$`I_{\alpha \beta s}(q)`$ $``$ $`[n_{\beta s}\mathrm{\Theta }(k_{\alpha s}k_{\beta s}q)`$ (205)
$`+n_{\alpha s}\mathrm{\Theta }(k_{\beta s}k_{\alpha s}q)]`$
$`+`$ $`\mathrm{\Theta }(k_{\alpha s}+k_{\beta s}q)\mathrm{\Theta }(q|k_{\alpha s}k_{\beta s}|)`$ (210)
$`\times {\displaystyle \frac{1}{\pi }}\{n_{\alpha s}[\mathrm{cos}^1\left({\displaystyle \frac{q+K_0}{2k_{\alpha s}}}\right)`$
$`\left({\displaystyle \frac{q+K_0}{2k_{\alpha s}}}\right)\sqrt{1\left({\displaystyle \frac{q+K_0}{2k_{\alpha s}}}\right)^2}]`$
$`+n_\beta [\mathrm{cos}^1\left({\displaystyle \frac{qK_0}{2k_{\beta s}}}\right)`$
$`\left({\displaystyle \frac{qK_0}{2k_{\beta s}}}\right)\sqrt{1\left({\displaystyle \frac{qK_0}{2k_{\beta s}}}\right)^2}]\}.`$
When $`\beta =\alpha `$, then $`k_{\beta s}=k_{\alpha s}`$, $`K_0=0`$, and Eq. (205) becomes
$`I_{\alpha \alpha s}(q)n_{\alpha s}\mathrm{\Theta }(2k_{\alpha s}q)`$ (211)
$`\times {\displaystyle \frac{2}{\pi }}\left[\mathrm{cos}^1\left({\displaystyle \frac{q}{2k_{\alpha s}}}\right)\left({\displaystyle \frac{q}{2k_{\alpha s}}}\right)\sqrt{1\left({\displaystyle \frac{q}{2k_{\alpha s}}}\right)^2}\right],`$ (212)
and the first exchange integral in Eq. (36) may be carried out explicitly:
$$\frac{1}{2L_xL_y}\underset{𝐪}{}V_{11}(q)I_{\alpha \alpha s}(q)=\frac{8}{3\sqrt{\pi }}\frac{e^2}{4\pi ϵ}n_{\alpha s}^{3/2}.$$
(213)
Equation (213) is just the exchange energy per unit area for a uniform spin-polarized two-dimensional electron gas of areal density $`n_{\alpha s}`$.
### 4 Interlayer exchange parameter
A key quantity in our discussion of the effects of interlayer exchange in double-layer systems is the interlayer exchange parameter $`\mathrm{\Gamma }`$, defined by
$`\mathrm{\Gamma }`$ $``$ $`{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐪s}{}}\left[{\displaystyle \frac{V_{11}(q)V_{12}(q)}{e^2d/2ϵ}}\right]`$ (215)
$`\times \left[{\displaystyle \frac{I_{aas}(q)+I_{bbs}(q)2I_{abs}(q)}{(n_an_b)^2}}\right],`$
which is positive and a monotonically decreasing function of the interlayer separation $`d`$:
$$0<\mathrm{\Gamma }(d>0)<\mathrm{\Gamma }(d0),$$
(216)
when the subband densities are regarded as fixed. We note that in the three-component phase ($`p=3`$), unlike the ($`p=1,2,4`$) phases, the equilibrium subband densities change with $`d`$, so that according to Eq. (66),
$$\underset{d\mathrm{}}{lim}\mathrm{\Gamma }_{\mathrm{eq}}(p=3)\frac{1}{(n_an_b)^2d}\frac{d}{a_0}\mathrm{},$$
(217)
when $`p_F=p_B`$, in apparent disagreement with Eq. (216). We stress that the inequality in Eq. (216) is true only when the subband densities $`n_{\alpha s}`$ are regarded as fixed, which is not the case in Eq. (217).
The interlayer exchange parameter $`\mathrm{\Gamma }`$ is important because it determines when SILC is possible (for $`\zeta <1\mathrm{\Gamma }_1`$ and $`p=1`$) – i.e., when $`\mathrm{sin}\theta 0`$. It also determines the value of the pseudospin Stoner enhancement factor $`I`$ (which depends on $`\mathrm{\Gamma }_0`$). $`\mathrm{\Gamma }`$ affects the state of the system (e.g., layer densities and Eisenstein ratio $`R_E`$) whenever $`\mathrm{sin}\theta 0`$.
#### a Inequality
Using the inequality
$$e^2d/2ϵ>V_{11}(q)V_{12}(q),$$
(218)
which is true for $`d>0`$, it follows from Eq. (215) that for $`d>0`$,
$`\mathrm{\Gamma }`$ $`<`$ $`{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐪s}{}}\left[{\displaystyle \frac{I_{aas}(q)+I_{bbs}(q)2I_{abs}(q)}{(n_an_b)^2}}\right]`$ (220)
$`={\displaystyle \underset{s}{}}{\displaystyle \frac{(n_{as}n_{bs})^2}{(n_an_b)^2}}=\mathrm{\Gamma }(d0),`$
where we have used the fact that
$$\underset{d0}{lim}\frac{[V_{11}(q)V_{12}(q)]}{e^2d/2ϵ}=1$$
(221)
and
$`{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐪s}{}}\left[I_{aas}(q)+I_{bbs}(q)2I_{abs}(q)\right]`$ (222)
$`={\displaystyle \frac{1}{(L_xL_y)^2}}{\displaystyle \underset{\mathrm{𝐪𝐊}s}{}}\left[\mathrm{\Theta }(k_{as}|𝐊+𝐪/\mathrm{𝟐}|)\mathrm{\Theta }(k_{bs}|𝐊+𝐪/2|)\right]`$ (223)
$`\times \left[\mathrm{\Theta }(k_{as}|𝐊𝐪/\mathrm{𝟐}|)\mathrm{\Theta }(k_{bs}|𝐊𝐪/2|)\right]`$ (224)
$`={\displaystyle \underset{s}{}}\left\{{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐤}{}}\left[\mathrm{\Theta }(k_{as}k)\mathrm{\Theta }(k_{bs}k)\right]\right\}^2`$ (225)
$`={\displaystyle \underset{s}{}}(n_{as}n_{bs})^2.`$ (226)
Thus the condition
$$(n_an_b)(n_an_b)0$$
(227)
is sufficient to guarantee that $`\mathrm{\Gamma }<1`$.
Equation (222) is true at finite temperature when $`I_{\alpha \beta s}(q)`$ is generalized appropriately. This is because
$$\underset{d0}{lim}\left[V_{11}(q)V_{12}(q)\right]=\frac{e^2d}{2ϵ}$$
(228)
is independent of the wave vector $`q`$. We may write
$`\mathrm{\Gamma }={\displaystyle \frac{1}{(L_xL_y)^2}}{\displaystyle \underset{𝐤_1,𝐤_2,s}{}}\left[{\displaystyle \frac{V_{11}(|𝐤_2𝐤_1|)V_{12}(|𝐤_2𝐤_1|)}{e^2d/2ϵ}}\right]`$ (229)
$`\times \left\{{\displaystyle \frac{\left(a_{𝐤_2s}^{}a_{𝐤_2s}b_{𝐤_2s}^{}b_{𝐤_2s}\right)\left(a_{𝐤_1s}^{}a_{𝐤_1s}b_{𝐤_1s}^{}b_{𝐤_1s}\right)}{(n_an_b)^2}}\right\},`$ (230)
which generalizes $`\mathrm{\Gamma }`$ to finite temperatures. In the limit $`d0`$, Eq. (228) shows that $`\mathrm{\Gamma }`$ approaches
$`{\displaystyle \frac{1}{(n_an_b)^2}}{\displaystyle \underset{s}{}}\left[{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐤}{}}\left(a_{𝐤s}^{}a_{𝐤s}b_{𝐤s}^{}b_{𝐤s}\right)\right]^2`$ (231)
$`={\displaystyle \frac{1}{(n_an_b)^2}}{\displaystyle \underset{s}{}}\left(n_{as}n_{bs}\right)^2`$ (232)
$`\{\begin{array}{cc}1,\hfill & p=1(n_a=n_T)\hfill \\ 1,\hfill & p=2a(n_a=n_b=n_T/2)\hfill \\ 1,\hfill & p=3(n_a=n_a=n_b=n_T/3)\hfill \\ 1/2,\hfill & p=2b(n_a=n_a=n_T/2)\hfill \\ 1/2,\hfill & p=4(n_a=n_a=n_b=n_b=n_T/4)\hfill \end{array}.`$ (238)
Empirically, we find that within the MFA, the spins in subbands $`a`$ and $`b`$ are either completely unpolarized (at higher densities) or fully polarized (at sufficiently low densities.) Therefore the only possible MFA configurations of spin and pseudospin that would not satisfy the inequality in Eq. (227) and that might therefore have $`\mathrm{\Gamma }>1`$ would be three-component ($`p=3`$) states in which subband $`a`$ (the majority subband) is spin-unpolarized and subband $`b`$ (the minority subband) is spin-unpolarized:
$$n_a=n_a,n_a>n_b>n_b=0.$$
(239)
#### b Pseudospin-unpolarized $`\mathrm{\Gamma }`$
It is useful to define and evaluate
$$\mathrm{\Gamma }_0\underset{n_an_b}{lim}\mathrm{\Gamma }.$$
(240)
We begin our calculation of $`\mathrm{\Gamma }_0`$ by noting that
$`I_{aas}(q)+I_{bbs}(q)2I_{abs}(q)=`$ (241)
$`{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐊}{}}\left[\mathrm{\Theta }(k_{as}|𝐊+𝐪/2|)\mathrm{\Theta }(k_{bs}|𝐊+𝐪/2|)\right]`$ (242)
$`\times \left[\mathrm{\Theta }(k_{as}|𝐊𝐪/2|)\mathrm{\Theta }(k_{bs}|𝐊𝐪/2|)\right]`$ (243)
$`{\displaystyle \frac{(k_F\mathrm{\Delta }n/n_T)^2}{L_xL_y}}{\displaystyle \underset{𝐊}{}}\delta (k_F|𝐊+𝐪/2|)`$ (244)
$`\times \delta (k_F|𝐊𝐪/2|)`$ (245)
$`=\left({\displaystyle \frac{k_F}{2\pi }}{\displaystyle \frac{\mathrm{\Delta }n}{n_T}}\right)^2{\displaystyle \frac{\mathrm{\Theta }(1x)}{x\sqrt{1x^2}}},`$ (246)
where $`\mathrm{\Delta }n(n_an_b)0`$, $`k_F=\sqrt{4\pi n_T/p}`$ is the Fermi wave vector per layer for the state with $`p`$ components ($`p=2,4`$), and $`xq/2k_F`$. Using Eq. (241), we obtain
$`\mathrm{\Gamma }_0`$ $`=`$ $`{\displaystyle \frac{2}{p}}{\displaystyle \frac{2/\pi }{e^2d/2ϵ}}{\displaystyle _0^1}𝑑x{\displaystyle \frac{[V_{11}(2k_Fx)V_{12}(2k_Fx)]}{\sqrt{1x^2}}}`$ (247)
$`=`$ $`{\displaystyle \frac{2}{p}}{\displaystyle \frac{2}{\pi z}}{\displaystyle _0^{\pi /2}}𝑑\theta {\displaystyle \frac{\left(1e^{z\mathrm{sin}\theta }\right)}{\mathrm{sin}\theta }},`$ (248)
where $`z=2k_Fd`$, and the second line is obtained by the substitution of variables $`x=\mathrm{sin}\theta `$. It is straightforward to obtain $`\mathrm{\Gamma }_0(z)`$ for small $`z0`$ by expanding the last line of Eq. (247) in powers of $`z`$,
$$\underset{z0}{lim}\mathrm{\Gamma }_0(z)=\frac{2}{p}\left[1\frac{1}{\pi }z+\frac{1}{12}z^2\right],$$
(249)
up to second order in $`z`$.
Obtaining $`\mathrm{\Gamma }_0`$ for $`z\mathrm{}`$ is more cumbersome. One way to proceed is to define a cutoff $`ϵ`$ that satisfies
$$\frac{1}{z}ϵ1$$
(250)
so that $`\mathrm{sin}\theta \theta `$ for $`\theta <ϵ`$ and $`z\mathrm{sin}\theta 1`$ for $`\theta >ϵ`$. Then
$`\mathrm{\Gamma }_0`$ $``$ $`{\displaystyle \frac{2}{p}}{\displaystyle \frac{2}{\pi z}}[{\displaystyle _0^ϵ}d\theta {\displaystyle \frac{\left(1e^{z\theta }\right)}{\theta }}`$ (252)
$`+{\displaystyle _ϵ^{\pi /2}}d\theta {\displaystyle \frac{1}{\mathrm{sin}\theta }}].`$
The first integral in Eq. (252) may be carried out using the identity
$$_0^R𝑑t\frac{\left(1e^t\right)}{t}=\mathrm{ln}(R)+\gamma +_R^{\mathrm{}}𝑑t\frac{e^t}{t},$$
(253)
where $`\gamma 0.5772`$ is Euler’s constant, and $`Rzϵ\mathrm{}`$, so that the last term of Eq. (253) can be dropped for large $`R`$. The second integral in Eq. (252) is well known:
$$𝑑\theta \frac{1}{\mathrm{sin}\theta }=\mathrm{ln}[\mathrm{tan}(\theta /2)].$$
(254)
Now, $`\mathrm{tan}(ϵ/2)ϵ/2`$ for $`ϵ1`$, so that the logarithmically divergent ($`\mathrm{ln}ϵ`$) parts of the two integrals in Eq. (252) cancel each other, leaving
$$\underset{z\mathrm{}}{lim}\mathrm{\Gamma }_0=\frac{2}{p}\frac{2}{\pi z}\left[\mathrm{ln}(2z)+\gamma \right].$$
(255)
#### c Pseudospin-polarized $`\mathrm{\Gamma }`$
We now compute
$$\mathrm{\Gamma }_1\underset{n_an_T}{lim}\mathrm{\Gamma }.$$
(256)
When the double-layer system is pseudospin polarized so that $`n_a=n_T`$, then $`I_{bbs}(q)=I_{abs}(q)=0`$, and it follows from Eqs. (211) and (215) that
$`\mathrm{\Gamma }_1`$ $`=`$ $`{\displaystyle \frac{1}{p}}{\displaystyle \frac{16}{\pi z}}{\displaystyle _0^1}𝑑x\left(1e^{zx}\right)\left[\mathrm{arccos}(x)x\sqrt{1x^2}\right]`$ (257)
$`=`$ $`{\displaystyle \frac{1}{p}}{\displaystyle \frac{32}{3\pi z}}\left[1S(z)\right],`$ (258)
where $`z2k_Fd`$, and
$`S(z)`$ $`=`$ $`{\displaystyle \frac{3}{2}}{\displaystyle _0^1}𝑑xe^{zx}\left[\mathrm{arccos}(x)x\sqrt{1x^2}\right]`$ (259)
$`=`$ $`{\displaystyle \frac{3\pi }{4z}}\left\{1{\displaystyle \frac{2}{z}}\left[I_1(z)L_1(z)\right]\right\}`$ (260)
$``$ $`\{\begin{array}{cc}1(3\pi /32)z+(1/15)z^2(\pi /256)z^3,\hfill & z0\hfill \\ 3\pi /4z,\hfill & z\mathrm{}\hfill \end{array}`$ (263)
so that
$`\mathrm{\Gamma }_1(z)`$ $``$ $`{\displaystyle \frac{1}{p}}\{\begin{array}{cc}1(32/45\pi )z+(1/24)z^2,\hfill & z0\hfill \\ (32/3\pi )/z8/z^2,\hfill & z\mathrm{}\hfill \end{array}`$ (266)
Here $`I_1`$ and $`L_1`$ are modified Bessel and modified Struve functions of the first kind, respectively:
$$I_n(z)=i^nJ_n(iz),L_n(z)=i^{(n+1)}H_n(iz),$$
(267)
where $`J_n`$ is the ordinary Bessel function of order $`n`$, and $`H_n`$ is the Hankel function of order $`n`$.
Obtaining Eq. (259) is somewhat involved. The first part of the integral in the first line of Eq. (259) can be obtained by writing
$$_0^1𝑑xe^{zx}\mathrm{arccos}(x)=_0^1𝑑xe^{zx}\left[\frac{\pi }{2}\mathrm{arcsin}(x)\right],$$
(268)
and using the identity
$`{\displaystyle _0^1}𝑑xe^{zx}\mathrm{arcsin}(x)`$ (269)
$`={\displaystyle \frac{\pi }{2z}}\left[I_0(z)L_0(z)e^z\right],`$ (270)
which corrects a misprint in Eq. (4.551.1) of Ref. . One then obtains
$`{\displaystyle _0^1}𝑑xe^{zx}\mathrm{arccos}(x)`$ (271)
$`={\displaystyle \frac{\pi }{2z}}\left[1+L_0(z)I_0(z)\right],`$ (272)
The second part of the integral in the first line of Eq. (259) can be obtained by writing
$`{\displaystyle _0^1}𝑑xe^{zx}x\sqrt{1x^2}={\displaystyle \frac{}{z}}{\displaystyle _0^1}𝑑xe^{zx}\sqrt{1x^2}`$ (273)
$`={\displaystyle \frac{}{z}}\left[1{\displaystyle _0^1}𝑑x{\displaystyle \frac{ze^{zx}}{\sqrt{1x^2}}}\right]`$ (274)
$`={\displaystyle \frac{\pi }{2}}{\displaystyle \frac{}{z}}\left[{\displaystyle \frac{L_1(z)I_1(z)}{z}}\right],`$ (275)
and by using the identities
$`{\displaystyle \frac{}{z}}\left[{\displaystyle \frac{I_1(z)}{z}}\right]`$ $`=`$ $`{\displaystyle \frac{I_2(z)}{z}}`$ (276)
$`{\displaystyle \frac{}{z}}\left[{\displaystyle \frac{L_1(z)}{z}}\right]`$ $`=`$ $`{\displaystyle \frac{L_2(z)}{z}}+{\displaystyle \frac{2}{3\pi }}`$ (277)
to obtain
$$_0^1𝑑xe^{zx}x\sqrt{1x^2}=\frac{1}{3}+\frac{\pi }{2z}\left[L_2(z)I_2(z)\right]$$
(278)
Combining Eqs. (271) and (278) and using the identities
$`I_0(z)I_2(z)`$ $`=`$ $`{\displaystyle \frac{2I_1(z)}{z}}`$ (279)
$`L_0(z)L_2(z)`$ $`=`$ $`{\displaystyle \frac{2L_1(z)}{z}}+{\displaystyle \frac{2z}{3\pi }}`$ (280)
gives the second line of Eq. (259). The third line of Eq. (259) follows from power series (for small $`z`$) and asymptotic (for large $`z`$) expansions of $`I_1(z)`$ and $`L_1(z)`$.
We note that for the same value of $`p`$, $`\mathrm{\Gamma }_1(z)\mathrm{\Gamma }_0(z)`$ for all $`z`$. This may be seen by the subsitution of variables $`x=\mathrm{sin}\theta `$ in Eq. (257):
$`\mathrm{\Gamma }_1(z)`$ $`=`$ $`{\displaystyle \frac{2}{p}}{\displaystyle \frac{2}{\pi }}{\displaystyle _0^{\pi /2}}𝑑\theta {\displaystyle \frac{\left(1e^{z\mathrm{sin}\theta }\right)}{z\mathrm{sin}\theta }}`$ (282)
$`\times \left\{\mathrm{sin}(2\theta )[(\pi 2\theta )\mathrm{sin}(2\theta )]\right\}\mathrm{\Gamma }_0(z),`$
where $`\mathrm{\Gamma }_0`$ is expressed as an integral over $`\theta `$ in the last line of Eq. (247). However, for differing values of $`p`$, $`\mathrm{\Gamma }_1(p=1)>\mathrm{\Gamma }_0(p=2)`$ for $`0<z<z_c44.09`$. Both $`\mathrm{\Gamma }_1(p=1)`$ and $`\mathrm{\Gamma }_0(p=2)`$ are plotted in Fig. 19. For $`z>z_c`$ (not shown), $`\mathrm{\Gamma }_1(p=1)<\mathrm{\Gamma }_0(p=2)`$.
### 5 Stoner enhancement factor
We now outline our calculation of the Stoner interaction parameter for $`t0`$. Consider a two- or four-component state with equal subband densities (pseudospin unpolarized), $`n_a=n_b=n_T/2`$. For small $`t0`$, imagine moving a small amount of charge $`\mathrm{\Delta }n/2`$ from subband $`b`$ to subband $`a`$, $`n_an_T/2+\mathrm{\Delta }n/2,n_bn_T/2\mathrm{\Delta }n/2`$, so that $`(n_an_b)=\mathrm{\Delta }n`$, and calculate the change in the total energy \[Eq. (36)\] as $`\mathrm{\Delta }n0`$. The effect of the change of densities on the last term of Eq. (36) may be calculating by using Eq. (247). The change in the energy per unit area due to $`\mathrm{\Delta }n`$ as $`\mathrm{\Delta }n0`$ is then given by
$`{\displaystyle \frac{\mathrm{\Delta }_0}{L_xL_y}}`$ $`=`$ $`{\displaystyle \frac{(\mathrm{\Delta }n)^2}{p\nu _0}}t\mathrm{\Delta }n{\displaystyle \frac{1}{\sqrt{\pi p}}}{\displaystyle \frac{e^2}{4\pi ϵ}}{\displaystyle \frac{(\mathrm{\Delta }n)^2}{\sqrt{n_T}}}`$ (283)
$`+`$ $`{\displaystyle \frac{e^2d}{2ϵ}}\left({\displaystyle \frac{\mathrm{\Delta }n}{2}}\right)^2\mathrm{\Gamma }_0.`$ (284)
Minimizing $`\mathrm{\Delta }_0/L_xL_y`$ with respect to $`\mathrm{\Delta }n`$ to solve for $`\mathrm{\Delta }n`$ and using the definition Eq. (87) of the Stoner enhancement $`I`$ gives
$`I`$ $`=`$ $`{\displaystyle \frac{\nu _0e^2}{2\pi ϵk_F}}\left(1p{\displaystyle \frac{\pi }{4}}k_Fd\mathrm{\Gamma }_0\right)`$ (285)
$`=`$ $`{\displaystyle \frac{\nu _0}{\pi }}\left\{2V_{11}(2k_F){\displaystyle _0^1}𝑑x{\displaystyle \frac{[V_{11}(2k_Fx)V_{12}(2k_Fx)]}{\sqrt{1x^2}}}\right\},`$ (286)
which is just the first line of Eq. (88). Equation (285) is equal to the $`t0`$ limit of the Stoner interaction parameter calculated in the GRPA, given in Eq. (14) of Ref. .
### 6 Intersubband gap
The subband transfer energy $`\mathrm{\Delta }_{ab}`$ for the pseudospin-polarized ($`n_a=n_T`$) $`p=1`$ or $`p=2b`$ phase is defined as the MFA energy required to move a particle from the occupied $`a`$ subband to the (otherwise empty) $`b`$ subband:
$$\mathrm{\Delta }_{ab}=\frac{\delta }{\delta n}\left(\frac{_0}{L_xL_y}\right),$$
(287)
where $`\delta _0`$ denotes the change of $`_0`$ under
$`\begin{array}{cc}n_a(1/p)(n_T\delta n),\hfill & n_b\delta n,\hfill \\ n_a(11/p)(n_T\delta n),\hfill & n_b=0,\hfill \end{array}`$ (290)
where $`\delta nn_T`$, and $`n_b=0`$ reflects the fact that at low densities ($`n_b=\delta n0`$), subband $`b`$ will be spin-polarized. In order to calculate $`\mathrm{\Delta }_{ab}`$ we first compute
$`{\displaystyle \frac{\delta }{\delta n}}{\displaystyle \underset{s}{}}\left[I_{aas}(q)+I_{bbs}(q)2I_{abs}(q)\right]`$ (291)
$`={\displaystyle \frac{\delta }{\delta n}}{\displaystyle \frac{1}{L_xL_y}}{\displaystyle \underset{𝐊}{}}\left[\mathrm{\Theta }(k_{as}|𝐊+𝐪/2|)\mathrm{\Theta }(k_{bs}|𝐊+𝐪/2|)\right]`$ (292)
$`\times \left[\mathrm{\Theta }(k_{as}|𝐊𝐪/2|)\mathrm{\Theta }(k_{bs}|𝐊𝐪/2|)\right]`$ (293)
$`{\displaystyle \frac{\delta }{\delta n}}{\displaystyle \frac{2}{L_xL_y}}{\displaystyle \underset{𝐊s}{}}\mathrm{\Theta }(k_{as}|𝐊+𝐪/2|)[\delta k_{as}\delta (k_{as}|𝐊𝐪/2|)`$ (294)
$`\mathrm{\Theta }(\delta k_{bs}|𝐊𝐪/2|)]`$ (295)
$`{\displaystyle \frac{\delta }{\delta n}}{\displaystyle \frac{1}{2\pi ^2}}{\displaystyle \underset{s}{}}[\delta k_{as}{\displaystyle }d^2K\mathrm{\Theta }(k_{as}|𝐊+𝐪|)\delta (k_{as}|𝐊|)`$ (296)
$`\mathrm{\Theta }(\delta k_{bs}|𝐊|)]`$ (297)
$`={\displaystyle \frac{\delta }{\delta n}}{\displaystyle \frac{1}{2\pi ^2}}{\displaystyle \underset{s}{}}[2k_{as}\delta k_{as}\mathrm{arccos}(q/2k_{as})\mathrm{\Theta }(2k_{as}q)`$ (298)
$`\pi (\delta k_{bs})^2\mathrm{\Theta }(k_{as}q)]`$ (299)
$`=\left[{\displaystyle \frac{2}{\pi }}\mathrm{\Theta }(1x)\mathrm{arccos}(x)+2\mathrm{\Theta }(1/2x)\right],`$ (300)
where $`k_{as}=\sqrt{4\pi n_{as}}`$ so that
$$\delta k_{as}=(2\pi /k_F)\delta n_{as},\delta k_b=\sqrt{4\pi \delta n},$$
(301)
and $`k_F=\sqrt{4\pi n_T/p}`$. In order to obtain the last line of Eq. (291), we expanded the bracketed terms to first order with respect to $`\delta n`$, and defined $`xq/2k_F`$.
The subband transfer gap is then given by
$`\mathrm{\Delta }_{ab}`$ $`=`$ $`{\displaystyle \frac{2n_T}{p\nu _0}}+2t\mathrm{sin}\theta +{\displaystyle \frac{4}{\sqrt{\pi }}}{\displaystyle \frac{e^2}{4\pi ϵ}}\sqrt{n_T/p}`$ (305)
$`{\displaystyle \frac{e^2dn_T}{2ϵ}}\mathrm{cos}\theta \left(\mathrm{cos}\theta \zeta \right)`$
$`{\displaystyle \frac{\mathrm{sin}^2\theta }{4}}{\displaystyle \frac{e^2k_F}{2\pi ϵ}}\{{\displaystyle \frac{\left[e^{z/2}(1z/2)\right]}{z/2}}`$
$`+{\displaystyle \frac{2}{\pi }}{\displaystyle _0^1}dx(1e^{zx})\mathrm{arccos}(x)\},`$
where $`\zeta =(p_Fp_B)/n_T`$ and $`z=2k_Fd`$. Equation (305) is expressed in dimensionless form in Eq. (172). |
warning/0003/cs0003040.html | ar5iv | text | # Implementing Integrity Constraints in an Existing Belief Revision SystemThis work was supported in part by the US Army Communications and Electronics Command (CECOM), Ft. Monmouth, NJ through a contract with CACI Technologies. CSE Technical Report 2000-03
## Introduction
Belief revision, or belief change, is the term used to describe any change in a knowledge base. The form of belief revision discussed in this paper is removal of propositions from a knowledge base that is known to be inconsistent in order to restore consistency. This is especially important to information fusion, where information is combined from multiple sources which might contradict each other. This paper describes some belief revision theories and the considerations that arose when they were implemented and added to the existing belief revision subsystem of a mature knowledge representation and reasoning system, SNePS(??).
These considerations center around the impossibility of implementing deductive closure and the proper weighting of belief revision guidelines. We address the need to formalize theories that take into account the fact that deductive closure cannot be guaranteed in a real-world, need-based, implemented system. We also explore one technique for following the belief revision guideline of minimizing damage to the belief space while retaining the most credible beliefs.
The next section provides the background necessary to understand the alterations we made to SNePS. Included are brief descriptions of the following: (a) the different groups doing belief revision research, (b) the integrity constraints we intend to implement, (c) SNePS and its belief revision sub-system, (d) the status of constraint adherence before system alterations, and (e) previous research to improve adherence.
The following two sections discuss the current changes made to improve adherence and implement automatic belief revision (autoBR). The latter gives the user a sense of non-monotonicity, because it is possible to add a belief to the belief space and, consequently, lose a previously held belief. However the underlying relevance-style, paraconsistent logic remains monotonic.
The final section contains conclusions and plans for future work.
## Background
### Theory vs. Implementation
Belief revision research can be divided into two groups: Theoretical vs. Implementations. These two groups differ in the amount of information assumed to be in a knowledge base. The theoretical researchers develop postulates about how a knowledge base should react during revision based upon a number of guidelines. One of these is the assumption of deductive closure, which means that everything derivable is contained in the knowledge base—a deductively closed belief space (DCBS), which is infinite in size and impossible to implement, although it can be simulated with limitations.
Implemented belief spaces must be finite and of reasonable size, and the reasoning operations performed on them must be done in a finite and reasonable time. Beliefs must be added gradually over time and not all derivable beliefs can be guaranteed to be present in the knowledge base. We refer to this kind of belief space as a deductively open belief space, or DOBS(?). In a DOBS, it is possible for a proposition to be derivable from the existing belief space without it being present in that belief space. If that proposition is the negation of another proposition derivable from the belief space, the belief space would be inconsistent, but not known to be inconsistent—i.e. the system would be unaware of the inconsistency. The normal meaning of the word *consistent* cannot, therefore, be applied to a DOBS. For the purpose of this paper, however, the term consistent will be used to describe a belief space or belief set that has no known contradictions (unless otherwise noted).
Even research on finite belief bases still refers to deductive closure in its determination of consistency or when considering belief revision postulates (????). Consistency is determined by the presence of a contradiction in the *implicit* beliefs of a belief base. How can this reliably be implemented? Even if it can be simulated on a small scale knowledge base, the theories developed might be suspect if applied to a very large system. Nebel voiced these concerns as well (?).
We address the need to formalize theories that take into account the fact that deductive closure cannot be guaranteed in a real-world, need-based, implemented system—even if the system is restricted by something as simple as the user needing a response within one minute. These theories need to suggest a belief revision technique that:
* takes time and complexity limitations into account
* recognizes that adhering to these limitations might result in revision choices that are poor in hindsight
* catches and corrects these poor choices as efficiently as possible.
### Integrity Constraints for Belief Revision
Gärdenfors and Rott (?) discuss postulates for belief revision using a coherence approach. These postulates, first presented in (?), are based on four integrity constraints (paraphrased below):
1. a knowledge base should be kept consistent whenever possible;
2. if a proposition can be derived from the beliefs in the knowledge base, then it should be included in that knowledge base;
3. there should be a minimal loss of information during belief revision;
4. if some beliefs are considered more important or entrenched than others, then belief revision should retract the least important ones.
These constraints come from the Theorist group, so strict adherence to constraints 1 and 2 in an implemented system is impossible. The proper weighting and combining of constraints 3 and 4 remains an open question in belief revision research and a challenge for both the Theorists and the Implementers.
### SNePS and SNeBR Before Alteration
#### SNePS
SNePS is a logic- and network-based knowledge representation, reasoning, and acting system designed to constitute the mind of a natural language competent cognitive agent. The underlying logic of SNePS is a monotonic, relevance-style, paraconsistent logic (?). One way that users can interact with SNePS is through the SNePSLOG interface—an interface which allows the user to input propositions in a style that uses “predicate calculus augmented with SNePS logical connectives” (??)—where propositions are expressed as well-formed formulas, or wffs.
Propositions added to the knowledge base by the user are called hypotheses. Propositions derived from those existing in the belief space are called derived propositions. The system records a justification for each proposition (whether it is a hypothesis or derived) by associating it with an *origin set* consisting of the hypotheses used in its derivation. An origin set for a belief is a set of hypotheses which is known to *minimally derive* that belief—i.e. no subset of a belief’s origin set is known to derive that belief.
This is along the style of an ATMS, short for “assumption-based” truth maintenance system —a term “introduced by (??), although similar ideas had been investigated earlier by (?).” (?) also mentions that (?) presented the “first description of as ATMS-like system.”
A hypothesis has a singleton origin set, containing only itself, but might also be derivable from other hypotheses. Multiple derivations of a single proposition can result in its having multiple origin sets. If the hypotheses in a proposition’s origin set are asserted, or believed, then the proposition is also believed and is part of the belief space. This situates SNePS firmly on the foundations side of the coherence/foundations belief revision divide, but a coherence approach can be simulated by additionally asserting each derived belief as a hypothesis.
In the SNePS terminology, the belief space is the set of believed propositions—both hypotheses and derived beliefs—which are supported by the current context. The current context is, intensionally, a named structure that contains a set of hypotheses. That set is the extensional context. When a new hypothesis is added, the intensional context now contains a different extensional set of hypotheses. When we refer to adding and removing hypotheses from “the context”, we are referring to the intensional context.
Unlike theoretical knowledge bases, implemented ones cannot promise deductive closure for a knowledge base, because of the space and time limitations of the real world. SNePS attempts to derive propositions as they are asked for—either by the user or by the system as it performs backward chaining or forward inference. This is typical of a DOBS as it is formalized in (?).
#### SNeBR
The SNePS belief revision sub-system, SNeBR (?), is activated when a derived proposition or a hypothesis is added to the belief space, *and* it explicitly contradicts a pre-existing belief. The following are examples of explicit contradictions in SNePS:
* `P` and `~P`
* `P` and `~(P` $``$ `Q)`
but *not*
* `Q` and `Q=>P` and `~P`
because, in this last case, `P` is only an implicit belief and must be derived before that contradiction can be detected.
The detection of an explicit contradiction is almost instantaneous, even in a knowledge base with thousands of nodes, due to the Uniqueness Principle (?), which states that no two SNePS terms denote the same entity.
> \[Therefore an\] explicit contradiction …in the belief space, is easily recognized by the system because …the data structure representing P is directly pointed to by the negation operator in the data structure representing `~P`. (?)
The user has the option of letting the belief space remain inconsistent or activating a manual version of belief revision to restore consistency. The latter is performed by forming a minimally-inconsistent set of hypotheses, which can be made consistent upon the removal of any one of its members. This set is the union of the origin sets for the contradicting beliefs—multiple origin sets for a belief result in multiple inconsistent sets to be revised. For example: If the contradictory propositions `P` and `~P` had one ($`\alpha `$) and two ($`\beta `$ and $`\gamma `$) origin sets respectively, then there would be two minimally-inconsistent sets formed for belief revision: ($`\alpha `$ $``$ $`\beta `$) and ($`\alpha `$ $``$ $`\gamma `$).
After forming the inconsistent sets, SNeBR prompts the user to remove at least one proposition from each set to restore consistency. It is up to the user to decide which beliefs should be removed (retracted, become unasserted).
### How SNeBR Adheres to Constraints 1 and 2
#### Adherence to Constraint 1
Because a SNePS belief space is a DOBS, it can only claim consistency in terms of not knowing of any contradictions. Immediately upon discovery of a contradiction, however, the system activates SNeBR to restore consistency. Therefore, the user always has the option to maintain consistency “whenever possible”.
#### Adherence to Constraint 2
Although SNePS cannot promise that all derivable beliefs are in the knowledge base, it does derive a proposition upon query if that proposition is derivable from the existing knowledge base. Therefore, SNePS follows an altered version of constraint 2: If a proposition can be derived from the beliefs in the knowledge base, then the system will produce it if it is asked for.
### Previous Attempts at Ordering Beliefs and AutoBR
The researchers described in this section chose to order beliefs based on relative credibility *as determined by the user*—i.e. the user decides which beliefs (or types of beliefs) are more credible than others. This not the only way to order beliefs as “more important or entrenched,” but, due to our interest in information fusion, this epistemic entrenchment is the way we, also, have chosen to order our beliefs. To this end, any reference in this paper to orderings of beliefs or sources should be assumed to mean ordering based on credibility (unless otherwise stated). This ordering was then used in the implementation of automatic belief revision (autoBR) which allowed the systems described to perform belief revision without user interaction.
Cravo and Martins (?) introduced an altered version of SNePS, called SNePSwD (SNePS with Defaults), that incorporated default reasoning. It also offered automatic belief revision based on ordering beliefs by credibility and specificity. The system allowed the user not only to order beliefs but to also order the orders. Ordering large amounts of information was tedious, however, and any new additions required updating relevant orderings.
Ehrlich (???) altered a version of SNePSwD that had this ordering capability. She chose to eliminate the laborious hand ordering by defining a predetermined group of “knowledge categories,” with a preset ordering. The category for a proposition was added as another argument of the proposition. She then ordered all her propositions based on the relative order of their knowledge categories. Although this saved her the manual ordering, there were several drawbacks: (a) she had to predetermine the knowledge categories that would be used, (b) no new ones could be added, and (c) the ordering hierarchy of the categories was fixed.
Both Cravo and Martins and Ehrlich developed their automatic belief revision processes (autoBR) to remove hypotheses based on credibility orderings—adhering to constraint 4. How to properly combine and weight both constraints 3 and 4 remains an open topic in belief revision. Our initial attempt to combine and weight them is detailed later.
## Alterations to Aid Adherence to Constraint 3
To minimize information loss, the total number of beliefs removed from the system during revision must be considered. Removal of some belief, `P`, will also remove any propositions that have `P` in all of their currently active origin sets. The revised SNeBR system orders the hypotheses in the inconsistent sets based on the number of derived propositions that they support.
It is also possible to make multiple inconsistent sets consistent with a single retraction in the case that a hypothesis common to those sets is the one chosen for removal. The revised system creates an ordered list of the hypotheses based on how many of the minimally-inconsistent sets they are in.
From these two orderings, two lists are formed: (a) the hypotheses supporting the fewest number of derived propositions, and (b) the hypotheses common to the largest number of inconsistent sets. These two lists are now considered during belief revision in the interest of adherence to constraint 3.
Unlike belief spaces created by deductive closure, our DOBS system builds beliefs as they are queried about, thus we can consider those beliefs to be of high interest to the user. This somewhat validates the connection between the cardinality of a belief set and the information it holds. Other researchers (?) also choose culprits based on set cardinality combined with credibility issues.
Our current approach to minimizing information loss involves counting believed propositions without regard to their internal form. For example, `A` and `B`$``$`C`$``$`D` are each counted as one proposition, as are both `Q(a)` and `all(x)(P(x) => Q(x))`, even thought the latter one of each pair clearly ”encodes” more information. Especially in the context of a DOBS, it would be important to try to assess the usefulness of a proposition in terms of the number of other propositions that might in the future be derived from it. We leave this assessment for future work.
## Epistemic Entrenchment Additions to SNeBR
### Sources and their information
As mentioned earlier, Ehrlich’s knowledge categories have to be determined and ordered off-line before running the system. This, plus the source information being included in each proposition as additional argument, forced a static treatment and implementation. Our revised SNePS system uses meta-propositions to assign sources, allowing dynamic source addition and ordering.
#### Problems with Source Information as an Added Argument
Representing the source information as an additional argument of the predicates has several problems associated with it. For example, if “Fran is smart.” were represented as `Smart(Fran)`, “The prof says that Fran is smart.” could be represented as `Smart(Fran, Prof)`, and the problems are:
1. The source of `Smart(Fran, Prof)` cannot be removed or changed without also removing or changing the belief that Fran is smart. Although that might immediately be reintroduced with `Smart(Fran, Nerd)`, belief revision may have had to be performed in the interim, wasting time and effort.
2. The proposition `~Smart(Fran, Sexist)` might represent either the belief that the sexist is the source of the information that Fran is not smart or the belief that the sexist is not the source of the information that Fran is smart. No matter which one it does represent, there is no obvious way to represent the other.
3. It is not clear how to ascribe a source to a rule, such as `all(x)(Grad(x) => Smart(x))`.
#### Benefits of Source Information in a Meta-Proposition
Representing the source information in a meta-proposition, is to represent it as a belief about the belief. For example, “The prof says that Fran is smart.” would be represented as `Source(Prof, Smart(Fran))`.<sup>1</sup><sup>1</sup>1This is syntactically and semantically correct, because propositions like Smart(Fran)— are, in SNePS, functional terms denoting propositions (?). This solves the three problems cited above:
1. The source of the belief that Fran is smart can be removed or changed, without removing or changing the belief that Fran is smart, by removing `Source(Prof, Smart(Fran))` without removing `Smart(Fran)`, and then, perhaps, introducing a different source, e.g. `Source(Nerd, Smart(Fran))`.
2. The belief that the sexist is the source of the information that Fran is not smart would be represented as `Source(Sexist, ~Smart(Fran))`, whereas the belief that the sexist is not the source of the information that Fran is smart would be represented as `~Source(Sexist, Smart(Fran))`.
3. The belief that the prof is the source of the rule that all grads are smart would be represented by `Source(Prof, all(x)(Grad(x) => Smart(x)))`.
As shown, this allows dynamic interaction, where the user can add, remove, and change source information about a proposition while the system is running and without affecting or entirely rewriting the actual proposition. Source orderings are also stored as propositions, such as `Greater(Prof,Nerd)`, which can be interacted with and reasoned about dynamically. New sources as well as their credibility orderings can be added at any time to the knowledge base. Propositions can even have more than one source, although SNeBR assumes a single source at this time.
### Ordering Propositions (Epistemic Entrenchment) and Sources
Our system currently depends on the user to determine the credibility of sources or beliefs directly. The user inputs information to the system declaring source credibility orders and/or belief credibility orders. These are partial orders that are qualitative and transitive.
The system currently assumes beliefs have, at most, one source, but future research will explore multiple source situations and their implementation. We are also assuming at this time that a more credible source delivers more credible information. E.g. Given the following:
* Lisa is more credible than Bart.
* Lisa tells us, “It is snowing.”
* Bart tells us, “Homer is fat.”
we consider “It is snowing” more credible than “Homer is fat.”
Since ordering can never be assumed complete or unchangeable, the system works with what it has—including propositions whose sources are unknown (assumed at this time to be more credible than beliefs that have recorded sources). An interesting issue to explore in the future would be to have the system dynamically establishing and adjusting source credibility information based on revision experiences.
### Recommendations and Automatic Belief Revision
The user can set the belief revision mode at any time from the top level of the SNePSLOG interface. The two modes are:
offers recommendations, but requires the user to revise (this is the default)
activates automatic belief revision, autoBR
From the union of all the inconsistent sets underlying the contradiction, SNeBR produces three lists that are used to create a recommended culprit list:
the least believed hypotheses
the hypotheses that are the most common (to the largest number of inconsistent sets)
the hypotheses that support the fewest beliefs in the knowledge base.
The first set provides possible culprits that support constraint 4 (remove the least important or least credible beliefs). Both the second and the third sets will provide possible culprits that support constraint 3 (minimal loss of information during belief revision). The culprit list is created by combining these lists using a method described in the next section.
Once the recommended culprit list is formed, the user is notified of the three lists as well as the culprit list. In manual mode, the user must then decide (with the help of those lists) which hypotheses to remove from the context. In auto mode, the system will perform an automatic retraction if the culprit list contains a single hypothesis. Otherwise it reverts to manual. In either case, any unresolved inconsistencies are dealt with manually.
### Culprit List Algorithm
Processing the three lists above to create a culprit list ($`CL`$) results in the smallest, non-empty intersection: $`CL=Minnot\mathrm{}((LBMCFS),(LBMC),(LBFS),(MCFS),LB,MC,FS)`$, where $`Minnot\mathrm{}`$ is a function that chooses the smallest non-empty set from a list of sets in decreasing order of importance (for tie-breaking purpose—e.g. choose $`LB`$ over $`MC`$).
Other factors being equal, removing a belief with low credibility (constraint 4) is currently preferred over one whose removal does the least damage (constraint 3), because credibility orders are not weakened by the absence of deductive closure. Regarding the information used to support constraint 3, the system prefers to remove a more common belief over one with the fewest supported nodes, because that will protect beliefs that have been derived in more ways. For example, given `A, B, D, A->C, B->C, and D->~C,` and the derivations of `C` (both ways) and `~C`, the minimally inconsistent sets would be `{A, A->C, D, D->~C}` and `{B, B->C, D, D->~C}` and both `D` and `D->~C` would be the most common hypotheses. Removal of either results in the loss of `~C` and the retention of `C`, which was derived two ways.
It should be emphasized, however, that the final determination of CL is the smallest, non-empty set found. In this sense, condition 3 regains some of its lost status. For example: if $`(LBFS)`$ contained three hypotheses and $`(MCFS)`$ contained two, the latter would be chosen over the former, even though only the former contained credibility information.
An edited sample of the autoBR output for a belief revision exercise is available in Appendix 1.
## Conclusion and Future Work
By considering the work of coherence and theoretical researchers, we were able to improve our existing belief revision system by adding information essential to improving culprit selection during revision. The revised system combines the constraints of minimal information loss and maximal credibility in its development of a culprit list during belief revision to return consistency to a knowledge base. It considers (a) the number of hypotheses and derived propositions that will be affected by the revision as well as (b) the relative credibility orderings of the hypotheses under consideration. These orderings are determined by both source credibility orderings and credibility orders directly assigned between hypotheses. All source and credibility information can be retracted, altered, added to, and reasoned about while the system is running.
Automatic belief revision is possible when the culprit list contains a single hypothesis. In this case, the system appears to the user to be non-monotonic: i.e. the user can add a proposition to an existing context, but end up with a belief space that is not a superset of the original belief space.
The issue of incorporating source information and credibility ordering into a knowledge base is key to maintaining the credibility of an information fusion system. Selection of a well-believed hypothesis as the culprit (due to other considerations like minimizing damage to the knowledge base) might also indicate a need to re-evaluate the reliability of its source. This might lead to development of a system that dynamically adjusts source and propositional credibility orders based on past performance.
Work for the immediate future will include dealing with (1) a proposition having multiple sources and (2) revising the inconsistent sets as a group. For the latter, we might first partition the inconsistent sets based on which of the contradictory nodes each hypothesis supports (or if it supports both), then analyzing the groups by their inconsistent set as well as by their partition. Then the system could better determine which of the contradictory nodes should be contracted and remove its supports efficiently. We hope that one result will be an improvement on safe contraction (?). For example: If retracting a proposition `P` with the two origin sets of `{A,B,C}` and `{B,D}` ordered in increasing credibility, then retracting the least-believed in each set (as per safe-contraction) would remove both `A` and `B`, when removal of `B` would be sufficient. This improvement might show up as the union of the least believed and the most common sets.
## Acknowledgements
The authors appreciate the insights and feedback of Bill Rapaport, Haythem Ismail, and the SNePS Research Group.
## Appendix 1
Below is a description of the information given to the knowledge base. WFF22 is added last with forward inferencing, which produces the contradictory propositions that trigger SNeBR. The five sources with their credibility orderings are:
| $`Holybook>Prof`$ | `WFF1: GREATER(HOLYBOOK,PROF)` |
| --- | --- |
| $`Prof>Nerd`$ | `WFF2: GREATER(PROF,NERD)` |
| $`Nerd>Sexist`$ | `WFF3: GREATER(NERD,SEXIST)` |
| $`Fran>Nerd`$ | `WFF4: GREATER(FRAN,NERD)` |
`WFF10: all(X)(JOCK(X) => (~SMART(X)))` `WFF11: SOURCE(NERD,WFF10)`
`WFF12: all(X)(FEMALE(X) => (~SMART(X)))` `WFF13: SOURCE(SEXIST,WFF12)`
`WFF14: all(X)(GRAD(X) => SMART(X))` `WFF15: SOURCE(PROF,WFF14)`
`WFF16: all(X)(OLD(X) => SMART(X))` `WFF17: SOURCE(HOLYBOOK,WFF16)`
```
WFF22: FEMALE(FRAN) and OLD(FRAN) and
GRAD(FRAN) and JOCK(FRAN)
```
`WFF23: SOURCE(FRAN,WFF22)`
The following code is an edited version of the system output showing the inconsistencies found and the hypotheses removed through autoBR. Author’s comments are in italics.
After all the information is in, one contradiction is detected:
The contradiction involves the newly derived proposition:
and the previously existing proposition:
To resolve the contradiction, SNePS Belief Revision (SNeBR) analyzes the inconsistent set formed by the union of the two Origin Sets for the contradictory propositions:
| (WFF16,WFF22) $``$ (WFF12,WFF22) $`=`$ |
| --- |
| | (WFF22,WFF16,WFF12) |
The three sets aiding culprit selection:
| The least believed hypothesis: |
| --- |
| | (WFF12) |
| The most common hypotheses: |
| | (WFF22 WFF16 WFF12) |
| The hypotheses supporting the fewest nodes: |
| | (WFF12 WFF16) |
| The system informs the user of its decision: |
| --- |
| | `I will remove the following node:` |
| | | `WFF12: all(X)(FEMALE(X) => (~SMART(X)))` |
The system continues reasoning and discovers the same contradiction (derived in new ways):
The contradiction involves the newly derived proposition:
and the previously existing proposition:
There are two known-to-be-inconsistent sets in the context, now:
The following sets are known to be inconsistent. To make the context consistent, remove at least one hypothesis from each of the sets:
The three sets aiding culprit selection:
| The least believed hypothesis: |
| --- |
| | (WFF10) |
| The most common hypotheses: |
| | (WFF22 WFF10) |
| The hypotheses supporting the fewest nodes: |
| | (WFF14 WFF10) |
The system’s decision in this case:
| I will remove the following node: |
| --- |
| | `WFF10: all(X)(JOCK(X) => (~SMART(X)))` | |
warning/0003/astro-ph0003157.html | ar5iv | text | # Very high energy gamma radiation associated with the unshocked wind of the Crab pulsar
## 1 Introduction
The Crab pulsar is a powerful nonthermal machine, accelerating plasma in the form of a relativistic wind that carries off most of the rotational energy of the pulsar.
At a distance of about $`r=r_\mathrm{S}0.1\mathrm{pc}`$ the wind is terminated by a standing reverse shock, which accelerates the electrons up to energies $`10^{15}\mathrm{eV}`$, and randomizes their pitch angles \[Rees & Gunn 1974, Kennel & Coroniti 1984\]. This results in formation of a bright synchrotron source in the region downstream of the shock. The synchrotron radiation of the Crab Nebula is well studied in a very broad frequency range, from radio to hard X-rays. Its general spectral and spatial characteristics are satisfactorily explained by the relativistic magnetohydrodynamics (MHD) model of Kennel & Coroniti \[Kennel & Coroniti 1984\]. Remarkably, the latter provides also a reasonable explanation, even in its simplified (spherically symmetric) form, for the detected very high energy (VHE) $`\gamma `$-rays as a result of inverse Compton (IC) scattering of relativistic electrons in the ambient low-frequency photon fields. This implies that the study of the TeV IC radiation of the Crab Nebula, combined with synchrotron X-ray emission, can yield unambiguous information about the relativistic electrons and the nebular magnetic field in the downstream region of the shock \[de Jager & Harding 1992, Stepanian 1995, Atoyan & Aharonian, 1996, Hillas et al. 1998\].
Although very important, this information unfortunately does not tell much about the origin and characteristics of the wind, i.e. about the region between the pulsar magnetosphere and the shock. It is generally believed that this region, where almost the whole rotational energy of the pulsar is somehow released in the form of kinetic energy of the wind, cannot be directly observed. This has a simple explanation. Although the wind electrons may have an energy as large as $`10^{13}`$ eV, they move together with the magnetic field and thus do not emit synchrotron radiation. This explains the fact that the region upstream of the shock is underluminous \[Kennel & Coroniti 1984\]. However, this statement is valid only for the synchrotron radiation of the wind. In fact, the wind could be directly observed through its IC radiation. Indeed, the IC $`\gamma `$-radiation of the wind electrons is unavoidable because of the illumination of the wind by external low-energy photons of different origin. There are three isotropic photon field populations that contribute effectively to the production of IC $`\gamma `$-radiation of the nebula in the downstream region: the nonthermal (synchrotron) and thermal (dust) radiation of the Crab Nebula itself, and the 2.7 K microwave background radiation (Atoyan & Aharonian 1996). Formally, all these photon fields could also serve as a target for the IC scattering of the wind before the shock. However, the fluxes of $`\gamma `$-rays emitted through these channels appear to be well below the sensitivity threshold of $`\gamma `$-ray telescopes. Meanwhile, in the pulsar vicinity, namely within approximately 100 light cylinders, the efficiency of production of IC $`\gamma `$-rays dramatically increases because of the existence of intense low energy radiation of the pulsar itself. The radiation consists of two, pulsed and unpulsed components, the latter being modulated at the period of the pulsar.
In this paper we show that at certain circumstances concerning the position of formation of the particle dominated wind, the geometry of the flow and the Lorentz-factor of the bulk motion of the wind, the fluxes of the IC $`\gamma `$-radiation of the wind could be sufficiently high to enable detection by present and forthcoming space-borne and ground-based $`\gamma `$-ray telescopes. Moreover, the distinct spectral features of this radiation could allow effective separation of the “wind” component of radiation from the heavy background, which consists of unpulsed radiation of the Crab Nebula at very high (TeV) energies and pulsed radiation at low (GeV) energies. We argue, that even upper limits obtained in such a study could provide unique information about the origin of the pulsar wind.
## 2 Characteristics of the wind
### 2.1 The total particle ejection rate of the wind
The wind from the Crab pulsar carries away most of the energy of rotation of the pulsar. The energy released in the form of electromagnetic emission of the pulsar, which peaks at gamma-ray energies, does not exceed 1 per cent of the total rotational losses \[Arons, 1996\], therefore it can be neglected in the energy balance of the wind.
The fluxes of the energy and the angular momentum of the wind consist of two parts. One of them corresponds to the matter and another corresponds to the electromagnetic field. The total flux of the kinetic energy of particles can be presented as
$$\dot{E}_{\mathrm{kin}}=\dot{N}mc^2<\gamma _w>,$$
(1)
while the flux of the angular momentum of the matter is equal to
$$\dot{L}_{\mathrm{kin}}=\dot{N}m<r\gamma _wv_\phi >.$$
(2)
Here $`\dot{N}`$ is the total rate of particle ejection from the pulsar magnetosphere, $`<\gamma _\mathrm{w}>`$ is the average Lorentz-factor of the wind, $`r`$ is the distance to the axis of rotation $`v_\phi `$ is the component of the velocity of plasma propagation in the direction of rotation. Hereafter we assume that the wind consists of only electrons and positrons. The total rate of particle ejection can be estimated within the models $`e^\pm `$ pair production in the pulsar magnetosphere:
$$\dot{N}_{\mathrm{gap}}=n_\pm cS_{\mathrm{cap}}\lambda ,$$
(3)
where $`S_{cap}=2\pi R_{}^3\mathrm{\Omega }/c`$ is the total area of the polar caps of the neutron star where the roots of the open magnetic field lines are placed, $`n_\pm `$ is the density of the particles in the primary beam, $`\mathrm{\Omega }`$ is the angular velocity of rotation of the pulsar, $`R_{}`$ is the radius of the pulsar and the factor $`\lambda `$ takes into account the multiplication of particles because the development of electromagnetic cascades in the magnetosphere. In the inner gap models, the electromagnetic cascade is initiated by a beam of electrons accelerated up to the Lorentz-factor $`\gamma _{\mathrm{gap}}210^7`$ \[Ruderman & Sutherland 1975, Arons 1983\]. The density of particles in the beam is of the order of Goldreich-Julian density $`n_\pm =n_{\mathrm{GJ}}`$ \[Goldreich & Julian 1969\] determined as
$$n_{GJ}=\frac{(𝛀𝐁)}{2\pi ec}.$$
(4)
Owing to the electromagnetic cascade in the pulsar magnetosphere, the number of particles increases by a factor of $`\lambda `$ and, correspondingly the Lorentz-factor of particles decreases by the same factor.
The calculations by Daugherty & Harding \[Daugherty & Harding 1982\] and Gurevich & Istomin \[Gurevich & Istomin 1985\] show that for the Crab pulsar this factor is of order of $`10^4`$. Note that these early calculations took into account only multiplication resulting from the cascades supported by two processes – the curvature radiation of electrons and $`e^\pm `$ pair production of $`\gamma `$-rays of these photons in the magnetic field. Meanwhile, the process of the Compton scattering of electrons on the soft thermal emission of the neutron star plays, most probably, a non-negligible role in the cascade development \[Arons 1998\], thus this effect should be taken into account in the estimates of $`\lambda `$. In any case, the uncertainties in the model parameters and assumptions does not allow an accurate theoretical estimate of $`\lambda `$, but instead give a broad range of possible values of $`\lambda `$ between $`10^3`$ and $`10^5`$. Remarkably, a rather accurate estimate of $`<\gamma _\mathrm{w}>`$ in the upstream flow can be derived from the analysis of the nonthermal high energy radiation of the downstream region. Indeed, the explanation of the spectrum of synchrotron X-ray emission by the wind electrons accelerated (redistributed) by the termination shock requires a power-law injection spectrum of the electrons $`Q(E)E^\alpha `$ with $`\alpha 2.4`$ and a cutoff of the spectrum below $`E_{}150200`$ GeV. Also, the interpretation of the X-ray and TeV $`\gamma `$-ray emissions within the synchro-Compton model of the Crab Nebula allows one to derive, with very good accuracy, the average magnetic field in the downstream region and the total luminosity in shock-accelerated electrons, $`B210^4`$ G, and $`\dot{W}3\times 10^{38}\mathrm{erg}/\mathrm{s}`$ (for review see Aharonian & Atoyan \[Aharonian & Atoyan 1998\]). The obvious conservation laws concerning both the number and the total energy of the of relativistic electrons in the downstream and upstream regions (i.e. before and after the shock acceleration) gives $`<\gamma _\mathrm{w}>=\frac{\alpha 1}{\alpha 2}(E_{}/m_\mathrm{e}c^2)1.310^6`$. The accuracy of this estimate depends on the possible range of variation of the parameters $`\alpha `$ and $`E_{}`$ that still fit the data, and is estimated smaller than factor of 4. Correspondingly we find the injection rate of the wind particles $`\dot{N}=\dot{W}/<\gamma _\mathrm{w}>m_\mathrm{e}c^2=2.810^{38}\mathrm{s}^1`$. This implies that for $`\dot{N}_{\mathrm{gap}}=210^{34}`$ the $`\lambda `$-factor should be close to $`10^4`$.
Thus, the cascade multiplication of the primary electrons accelerated in the pulsar magnetosphere leads to the formation of an $`e^\pm `$ plasma with parameters
$$\dot{N}=\frac{\lambda B_0\mathrm{\Omega }^2R_{}^3}{ec^2},$$
(5)
and
$$\gamma _{\mathrm{w0}}=\frac{\gamma _{\mathrm{gap}}}{\lambda }.$$
(6)
The Lorentz factor of this initial wind is close to $`\gamma _{w0}10^3`$ while the kinetic energy flux of the wind is
$$\dot{E}_0\gamma _{wo}mc^2\dot{N}$$
(7)
For the Crab pulsar this flux is estimated as $`1.6\times 10^{35}`$ erg/s. It is much lower than the total rotational losses of the pulsar $`\dot{E}_{rot}=4\times 10^{38}`$ erg/s (see e.g. Shklovsky \[Shklovsky 1968\]). This leads to a conclusion that the most part of pulsar’s rotational energy is carried off by the electromagnetic field (see e.g. Arons \[Arons, 1996\]). The state of the wind is characterized by so-called $`\sigma `$-parameter determined as the ratio of the electromagnetic energy flux to the kinetic energy flux of particles in the wind. At $`\sigma _w1`$ the wind is Poynting-flux dominated. At $`\sigma _w1`$ the wind is kinetic-energy-dominated. Initially the wind ejected from the pulsar magnetosphere is Poynting flux dominated, because $`\sigma _{\mathrm{w0}}=\dot{E}_{\mathrm{rot}}/\dot{E}_02.510^3`$. The estimates of the wind parameters in the outer gap model gives essentially the same magnetization parameter \[Cheung & Cheng 1994, Coroniti 1990\].
On the other hand, the explanation of the characteristics of the nonthermal radiation of the Crab Nebula requires that $`\sigma _w`$ be between $`10^3`$ and $`10^2`$ in the wind region upstream of the termination shock \[Rees & Gunn 1974, Kennel & Coroniti 1984\]. Thus the magnetization parameter $`\sigma _w`$ decreases by several orders of magnitude on the way from the light cylinder to the pre-shock region. Therefore it is difficult to avoid a conclusion that the wind is additionally accelerated in a some region beyond the light cylinder.
The theory by Kennel & Coroniti \[Kennel & Coroniti 1984\] is based on the assumption of ideal MHD flow of the plasma after the terminating shock wave. All dissipative processes are neglected, with exception of the cooling of the plasma due to the synchrotron emission. Lyubarskii \[Lyubarskii 1992\] and Begelman \[Begelman 1998\] argued that the theory of Kennel & Coroniti \[Kennel & Coroniti 1984\] is likely not complete. The magnetic field after the terminating shock is mainly toroidal. It is very well known in plasma physics that such configuration is strongly unstable \[Bateman 1980\]. The instabilities can essentially change the physics of the flow in the Nebula. The basic process is the fast dissipation of the toroidal magnetic field with conversion of its energy into the energy of the relativistic particles. The instability drives the plasma towards equipartition of energy between the magnetic field and the matter. This process will inevitably be accompanied by acceleration of particles. The dissipative processes provide the dynamics of the plasma, in good agreement with the observed velocity of the expansion of the outer edge of the nebula, even for the parameter $`\sigma _\mathrm{w}1`$. We emphasize that, although very reasonable, this argument nevertheless does not solve the problem of the wind acceleration, because even in this case we have to transform wind with $`\sigma _{\mathrm{w0}}10^3`$ into wind with $`\sigma _\mathrm{w}1`$. Moreover, the wind with $`\sigma _\mathrm{w}1`$ has almost the same characteristics as we presented above. The only exception is that the Lorentz factor of the wind with $`\sigma _\mathrm{w}1`$ is twice as small as that of the wind with $`\sigma _\mathrm{w}10^3`$. In the limits of uncertainty of the multiplication factor $`\lambda `$ this difference is not important, however
### 2.2 The energy spectrum of the wind electrons
Here we assume that the wind with $`\sigma 1`$ is formed in some ‘acceleration region’ at a distance to the axis of rotation $`R_w`$. We also assume that the plasma axially isotropically fills all the open field lines. Thus, the wind is not modulated in the azimuthal direction.
For calculation of the spectra of emission we need information about spatial and energetic distribution of particles of the wind. Let us first summarize the information about these characteristics following directly from observations. The average Lorentz-factor of the wind in the regime of $`\sigma 1`$ is determined as
$$<\gamma _\mathrm{w}>=\frac{\dot{E}_{\mathrm{rot}}}{\dot{N}},$$
(8)
with a typical value $`10^6`$.
The kinetic-energy-dominated wind is believed to be cold, because the region of the flow of this wind is observed as an ‘underluminous’ region (see Kennel & Coroniti \[Kennel & Coroniti 1984\] and referenced literature). Otherwise, a hot wind would produce remarkable synchrotron emission, which would contradict the existence of underluminous region.
There is definite latitudinal inhomogeneity of the wind. The observations by ROSAT of the torus of X-ray emission in the Crab Nebula clearly demonstrate that energy flux in the wind varies with latitude \[Hester et al. 1995\]. This fact means that the density of the wind, its Lorentz-factor and the toroidal magnetic field should depend on the latitude.
It is believed that the existence of an X-ray torus implies that the wind flows predominantly in narrow disc-like sector along the equator with opening angle less than $`30^0`$ \[Hester et al. 1995\]. However, one should take into account the fact that optical emission is uniformly distributed in the Nebula. This means that there should be a component of the wind at high latitudes. As the luminosities in optical and in X-rays are comparable the energetics of this high latitude wind component is comparable with the energetics of the equatorial component which produces the X-ray torus. Moreover, the evidence that the wind at high latitudes exists close to the pulsar follows also from HST observations \[Hester et al. 1995\]. In our calculations we assume that the particle flux in the wind is isotropic. However to be consistent with observations in X-rays we assume that the Lorentz-factor and the toroidal magnetic field of the wind depend on latitude.
There are no model independent estimates of the latitudinal dependence of the Lorentz-factor of the wind and toroidal magnetic field. We use here the characteristics of the wind obtained in the model of an axisymmetric rotator \[Bogovalov 1997\]. These characteristics are also valid for oblique rotators \[Bogovalov 1999\].
Below we assume that
$$\gamma _\mathrm{w}=\gamma _{\mathrm{w0}}+\gamma _{\mathrm{max}}\mathrm{cos}^2(\alpha ),$$
(9)
where $`\gamma _{\mathrm{max}}=\sigma _{\mathrm{w0}}\gamma _0`$ is the maximum Lorentz-factor of the plasma on the equator($`\alpha =0`$); $`\gamma _010^3`$ (see equation 6) and $`\sigma _{w0}10^3`$ are the Lorentz-factor and the magnetization parameter of the wind near the light cylinder. The wind is assumed to be monoenergetic at a given latitude. It follows from equation (9) that
$$<\gamma _w>=\frac{2}{3}\gamma _{max}.$$
(10)
In the particular case of the axisymmetric rotator \[Bogovalov & Kotov 1992\],
$$\gamma _{max}=\frac{eB_0R_{}}{2mc^2\lambda }(R_{}\mathrm{\Omega }/c)^2.$$
(11)
### 2.3 The geometry of the wind flow
The geometry of the flow and of the IC process are drawn in fig. 1. The neutron star is placed on the axis of rotation and ejects the wind radially. The dash-dotted vertical lines show the light cylinder. The wind is accelerated in the acceleration zone by some unspecified mechanism and at the distance $`R_W`$ it has characteristics discussed above. The acceleration is completed at $`R_W`$. Beyond $`R_W`$ particles in the wind move along straight lines without further acceleration. The equatorial plane of the pulsar is inclined in relation to the observer at the angle $`\alpha =33^0`$ \[Hester et al. 1995\]. IC photons move along the direction of motion of electrons. Therefore only the particles of the wind directed towards Earth can produce observable emission. The lines of flow of the cold kinetic energy dominated wind after acceleration are not exactly radial. Classical mechanics provides a simple relationship between the rates of the rotational energy losses of the pulsar, $`\dot{E}_{rot}=I\mathrm{\Omega }\dot{\mathrm{\Omega }}`$, and the angular momentum losses, $`\dot{L}=I\dot{\mathrm{\Omega }}`$:
$$\dot{E}_{\mathrm{rot}}=\dot{L}\mathrm{\Omega }.$$
(12)
Here $`I`$ is the momentum of inertia of the neutron star. As the electromagnetic field carries off practically no energy in the kinetic-energy-dominated wind, it does not carry the angular momentum either. In the theory of plasma flow in the magnetosphere of an axisymmetric rotator this statement is verified immediately \[Bogovalov 1997\]. Under this condition it follows from equations (1),(2), and (12) that after the acceleration the azimuthal velocity of the wind $`v_\phi `$ is connected with $`R_W`$ as
$$\frac{v_\phi }{c}=\frac{R_\mathrm{L}}{R_\mathrm{W}}.$$
(13)
From this relationship and from Fig. 1 it follows that the projection of the vector of the velocity of the plasma after acceleration lies on the line tangential to the light cylinder. The angle $`\theta `$ between the direction of the motion of relativistic particles in the wind and the soft thermal photons emitted from the pulsar depends on distance $`r`$ to the axis of rotation:
$$\mathrm{sin}\theta =\mathrm{cos}\alpha \frac{R_\mathrm{L}}{r}.$$
(14)
It is seen from this relationship that at any distance from the pulsar there is a non-zero angle between the radial direction and the velocity of the wind. The Inverse Compton scattering of the wind electrons on soft photons emitted as a fan-like beam from the inner magnetosphere of the pulsar results in the production of hard $`\gamma `$\- ray photons. Equations (14) and (13) remain valid for condition $`\gamma _w\gamma _{w0}`$ \[Bogovalov 1997\], also fulfilled in the wind with $`\sigma _w1`$. Below we will not distinguish between the cases with $`\sigma _w1`$ and $`\sigma _w1`$ .
The angular distribution of soft emission close to the pulsar is not well known. In the outer gap model the optical and soft x-ray emission is generated inside, but not far from the light cylinder \[Romani & Yadigaroglu 1995\]. There are several sources of soft emission in the magnetosphere corresponding to the position of the outer gaps. In this model, the projection of the motion of photons from these sources on the equatorial plane is predominantly directed radially. Then the angle between direction of motion of the wind and soft non-thermal photons can also be estimated by equation (14). Note that at $`R_WR_L`$ the IC flux does not depend on the position of the source inside the magnetosphere if the soft photons are emitted radially.
Our calculations of IC radiation are based on the assumption that the wind is illuminated only by the emission that is directed to Earth. However, as we can not exclude the existence of an additional component of optical emission not directed to the observer, but illuminating the wind, our estimates of $`\gamma `$ -ray flux could be considered only as a lower limit.
## 3 Inverse Compton radiation of the wind
### 3.1 The fields of soft photons
The soft emission from the Crab pulsar consists of two, pulsed and unpulsed, components. The recent ROSAT observations allowed one to distinguish the contribution of the pulsar in the observed unpulsed flux from the emission of the nebula. The main contribution to the unpulsed radiation of the pulsar comes, most probably, from the thermal emission of the hot surface of the neutron star. According to the ROSAT observations, the total energy flux of the unpulsed emission in the range 0.1 - 2.4 keV is $`10^{34}`$ erg/s \[Becker & Trümper 1997\]. The latter could be approximated by black-body spectrum with a temperature $`1.910^6K`$ and total luminosity $`10^{34}`$ erg/s.
The pulsed soft radiation of the pulsar is dominated by nonthermal processes in the magnetosphere. The photons in the optical to X-ray band of this radiation play the most important role in the production of inverse Compton $`\gamma `$-rays. Measurements of the spectra of the pulsed optical emission by different groups \[Nasuti et al. 1996, Oke 1969, Percival et.al. 1993\] give the following spectra of optical photons in the range 1680 - 7400 Å
$$F_\nu =3.1(\frac{\nu }{\nu _0})^{0.11}\mathrm{mJy},$$
(15)
where $`\nu _0=6.8210^{14}\mathrm{Hz}`$.
The X-ray data in the range of 0.1 - 2.4 keV obtained by ROSAT can be approximated by a power-law with the photon index 1.5 and luminosity (assuming that the radiation is emitted isotropically) $`7.110^{35}`$ erg/s \[Becker & Trümper 1997\].
The extrapolation of the soft x-ray emission spectrum to the optical range gives lower flux than is observed \[Knight 1984\]. This means that the optical emission has a cutoff in the ultraviolet region. Unfortunately there are no data for the pulsating emission in this region. Therefore we assume an exponential cutoff in the spectrum above 0.05 keV. The emission is also strongly suppressed at infrared wavelengths \[Middleditch et al. 1983\]. Below we use the following approximation of the averaged density of non-thermal photons near the light cylinder
$$n(ϵ)=Z(ϵ)\mathrm{exp}\left[\left(\frac{810^4\mathrm{keV}}{ϵ}\right)^2\right]\frac{\mathrm{photon}}{\mathrm{cm}^3\mathrm{keV}},$$
(16)
where $`ϵ`$ is the energy of photons in keV and
$`Z(ϵ)=1.8210^{16}ϵ^{1.5}+`$ (17)
$`+1.2510^{19}ϵ^{1.11}\mathrm{exp}\left[\left({\displaystyle \frac{ϵ}{510^3\mathrm{keV}}}\right)\right]{\displaystyle \frac{\mathrm{ph}}{\mathrm{cm}^3\mathrm{keV}}}.`$
The photon density of thermal and non-thermal components of low-frequency radiation on the light cylinder is shown in Fig. 2. We assume that the density of the both photon fields fall off with distance to the pulsar center $`R`$ as $`1/R^2`$.
### 3.2 The fluxes and spectra of gamma radiation
The optical depth characterizing the Compton scattering of the wind electrons propagating through the radiation fields of the pulsar is defined as
$$\tau =\sigma (\omega ,\gamma ,𝐯,𝐜)(1\mathrm{𝐯𝐜})n_{ph}(\omega ,𝐜,r)𝑑l𝑑\omega ,$$
(18)
where $`\sigma (\omega ,\gamma ,𝐯,𝐜)`$ is the total invariant cross-section of the IC scattering, $`\omega `$ is the energy of soft photons in $`mc^2`$. Integration of equation (18) over $`dl`$ is performed along the trajectory of a wind electron (see Fig. 1) from the wind formation position to infinity.
The $`\gamma `$-ray energy flux at the angle $`\alpha `$ to the plane of the pulsar equator is equal to
$`L_{\mathrm{IC}}(E_\gamma )={\displaystyle \frac{\dot{N}}{4\pi D^2}}\times `$ (19)
$`\times {\displaystyle }{\displaystyle }E_\gamma {\displaystyle \frac{d\sigma }{dE_\gamma }}(\gamma _\mathrm{w}(\alpha ),\omega ,E_\gamma ,\theta )n_{ph}(\omega ,𝐜,𝐫)d\omega dl,`$
where $`D=2`$ kpc is the distance to the source, $`\frac{d\sigma }{dE_\gamma }(\gamma _\mathrm{w}(\xi ),\omega ,E_\gamma ,\theta )`$ is the differential cross-section of the Compton scattering of a photon with energy $`\omega `$ and electron with $`\gamma _w`$ encountering under angle $`\theta `$ and producing a photon with energy $`E_\gamma `$. This cross-section was obtained after integration of the differential cross-section \[Berestetskii, Lifshitz & Pitaevskii 1971\] over the emission angle
$`{\displaystyle \frac{d\sigma }{dE_\gamma }}(\gamma ,\omega ,E_\gamma ,\theta )={\displaystyle \frac{\pi r_\mathrm{e}^2}{(pk)^2}}\{{\displaystyle \frac{1}{(pk)^2}}{\displaystyle \frac{2I}{(pk)\gamma _c\omega _c}}+`$
$`+{\displaystyle \frac{I^3(1U\delta \beta )}{(\gamma _c\omega _c)^2}}+2({\displaystyle \frac{1}{(pk)}}{\displaystyle \frac{I}{\gamma _c\omega _c}})+`$
$`+({\displaystyle \frac{(pk)I}{\gamma _c\omega _c}}+{\displaystyle \frac{\gamma _c\omega _c(1U\delta \beta )}{(pk)}})\}{\displaystyle \frac{\omega _c}{\mathrm{\Gamma }V_c}},`$ (20)
where
$`(pk)=\gamma \omega (1v\mathrm{cos}\theta ),`$ $`V_c={\displaystyle \frac{\sqrt{\omega ^2+(\gamma v)^2+2\gamma v\omega \mathrm{cos}(\theta )}}{(\gamma +\omega )}}`$ ,
$`\mathrm{\Gamma }={\displaystyle \frac{\gamma +\omega }{\sqrt{1+2(pk)}}},`$ $`\omega _c={\displaystyle \frac{(pk)}{\sqrt{1+2(pk)}}},`$
$`\gamma _c={\displaystyle \frac{1+(pk)}{\sqrt{1+2(pk)}}},`$ $`U={\displaystyle \frac{(pk)}{1+(pk)}},`$
$`\delta ={\displaystyle \frac{1}{V_c}}({\displaystyle \frac{E_\gamma }{\mathrm{\Gamma }\omega _c}}1),`$ $`\beta ={\displaystyle \frac{1}{UV_c}}({\displaystyle \frac{\gamma }{\mathrm{\Gamma }\gamma _c}}1).`$
Function $`I`$ in equation (20) is defined as follows
$$I=(1U^22U(1U)\beta \delta +U^2(\delta \beta )^2)^{1/2}.$$
(21)
Integration of equation (19) is performed along the trajectory of a electron and over the spectra of soft photons. It is assumed that the observer detects photons from a monoenergetic beam of electrons from the wind moving to the observer. We take $`\gamma _{\mathrm{max}}=\frac{3}{2}\dot{E}_{\mathrm{rot}}/\dot{N}`$, with $`\gamma _\mathrm{w}(\alpha )=\gamma _{\mathrm{max}}\mathrm{cos}^2(33^0)`$, since the plane of the pulsar equator is inclined to the observer at the angle $`33^0`$ \[Hester et al. 1995\].
### 3.3 IC photons from thermal isotropic radiation
The optical depth $`\tau `$ , and therefore the spectra of IC photons, depends strongly on the distance $`R_\mathrm{w}`$ at which the kinetic energy dominated wind is formed.
The dependence of the optical depth on $`R_\mathrm{w}`$ is shown in Fig. 3 for two different values of the Lorentz factors of the wind $`\gamma _\mathrm{w}=410^6`$ and $`410^7`$
Strong dependence of $`\tau `$ on $`\gamma _\mathrm{w}`$ is explained by the fact that the optical depth is dominated by IC scattering at small distances from the pulsar where $`2\omega \gamma (1\mathrm{cos}(\theta ))1`$, i.e. the Compton scattering takes place in the Klein-Nishina regime. In this regime the IC cross-section decreases with the electron energy as $`\gamma _w^1`$, which lead to larger optical depth for smaller values of the Lorentz-factor of the wind $`\gamma _w`$. This effect is especially strong for small values of $`R_\mathrm{w}`$. In this case the optical depth is accumulated by IC scattering close to the light cylinder where the collision angle $`\theta `$ is large, and the IC scattering takes place in deep Klein-Nishina regime.
In many “standard” astrophysical situations the efficiency of the $`\gamma `$-ray production in the Klein-Nishina regime is significantly suppressed because of the synchrotron cooling of electrons. In the case of a cold wind we deal with a unique situation when the synchrotron losses of electrons are completely suppressed, and thus the wind electrons lose their energy only through the inverse Compton radiation. Note that in deep Klein-Nishina regime, namely when $`2\gamma _w\omega (1\mathrm{cos}\theta )10^4`$, the triplet pair production dominates over the inverse Compton scattering \[Mastichiadis 1991, Dermer & Schlikeiser 1991\]. This process results in production of new electrons. However, as in all interesting cases the optical depth $`\tau 1`$, the secondary electrons do not increase significantly the density of the wind. Therefore we ignore this process here. The same is true also for another process connected with absorption of TeV $`\gamma `$ -rays in the magnetic field of the wind (see Appendix).
In fig. 4 we present the expected $`\gamma `$-ray fluxes of the wind.
Solid, dashed and dotted lines correspond to the fluxes produced by a wind originated at $`R_\mathrm{w}/R_\mathrm{L}`$=1, 5 and 10. Remarkably, the spectrum of IC $`\gamma `$ -rays has a specific line feature since the radiation is produced in the Klein-Nishina regime. Because of this feature, the IC radiation of the wind can be easily distinguished from the smooth spectra of the Crab Nebula. The corresponding integral fluxes are shown in Fig. 5 for different $`\gamma _w`$ and $`R_w`$.
The comparison of the calculated spectra with the observed TeV $`\gamma `$-ray fluxes of the Crab Nebula (for review see Weekes et al. 1997) leads to an interesting conclusion that for any reasonable Lorentz-factor of the wind , the calculated $`\gamma `$-ray fluxes significantly exceed the observed fluxes unless the wind is formed well beyond the light cylinder. This implies a meaningful constraint on the ‘birthplace’ of the kinetic energy dominated wind with $`\gamma _w10^6`$: $`R_\mathrm{w}5R_\mathrm{L}`$. For $`\gamma 1.210^5`$, a similar conclusion is imposed by the EGRET \[Nolan et al. 1993\] observations of unpulsed radiation above 1 GEV. The lack of measurements in the energy region between 10 GeV and 300 GeV does not completely exclude a possibility of formation of the particle dominated wind close to the light cylinder, if one assumes that $`10^5<\gamma _{max}<10^6`$. Although the analysis of the observed synchrotron and IC components of the nonthermal radiation of the Crab Nebula gives certain preference to larger wind Lorentz factors, $`\gamma _\mathrm{w}10^6`$, it is important to have observational constraints on $`R_\mathrm{w}`$ also for $`\gamma _\mathrm{w}`$ in the region between $`10^5`$ to $`10^6`$, which is presently missing This should be provided in the near future by measurements of the Crab spectrum between 10 and 300 GeV by the new generation of low-threshold atmospheric Cherenkov detectors as well as by the GLAST.
### 3.4 The interaction of the wind with the nonthermal emission
The IC $`\gamma `$ radiation caused by illumination of the wind by pulsed soft photon emission should be modulated at the same pulsation period. This conclusion is true even for the isotropic wind. For calculation of the IC fluxes it is assumed that the nonthermal source of soft emission is located inside the light cylinder (see Fig. 1). It is easy to show that under realistic conditions the phase curve of the gamma-rays should be close to the phase curve of the soft emission. Indeed, let us assume for simplicity that the beam is very narrow (delta-function like). Owing to the rotation of the pulsar, it will form a spiral. The point where this spiral crosses the line tangential to the light cylinder and directed towards the observer is the source of $`\gamma `$-rays (see Fig. 6). The width of the $`\gamma `$-ray pulse is determined by the delay in the arrival time $`\delta t`$ of optical and gamma-ray emission to the observer:
$$\delta t\frac{\theta T}{2\pi },$$
(22)
where $`\theta `$ is the maximal value of the angle between the wave vector of photons and the velocity of the particles in the wind, $`T`$ is the period of rotation. In the most interesting case, $`\theta 1`$, the phase curves of gamma-ray emission produced in the process under consideration and those of the optical emission illuminating the wind are expected to be similar. However, because we observe only the pulsed soft emission directed to us and do not observe other possible part of the emission (not directed towards the observer) that can illuminate the wind, there could be a difference between the $`\gamma `$-ray and directly observed soft emission light curves.
It is seen from Fig. 2 that the average density of nonthermal photons is several orders of magnitude larger than the density of the thermal photons. Therefore, the IC optical depth for an electron is much larger than in the case of thermal radiation (see Fig.7). In particular, at $`R_W5R_L`$ the optical depth $`\tau 1`$. As the IC scattering takes place in the Klein-Nishina regime, one or two interactions are sufficient to destroy the wind, and the whole energy of the wind electrons would be transferred to $`\gamma `$ ray emission with huge luminosity $`10^{38}`$ erg/s. This very fact excludes the possibility of formation of the kinetic dominated wind within $`5R_L`$. Moreover, this conclusion can be extended to larger distances. Fig. 8 shows the spectra of the emission for two maximal energies of the electrons in the wind. Solid lines show the spectra for $`\gamma _{max}=310^6`$ and dashed lines show the spectra for $`\gamma _{max}=310^7`$. In contrast to the IC $`\gamma `$-rays from interaction with thermal emission, there are no lines in the spectra of the emission because the IC scattering takes place in the Thomson regime. The spectra consist of two well separated broad components, corresponding to two components in the spectra of the soft nonthermal emission presented in Fig. 2. The comparison of calculated $`\gamma `$-ray fluxes with upper limits on pulsed emission reported by the Whipple \[Weekes et al. 1998\] and HEGRA \[Aharonian et al. 1999\] groups shows that the wind must be formed at relatively large distances from the pulsar, $`R_W>30R_L`$.
The spectra shown in Fig. 8 were calculated at the assumption that the nonthermal source emits photons along the radial direction from the pulsar as it happens in the outer gap model \[Romani & Yadigaroglu 1995\]. Self-consistent MHD solutions confirm that the plasma should move predominantly radially in the pulsar magnetosphere \[Bogovalov 1997\]. Therefore the source of nonthermal photons should indeed emit along the radial direction from the pulsar. In the model by Lyubarskii \[Lyubarskii 1996\] the source of the soft nonthermal photons emits in the direction opposite to the direction of rotation of the pulsar. If so, the $`\gamma `$ -ray emission would be even higher than shown in Fig. 4.
We are not aware of models in which the source of photons corotates with the pulsar and emits photons tangentially along the direction of rotation. In this case the angle of interaction $`\theta `$ is minimum and correspondently the IC flux is expected to be reduced. Although it is almost unlikely that the plasma (and source of soft photons ) can corotate with the pulsar, it is perhaps wise to consider this limiting case from pedagogical point of view in some detail.
Optical and soft X-ray emission can be produced by plasma moving with a Lorentz-factor of order $`\gamma _e10^210^3`$ in the pulsar magnetosphere. At corotation with the pulsar the plasma have azimuthal velocity $`v_\phi =\frac{r\mathrm{\Omega }}{c}`$, where $`r`$ is the radius in the cylindrical system of coordinates. Since these particles should also be directed towards the observer, they have the component of the velocity $`v_z=c\mathrm{cos}\alpha `$. From the relativistic relationship it follows that
$$\gamma _e^2=1+\gamma _e^2[cos^2\alpha +(\frac{r_s\mathrm{\Omega }}{c})^2+(\frac{v_r}{c})^2],$$
(23)
where $`r_s`$ is the distance from the source to the center of the pulsar, $`v_r`$ is the radial component of the velocity. The soft photons are emitted at the angle $`\psi `$ as shown in Fig. 6. Neglecting the term $`1/\gamma _e^2`$ we obtain that
$$\mathrm{sin}\psi =\frac{r_s\mathrm{\Omega }}{c\mathrm{sin}\alpha }.$$
(24)
This relationship shows that a source that emits soft photons in direction to the observer and corotating with the pulsar can not be located exactly on the light cylinder.
Fig. 9 demonstrates how the spectra of IC radiation depend on the position of the corotating source of the soft photons inside the pulsar magnetosphere for parameters of the kinetic energy dominated wind at $`R_w=40R_L`$. Curve 1 corresponds to the emission from the corotating source located on the surface of the pulsar. The same flux is produced by the source of soft photons with the beam directed radially from the pulsar; the IC flux generated from the last source does not depend on $`r_s`$. Therefore comparison of the curve 1 with others allows one to compare the fluxes of IC photons produced at the scattering of electrons on the soft emission from the corotating source and the source with the soft photons emitted radially, but located on the same distance $`r_s`$. It follows from this figure that the flux of the IC photons can be reduced to zero only at the position of the source of soft photons not far from the light cylinder. At any other position of the source the efficiency of generation remains high at $`r_s<0.6R_L`$. The existing upper limit on the pulsating flux from the Crab pulsar means that either the wind is formed beyond 30$`R_L`$ or the wind is formed close to the light cylinder but the pulsating source of soft nonthermal photons corotates with the pulsar and is located close to the light cylinder.
## 4 Summary
The Crab Nebula is a unique cosmic laboratory with an unprecedentedly broad spectrum of the observed nonthermal radiation that extends throughout 21 decades of frequency – from radio wavelengths to very high energy $`\gamma `$-rays (for review see e.g. Aharonian & Atoyan 1998) It is commonly accepted that the synchrotron nebula is powered by the relativistic wind of electrons generated at the pulsar and terminated by a standing reverse shock wave at a distance $`r_\mathrm{s}0.1\mathrm{pc}`$ \[Rees & Gunn 1974\]. The relativistic MHD models, even in their simplified form (e.g. ignoring the axisymmetric structure of the wind and its interaction with the optical filaments), successfully describe the general characteristics of the synchrotron nebula, and predict realistic distributions of relativistic electrons and magnetic field in the downstream region behind the shock \[Kennel & Coroniti 1984\]. Meanwhile our knowledge of the unshocked wind, i.e. about the region between the pulsar magnetosphere and the shock is based only on theoretical speculations. Moreover, it is commonly thought that the wind could not be visible in the region upstream of the termination shock because the relativistic electrons and magnetic field in wind move together, thus the unshocked wind does not produce synchrotron radiation. In this paper we show that the kinetic energy dominated wind nevertheless could be directly observed through its IC radiation because of the illumination of the wind by low-energy radiation of the pulsar. The $`\gamma `$-ray emission consists of two components, pulsed and unpulsed, associated with the nonthermal and thermal low-energy radiation components of the pulsar, respectively.
The unpulsed component of $`\gamma `$-ray emission associated with thermal radiation of the pulsar with temperature $`210^6\mathrm{K}`$ is produced in deep the Klein-Nishina regime, and therefore has a very sharp (line-like) spectral feature which peaks at energy $`E\gamma _w\mathrm{m}_\mathrm{e}\mathrm{c}^2`$. Detection of this component would therefore result in unique information about the Lorentz-factor of the bulk motion of the wind. The nonthermal radiation of the pulsar has rather broad energy spectrum which extends to optical and infrared wavelengths, and therefore the IC $`\gamma `$-ray emission associated with this component takes place, to a large extent, in the Thomson regime. This results in a broad $`\gamma `$-ray spectrum with a sharp cutoff at $`E\gamma _w\mathrm{m}_\mathrm{e}\mathrm{c}^2`$.
The absolute $`\gamma `$-ray fluxes of both components depend strongly on the site of formation of the kinetic dominated wind, as well as the Lorentz-factor and the geometry of propagation of the wind. Thus even the flux upper limits of these $`\gamma `$-ray components should provide important constraints on the wind parameters. In particular, we show that the comparison of the calculated flux of the unpulsed inverse IC emission with the measured $`\gamma `$-ray flux of the Crab Nebula excludes the possibility of formation of the kinetic-energy-dominated wind within 5 light cylinder radii of the pulsar, $`R_\mathrm{w}5R_\mathrm{L}`$. The analysis of the pulsed IC emission, calculated under reasonable assumptions concerning the production site and angular distribution of the optical pulsed radiation, yields even tighter restrictions, namely $`R_\mathrm{w}30R_\mathrm{L}`$.
The mechanism of $`\gamma `$-radiation of the wind of the Crab pulsar discussed in this paper should certainly take place in other pulsars as well. However, from the point of view of detection of this radiation, the Crab is a unique object due to its very powerful wind and relatively high luminosity of thermal and nonthermal low-energy radiation, which provides seed photons for the IC scattering. In other pulsars the IC $`\gamma `$-ray fluxes of unshocked winds are expected to be below the detection threshold of current $`\gamma `$-ray instruments, unless the kinetic energy dominated winds of pulsars are produced very close to the light cylinder. The situation could be different in binary systems containing a pulsar and luminous optical companion, the latter being an effective supplier of seed photons for IC scattering. For example, the pulsar/Be star binary system PSR 1259-63 seems to be a unique object for the search for IC TeV radiation from both shocked \[Kirk, Ball & Skjæraasen 1999\] and unshocked \[Kirk & Ball 1999\] winds of the pulsar.
Acknowledgments
We thank the Astrophysics group of the MPI für Kernphysik, in particular, H. J. Völk, A.M. Atoyan, J.G. Kirk, as well as L. Ball and Yu. Lyubarskii for many fruitful discussions. SB thanks MPI für Kernphysik for warm hospitality and support during his work on this project.
## Appendix A Opacity of the magnetic field of the wind for the TeV Photons.
In the paper of \[Cheung & Cheng 1994\] it was argued that TeV photons produced near the light cylinder of the Crab pulsar will be absorbed, because of conversion of these photons in $`e^\pm `$ pairs in nonuniform magnetic field. We show here that accurate estimate of the rate of the conversion of photons in pairs taking into account electric field existing in the wind give negligible absorption of TeV gamma-rays produced via IC scattering.
The motion of the wind occurs under frozen in condition. Particles move not only along field lines, but there is component of the velocity $`𝐕_𝐝`$ directed perpendicular to the magnetic field line. This is so called drift velocity defined by the expression
$$𝐕_𝐝=\frac{𝐄\times 𝐁}{B^2}$$
(25)
$`V_d`$ is comparable with c beyond the light cylinder, where the electric and magnetic fields are of the same order. The schematic relationship between the magnetic and electric fields and the velocity of plasma is shown in Fig. 10. In this situation an energetic photon is emitted along the velocity vector of the particle at large angle to the magnetic field. However, this happens in the region of crossed electric and magnetic fields, where the coefficient of absorption of the photon is modified by the electric field. The probability of conversion of the photon on unit length, taking into account the electric field, was defined by Daugherty & Lerche: \[Daugherty & Lerche,1975\]
$$\xi =0.23\alpha \frac{mc}{\mathrm{}}\frac{B\chi }{B_{cr}}\frac{(1E^2/B^2)}{(1E\eta _\mathrm{x}/B)}\mathrm{exp}\left(\frac{8}{3}\frac{B_{\mathrm{cr}}}{E_\gamma \chi B}\right),$$
(26)
where
$$\chi =\sqrt{(\eta _x\frac{E}{B})^2+\eta _y^2(1\frac{E^2}{B^2})},$$
(27)
and $`\alpha =1/137`$. $`\eta _\mathrm{x}`$ and $`\eta _\mathrm{y}`$ are the components of the unit vector directed along the velocity of the photon as it is shown in Fig. 11 and $`E_\gamma `$ is the energy of photon expressed in units of $`mc^2`$.
The components of the velocity of the particle in the system of coordinates presented in Fig. 11 are as follows $`v_x=\frac{cE}{B}`$, $`v_y=0`$, $`v_z=c\sqrt{1\frac{E^2}{B^2}}`$. The corresponding components of the unit vector $`\eta `$ of the photon at the place of the emission will be $`\eta _\mathrm{x}=\frac{v_\mathrm{x}}{v}=\frac{Ec}{vB}`$, $`\eta _y=0`$, $`\eta _\mathrm{z}=\sqrt{1\eta _\mathrm{x}^2}`$. Substitution of these components in equation (27) gives us
$$\chi =\frac{E}{B}\frac{c}{\gamma ^2(1+v/c)v}$$
(28)
This factor is extremely small, $`\chi 10^{14}`$, for the expected parameters of the wind .
Photons propagate along straight lines. Charged particles move in electromagnetic fields and their trajectories diverge from straight lines. If photon would moved along the trajectory of the charged particles, they would never convert in pairs. Therefore it is clear that the probability of conversion of a photon into a pair basically depends on the radius of curvature $`R_c`$ of the trajectory of the charged particles, but not the magnetic field.
The angle $`\vartheta `$ between the direction of propagation of the photon and direction of propagation of the emitting electron depends on the path length of the photon $`l`$;
$$\vartheta =\frac{l}{R_c}.$$
(29)
The dependence of components of the vector $`\eta `$ on $`\theta `$ is as follows
$$\eta _x=\eta _{x0}\mathrm{cos}\vartheta +\eta _{z0}\mathrm{sin}\vartheta \mathrm{sin}\phi $$
(30)
$$\eta _y=\mathrm{sin}\vartheta \mathrm{cos}\phi $$
(31)
Neglecting by terms of the order $`1/\gamma ^2`$ and $`\mathrm{sin}^n\vartheta `$ in powers higher than 2, we obtain the following estimate for $`\chi `$ at small $`\theta `$
$$\chi \frac{l}{R_\mathrm{c}}\sqrt{1\left(\frac{E}{B}\right)^2}.$$
(32)
As we are interested in the maximal values of $`\chi `$, we can neglect the term $`\sqrt{1(\frac{E}{B})^2}`$ in this expression. Assuming that after the light cylinder the total magnetic field decreases as $`1/r`$, we obtain for the function $`q=\frac{8B_{\mathrm{cr}}}{3E_\gamma B\chi }`$ the following estimate of the upper limit
$$q=\frac{8B_{\mathrm{cr}}R_\mathrm{c}}{3E_\gamma B_{\mathrm{lc}}r_\mathrm{L}},$$
(33)
and for the probability of conversion of the photon into a pair we obtain finally the estimate
$$\xi 0.23\alpha \frac{mc}{\mathrm{}}\frac{B_{\mathrm{lc}}r_\mathrm{L}}{B_{\mathrm{cr}}R_\mathrm{c}}\mathrm{exp}(\frac{8B_{\mathrm{cr}}R_\mathrm{c}}{3E_\gamma B_{\mathrm{lc}}r_\mathrm{L}})$$
(34)
where $`B_{\mathrm{lc}}`$ is the magnetic field on the light cylinder and $`R_\mathrm{c}`$ is also taken on the light cylinder.
The solution of the problem of the relativistic wind flows in the model of an axisymmetrically rotating star allows us to estimate the radius of curvature $`R_\mathrm{c}`$ of the trajectories of the particles in the wind \[Bogovalov 1997\] as
$$\frac{1}{R_\mathrm{c}}=\frac{\sigma _\mathrm{w}}{\gamma ^2r}.$$
(35)
This estimate shows that the radius of curvature of the trajectories of particles is much more than the curvature radius of the magnetic field line, which is of the order $`r`$. Such a large curvature radius gives $`q1`$ for almost any parameters of the wind. This is why the IC photons produced in the wind are not converted into pairs even at energies $`100`$ TeV and $`\sigma _\mathrm{w}1`$. |
warning/0003/math0003115.html | ar5iv | text | # NNR Revisited
## 0. Introduction
One of the major problems in the theory of iterated forcing is to prove some preservation theorems. One example is the preservation of cardinals or cofinalities. By Solovay-Tennenbaum , finite support iteration of c.c.c. forcing notions is c.c.c. and hence it preserves all cardinals and cofinalities. In the case of countable support iterations, Shelah proved that the countable support iteration of proper forcing notions is again proper, and hence it preserves $`\mathrm{}_1`$.
In this paper we are interested in “not adding new reals”, abbreviated NNR, in the case of countable support iteration of proper forcing notions. There is a lot of work in this direction, see for example \[6, Ch.V,§7, Ch.VIII,§4, Ch.XVIII,§1,§2\], \[9, §3\] and the references there. In this paper we present a more general preservation theorem, and as an application we prove the consistency with ZFC + GCH of Souslin’s hypothesis with non-club guessing. We also prove the consistency with GCH of further cases of “strong failure of club guessing”, in particular, we answer a question of Moore.
There are several limitations in preserving NNR at limit stages of countable support iterations that we will discuss some of them in Section 2. By results of Shelah, if, e.g. $`𝕍=𝕃`$, then the preservation of NNR fails. Indeed, assuming $`𝕍=𝕃`$, Shelah has built a countable support iteration of length $`\omega ^2`$ of NNR forcing notions of size $`\mathrm{}_1`$ such that the limit adds a new real (see \[6, XVIII. Lemma 1.1\]). Justin Moore , solving a problem from \[9, §3\], proved that the continuum hypothesis implies the negation of the forcing axiom for the class of completely proper forcing notions. Indeed he proved in ZFC + CH that some proper forcing notion not adding reals and satisfying a “strong form of the medicine against weak diamond” has no generic, in fact, is a tree with no branches.
The paper is organized as follows. In Section 1, we present some definitions and results which are needed for the rest of the paper. In Section 2, we discuss some obstacles about preservation of NNR in countable support iterations and suggest some ideas about how to overcome them.
In Section 3, we present sufficient conditions for countable support iteration of proper forcing notions, not to add reals. For this we define “reasonable parameters $`𝔭`$” and we have two main demands. One (clause (c) of Definition 3.9) is a weakening of “$`\alpha `$-proper for every $`\alpha <\omega _1`$”. This time it has the form (on $`_i`$), $`𝔭`$-proper which informally says that: if $`𝔭N,Y\{MN:M\text{ appropriate}\}`$ is $`\alpha `$-large, then for some $`(N,_i)`$-generic condition $`qp,q`$ forces that {MY:M[𝔾
~
i]𝕍=M}conditional-set𝑀𝑌𝑀delimited-[]subscript𝔾
~
subscript𝑖𝕍𝑀\{M\in Y:M[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{Q}}_{i}}]\cap\mathbb{V}=M\} is $`\alpha `$-large (the meaning of $`\alpha `$-large depends on $`𝔭`$). The other main demand (clause (d) of Definition 3.9) is a “weak diamond preventive”.
We then show that $`\alpha `$-properness for $`\alpha <\omega _1`$ is sufficient for the first main demand (in Lemma 3.17(3)). The demand on the games for $`𝔭`$ helps to prove the preservation of $`𝔭`$-properness.
The preservation theorem in Section 3 does not, for standard $`𝔭`$, cover shooting a club $`C\omega _1`$ running away for $`C_\delta \delta =sup(C_\delta ),C_\delta `$ small. For this we will use, in Section 4, $`(𝔭,\alpha ,\beta )`$-proper for enough pairs $`\alpha \beta <\mathrm{}g(𝔭)`$ (so starting from $`\beta `$-large we get $`\alpha `$-large; for many $`\alpha `$ we can choose $`\beta =\alpha `$, but during the inductive proof we pass through cases of $`\alpha <\beta `$). Here we introduce various definitions and basic facts needed.
In Section 5, we present the natural forcing showing $`\kappa =2`$ is interesting (not only $`\kappa =\mathrm{}_0`$) (from \[5, Ch.VIII,§4\]). We show that the natural forcing (see above) for running away from $`C_\delta \delta `$, of small order type (see \[6, Ch.XVIII,§2\]) falls under our framework for delayed properness. We give examples: running away from $`C_{\delta ,0},C_{\delta ,1}:\delta <\omega _1\text{ limit},C_{\delta ,0},C_{\delta ,1}`$ are disjoint closed subsets of $`\delta `$ with no restrictions on their order type so we ask for $`C,CC_{\delta ,0}`$ or $`CC_{\delta ,1}`$ to be bounded in $`\delta `$ and more.
In Section 6, we give a sufficient condition for the limit forcing not to add reals. We here are weakening the demand “$`𝔭`$-proper”, using $`(𝔭,\alpha ,f(\alpha ))`$-proper instead of $`(𝔭,\alpha ,\alpha )`$-proper, what we called delayed properness. The price is that here $`𝔭`$ has length of large cofinality, so essentially we catch our tails on a club of it. Also the results here cover the examples.
In Section 7, we derive some forcing axioms from our preservation theorems and give several examples that fit into our axioms.
Finally in Section 8, we answer a question of Justin Moore, which is related to the failure of weak club guessing at $`\omega _1`$ in the presence of $`\mathrm{CH}`$.
The results and methods in this paper are all due to the second author. The first author’s contribution was to fill in some details and to write the paper.
## 1. Some preliminaries
In this section we present some preliminaries that are needed for the rest of the paper. We assume familiarity with the theory of iterated forcing and countable support iterations. For a forcing notion $``$ and conditions $`p,q,`$ we say $`q`$ is stronger than $`p`$ if $`qp`$.
###### Definition 1.1.
1. $``$ is $`\alpha `$-proper if whenever $`\chi `$ is large enough regular, $`\overline{N}=N_i:i\alpha `$ is an increasing and continuous chain of countable elementary submodels of $`((\chi ),)`$ with $`\alpha ,N_0`$ and $`\overline{N}(i+1)N_{i+1}`$, if $`pN_0`$, then there is $`q,pq`$ such that $`q`$ is $`(N_i,)`$-generic for each $`i\alpha `$.
2. We say $``$ is $`(<\omega _1)`$-proper if $``$ is $`\alpha `$-proper for any $`\alpha <\omega _1`$.
3. We say $``$ is $`(<^+\omega _1)`$-proper if it satisfies clause (1) for any $`\alpha <\omega _1`$ even omitting “$`\alpha N_0`$”.
###### Definition 1.2.
Suppose $``$ is a forcing notion, $`p`$ and $`N`$ is a model with $`N`$. Then
1. $`\mathrm{Gen}(N,)=\{𝔾N:𝔾\text{ is a }N\text{-generic filter over }N\}.`$
2. $`\mathrm{Gen}^+(N,)=\{𝔾\mathrm{Gen}(N,):G`$ has an upper bound in $`\}`$.
3. $`\mathrm{Gen}(N,,p)=\{𝔾\mathrm{Gen}(N,):p𝔾\}`$.
One important notion that is useful in proofs for showing that certain countable support iterations do not add reals is Shelah’s notion of completeness system.
###### Definition 1.3.
(\[6, Ch. V, Definition 5.2\]) A *completeness system* for a forcing notion $``$ is a function $`𝔻`$ such that the following statements hold:
1. For a sufficiently large $`\theta ,`$ the domain of $`𝔻`$ consists of pairs $`(N,p),`$ where $`N(H(\theta ),)`$ is countable, $`N`$ and $`pN,`$
2. For every $`(N,p)\mathrm{dom}(𝔻),`$ $`𝔻(N,p)`$ is a collection of subsets of $`\mathrm{Gen}(N,,p)`$.
###### Definition 1.4.
(\[6, Ch. V, Definition 5.2\]) Suppose $`\kappa `$ is a cardinal. We say $`𝔻`$ is a $`\kappa `$-completeness system for $``$, if it is a completeness system for $``$ and for every $`(N,p)\mathrm{dom}(𝔻),`$ the intersection of fewer than $`1+\kappa `$ elements of $`𝔻(N,p)`$ is nonempty.
###### Definition 1.5.
(\[6, Ch. V, Definition 5.4\]) A completeness system $`𝔻`$ for $``$ is *simple* if there is a second order formula $`\mathrm{\Psi }`$ such that $`𝔻(N,p)=\{𝒢_X:XN\}`$, where
$$𝒢_X=\{𝔾\mathrm{Gen}(N,,p):(N,,N)\mathrm{\Psi }(𝔾,X)\}.$$
###### Definition 1.6.
(\[6, Ch. V, Definition 5.3\]) Suppose $`𝔻`$ is a simple completeness system for $``$. Then $``$ is said to be $`𝔻`$-complete, if for every $`(N,p)\mathrm{dom}(𝔻),`$ $`\mathrm{Gen}^+(N,,p)`$ contains an element of $`𝔻(N,p)`$.
The next theorem of Shelah gives a sufficient condition for a countable support iteration of forcing notions to not add new reals.
###### Theorem 1.7.
(\[6, Ch. VIII, Theorem 4.5\]) A countable support iteration of forcing notions which are $`<\omega _1`$-proper and $`𝔻`$-complete with respect to a simple 2-completeness system does not introduce reals.
###### Definition 1.8.
(\[6, Ch. VII, Definition 1.2\]) The forcing notion $``$ satisfies the $`\kappa `$-e.c.c. ($`\kappa `$-extra chain condition), if there is a binary relation $`R`$ on $``$ such that:
* For any sequence $`p_i:i<\kappa `$ of elements of $`,`$ there are pressing down functions $`f_n:\kappa \kappa `$ (i.e., for all $`\alpha <\kappa ,f_n(\alpha )<1+\alpha `$) for $`n<\omega `$ such that for all $`0<i,j<\kappa `$,
$$\underset{n<\omega }{}(f_n(i)=f_n(j))p_iRp_j.$$
* If $`p_i:i\omega `$ and $`q_i:i\omega `$ are increasing sequences in $``$ and for all $`n<\omega ,p_nRq_n`$, then there is an $`r`$ such that for all $`n<\omega ,rp_n,q_n`$.
###### Definition 1.9.
(\[6, Ch. VIII, Definition 2.1\]) The forcing notion $``$ satisfies the $`\kappa `$-pic ($`\kappa `$-properness isomorphism condition), if the following holds for any large enough regular cardinal $`\lambda `$: Suppose $`i<j<\kappa ,N_i,N_j((\lambda ),,_\lambda )`$ (where $`_\lambda `$ is a well-ordering of $`(\lambda )`$) are countable such that $`\kappa ,N_iN_j`$, $`iN_i,jN_j,N_i\kappa j,N_ii=N_jj,pN_i`$ and $`h:N_iN_j`$ is such that $`hN_iN_j`$ is identity and $`h(i)=j.`$ Then there exists $`q`$ such that:
* $`qp,h(p)`$ and for every maximal antichain $`N_i`$ of $``$, we have that $`N_i`$ is predense above $`q`$ and similarly for $`N_j`$,
* for every $`rN_i`$ and $`q^{}q,`$ there is $`q^{\prime \prime }q^{}`$ such that
$$rq^{\prime \prime }h(r)q^{\prime \prime }.$$
See \[6, Ch.VII, §1\] and \[6, Ch.VIII, §2\] for more information about the above two defined notions.
###### Theorem 1.10.
Assume CH holds.
1. If $``$ is a countable support iteration of length at most $`\omega _2`$ whose iterands are $`<\omega _1`$-proper, $`𝔻`$-complete for some $`\mathrm{}_1`$-completeness system from $`𝕍`$ and satisfy the $`\mathrm{}_2`$-e.c.c, then $``$ satisfies the $`\mathrm{}_2`$-c.c. The same result holds if we replace “$`\mathrm{}_1`$-completeness system” by “$`\mathrm{}_0`$-completeness system” or by “2-completeness system”.
2. If $``$ is a countable support iteration of length at most $`\omega _2`$ whose iterands satisfy the $`\mathrm{}_2`$-pic, then $``$ satisfies the $`\mathrm{}_2`$-c.c.
###### Proof.
(1). For the case of $`\mathrm{}_1`$-completeness system see \[6, Ch.VII, Lemmas 1.3\]. The case of $`\mathrm{}_0`$-completeness system follows from \[6, Ch.VII, Lemmas 1.6\] and the case of 2-completeness system follows from the above resuls combined with \[6, Ch.VIII, Theorem 4.5 and Lemma 4.13\].
(2). See \[6, Ch.VIII, Lemma 2.4\]. ∎
## 2. Obstacles for NNR preservation
In this section we give lengthy explanation of the problems and proofs for NNR countable support iterations of proper forcing notions, and suggest some ideas about how to overcome them. These ideas will be made precise in the later sections of the paper.
###### Definition 2.1.
1. Let $`K_0`$ be the family of countable support iterations ¯=i,
~
i:i<α\bar{{\mathbb{Q}}}=\langle{\mathbb{P}}_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}:i<\alpha\rangle. We denote $`_\alpha =\mathrm{Lim}(\overline{})`$.
2. We say $`\overline{}K_0`$ is proper if for each i<α,i
Q
~
ii<\alpha,~{}\Vdash_{{\mathbb{P}}_{i}}\text{``}\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} is proper”. Note that it follows that j/G
~
isubscript𝑗subscript𝐺
~
subscript𝑖{\mathbb{P}}_{j}/\mathchoice{\oalign{$\displaystyle G$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle G$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle G$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle G$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}} is proper for $`i<j\alpha `$ (see or ).
3. We say $`\overline{}K_0`$ is $`{}_{}{}^{\omega }\omega `$-bounding if for each i<α,i
Q
~
ii<\alpha,~{}\Vdash_{{\mathbb{P}}_{i}}\text{``}\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} is $`{}_{}{}^{\omega }\omega `$-bounding” <sup>1</sup><sup>1</sup>1The forcing notion $``$ is $`{}_{}{}^{\omega }\omega `$-bounding (in the universe $`𝕍`$) if every $`f({}_{}{}^{\omega }\omega )^𝕍^{}`$ is bounded by some $`g({}_{}{}^{\omega }\omega )^𝕍`$.. It again follows that $`_j/\dot{G}__i`$ is $`{}_{}{}^{\omega }\omega `$-bounding for $`i<j\alpha `$.
4. We say $`\overline{}`$ is NNR if for each $`i<\alpha ,`$ the forcing notion $`_{i+1}`$ adds no reals, or equivalently if for i<α,i
Q
~
ii<\alpha,\Vdash_{{\mathbb{P}}_{i}}\text{``}\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} adds no reals” and for each $`\beta <\alpha ,_\beta `$ adds not reals.
It would be nice if also NNR is preserved in limit stages of the iteration. But this is wrong for at least two known reasons, explained below:
1. weak diamond
2. existence of clubs.
Let us explain these obstacles in more details and the way to avoid them.
Weak diamond:
Let us first explain the obstacle arising from weak diamond. Given a stationary set $`S\omega _1,`$ recall that the weak diamond $`\mathrm{\Phi }_S`$ says: for each function $`F:`$$`{}_{}{}^{<\omega _1}22`$, there exists $`g:\omega _12`$ such that for each $`f:\omega _12`$, the set
$$\{\delta S:g(\delta )=F(f\delta )\}$$
is stationary. By Devlin-Shelah (see also \[5, Ch.XII,§1\] or \[6, AP,§1\]), $`\mathrm{\Phi }_{\omega _1}`$ is equivalent to $`2^\mathrm{}_0<2^\mathrm{}_1`$.
Now let $`\overline{\eta }=\eta _\delta :\delta <\omega _1,\delta \text{ limit}`$ be a ladder system, where $`\eta _\delta =\eta _\delta (n):n<\omega `$ is an increasing $`\omega `$-sequence of ordinals cofinal in $`\delta `$. Let $`D`$ be a non-principal ultrafilter on $`\omega `$. For $`f{}_{}{}^{\omega _1}2`$ and a limit ordinal $`\delta <\omega _1`$ let
$$\text{Av}_D(f,\eta _\delta )=\mathrm{}\{n:f(\eta _\delta (n))=\mathrm{}\}D.$$
Consider the following natural question:
###### Question 2.2.
(CH) Given $`\overline{e}=e_\delta :\delta <\omega _1,\delta \text{ limit},e_\delta \{0,1\}`$, is there $`f{}_{}{}^{\omega _1}2`$ such that for a club of $`\delta <\omega _1`$ we have $`e_\delta =\mathrm{Av}_D(f,\eta _\delta )`$?
Naturally, trying to prove the consistency of this statement, we should use a countable support iteration ¯=i,
~
i:i<ω2\bar{{\mathbb{Q}}}=\langle{\mathbb{P}}_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}:i<\omega_{2}\rangle, where for each $`i<\omega _2`$, for some $`\overline{e}V^_i`$ as in Question 2.2, $`_i`$ is defined in $`V^_i`$ as $`_i=_{\overline{e}}`$, where
$$\begin{array}{c}\hfill _{\overline{e}}=\{f:\text{ for some }\zeta <\omega _1,f{}_{}{}^{\zeta }2\text{ and for every limit ordinal}\\ \hfill \delta \zeta \text{ we have }\mathrm{Av}_D(f,\eta _\delta )=e_\delta \}.\end{array}$$
This is a very nice forcing notion, it is proper (even $`<\omega _1`$-proper, see below) and NNR. For example, let us show that $`_{\overline{e}}`$ is NNR.
Thus let $`p_{\overline{e}}`$, τ
~
Ve¯𝜏
~
superscript𝑉subscript¯𝑒\mathchoice{\oalign{$\displaystyle\tau$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\in V^{{\mathbb{Q}}_{\bar{e}}} and p
τ
~
:ωOrd:forces𝑝
τ
~
𝜔Ordp\Vdash\text{``}\mathchoice{\oalign{$\displaystyle\tau$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}:\omega\rightarrow{\rm Ord} is a function”. Let $`\chi `$ be a large enough regular cardinal and let $`\overline{N}=N_i:i\omega ^2`$ be an increasing and continuous chain of countable elementary submodels of $`((\chi ),)`$ with e¯,e¯,p,τ
~
N0¯𝑒subscript¯𝑒𝑝𝜏
~
subscript𝑁0{\bar{e}},{\mathbb{Q}}_{\bar{e}},p,\mathchoice{\oalign{$\displaystyle\tau$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\in N_{0} and $`\overline{N}(i+1)N_{i+1}`$. Let $`\delta (i)=N_i\omega _1`$. So $`\delta (i):i\omega ^2`$ is a strictly increasing and continuous sequence of countable ordinals. Since $`\eta _{\delta (\omega ^2)}`$ has order type $`\omega `$,
$$W=\{i<\omega ^2:n(\delta _i\eta _{\delta (\omega ^2)}(n)<\delta _{i+1})\}$$
has order type $`\omega `$ as well. So, for each $`n<\omega ,`$ we can find $`\mathrm{}_n<\omega `$ such that
$$\underset{n<\omega }{}\underset{m<\omega }{}\omega n+\mathrm{}_n+mW.$$
We choose, by induction on $`n<\omega ,p_n_{\overline{e}}`$ and $`a_n\mathrm{Ord}`$ such that:
* $`pp_n,`$
* $`p_{n1}p_n,`$
* $`p_nN_{\omega n+\mathrm{}_n+1},`$
* $`p_n`$ forces some value for τ
~
(n)𝜏
~
𝑛\mathchoice{\oalign{$\displaystyle\tau$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(n), say pnτ
~
(n)=aˇnforcessubscript𝑝𝑛𝜏
~
𝑛subscriptˇ𝑎𝑛p_{n}\Vdash\mathchoice{\oalign{$\displaystyle\tau$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(n)=\check{a}_{n}
* On $`[\delta _{\omega m+\mathrm{}_m},\delta _{\omega (m+1)+\mathrm{}_{m+1}})\mathrm{Rang}(\eta _\delta )\backslash \mathrm{dom}(p)`$, $`p_n`$ agrees with $`e_{\delta (\omega ^2)}`$.
This is easily seen to be possible by the choice of $`\mathrm{}_n`$’s. Then $`q=\underset{n<\omega }{}p_n`$ is a condition in $`_{\overline{e}}`$ and it forces τ
~
=aˇn:n<ωVˇ.\mathchoice{\oalign{$\displaystyle\tau$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}=\langle\check{a}_{n}:n<\omega\rangle\in\check{V}.
Also note that for every $`\alpha <\omega _1,_\alpha =\{f_{\overline{e}}:\alpha \mathrm{dom}(f)\}`$ is a dense open subset of $`_{\overline{e}}`$, and hence if $`𝔾`$ is $`_{\overline{e}}`$-generic over $`V`$, then $`f=\underset{f𝔾}{}f:\omega _12`$ is as requested in Question 2.2, for $`\overline{e}`$. But clearly the weak diamond tells us for this case that the answer is no, that is:
$$\overline{e}f{}_{}{}^{\omega _1}2^{\mathrm{stat}}\delta (e_\delta \mathrm{Av}_D(f,\eta _\delta )).$$
In fact this holds for any function $`\mathrm{Av}^{}:\underset{\delta <\omega _1}{}{}_{}{}^{\delta }2\{0,1\}`$. So if $`\overline{}`$ is going to preserve NNR, the desired demand on $`_i`$’s should exclude the $`_{\overline{e}}`$’s. We now explain a way to overcome the above difficulty. Let us first give a definition.
###### Definition 2.3.
1. Let $`K_1`$ be the class of proper $`{}_{}{}^{\omega }\omega `$-bounding iterations $`\overline{}K_0`$.
2. Let $`K_2`$ be the class of NNR iterations $`\overline{}K_1`$.
3. Let $`K_3`$ be the class of $`\overline{}K_2`$ such that if
1. $`\chi `$ is a large enough regular cardinal,
2. $`N((\chi ),)`$ is countable,
3. $`\overline{}N`$,
4. $`i\mathrm{}g(\overline{})N`$,
5. $`p_{i+1}N`$,
6. $`q_0,q_1_i`$ are $`(N,_i)`$-generic (i.e. q``N[𝔾
~
i]𝕍=N"forcessubscript𝑞``𝑁delimited-[]subscript𝔾
~
subscript𝑖𝕍𝑁"q_{\ell}\Vdash``N[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}]\cap\mathbb{V}=N"),
7. q``𝔾
~
iN=𝔾"forcessubscript𝑞``subscript𝔾
~
subscript𝑖𝑁superscript𝔾"q_{\ell}\Vdash``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}\cap N=\mathbb{G}^{*}",
8. $`piq_{\mathrm{}}`$,
then we can find $`q_0^{},q_1^{}`$ and $`𝔾^{}`$ such that for $`\mathrm{}=1,2`$ we have
1. $`q_{\mathrm{}}q_{\mathrm{}}^{}_{i+1}`$,
2. $`pq_{\mathrm{}}^{}`$,
3. q``𝔾
~
i+1N=𝔾"forcessubscriptsuperscript𝑞``subscript𝔾
~
subscript𝑖1𝑁superscript𝔾absent"q^{\prime}_{\ell}\Vdash``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i+1}}\cap N=\mathbb{G}^{**}",
4. $`q_{\mathrm{}}^{}`$ is $`(N,_{i+1})`$-generic, so $`𝔾^{}_{i+1}N`$ is generic over $`N`$.
Clause (3) of the above definition tries to say the following. We know 𝔾
~
iNsubscript𝔾
~
subscript𝑖𝑁\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}\cap N (as being $`𝔾^{}`$) and we are looking at $`N[𝔾^{}]`$ (formally, only its isomorphism type). So we know
~
iN[𝔾]subscriptsuperscript
~
𝑁𝑖delimited-[]superscript𝔾\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{N}_{i}[\mathbb{G}^{*}]. We would like to find 𝔾
~
iN[𝔾]superscript𝔾subscriptsuperscript
~
𝑁𝑖delimited-[]superscript𝔾\mathbb{G}^{\prime}\subseteq\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{N}_{i}[\mathbb{G}^{*}] generic over $`N[𝔾^{}]`$, so that $`𝔾^{},𝔾^{}`$ will determine $`𝔾^{}`$. But we need a guarantee that $`𝔾^{}`$ will have an upper bound in
~
i[𝔾i]subscript
~
𝑖delimited-[]subscript𝔾subscript𝑖\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}[\mathbb{G}_{{\mathbb{P}}_{i}}]. If we know $`𝔾__i`$, this is fine; but in a sense, we are given 2 candidates by $`q_0,q_1`$ and can increase them to $`q_0^{}i,q_1^{}i`$, and have to find $`𝔾^{}`$ “accepted” by both.
The weak diamond obstacle was overcame in \[6, Ch.V,§7\] using $`\mathrm{}_1`$-completeness systems and in \[6, Ch.XVIII,§4\] using 2-completeness systems. Here we show that being in $`K_3`$ is sufficient to overcome this difficulty. Indeed, we will assume something like the following. Many times in some sense $`q_0,q_1_i`$ are $`(N,_i)`$-generic, p
~
iN,qi``𝔾
~
iN=𝔾"formulae-sequence𝑝subscript
~
𝑖𝑁subscriptforcessubscript𝑖subscript𝑞``subscript𝔾
~
subscript𝑖𝑁superscript𝔾"p\in\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}\cap N,q_{\ell}\Vdash_{{\mathbb{P}}_{i}}``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}\cap N=\mathbb{G}^{*}" and for some $`𝔾^{},q_0^{}q_0,q_1^{}q_1`$ in $`_{i+1}`$ we have 𝔾(
~
iN)[𝔾]superscript𝔾subscript
~
𝑖𝑁delimited-[]superscript𝔾\mathbb{G}^{\prime}\subseteq(\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}\cap N)[\mathbb{G}^{*}] and qi``𝔾
~
iN[𝔾]=𝔾"subscriptforcessubscript𝑖subscriptsuperscript𝑞``subscript𝔾
~
subscript𝑖𝑁delimited-[]superscript𝔾superscript𝔾"q^{\prime}_{\ell}\Vdash_{{\mathbb{P}}_{i}}``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{Q}}_{i}}\cap N[\mathbb{G}^{*}]=\mathbb{G}^{\prime}" and $`p𝔾^{}`$.
Unfortunately, this is not sufficient to overcome with the other obstacle $`_2`$. There is an example where for some incomparable $`q_0`$ and $`q_1`$ in i,E
~
isubscript𝑖subscript𝐸
~
𝑖{\mathbb{Q}}_{i},\mathchoice{\oalign{$\displaystyle E$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle E$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle E$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle E$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} a $`_i`$-name of a club and for some $`\alpha (q_0,q_1)`$ we have:
$$q_{\mathrm{}}q_{\mathrm{}}^{}$$
q``E
~
iδ=EiδEiδ0E1δ1\α(q0,q1) is finite".forcessubscriptsuperscript𝑞``subscript𝐸
~
𝑖𝛿subscriptsuperscript𝐸𝛿𝑖subscriptsuperscript𝐸subscript𝛿0𝑖\subscriptsuperscript𝐸subscript𝛿11𝛼subscript𝑞0subscript𝑞1 is finite"q^{\prime}_{\ell}\Vdash``\mathchoice{\oalign{$\displaystyle E$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle E$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle E$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle E$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}\cap\delta=E^{\delta}_{i}\Rightarrow E^{\delta_{0}}_{i}\cap E^{\delta_{1}}_{1}\backslash\alpha(q_{0},q_{1})\text{ is finite}".
This leads us to suggest another idea to overcome the obstacle which arises from the existence of clubs.
Existence of clubs:
Let us now explain the obstacle that arises from working with clubs. This problem was already overcame either by using $`(<\omega _1)`$-properness or by a kind of “finite powers are proper”.
As is shown in , , $`(<\omega _1)`$-properness is an antidote to such problems, i.e. against $`_2`$. Though this is fine for many applications, like specializing an Aronszajn tree and many others, but this requirement is too strong (see ). For example consider the following question.
###### Question 2.4.
Let $`\overline{C}=C_\delta :\delta <\omega _1,\delta \text{ limit}`$ where $`C_\delta \delta =sup(C_\delta ),\mathrm{otp}(C_\delta )=\omega `$ or at least $`<\delta `$. Is there a club $`E`$ of $`\omega _1`$ such that for each limit ordinal $`\delta <\omega _1,`$ we have $`\delta >sup(C_\delta E)`$ (i.e. is this consistent with CH)?
The natural forcing notion for adding such a club is given by
$$\begin{array}{c}\hfill _{\overline{C}}^1=\{f:\text{ for some non-limit }\alpha <\omega _1\text{ we have }f{}_{}{}^{\alpha }2,f^1(\{1\})\\ \hfill \text{is closed and }\delta <\alpha \text{ limit }sup(f^1(\{1\})C_\delta )<\delta \}.\end{array}$$
This forcing notion is not even $`\omega `$-proper, for if $`N_i:i\omega `$ satisfies $`C_{N_\omega \omega _1}=\{N_i\omega _1:i<\omega \}`$, then no $`f_{\overline{c}}^1`$ is $`(N_i,_{\overline{C}}^1)`$-generic, for infinitely many $`i`$’s.
A solution to this problem was suggested in \[6, Ch.XVIII,§2\]) by demanding that each $`_i\times _i`$ is proper for $`i<\mathrm{}g(\overline{})`$. While this is fine for $`_{\overline{C}}^1`$, this seems to exclude specializing an Aronszajn tree without adding reals.
In the proofs, we usually arrive to a situation as follow:
* $`\overline{}N_0N`$,
* $`N_0((\chi ),)`$ and $`N((\chi ),)`$ are countable,
* $`q_{\mathrm{}}`$ is $`(N,_i)`$-generic and $`(N_0,_i)`$-generic (for $`\mathrm{}<2`$),
* $`q_{\mathrm{}}`$ forces that 𝔾
~
iN=𝔾subscript𝔾
~
subscript𝑖𝑁subscript𝔾\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}\cap N=\mathbb{G}_{\ell} (for $`\mathrm{}<2`$),
* $`𝔾^{}=𝔾_1N_0=𝔾_2N_0`$,
* $`i,j,pN_0[𝔾^{}],ij\mathrm{}g(\overline{})`$,
* $`pP_j,pi𝔾^{}`$ (and possibly more).
We would like to find $`𝔾^{}_j^N/𝔾^{}`$ generic over $`N_0`$ such that $`q_0`$ and $`q_1`$ both force that it has an upper bound in Pj/𝔾
~
isubscript𝑃𝑗subscript𝔾
~
subscript𝑖P_{j}/\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}. If $`j=i+1`$ this means 𝔾
~
i[𝔾]superscript𝔾subscript
~
𝑖delimited-[]superscript𝔾\mathbb{G}^{\prime}\subseteq\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}[\mathbb{G}^{*}] is generic over $`N_0`$ such that $`q_0,q_1`$ both force that $`𝔾^{}`$ has an upper bound in
~
i[𝔾i]subscript
~
𝑖delimited-[]subscript𝔾subscript𝑖\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}[\mathbb{G}_{{\mathbb{P}}_{i}}].
It is natural to demand $`𝔾^{}N`$, as otherwise the two possible generic extensions (for $`q_0`$ and $`q_1`$) become unrelated. For the case $`j=i+1`$, the medicine against $`_1`$ should help us. But we need it for every $`j`$. Naturally we prove it by induction on $`j`$, and the successor case can be reduced to the case $`j=i+1`$.
But to continue to a limit case, we need $`𝔾^{}N`$ and more: for some intermediate $`N_1`$ with $`N_0N_1N`$, we also need [qN1[𝔾
~
i]𝕍=N1]subscriptdelimited-[]forcessubscript𝑞subscript𝑁1delimited-[]subscript𝔾
~
subscript𝑖𝕍subscript𝑁1\bigwedge\limits_{\ell}[q_{\ell}\Vdash N_{1}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}]\cap\mathbb{V}=N_{1}]. So the clubs of elementary submodels which $`q_0,q_1`$ induce on $`\{MN:MN\}`$ should have non-trivial intersection. This is a major point and it has always appeared in some form. Here the medicine against $`_2`$ should help, in some way there will be many possible $`N_1`$’s; but its help has a price, that is we have to carry it during the induction. On the other hand the models playing the role of $`N_1`$ may change, we may “consume it and discard it”.
Note that the discussion is on two levels. Necessary limitations of universes with $`\mathrm{CH}`$ on the one hand, and how we try to carry the inductive proof on appropriate iterations on the other hand; the connection though is quite tight.
So we shall try for $`j\mathrm{}g(\overline{})N_0`$ to extend the situation with $`i`$ being replaced by $`j`$ while $`𝔾^{}`$ is being increased to $`𝔾^{}`$. We shall prove by induction some suitable facts, with $`𝔾^{}`$ the object we are really interested in. We are given $`q_1,q_2_i`$ and would like to find suitable $`q_1^{},q_2^{}_j`$ such that $`q_{\mathrm{}}^{}i=q_{\mathrm{}}`$. This last requirement helps us in limit steps to find an upper bound.
So the real action occurs for $`j`$ limit, hence we choose $`\zeta _nN[i,j)`$ such that $`\zeta _0=i,\zeta _n<\zeta _{n+1}`$ (sometimes better to have $`i`$ and each $`\zeta _n`$ non-limit) and $`\underset{n<\omega }{}\zeta _n=sup(jN)`$. You can think of:
1. in each case of limit $`j`$, proving the inductive statement, we choose a
2. “surrogate” for $`N`$ called $`N_1`$, during the induction it serves like $`N`$, in
3. the limit dealing with $`\zeta _0,\zeta _1,\mathrm{}`$ using the induction hypothesis on $`N_1`$
4. we get $`𝔾^{}`$ which may not be in $`N_1`$ but is in $`N`$.
So we try to choose by induction on $`n,`$ the conditions $`q_{0,n},q_{1,n}`$ and $`𝔾_n^{}`$ such that:
* $`q_{\mathrm{},n}_{\zeta _n}`$ is $`(N,_{\zeta _n})`$-generic,
* $`q_{\mathrm{},0}=q_{\mathrm{}},`$
* $`q_{\mathrm{},n+1}\zeta _n=q_{\mathrm{},n},`$
* $`𝔾_n^{}N_1,`$
* $`𝔾_n^{}P_{\zeta _n}N`$ is generic over $`N`$, and
* q,n``𝔾
~
ζnN=𝔾nforcessubscript𝑞𝑛``subscript𝔾
~
subscriptsubscript𝜁𝑛𝑁subscriptsuperscript𝔾𝑛q_{\ell,n}\Vdash``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{\zeta_{n}}}\cap N=\mathbb{G}^{*}_{n}.
The construction of the $`𝔾_n^{}`$ should use little information on the actual $`q_{\mathrm{},n}`$ so that the choices of the $`𝔾_n^{}`$ can be carried say inside $`N_1`$ so that $`𝔾_n^{}:n<\omega N`$. In fact several models will play a role like $`N_1`$.
By the proof of the preservation of $`{}_{}{}^{\omega }\omega `$-bounding we can choose some $`N_1`$ and demand “$`q_{\mathrm{},n}`$ gives to each $`_{\zeta _n}`$-name of an ordinal τ
~
nN1subscript𝜏
~
𝑛subscript𝑁1\mathchoice{\oalign{$\displaystyle\tau$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\tau$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n}\in N_{1}, only finitely many possibilities”.
Let us now explain how $`(<\omega _1)`$-properness or remaining proper under products can help in such arguments. If the forcings are $`(<\omega _1)`$-proper , then we can assume in the beginning that $`N_{1,\gamma }:\gamma AN`$ is an increasing and continuous chain of countable elementary submodels of some $`(H(\chi ),)`$, $`N_0N_{1,\gamma }N,N_{1,\gamma }:\gamma \beta N_{\beta +1}`$ with $`A=(j+1)N\backslash i`$ and assume $`q_{\mathrm{}}`$ is $`(N_{1,\gamma },_i)`$-generic for $`\gamma A`$ (similarly for $`q_0^{},q_1^{},j`$ in the conclusion) and demand $`q_{\mathrm{},n}`$ is $`(N_{1,\gamma },_{\zeta _n})`$-generic for $`n<\omega `$ and $`\gamma A\backslash \zeta _n`$.
If components of the iteration remain proper under products, then we demand things like “$`(q_0,q_1)`$ is $`(N_1,_i\times _i)`$-generic” so this gives many common $`N_1`$’s, but to preserve this we need more complicated situations. Instead of a “tower” of models of countable length, we have a finite tower of models where on the bottom we are computing $`𝔾^{}_{\zeta _n}`$ and as we go up, less and less is demanded.
In this paper, we will deal with a condition which follows from both “$`(<\omega _1)`$-properness” and (essentially) “the square of the forcing notion is proper”. We call this $`𝔭`$-properness where “$``$ is $`𝔭`$-proper” says that if $`Y`$ is a large family of $`MN`$ and if $`pN`$ and $`N`$, then for some $`q`$ we have $`qp`$ is $`(N,)`$-generic and q``{MY:M[𝔾
~
]𝕍=M}forces𝑞``conditional-set𝑀𝑌𝑀delimited-[]subscript𝔾
~
𝕍𝑀q\Vdash``\{M\in Y:M[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{Q}}}]\cap\mathbb{V}=M\} is large”.
## 3. Preservation of not adding reals
In this section we define the notion of $`𝔭`$-properness, for a reasonable parameter $`𝔭`$, and prove some preservation theorems.
###### Definition 3.1.
1. A pseudo-filter on a set $`N`$ is a family $`D`$ of subsets of $`N`$ which is closed under supersets. If $`D`$ is a pseudo-filter on $`N`$, then we set $`D^{}=𝒫(N)D`$.
2. If $`D`$ is a filter on $`N`$, then set $`D^+=\{XN:NXD\}`$.
###### Definition 3.2.
We say $`𝔭=(\overline{\chi },\overline{R},\overline{},\overline{D})=(\overline{\chi }^𝔭,\overline{R}^𝔭,\overline{}^𝔭,\overline{D}^𝔭)`$ is a reasonable parameter, when for some ordinal $`\alpha ^{}`$, denoted $`\mathrm{}g(𝔭)`$, we have:
1. $`\overline{\chi }=\chi _\alpha :\alpha <\alpha ^{},`$ where $`\chi _\alpha `$ is a regular cardinal and $`((\underset{\beta <\alpha }{}\chi _\beta )^+)(\chi _\alpha )`$.
2. $`\overline{R}=R_\alpha :\alpha <\alpha ^{},`$ where $`R_\alpha (\chi _\alpha )`$.
3. $`\overline{}=_\alpha :\alpha <\alpha ^{}`$, where $`_\alpha [(\chi _\alpha )]^\mathrm{}_0`$ is stationary.
4. $`\overline{D}=D_\alpha :\alpha <\alpha ^{},`$ where $`D_\alpha `$ is a function with domain $`_\alpha ,`$ and for $`a_\alpha ,D_\alpha (a)`$ is a pseudo-filter on $`a`$.
5. for $`\alpha <\alpha ^{}`$ set $`𝔭^{[\alpha ]}=:\overline{\chi }\alpha ,\overline{R}(\alpha +1),\overline{}\alpha ,\overline{D}\alpha `$, so it belongs to $`(\chi _\alpha )`$.
6. if $`a_\alpha `$, then for some countable $`N((\chi _\alpha ),)`$, $`a`$ is the universe of $`N`$, so we may write $`D_\alpha (N)`$ instead of $`D_\alpha (a)`$ and $`N_\alpha `$ instead of $`|N|_\alpha `$.
7. if $`\alpha <\alpha ^{}`$ and $`N_\alpha `$, then $`𝔭^{[\alpha ]}N`$, so $`\alpha N`$.
8. for $`N_\alpha `$ and $`XN`$ we have:
$`XD_\alpha (N)(\underset{\beta <\alpha }{}_\beta )XD_\alpha (N).`$
9. if $`N_\alpha ,XD_\alpha (N),\beta \alpha N`$ and $`yN(\chi _\beta )`$, then for some $`M_\beta X`$ we have $`XMD_\beta (M)`$ and $`yM`$
Let us explain a little about the intended meaning of the above definition. The requirement (a) is just technical. About $`R_\alpha `$, we could require $`R_\alpha `$ is a relation on $`(\chi _\alpha )`$, in a sense it codes a club of $`[(\chi _\alpha )]^\mathrm{}_0`$. In clause (e), we considered $`\overline{R}(\alpha +1)`$ and not $`\overline{R}\alpha `$. This makes it an easy demand on $`_\alpha ,`$ i.e., if $`N_\alpha `$, then $`R_\alpha N`$. Clause (h) says that each $`D_\alpha (N)`$ has concentrated on $`\underset{\beta <\alpha }{}_\beta `$, and the last clause (i) is some kind of density, as it implies that $`N(\chi _\beta )\{M:M_\beta X\}`$.
###### Remark 3.3.
1. Note that $`_\alpha :\alpha <\mathrm{}g(𝔭)`$ are pairwise disjoint by items (g) and (e), so $`D(N)`$ can be well defined as $`D_\alpha (N)`$ for the unique $`\alpha `$ such that $`N_\alpha `$.
2. Clearly, by clause (h), only $`D_\alpha (N)𝒫(\underset{\beta <\alpha }{}_\beta )`$ matters.
###### Remark 3.4.
Some natural choices for $`D(N)`$ are as follows:
1. $`D(N)`$ is a filter on $`N`$.
2. $`D(N)=\{XN:X\mathrm{}modF\}`$ for some filter $`F`$ on $`N`$.
3. $`D(N)=F^+`$ for a filter $`F`$ on $`N`$.
###### Definition 3.5.
Suppose $`𝔭=(\overline{\chi },\overline{R},\overline{},\overline{D})`$ is a reasonable parameter as in Definition 3.2.
1. We say $`\overline{D}`$ is standard, if for every $`\alpha <\alpha ^{}(=\mathrm{}g(𝔭))`$ and $`N_\alpha `$ we have
$$\begin{array}{cccc}D_\alpha (N)=\{XN:& \text{ for every }\gamma N\alpha \text{ and}\hfill & & \\ & yN\{(\chi _\beta ):\beta N\alpha \},\hfill & & \\ & \text{for some }\beta N(\alpha \gamma )\text{ and}\hfill & & \\ & MX_\beta ,\text{ we have }yM\hfill & & \\ & \text{and }XMD_\beta (M)\}.\hfill & & \end{array}$$
2. We say $`𝔭`$ is standard if $`\overline{D}`$ is standard.
3. We define the partial order $`_𝔭`$ on $`\alpha ^{}=\mathrm{}g(𝔭)`$ as follows: $`\alpha _𝔭\beta `$ iff
1. $`\alpha \beta ,`$
2. $`N_\beta \alpha NN(\chi _\alpha )_\alpha `$,
3. $`N_\beta \alpha NYD_\beta (N)Y\underset{\gamma <\alpha }{}_\gamma D_\alpha (M)`$, where $`M=N(\chi _\alpha )`$.
4. We say $`𝔭`$ is simple if $`\alpha \beta <\alpha ^{}\alpha _𝔭\beta `$.
5. If $`N((\chi ),)`$ and $`N(\chi _\alpha )_\alpha `$, (hence $`\alpha ,𝔭\alpha ,R_\alpha N`$), then we let $`D_\alpha (N)=D_\alpha ^𝔭(N)`$ to be $`D_\alpha (N(\chi _\alpha ))`$.
When $`𝔭`$ is standard, we may drop $`\overline{D}^𝔭`$ and just write $`𝔭=(\overline{\chi }^𝔭,\overline{R}^𝔭,\overline{}^𝔭)`$. Also if $`𝔭`$ is clear from the context, we may remove the superscript $`𝔭`$. We now define several games related to a reasonable parameter $`𝔭`$.
###### Definition 3.6.
Suppose $`𝔭`$ is a reasonable parameter.
1. For $`0<\alpha <\mathrm{}g(𝔭)`$ and $`N_\alpha ^𝔭`$, the game $`\mathrm{}_\alpha (N,𝔭)`$ is defined as follows. The play lasts $`\omega `$ moves, in the $`n`$-th move:
1. the challenger chooses $`X_nD_\alpha (N)`$ such that $`m<nX_nX_m`$
2. the chooser chooses $`M_nX_n`$ and $`Y_nM_nX_n`$ satisfying $`Y_nD(M_n)N`$
3. the challenger chooses $`Z_nY_n`$ such that $`Z_nD(M_n)`$.
At the end, the chooser wins if $`\{\{M_n\}Z_n:n<\omega \}D_\alpha (N)`$.
2. Assume $`NN^{}((\chi ),)`$, $`𝔭\alpha N^{}`$ and $`NN^{}`$ are countable. The game $`\mathrm{}_\alpha ^{}(N,N^{},𝔭)`$ is defined similar to $`\mathrm{}_\alpha (N,𝔭)`$, but during the $`n`$-th move, we demand that all the chosen objects belong to $`N^{}`$ (this means only then $`X_nN^{}`$), and at the end of the $`n`$-th move, the chooser also chooses $`X_n^{}X_n,X_n^{}D_\alpha (N)N^{}`$ and the challenger in the next move has to satisfy $`X_{n+1}X_n^{}`$.
3. Omitting $`N^{}`$, i.e., writing $`\mathrm{}_\alpha ^{}(N,𝔭)`$ we mean: for any $`N^{}`$ as in (2), the demand $`\mathrm{}_\alpha ^{}(N,N^{},𝔭)`$ holds.
4. We say that $`𝔭`$ is a winner or a $`\mathrm{}`$-winner (resp. $`\mathrm{}^{}`$-winner), if for every $`0<\alpha <\mathrm{}g(𝔭)`$ and $`N_\alpha ^𝔭`$, the chooser has a winning strategy in the game $`\mathrm{}_\alpha (N,𝔭)`$ (resp. $`\mathrm{}_\alpha ^{}(N,𝔭)`$).
5. We say that $`𝔭`$ is a non-$`\mathrm{}`$-loser (resp. a non-$`\mathrm{}^{}`$-loser) if for $`0<\alpha <\mathrm{}g(𝔭)`$ and $`N_\alpha `$ the challenger has no winning strategy in $`\mathrm{}_\alpha (N,𝔭)`$ (resp. $`\mathrm{}_\alpha ^{}(N,𝔭)`$).
###### Lemma 3.7.
1. If $`𝔭`$ is a reasonable parameter with the standard $`\overline{D}^𝔭`$, then $`𝔭`$ is a winner.
2. If $`𝔭`$ is a $`\mathrm{}_\alpha `$-winner, then $`𝔭`$ is a $`\mathrm{}_\alpha ^{}`$-winner. If $`𝔭`$ is a $`\mathrm{}`$-winner, then $`𝔭`$ is a $`\mathrm{}^{}`$-winner. Similarly for a non-loser.
###### Proof.
(1). Suppose $`𝔭`$ is a standard reasonable parameter. Let $`0<\alpha <\mathrm{}g(𝔭)`$ and $`N_\alpha ^𝔭.`$ Let $`y_n:n<\omega `$ be an enumeration of $`N\{(\chi _\beta ):\beta \alpha N\}`$ such that each $`yN\{(\chi _\beta ):\beta \alpha N\}`$ appears infinitely often in the enumeration and let $`\gamma _n:n<\omega `$ be an increasing sequence of ordinals in $`N\alpha `$ with $`\underset{n<\omega }{sup}\gamma _n=sup(N\alpha ).`$ We define the following winning strategy for chooser in the game $`\mathrm{}_\alpha (N,𝔭)`$: in the $`n`$-th move, the challenger chooses some $`X_nD_\alpha ^𝔭(N)`$. In particular, we can find $`\beta _nN\alpha \gamma _n`$ and $`M_nX_n_{\beta _n}^𝔭`$ such that $`y_nM_n`$ and $`M_nX_nD_{\beta _n}^𝔭(M_n)`$. Set also $`Y_n=M_nX_n.`$ Then the challenger choose some $`Z_nY_n.`$
We show that
$$X=\{\{M_n\}Z_n:n<\omega \}D_\alpha ^𝔭(N).$$
Thus let $`\gamma N\alpha `$ and $`yN\{(\chi _\beta ):\beta \alpha N\}`$. Pick $`n<\omega `$ such that $`\gamma _n>\gamma `$ and $`y=y_n`$. Then $`\beta _n`$ and $`M_n`$ are such that $`\beta _nN\alpha \gamma `$, $`M_nX_{\beta _n}^𝔭`$ and we have $`y_nM_n`$ and $`XM_nD_{\beta _n}^𝔭(M_n)`$. Thus $`XD_\alpha ^𝔭(N)`$. Hence the above process defines a winning strategy for chooser, as required.
(2). Suppose $`𝔭`$ is a $`\mathrm{}_\alpha `$-winner. Let $`N_\alpha `$ and assume that $`N^{}`$ is such that $`NN^{}((\chi ),)`$, $`𝔭\alpha N^{}`$ and $`N^{}`$ is countable. We define a winning strategy for chooser in the game $`\mathrm{}_\alpha ^{}(N,N^{},𝔭)`$.
Let $`\sigma `$ be a winning strategy for chooser in the game $`\mathrm{}_\alpha (N,𝔭).`$ By elementarity, we may assume that $`\sigma `$ is in $`N^{}`$. We define the strategy $`\sigma ^{}`$ for chooser in the game $`\mathrm{}_\alpha ^{}(N,N^{},𝔭)`$ as follows. At the $`n`$-th move, the challenger chooses some $`X_nN^{}`$. Then chooser picks the sets $`M_n`$ and $`Y_n`$ via the strategy $`\sigma `$ and he also takes $`X_n^{}`$ to be $`X_n.`$ As $`\sigma `$ is in $`N^{}`$, all these objects are also in $`N^{}`$. Then challenger chooses some $`Z_nN^{}.`$ It is evident that $`\sigma ^{}`$ is a winning strategy for chooser in the game $`\mathrm{}_\alpha ^{}(N,N^{},𝔭)`$, as required. The other cases of the lemma can be proved in a similar way. ∎
###### Definition 3.8.
Assume $`𝔭`$ is a reasonable parameter, $`\alpha <\mathrm{}g(𝔭)`$, $`N_\alpha ^𝔭,yN`$ and $`N`$ is a forcing notion. Set
[𝔾
~
,N,y]={MN:,yM and
G
~
M is (M)-generic over M},subscriptsubscript𝔾
~
𝑁𝑦conditional-set𝑀𝑁𝑦𝑀subscript and
G
~
𝑀 is 𝑀-generic over 𝑀{{\mathscr{M}}}_{{\mathbb{P}}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}},N,y]=\{M\in N:{\mathbb{P}},y\in M\text{ and }\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}}\cap M\text{ is }({\mathbb{P}}\cap M)\text{-generic over }M\},
where 𝔾
~
subscript𝔾
~
\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}} is the canonical $``$-name for the generic filter.
We consider [𝔾
~
,N,y]subscriptsubscript𝔾
~
𝑁𝑦{{\mathscr{M}}}_{{\mathbb{P}}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}},N,y] as a $``$-name and then $`_{}[𝔾,N,y]`$ is well defined for any $``$-generic filter $`𝔾`$. If $``$ is clear from the context, we may omit it. Note that $`_{}[𝔾,N,y]=_{}[𝔾N,N,y]`$, so we may write $`𝔾N`$ instead of $`𝔾`$. If $`y=\mathrm{}`$ we may omit it.
###### Definition 3.9.
We say $`\overline{}K_0`$ is a $`𝔭`$NNR$`{}_{\mathrm{}_0}{}^{}{}_{}{}^{0}`$ iteration if the following conditions are satisfied:
1. ¯=i,
~
i:i<j()\bar{{\mathbb{Q}}}=\langle{\mathbb{P}}_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}:i<j(*)\rangle is a countable support iteration of proper forcing notions such that $`\overline{},𝒫(\mathrm{Lim}(\overline{}))(\chi _0^𝔭)`$.
2. forcing with $`_{j()}=\mathrm{Lim}(\overline{})`$ does not add reals.
3. (long properness) suppose that:
1. $`(\alpha )ijj(),\alpha <\mathrm{}g(𝔭)`$,
2. $`(\beta )N_\alpha ^𝔭,\{i,j,\overline{}\}N`$,
3. $`(\gamma )`$ the condition $`q_i`$ is $`(N,_i)`$-generic,
4. $`(\delta )q`$𝔾
~
iN=𝔾subscript𝔾
~
subscript𝑖𝑁𝔾\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}\cap N=\mathbb{G}”,
5. $`(\epsilon )p_jN`$ and $`pi𝔾`$,
6. $`(\zeta )Y__i[𝔾,N,y]`$ where $`y=\overline{},i,j`$ and $`YD_\alpha ^𝔭(N)`$ <sup>2</sup><sup>2</sup>2Note that the ordinal $`i`$ is reconstructible from $`\overline{}`$ and $`𝔾`$.,
then there are $`𝔾^{},q^{}`$ such that:
1. $`(\eta )q^{}_j,pq^{}`$ and $`qq^{}i`$,
2. $`(\theta )q^{}`$ is $`(N,_j)`$-generic,
3. $`(\iota )q^{}`$𝔾
~
jN=𝔾subscript𝔾
~
subscript𝑗𝑁superscript𝔾\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{j}}\cap N=\mathbb{G}^{\prime}”,
4. $`(\kappa )Y__j[𝔾^{},N,y]D_\alpha ^𝔭(N)`$.
4. (anti weak diamond or anti-w.d.) suppose that:
1. $`(\alpha )ijj()`$ and $`\alpha <\mathrm{}g(𝔭),`$
2. $`(\beta )N_0N_1_\alpha ^𝔭,N_0\underset{\beta <\alpha }{}_\beta ^𝔭,`$
3. $`(\gamma )\mathrm{otp}(N_0[i,j))<\alpha `$ <sup>3</sup><sup>3</sup>3so naturally $`\mathrm{}g(𝔭)=\omega _1`$. We use the parallel of “$`\mathrm{}_0`$-completeness system” rather than “2-completeness system” of as things are complicated enough anyhow; see Definition 3.14 and Theorem 3.15.,
4. $`(\delta )n<\omega `$ and for $`\mathrm{}<n`$ we have $`q_{\mathrm{}}_i`$ is $`(N_1,_i)`$-generic,
5. $`(ϵ)q_{\mathrm{}}`$𝔾
~
iN1=𝔾subscript𝔾
~
subscript𝑖subscript𝑁1superscript𝔾\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}\cap N_{1}=\mathbb{G}^{\ell}”,
6. $`(\zeta )\underset{\mathrm{}<n}{}[𝔾^{\mathrm{}}N_0=𝔾^{}]`$ where $`𝔾^{}_iN_0`$ is generic over $`N_0`$,
7. $`(\eta )Y=:\underset{\mathrm{}<n}{}__i[𝔾^{\mathrm{}},N_1]D_\alpha ^𝔭(N_1)`$,
8. $`(\theta )p_jN_0`$ is such that $`pi𝔾^{}`$.
Then:
1. for some $`𝔾^{}_jN_0`$ generic over $`N_0`$ we have $`p𝔾^{}N_1`$ and $`\underset{\mathrm{}<n}{}\underset{qG^{\mathrm{}}}{}[q\mathrm{`}\mathrm{`}𝔾^{}`$ has an upper bound in j/𝔾
~
i"]{\mathbb{P}}_{j}/\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}"].
###### Remark 3.10.
We may like to phrase clause (c) as a condition on each
~
isubscript
~
𝑖\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}, for this see Definitions 4.7, 4.9 and 4.12.
We now state and prove the main result of this section.
###### Theorem 3.11.
Assume $`𝔭`$ is a reasonable parameter of length $`\omega _1`$, $`\overline{}`$ is a countable support iteration such that $`\overline{},𝒫(\mathrm{Lim}\overline{})(\chi _0^𝔭)`$, $`\delta =\mathrm{}g(\overline{})`$ is a limit ordinal and for every $`\alpha <\delta ,\overline{}\alpha `$ is a $`𝔭\text{NNR}_\mathrm{}_0^0`$ iteration and $`𝔭`$ is a $`\mathrm{}`$-winner. Then $`\overline{}`$ is a $`𝔭\text{NNR}_\mathrm{}_0^0`$ iteration.
###### Proof.
We show that items (a)-(d) of Definition 3.9 are satisfied by $`\overline{}`$.
Proof of clause (a): This holds trivially by our assumptions.
Proof of clause (b): This follows from clause (d) of Definition 3.9 proved below. To see this, let $`p_\delta =\mathrm{Lim}\overline{},`$ $`r`$ $`\stackrel{~}{}`$ be a $`_\delta `$-name and suppose that $`p`$ $`r`$ $`\stackrel{~}{}`$ is a real”. In Definition 3.9(d) set $`i=0,j=\delta ,n=1`$ and pick $`N_0N_1`$ so that the hypotheses in $`()_1`$ are satisfied and r
~
N0𝑟
~
subscript𝑁0\mathchoice{\oalign{$\displaystyle r$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle r$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle r$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle r$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\in N_{0}. Thus, by $`()_2,`$ we can find $`𝔾^{}_\delta N_0`$ which is generic over $`N_0`$, $`p𝔾^{}`$ and $`𝔾^{}`$ has an upper bound $`qp`$ in $`_\delta `$. By genericity of $`𝔾^{}`$, $`q`$ decides $`r`$ $`\stackrel{~}{}`$ to be a real in $`V`$, and we are done.
We first prove clause (d) and then return to clause (c).
Proof of clause (d): Let $`i,j,\alpha ,N_0,N_1,n,q_0,\mathrm{},q_{n1},G^0,\mathrm{},G^{n1},G^{}`$ and $`p`$ be as in the assumptions of clause (d). Let $`\alpha ^{}=\mathrm{otp}(N_0[i,j))`$. Then $`\alpha ^{}<\omega _1`$, so $`\alpha ^{}N_1`$. If $`j<\delta `$, then by our assumption $`\overline{}j`$ is a $`𝔭`$NNR$`{}_{\mathrm{}_0}{}^{}{}_{}{}^{0}`$ iteration and the result follows. Thus assume that $`j=\delta `$. If $`i=j`$, the conclusion is trivial, so assume $`i<j`$.
Let $`i_mN_0j`$ be such that $`i_0=i`$ and $`i_m:m<\omega N_1`$ is an increasing sequence with $`\underset{m<\omega }{}i_m=sup(N_0j)`$. Choose $`M_k:k<5`$ such that:
* $`y^{}:=\{i,j,\alpha ,\alpha ^{},\overline{},N_0,i_m:m<\omega \}M_k`$,
* $`M_k_\alpha ^{}^𝔭N_1\underset{\mathrm{}<n}{}__i[G^{\mathrm{}},N_1]`$,
* $`M_0M_1M_2M_3M_4`$,
* $`\underset{\mathrm{}<n}{}__i[𝔾^{\mathrm{}},M_0,y^{}]D_\alpha ^𝔭(M_0)`$.
Note that for $`\mathrm{}<n`$ and $`k<5`$, $`N_0M_kN_1`$ and $`𝔾^{\mathrm{}}M_k`$ is a generic subset of $`_iM_k`$. Now for $`\mathrm{}<n`$ we can choose $`q_{\mathrm{}}^{}𝔾^{\mathrm{}}M_4`$ such that:
* $`q_{\mathrm{}}^{}`$ forces (for $`_{i_0}=_i`$) a value for 𝔾
~
i0M3subscript𝔾
~
subscriptsubscript𝑖0subscript𝑀3\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{0}}}\cap M_{3}, which necessarily is $`𝔾^{\mathrm{}}M_3`$,
* $`q_{\mathrm{}}q_{\mathrm{}}^{}`$,
* $`q_{\mathrm{}}^{}`$ is $`(M_k,_{i_0})`$-generic, forcing 𝔾
~
i0Mk=𝔾Mksubscript𝔾
~
subscriptsubscript𝑖0subscript𝑀𝑘superscript𝔾subscript𝑀𝑘\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{0}}}\cap M_{k}=\mathbb{G}^{\ell}\cap M_{k} for $`k=0,1,2,3`$,
* $`q_{\mathrm{}}^{}`$ forces 𝔾
~
i0N0=𝔾subscript𝔾
~
subscriptsubscript𝑖0subscript𝑁0superscript𝔾\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{0}}}\cap N_{0}=\mathbb{G}^{*}.
Let $`_m^{}:m<\omega M_0`$ list the maximal antichains of $`_j`$ that belongs to $`N_0`$. We choose, by induction on $`m<\omega `$, the objects $`r_m,p_m,n_m,𝔾_m^{}`$, $`𝔾_m^{\mathrm{}}:\mathrm{}<n_m`$ and $`Y_m`$ such that:
1. $`(a)r_m_{i_m}M_4`$,
2. $`(b)\mathrm{dom}(r_m)[i,i_m)`$,
3. $`(c)r_{m+1}i_m=r_m`$,
4. $`(d)q_{\mathrm{}}^{}r_m_{i_m}`$ and is $`(M_k,_{i_m})`$-generic for $`k=0,1,2,3`$ and is $`(N_0,_{i_m})`$\- generic,<sup>4</sup><sup>4</sup>4note that $`q_{\mathrm{}}^{}`$ and $`r_m`$ have disjoint domains.
5. $`(e)`$ for every predense subset $``$ of $`_{i_m}`$ which belongs to $`M_2`$, for some finite $`𝒥M_2`$, the set $`𝒥`$ is predense above $`q_{\mathrm{}}^{}r_m,`$ for each $`\mathrm{}<n`$,
6. $`(f)n_m<\omega `$ and for $`\mathrm{}<n_m`$ $`𝔾_m^{\mathrm{}}M_1`$ is a subset of $`_{i_m}M_0`$ generic over $`M_0,`$
7. $`(g)`$ if $`\mathrm{}<n_{m+1}`$, then $`𝔾_{m+1}^{\mathrm{}}_{i_m}\{𝔾_m^k:k<n_m\}`$,
8. $`(h)n_0=n`$ and $`𝔾_0^{\mathrm{}}=𝔾^{\mathrm{}}M_0`$,
9. (i)qrmim``𝔾
~
imM0{𝔾m:<nm}subscriptforcessubscriptsubscript𝑖𝑚𝑖subscriptsuperscript𝑞subscript𝑟𝑚``subscript𝔾
~
subscriptsubscript𝑖𝑚subscript𝑀0conditional-setsubscriptsuperscript𝔾𝑚subscript𝑛𝑚(i)\quad q^{\prime}_{\ell}\cup r_{m}\Vdash_{{\mathbb{P}}_{i_{m}}}``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{0}\in\{\mathbb{G}^{\ell}_{m}:\ell<n_{m}\}”,
10. $`(j)𝔾_m^{}`$ is a subset of $`_{i_m}N_0`$ generic over $`N_0`$,
11. $`(k)𝔾_m^{}𝔾_m^{\mathrm{}}`$ for $`\mathrm{}<n_m`$,
12. $`(l)p_m`$ is such that:
13. $`(l`$-$`1)p_m_jN_0`$,
14. $`(l`$-$`2)p_mi_m𝔾^{}_m,`$
15. $`(l`$-$`3)p_{m+1}^{}_m,`$
16. $`(l`$-$`4)p_0=p,`$
17. $`(l`$-$`5)p_mp_{m+1}`$.
18. $`(m)Y_m=\underset{\mathrm{}<n_m}{}_{_{i_m}}[G_m^{\mathrm{}},M_0,y^{}]D_\alpha ^{}(M_0)`$ where $`y^{}=\{N_0,i_m:m<\omega ,\overline{},i,j\}`$.
The construction is clear for $`m=0`$. So suppose that we have it for $`m`$ and we shall choose for $`m+1`$. We do this is several steps.
Stage A: Choose $`p_{m+1}N_0_m^{}`$ such that $`p_mp_{m+1}`$ and $`p_{m+1}i_m𝔾_m^{}`$.
Stage B: Choose $`𝔾_{m+1}^{}_{i_{m+1}}N_0`$ generic over $`N_0`$ such that $`𝔾_m^{}𝔾_{m+1}^{}M_0,p_{m+1}i_{m+1}𝔾_{m+1}^{}`$ and
$`\underset{\mathrm{}<n_m}{}\underset{r𝔾_m^{\mathrm{}}}{}[r_{_{i_m}}\mathrm{`}\mathrm{`}𝔾_{m+1}^{}`$ has an upper bound in im+1/𝔾
~
im]{\mathbb{P}}_{i_{m+1}}/\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}].
This is easy by applying clause (d) of the Definition 3.9 for $`i_m,i_{m+1},\alpha ^{},p_{m+1}i_m,𝔾_m^{},𝔾_m^{\mathrm{}}:\mathrm{}<n_m,N_0,M_0`$, for the forcing notion $`\overline{}i_{m+1}`$, which is, by induction hypothesis, a $`𝔭NNR_\mathrm{}_0^0`$ iteration, . We also use the fact that $`\mathrm{otp}(N_0[i_m,i_{m+1}))<\mathrm{otp}(N_0[i_m,j))=\alpha ^{}`$.
Stage C: As $`_{i_m}`$ is proper and adds no new reals, 𝔾
~
imM1subscript𝔾
~
subscriptsubscript𝑖𝑚subscript𝑀1\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{1} is a $`_{i_m}`$-name of an object from $`𝕍`$, so
={pim:p forces a value for
G
~
imM1 in 𝕍}conditional-set𝑝subscriptsubscript𝑖𝑚𝑝subscript forces a value for
G
~
subscriptsubscript𝑖𝑚subscript𝑀1 in 𝕍{{\mathscr{I}}}=\{p\in{\mathbb{P}}_{i_{m}}:p\text{ forces a value for }\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{1}\text{ in }\mathbb{V}\}
is a dense open subset of $`_{i_m}`$ and $`M_2`$. By clause $`()_1(e)`$ of the induction hypothesis, there is a finite $`𝒥M_2`$ such that: $`\mathrm{}<n𝒥`$ is predense above $`q_{\mathrm{}}^{}r_m`$. Without loss of generality $`𝒥`$ is minimal. Let $`n_{m+1}=|𝒥|`$.
Let $`𝒥=\{p_m^{\mathrm{}}:\mathrm{}<n_{m+1}\}`$ and for each $`\mathrm{}<n_{m+1}`$ choose $`H_m^{\mathrm{}}M_2`$ such that pm``𝔾
~
imM1=Hmforcessubscriptsuperscript𝑝𝑚``subscript𝔾
~
subscriptsubscript𝑖𝑚subscript𝑀1subscriptsuperscript𝐻𝑚p^{\ell}_{m}\Vdash``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{1}=H^{\ell}_{m}”. As $`𝒥`$ is minimal, $`H_m^{\mathrm{}}M_0\{𝔾_m^{\mathrm{}}:\mathrm{}<n_m\}`$ so for some $`h:n_{m+1}n_m`$ and every $`\mathrm{}<n_{m+1}`$ we have $`H_m^{\mathrm{}}M_0=𝔾_m^{h(\mathrm{})}`$.
Let $`Y=\underset{\mathrm{}<m_n}{}_{_{i_m}}[𝔾_m^{\mathrm{}},M_0,y^{}]D_\alpha ^{}^𝔭(M_0)`$. Now we choose by induction on $`\mathrm{}n_{m+1}`$ a condition $`r_m^{\mathrm{}}M_1`$ such that:
1. $`(\alpha )r_m^{\mathrm{}}_{i_{m+1}}M_1`$ and $`r_m^{\mathrm{}}i_mH_m^{\mathrm{}}`$,
2. $`(\beta )r_m^{\mathrm{}}`$ is $`(M_0,_{i_m})`$-generic and forces a value for 𝔾
~
imM0subscript𝔾
~
subscriptsubscript𝑖𝑚subscript𝑀0\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{0}, say $`𝔾_{m+1}^{\mathrm{}}`$,
3. $`(\gamma )r_m^{\mathrm{}}`$ is above $`𝔾_{m+1}^{}`$, and moreover above $`p_m^{h(\mathrm{})}`$,
4. $`(\delta )Y\underset{k<\mathrm{}}{}_{_{i_m}}[𝔾_{m+1}^k,M_0,y^{}]D_\alpha ^{}^𝔭(M_0)`$.
The construction can be easily done by applying clause (c) of Definition 3.9 to $`i_m,i_{m+1},\alpha ^{},M_0`$, large enough member of $`H_m^{\mathrm{}},p_m^{h(\mathrm{})}`$ and $`Y_m^{\mathrm{}}=Y\underset{k<\mathrm{}}{}[𝔾_{m+1}^k,M_0,y^{}]M_1,`$ for the $`𝔭NNR_\mathrm{}_0^0`$-iteration $`\overline{}i_{m+1}`$.
Stage D: By \[6, Ch.XVIII, Claim 2.6\], we can choose $`r_{m+1}`$ as required such that $`\{r_m^{\mathrm{}}:\mathrm{}<n_{m+1}\}`$ is predense over it.
This completes the inductive construction. Let us now show that this is sufficient to get clause (d). Let $`^{}N_1`$ be a well-ordering of $`M_4`$. During the construction above we chose inductively members of $`M_4`$ and all the parameters used are from $`M_4`$, so if we always choose the $`^{}`$-first object, the construction is determined and is in $`N_1`$. By clause $`()_1(c)`$, we have
1. Let $`r=\underset{m}{}r_m`$ be the unique $`r_j`$ satisfying $`m<\omega ri_m=r_m`$. Then $`r_jN_1`$.
Also, by the choice of $`_m^{}:m<\omega `$ and clause $`()_1(k)`$, we have
1. $`𝔾^{}=\{p^{}_jN_0:\underset{m<\omega }{}[p^{}p_m]\}`$ belongs to $`M_4`$ and is a subset of $`_jN_0`$ generic over $`N_0`$.
It is also clear from $`()_1(\mathrm{})`$ that
1. $`q_{\mathrm{}}^{}r`$ is above $`𝔾^{}`$ (in $`_j`$).
So we have finished proving clause (d).
Proof of clause (c): We prove this by induction on $`\alpha `$. Let $`i,j,\alpha ,N,p,q`$ and $`Y`$ be as there. If $`j<\delta `$ we can apply “$`\overline{}j`$ is a $`𝔭NNR_\mathrm{}_0^0`$ iteration”, so without loss of generality $`j=\delta `$. If $`i=j`$ the statement is trivial, so assume $`i<j`$.
Choose $`i_nNj`$, for $`n<\omega `$, such that $`i_0=i,i_n<i_{n+1}`$ and $`\underset{n<\omega }{}i_n=sup(jN)`$. Let $`(y_n,\beta _n):n<\omega `$ list the pairs $`(y,\beta )N\times (\alpha N)`$ such that $`y(\chi _\beta ^𝔭)`$. Let $`\sigma `$ be a winning strategy for the chooser in the game $`\mathrm{}_\alpha (N,𝔭)`$ and let $`_n:n<\omega `$ list the dense open subsets of $`_j`$ which belong to $`N`$.
We choose by induction on $`n<\omega `$, the objects qn,p
~
n,M
~
nsubscript𝑞𝑛subscript𝑝
~
𝑛subscript𝑀
~
𝑛q_{n},\mathchoice{\oalign{$\displaystyle p$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle p$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n},\mathchoice{\oalign{$\displaystyle M$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle M$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n} and Y
~
nsubscript𝑌
~
𝑛\mathchoice{\oalign{$\displaystyle Y$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle Y$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle Y$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle Y$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n} such that:
1. $`(a)q_n_{i_n}`$ with $`q_0=q`$,
2. $`(b)q_n`$ is $`(N,_{i_n})`$-generic,
3. $`(c)q_{n+1}i_n=q_n`$,
4. (d)p
~
n𝑑subscript𝑝
~
𝑛(d)\quad\mathchoice{\oalign{$\displaystyle p$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle p$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n} is a $`_{i_n}`$-name of a member of (j/𝔾
~
in)Nsubscript𝑗subscript𝔾
~
subscriptsubscript𝑖𝑛𝑁({\mathbb{P}}_{j}/\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{n}}})\cap N
5. (e)p
~
n𝑒subscript𝑝
~
𝑛(e)\quad\mathchoice{\oalign{$\displaystyle p$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle p$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n} is forced to belong to $`_n`$,
6. (f)M
~
n𝑓subscript𝑀
~
𝑛(f)\quad\mathchoice{\oalign{$\displaystyle M$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle M$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n} is a $`_{i_n}`$-name of a member of $`_{\beta _n}^𝔭N`$,
7. $`(g)`$if $`𝔾_j_j`$ is generic over $`𝕍`$ such that qn𝔾j,p
~
n[𝔾jin]𝔾jformulae-sequencesubscript𝑞𝑛subscript𝔾𝑗subscript𝑝
~
𝑛delimited-[]subscript𝔾𝑗subscriptsubscript𝑖𝑛subscript𝔾𝑗q_{n}\in\mathbb{G}_{j},\mathchoice{\oalign{$\displaystyle p$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle p$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n}[\mathbb{G}_{j}\cap{\mathbb{P}}_{i_{n}}]\in\mathbb{G}_{j} and if M=M
~
n[𝔾j]𝑀subscript𝑀
~
𝑛delimited-[]subscript𝔾𝑗M=\mathchoice{\oalign{$\displaystyle M$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle M$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n}[\mathbb{G}_{j}], then
1. $`(\alpha )𝔾_jM`$ is a subset of $`_jM`$ generic over $`M`$,
2. $`(\beta )__j[𝔾_jM,M,y^{}]YD_{\beta _n}^𝔭[M]`$,
3. (γ)p
~
n[𝔾j]𝛾subscript𝑝
~
𝑛delimited-[]subscript𝔾𝑗(\gamma)\quad\mathchoice{\oalign{$\displaystyle p$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle p$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle p$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n}[\mathbb{G}_{j}] belongs to $`M`$,
8. (h)Ymim[𝔾
~
im,M
~
m,y],Y
~
m,im,M
~
m:mn(h)\quad\langle Y_{m}\cap{{\mathscr{M}}_{{\mathbb{P}}_{i_{m}}}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i_{m}},\mathchoice{\oalign{$\displaystyle M$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle M$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{m},y^{*}],\mathchoice{\oalign{$\displaystyle Y$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle Y$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle Y$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle Y$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{m},{\mathbb{P}}_{i_{m}},\mathchoice{\oalign{$\displaystyle M$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle M$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{m}:m\leq n\rangle is forced by $`q_n`$ to be an initial segment of a play of the game $`\mathrm{}_\alpha (N)`$ in which the chooser uses the fixed winning strategy $`\sigma `$.
The proof is straight by the induction hypothesis on $`\beta `$, the fact that M
~
n,Y
~
nsubscript𝑀
~
𝑛subscript𝑌
~
𝑛\mathchoice{\oalign{$\displaystyle M$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle M$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle M$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n},\mathchoice{\oalign{$\displaystyle Y$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle Y$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle Y$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle Y$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{n} are $`_{i_n}`$-names of objects from $`𝕍`$ and $`\overline{}i_n`$ is a $`𝔭NNR_\mathrm{}_0^0`$-iteration. Now let $`q^{}=\underset{n<\omega }{}q_n`$ and $`𝔾^{}=\{p^{}_jN:\underset{n<\omega }{}p^{}q_n\}`$. Then $`𝔾^{}`$ and $`q^{}`$ are as required; see also the proof of clause $`(c)^{}`$ of Theorem 6.3 for more details, where a more general result is proved.
The theorem follows. ∎
###### Remark 3.12.
1. It is possible to use “adding no reals+ clause (d)” in the proof of clause (c) in order to weaken “winner” to “not loser”. Also we can use $`\mathrm{}_\alpha ^{}(N,N^{},)`$, see Section 6. The assumption “$`N_0\underset{\beta <\alpha }{}_\beta `$” can be replaced by $`\mathrm{`}\mathrm{`}N_0_0^{}`$ with $`_0^{}[(\chi _0^𝔭)]^\mathrm{}_0`$ stationary”. We can also put extra restrictions on $`𝔾^{}`$ (and $`𝔾^{}`$), for example we can require $`[𝔾^{},N_0,y^{}]`$ is large.
2. The use of $`\chi _\alpha :\alpha <\mathrm{}g(𝔭)`$ is not really necessary as all the properties depend just on $`N_{\mathrm{}}𝒫(_{\mathrm{}g(\overline{})})`$.
3. The proof of clause (c) being preserved can be applied to any $`\overline{}`$ satisfying (a) + (c) of Definition 3.9 (so possibly adding reals), but then we have to replace $`D_\alpha ^𝔭(N)`$ by a definition of such pseudo filters with the winning strategy being absolute enough, e.g. for standard $`\overline{D}`$.
###### Definition 3.13.
Let $`\overline{}`$ be a countable support iteration of forcing notions. It will be called $`𝔭`$-proper if it satisfies items (a) and (c) of Definition 3.9.
We like to consider the parallel of having 2-completeness systems and also to demand only non-losing rather than winning in the assumption of Theorem 3.11.
###### Definition 3.14.
Let $`\kappa [2,\omega ]`$. We say that $`\overline{}`$ is a $`𝔭NNR_\kappa ^0`$-iteration if items (a)-(c) of Definition 3.9 are satisfied and clause (d) is replaced by $`\kappa `$-anti w.d., which is the same as in (d) there, but with $`n<1+\kappa `$ and $`N_0_0^𝔭`$.
The next theorem is a natural generalization of Theorem 3.11.
###### Theorem 3.15.
Assume $`𝔭`$ is a reasonable parameter of length $`\omega _1`$ which is a non-$`\mathrm{}^{}`$-loser, $`2\kappa <\mathrm{}_0,\overline{}`$ is a countable support iteration with $`𝒫(\mathrm{Lim}(\overline{}))(\chi _0^𝔭),`$ $`\delta =\mathrm{}g(\overline{})`$ is a limit ordinal and for $`i<\delta ,\overline{}i`$ is a $`𝔭NNR_\kappa ^0`$-iteration. Then $`\overline{}`$ is a $`𝔭NNR_\kappa ^0`$-iteration.
###### Proof.
The proof is similar to the proof of Theorem 3.11, with some changes, as in \[5, Ch.VIII, Claim 4.10\] and \[6, Ch.XVIII, Proof 2.10C\], so we do not give the details. The only main change is that during the proof of clause (d), we add the following extra conditions to items (a)-(m):
1. $`n_m`$ is a power of 2, say $`2^{n_m^{}}`$ and so we can rename $`\{𝔾_m^{\mathrm{}}:\mathrm{}<n_m\}`$ as $`\{𝔾_m^\eta :\eta {}_{}{}^{n_m^{}}2\}`$,
2. The following conditions are satisfied:
3. $`(\alpha )`$ for $`\eta {}_{}{}^{(n_m^{})}2`$, $`M_\eta M_1_{j_\eta }^𝔭`$, where $`j_0=\mathrm{otp}([i,j)N_0)`$ and if $`\eta =\nu ^{}i,`$ then $`j_\eta =\mathrm{otp}([i,j)M_\nu )`$,
4. $`(\beta )M_{}=N_0`$,
5. $`(\gamma )M_\eta M_{\eta ^{}0}M_{\eta ^{}1}`$,
6. $`(\delta )\eta \nu _1{}_{}{}^{n_m^{}}2\eta \nu _2{}_{}{}^{n_m^{}}2𝔾_m^{\nu _1}M_\eta =𝔾_m^{\nu _2}M_\eta `$ so we call it $`K_m^\eta `$
7. $`(\epsilon )`$ for $`\eta {}_{}{}^{n_m^{}>}2`$, $`M_{\eta ^{}0}=M_{\eta ^{}1}`$, call it $`N_\eta `$,
8. $`(\zeta )`$ for $`\eta {}_{}{}^{(m_m^{}>)}2`$ and $`\mathrm{}<2`$, $`N_\eta _{j_\eta ^{}\mathrm{}}^𝔭`$,
9. $`(\eta )Y_m^\eta =[K_m^{\eta ^{}0},N_\eta ][K_m^{\eta ^{}1},N_\eta ]D_{j_{\eta ^{}0}}(N_\eta )`$.
The rest of the argument is essentially the same as before. ∎
The following is an immediate consequence of Theorems 3.11 3.15.
###### Conclusion 3.16.
Suppose $`𝔭`$ is a non-$`\mathrm{}^{}`$-loser reasonable parameter with $`\mathrm{}g(𝔭)=\omega _1`$, $`\overline{}`$ is a countable support iteration and $`2n()\mathrm{}_0`$. Then, $`\overline{}`$ is a $`𝔭NNR_{n()}^0`$-iteration iff for each $`i<\mathrm{}g(\overline{})`$
1. ~
isubscript
~
𝑖\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} is a proper forcing and i,
~
isubscript𝑖subscript
~
𝑖{\mathbb{P}}_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} satisfy clauses (d) + (c) of the Definition “$`𝔭NNR_{n()}^0`$-iteration” with $`i,i+1`$ here standing for $`i,j`$ there.
###### Proof.
By induction on $`j=\mathrm{}g(\overline{})`$. For $`j=0`$ there is nothing to prove and for $`j`$ a successor ordinal, this follows easily from the definitions. For $`j`$ a limit ordinal, the result follows from Theorem 3.11 (for $`n()=\mathrm{}_0`$) or Theorem 3.15 (for $`2n()<\mathrm{}_0`$). ∎
We point out here that Clause (c) of Definitions 3.9 and 3.14 really follows from earlier properties which play parallel roles.
###### Lemma 3.17.
1. Assume that $`𝔭`$ is a standard reasonable parameter, $`\alpha <\mathrm{}g(𝔭),N_\alpha ^𝔭,YD_\alpha ^𝔭(N)`$ and $`\delta \omega _1N`$ is a limit ordinal. Then we can find sequences $`\overline{N}=N_i:i<\delta `$ and $`\overline{\gamma }=\gamma _i:i<\delta `$ such that:
1. $`N_iN`$ is countable, $`N\alpha N_i`$ and $`N_iY`$ (for $`i<\delta `$),
2. $`N_i\underset{\beta \alpha N}{}((\chi _\beta ^𝔭),)`$, and $`\beta \alpha N_iN_i(\chi _\beta ^𝔭)((\chi _\beta ^𝔭),)`$,
3. $`i<jN_iN_j`$,
4. if $`i`$ is a limit ordinal, then $`N_i=\underset{j<i}{}N_j`$ and $`N\{(\chi _\beta ^𝔭):\beta \alpha N\}=\underset{j<\delta }{}N_j`$, so we can stipulate $`N_\delta =N`$,
5. $`\beta \alpha N(\chi _\beta ^𝔭)N_j:jiN_{i+1}`$,
6. $`\gamma _iN_i\alpha ,N_i(\chi _{\gamma _i}^𝔭)_{\gamma _i}^𝔭Y`$ and $`\overline{\gamma }(i+1)N_{i+1}`$,
7. if $`i\delta `$ is a limit ordinal and either $`(i=\delta `$ and $`\beta \alpha N_i)`$ or $`(i<\delta `$ and $`\beta \gamma _iN_i)`$, then for some $`j<i,\gamma _j=\beta `$ and $`yN_j`$.
8. if $`i\delta `$ is a limit ordinal, then $`\{N_j(\chi _{\gamma _j}^𝔭):j<i\text{ and }\gamma _j<\gamma _i\}D_{\gamma _i}^𝔭(N)=D_{\gamma _i}^𝔭(N(\chi _{\gamma _i}^𝔭)).`$
9. if $`\delta <N\omega _1`$ then $`\delta N_0`$, if $`\delta =N\omega _1`$ then $`i<\delta iN_0`$.
2. If $`𝔭`$ is a standard reasonable parameter, $`\overline{}`$ is a countable support iteration, $`\mathrm{}g(\overline{})=\beta +1,\overline{}\beta `$ is $`𝔭NNR_{k()}^0`$-iteration and β``
~
βsubscriptforcessubscript𝛽absent``subscript
~
𝛽\Vdash_{{\mathbb{P}}_{\beta}}``\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\beta} is proper and $`(<^+\omega _1)`$-proper”, then $`\overline{}`$ is $`𝔭`$-proper (see Definition 3.13).
3. If $`\mathrm{}g(𝔭)=\omega _1`$, then in part $`(2)`$, it suffices to assume i``
~
isubscriptforcessubscript𝑖absent``subscript
~
𝑖\Vdash_{{\mathbb{P}}_{i}}``\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} is $`(<\omega _1)`$-proper”.
###### Proof.
(1). By induction on $`i<\delta `$ we prove that there are sequences $`N_j:j<iN`$ and $`\gamma _j:j<i`$ satisfying the relevant requirements, so that for some sequence $`N_j^{}:j<iN`$ with $`N_j^{}((\chi _\alpha ^𝔭),),`$ $`N_j=N_j^{}\underset{\beta \alpha N}{}((\chi _\beta ^𝔭),)`$.
For $`i=0`$, let $`N_0^{}((\chi _\alpha ^𝔭),)`$ be a countable model such that $`N_0^{}NY`$ and such that $`N_0^{}`$ contains all relevant information, in particular, $`N\alpha N_0^{},\delta N_0^{}`$. Let $`N_0=N_0^{}\underset{\beta \alpha N}{}((\chi _\beta ^𝔭),).`$ Pick also $`\gamma _0`$ such that $`\gamma _0N_0\alpha `$ and $`N_0(\chi _{\gamma _0}^𝔭)_{\gamma _0}^𝔭Y`$. Such $`\gamma _0`$ exists as $`𝔭`$ is standard.
If $`i=j+1`$ is a successor ordinal, let $`N_i^{}((\chi _\alpha ^𝔭),)`$ be a countable model such that:
* $`N_i^{}NY`$,
* $`N_j^{}N_i,`$
* $`\beta \alpha N(\chi _\beta ^𝔭)N_k:kjN_i`$,
* $`\gamma _k:kjN_i`$.
Then, using $`𝔭`$ is standard, take $`\gamma _iN_i\alpha `$ such that $`N_i(\chi _{\gamma _i}^𝔭)_{\gamma _i}^𝔭Y`$.
For limit $`i`$ set $`N_i^{}=\underset{j<i}{}N_j`$ and $`N_i=N_i^{}\underset{\beta \alpha N}{}((\chi _\beta ^𝔭),)`$. Let also $`\gamma _i`$ be such that $`\{N_j(\chi _{\gamma _j}^𝔭):j<i\text{ and }\gamma _j<\gamma _i\}D_{\gamma _i}^𝔭(N)=D_{\gamma _i}^𝔭(N(\chi _{\gamma _i}^𝔭)).`$
As $`N`$ is countable, we can choose $`N_j^{}`$’s such that clause (d) holds as well. This completes our inductive construction.
(2) We show that $`\overline{}`$ satisfies clause (c) of Definition 3.9. Thus let $`i,j,\alpha ,N,q,p,𝔾`$ and $`Y`$ be as there. Without loss of generality, $`i=\beta `$ and $`j=\beta +1`$.
By the definition of $`(<^+\omega _1)`$-proper, if $`𝔾_\beta _\beta `$ is generic over $`N,`$ then $`__\beta [𝔾_\beta ,N,y^{}]D_\beta ^𝔭(N)`$. Let $`\delta =N\omega _1`$ and let $`N_i:i<\delta `$ be as in $`(1)`$. Without loss of generality p(β)N0
~
β[𝔾β]𝑝𝛽subscript𝑁0subscript
~
𝛽delimited-[]subscript𝔾𝛽p(\beta)\in N_{0}\cap\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\beta}[\mathbb{G}_{\beta}]. Let $`q^{}p`$ be (Ni,
~
β[𝔾β])subscript𝑁𝑖subscript
~
𝛽delimited-[]subscript𝔾𝛽(N_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\beta}[\mathbb{G}_{\beta}])-generic for every $`i<\delta `$ <sup>5</sup><sup>5</sup>5formally we only need to look at $`\overline{N}^{}=N_i^{}:i<\delta ,N_i^{}=N_i(\chi _0^𝔭)`$ and apply the $`(<^+\omega _1)`$-properness to it. so that $`q^{}\beta q.`$ Let also 𝔾(β)
~
β[𝔾β]𝔾𝛽subscript
~
𝛽delimited-[]subscript𝔾𝛽\mathbb{G}(\beta)\subseteq\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\beta}[\mathbb{G}_{\beta}] be generic over $`N`$ such that $`q^{}\beta `$ forces $`𝔾(\beta )`$ has an upper bound in
~
β[𝔾β]subscript
~
𝛽delimited-[]subscript𝔾𝛽\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\beta}[\mathbb{G}_{\beta}]. Set $`𝔾^{}=𝔾_\beta 𝔾(\beta ).`$ Then $`q^{},𝔾^{}`$ are as required.
(3) Follows from (2), as $`\alpha N\omega _1\delta =\omega \alpha N\omega _1`$. ∎
###### Remark 3.18.
1. The results of this section include as special cases \[5, Ch.V, §5, §7\]. There is no direct comparison with \[5, Ch.VIII, §4\], \[6, Ch.VIII, §4\], but we can make the notion somewhat more complicated, to include the theorems there in our context, but this is not really needed for the examples discussed there (see Section 5). The condition in \[5, Ch.VIII, §4\] and \[6, Ch.VIII, §4\] involves having many sequences $`N_i:i\delta `$ such that if $`p_0,p_1_\alpha ,p_{\mathrm{}}`$ is $`(N_i,)`$-generic for i,p``𝔾
~
αN0=𝔾forces𝑖subscript𝑝``subscript𝔾
~
subscript𝛼subscript𝑁0superscript𝔾i,p_{\ell}\Vdash``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{\alpha}}\cap N_{0}=\mathbb{G}^{*}”, then there is 𝔾Gen(N0[𝔾],
~
α[𝔾]),𝔾N0,formulae-sequencesuperscript𝔾Gensubscript𝑁0delimited-[]superscript𝔾subscript
~
𝛼delimited-[]superscript𝔾superscript𝔾subscript𝑁0\mathbb{G}^{\prime}\subseteq{\rm Gen}(N_{0}[\mathbb{G}^{*}],\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}[\mathbb{G}^{*}]),\mathbb{G}^{\prime}\in N_{0}, such that α``𝔾
~
subscriptnot-forcessubscript𝛼absent``fragmentsG
~
\nVdash_{{\mathbb{P}}_{\alpha}}``\mathchoice{\oalign{$\displaystyle\mathbb{G}^{\prime}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}^{\prime}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}^{\prime}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}^{\prime}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}} has no bound in
~
αsubscript
~
𝛼\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}”. This speaks on a family of sequences from $`[(\chi )]^\mathrm{}_0`$ rather than members of $`(\chi )`$.
2. For \[6, Ch.XVIII, §2\], the comparison is not so easy. Our problem is to “carry” good $`(N,_i,𝔾_{\mathrm{}}:\mathrm{}<n),𝔾_{\mathrm{}}\mathrm{Gen}(N,_i)`$ with a bound, such that we can “increase $`i`$” and we can find $`N^{},yN^{}N,N^{}N`$ such that $`(N^{},P_i,𝔾_iN:\mathrm{}<n)`$ is good enough. In \[6, Ch.XVIII\] we are carrying genericity in some $`_{\overline{\alpha }},`$ where $`\overline{\alpha }\text{ trind}(i)`$<sup>6</sup><sup>6</sup>6see \[6, Ch.XVIII, Definiton 2.1\]., but here we have much less. But what we need is the implication “if $`(N,_i,\overline{G})`$ is good we can extend it to good $`(N,_{i+1},\overline{G}^{})`$”, so making good weaker generates an incomparable notion and clearly there are other variants.
3. The iteration theorems proved in this section can be used to give alternative proofs of the consistency results in \[6, Ch.XVIII, §1\] (see Section 5).
## 4. Delayed properness
In this section we introduce several notions that will be used in sections 5 and 6. We concentrate on simple reasonable parameters and we present two versions for it. The simpler one is version 2 for which simplicity is a very natural demand. The proof of the next lemma is straightforward in which a general way to create simple reasonable parameters is introduced.
###### Lemma 4.1.
1. Assume
1. $`\overline{\chi }=\chi _\alpha :\alpha <\alpha ^{}`$ increases fast enough, so that $`((\underset{\beta <\alpha }{}\chi _\beta )^+)(\chi _\alpha )`$,
2. $`_\alpha \{N:N`$ is a countable elementary submodel of $`((\chi _\alpha ),)\}`$ is stationary,
3. $`R_\alpha (\chi _\alpha )`$ and $`N_\alpha `$ implies $`\chi _\beta :\beta <\alpha N,R_\beta :\beta \alpha N`$ and $`_\beta :\beta <\alpha N`$.
Then there is a standard reasonable parameter $`𝔭`$ with $`\mathrm{}g(𝔭)=\alpha ^{},\chi _\alpha ^𝔭=\chi _\alpha ,_\alpha ^𝔭=_\alpha `$ and $`R_\alpha ^𝔭=R_\alpha `$.
2. If in addition clause (d) below holds, then $`𝔭`$ is a simple standard reasonable parameter (recall Definition 3.5(4)), where
1. $`\beta N_\alpha ,\beta <\alpha N(\chi _\beta )_\beta `$.
3. If $`\chi _\alpha =(\mathrm{}_{2\alpha +1})^+`$ for $`\alpha <\alpha ^{},R_\alpha (\chi _\alpha )`$, then $`\chi _\alpha `$ increases fast enough. If $`\chi _\alpha :\alpha <\alpha ^{},R_\alpha :\alpha <\alpha ^{}`$ are as in part (1), $`\chi \chi _0,[(\chi )]^\mathrm{}_0`$ stationary and we let $`_\alpha =\{N:N`$ is a countable elementary submodel of $`((\chi _\alpha ),)`$ and $`\chi _\beta :\beta <\alpha ,R_\beta :\beta \alpha ,`$ belong to $`N`$ and $`N(\chi )\}`$, then the assumptions of parts (1) and (2) above hold.
###### Proof.
(1). Let $`𝔭=\overline{\chi },\overline{R},\overline{},\overline{D}`$, where $`\overline{\chi },\overline{R}`$ and $`\overline{}`$ are given as above and $`\overline{D}`$ is defined as in Definition 3.5(1). Then $`𝔭`$ is easily seen to be a standard reasonable parameter as required.
Items (2) and (3) are clear. ∎
We now define an extension of the games $`\mathrm{}_\alpha (N,𝔭)`$ and $`\mathrm{}_\alpha (N,N^{},𝔭)`$ given in Definition 3.6.
###### Definition 4.2.
Let $`𝔭`$ be a reasonable parameter and $`\alpha \beta <\mathrm{}g(𝔭)`$.
1. For $`N_\beta ^𝔭`$ such that $`\alpha N`$, we define a game $`\mathrm{}_{\alpha ,\beta }(N)=\mathrm{}_{\alpha ,\beta }(N,𝔭)`$ of length $`\omega `$ as follows. In the $`n`$-th move:
1. the challenger chooses $`X_nD_\beta ^𝔭(N)`$ such that $`m<nX_nX_m`$,
2. the chooser chooses $`\alpha _n\alpha N`$,
3. the challenger chooses $`\beta _n^{}\beta N`$ and $`y_n^{}N(\chi _{\alpha _n}^𝔭)`$,
4. the chooser chooses $`\beta _n\beta N\backslash \beta _n^{}`$ together with $`M_nX_n_{\beta _n}^𝔭`$, $`y_nM_n(\chi _{\alpha _n}^𝔭)`$ and $`Y_nD_{\beta _n}^𝔭(M_n)`$ satisfying:
* $`\alpha _n\beta _n,`$
* $`y_n^{}M_n,`$
* $`\alpha _nM_n`$,
* $`Y_nX_n`$ and $`Y_nN`$
5. the challenger chooses $`M_n^{}Y_n_{\alpha _n}^𝔭\left(M_n\{M_n(\chi _{\alpha _n}^𝔭)\}\right)`$ satisfying $`y_n,y_n^{}M_n^{}`$ and chooses $`Z_nD_{\alpha _n}^𝔭(M_n^{})=D_{\alpha _n}^𝔭(M_n^{}(\chi _{\alpha _n}^𝔭))`$ such that $`Z_nY_n`$.
At the end, the chooser wins the play if
$$\{Z_n\{M_n^{}\}:n<\omega \}D_\alpha ^+(N)=D_\alpha ^+(N(\chi _\alpha ^𝔭)),$$
where $`D_\alpha =D_\alpha ^𝔭`$.
2. We call $`\mathrm{}_{\alpha ,\beta }(N)=\mathrm{}_{\alpha ,\beta }(N,𝔭)`$, defined in clause (1), version 1 of the game. Version 2 of the game is defined similarly, where
* in clause (d) we require $`\alpha _n_𝔭\beta _n`$.
* in clause (e), we add the requirement $`M_n^{}=M_n(\chi _{\alpha _n}^𝔭)`$,
If we do not mention the version, it means that it holds for both versions.
3. Assume $`NN^{}((\chi ),)`$. We define the game $`\mathrm{}_{\alpha ,\beta }^{}(N,N^{},𝔭)`$ similarly, where items (a) - (e) are replaced by:
1. the challenger chooses $`X_nD_\beta ^𝔭(N)N^{}`$ such that $`m<nX_nX_m^{}`$,
2. the chooser chooses $`\alpha _n\alpha N`$ and $`X_n^{}X_n`$ such that $`X_n^{}D_\beta ^𝔭(N)N^{}`$,
3. like (c) above,
4. like (d) above, but we replace “$`Y_nN`$” by “$`Y_nN^{}`$”,
5. like (e) above, but add $`Z_nN^{}`$.
Note that every proper initial segment of a play belongs to $`N^{}`$.
###### Definition 4.3.
Let $`𝔭`$ be a reasonable parameter.
1. For $`\alpha \beta <\mathrm{}g(𝔭)`$, we say $`𝔭`$ is a $`\mathrm{}_{\alpha ,\beta }`$-winner (resp. non-$`\mathrm{}_{\alpha ,\beta }`$-loser), if for some $`x(\chi _\beta ^𝔭)`$ we have:
1. if $`\{x,\alpha \}N_\beta ^𝔭`$, then the chooser wins the game $`\mathrm{}_{\alpha ,\beta }(N,𝔭)`$
2. (resp. the challenger does not win in the game $`\mathrm{}_{\alpha ,\beta }(N,𝔭)`$).
2. Similarly we can define when $`𝔭`$ is a $`\mathrm{}_{\alpha ,\beta }^{}`$-winner (resp. non-$`\mathrm{}_{\alpha ,\beta }^{}`$-loser).
3. For any function $`f:\mathrm{}g(𝔭)𝒫(\mathrm{}g(𝔭))`$ we can replace $`\alpha ,\beta `$ by $`f`$, so that $`𝔭`$ is a $`\mathrm{}_f`$-winner (resp. non-$`\mathrm{}_f`$-loser) if for every $`\alpha <\mathrm{}g(𝔭)`$ and $`\beta f(\alpha )`$, $`𝔭`$ is a $`\mathrm{}_{\alpha ,\beta }`$-winner (resp. non-$`\mathrm{}_{\alpha ,\beta }`$-loser).
Given a reasonable parameter $`𝔭`$, we define some families of functions as follows.
###### Definition 4.4.
Let $`𝔭`$ be a reasonable parameter.
1. Let $`^𝔭`$ be the family of functions $`f`$ from $`\mathrm{}g(𝔭)`$ to $`𝒫(\mathrm{}g(𝔭))`$ such that for each $`\alpha <\mathrm{}g(𝔭),f(\alpha )`$ is a nonempty subset of $`[\alpha ,\mathrm{}g(𝔭))`$.
2. Let $`_{\mathrm{club}}^𝔭`$ be the set of $`f^𝔭`$ such that for each $`\alpha <\mathrm{}g(𝔭),f(\alpha )`$ is a club of $`\mathrm{}g(𝔭)`$.
3. Let $`_{\mathrm{nd}}^𝔭`$ be the set of $`f^𝔭`$ such that for each $`\alpha <\mathrm{}g(𝔭),f(\alpha )`$ is an end segment of $`\mathrm{}g(𝔭)`$, we then may identify $`f(\alpha )`$ with $`\mathrm{min}(f(\alpha ))`$.
4. Call $`f^𝔭`$ decreasing continuous if
1. $`\alpha <\beta <\mathrm{}g(𝔭)f(\alpha )f(\beta )`$,
2. for limit $`\delta <\mathrm{}g(𝔭)`$ we have $`f(\delta )=\{f(\alpha ):\alpha <\delta \}`$.
Let also $`fg`$ mean that $`(\alpha <\mathrm{}g(𝔭))(g(\alpha )f(\alpha ))`$.
5. Let $`_{\mathrm{dc}}^𝔭`$ be the set of decreasing continuous functions $`f_{\mathrm{club}}^𝔭`$.
We are interested in dealing with winning strategies for the games $`\mathrm{}_f`$, where $`f`$ is a function coming from one of the above family of functions. The next lemma gives some obvious monotonicity properties for these classes of functions.
###### Lemma 4.5.
Assume $`𝔭`$ is a reasonable parameter.
1. If $`\alpha _𝔭\alpha ^{}\beta =\beta ^{}<\mathrm{}g(𝔭)`$, and $`𝔭`$ is a $`\mathrm{}_{\alpha ,\beta }`$-winner, then it is $`\mathrm{}_{\alpha ^{},\beta ^{}}`$-winner. Similarly for $`\mathrm{}^{}`$-winner, non-$`\mathrm{}`$-loser and non-$`\mathrm{}^{}`$-loser.
2. If $`𝔭`$ is a $`\mathrm{}_{\alpha ,\beta }`$-winner, then $`𝔭`$ is a $`\mathrm{}_{\alpha ,\beta }^{}`$-winner and a non-$`\mathrm{}_{\alpha ,\beta }`$-loser. If $`𝔭`$ is a $`\mathrm{}_{\alpha ,\beta }^{}`$-winner or non-$`\mathrm{}_{\alpha ,\beta }`$-loser, then $`𝔭`$ is non-$`\mathrm{}_{\alpha ,\beta }^{}`$-loser.
3. Assume $`f,g^𝔭`$ and $`fg`$. If $`𝔭`$ is a $`\mathrm{}_f`$-winner (or $`\mathrm{}_f^{}`$-winner) (or non-$`\mathrm{}_f`$-loser) (or non-$`\mathrm{}_f^{}`$-loser), then $`𝔭`$ is a $`\mathrm{}_g`$-winner (or $`\mathrm{}_g^{}`$-winner) (or non-$`\mathrm{}_g`$-loser) (or non-$`\mathrm{}_g^{}`$-loser).
###### Proof.
(1) Suppose $`\sigma `$ is a winning strategy for chooser in the game $`\mathrm{}_{\alpha ,\beta }`$. We show that it is a winning strategy for chooser in the game $`\mathrm{}_{\alpha ^{},\beta ^{}}`$ as well. Suppose not, so, following the notation of Definition 4.2,
$$\{Z_n\{M_n^{}\}:n<\omega \}D_\alpha ^{}^+(N)(=D_\alpha ^{}^+(N(\chi _\alpha ^{}^𝔭))).$$
It then follows that
$$N(\chi _\alpha ^{}^𝔭)\{Z_n\{M_n^{}\}:n<\omega \}D_\alpha ^{}(N).$$
But then, as $`\alpha _𝔭\alpha ^{},`$
$$N(\chi _\alpha ^𝔭)\{Z_n\{M_n^{}\}:n<\omega \}D_\alpha (N).$$
But, as $`\sigma `$ is a winning strategy for chooser in the game $`\mathrm{}_{\alpha ,\beta }`$, we have
$$\{Z_n\{M_n^{}\}:n<\omega \}D_\alpha ^+(N),$$
a contradiction.
The proof of clause (2) is similar to the proof of Lemma 3.7(2) and the proof of clause (3) is straightforward. ∎
###### Lemma 4.6.
1. Assume $`𝔭`$ is a standard reasonable parameter. Then $`𝔭`$ is a winner.
2. If $`𝔭`$ is a reasonable parameter and it is a winner, then $`𝔭`$ is a $`\mathrm{}_{\alpha ,\alpha }`$-winner.
3. If $`𝔭`$ is a reasonable parameter and it is a winner and $`\alpha \beta <\mathrm{}g(𝔭)`$, then $`𝔭`$ is a $`\mathrm{}_{\alpha ,\beta }`$-winner (hence $`\mathrm{}_f`$-winner for any $`f:\mathrm{}g(𝔭)𝒫(\mathrm{}g(𝔭))`$).
4. Similarly with $`\mathrm{}^{}`$-winner, $`\mathrm{}_{\alpha ,\beta }^{}`$ winner and/or with the “non-loser” cases.
###### Proof.
Clause (1) follows from Lemma 3.7. The proof of items (2)-(4) is easy and follows from the Definitions 3.6 and 4.2. ∎
We now define an interpretation of reasonable parameters in the forcing extensions and show that these interpretations are reasonable parameters in the corresponding extension.
###### Definition 4.7.
Let $`𝔭`$ be a reasonable parameter and let $``$ be a proper forcing notions which adds no new reals. Suppose that $`𝒫()(\chi _0^𝔭)`$ and $`𝔾_{}`$ is generic over $`𝕍`$. We interpret $`𝔭`$ in $`𝕍^{}`$ as $`𝔭^{}=𝔭^{𝕍[𝔾_{}]}`$, defined as follows:
1. $`\chi _\alpha ^𝔭^{}=\chi _\alpha ^𝔭`$,
2. $`R_\alpha ^𝔭^{}=R_\alpha ^𝔭,,𝔾_P`$,
3. $`_\alpha ^𝔭^{}=\{N[𝔾_P]:N_\alpha ^𝔭`$, $`N`$ and $`N[𝔾_{}]V=N\}`$,
4. $`D_\alpha ^𝔭^{}(N[𝔾_{}])=\{\{M[𝔾_{}]_\alpha ^𝔭^{}:MY\underset{\beta <\alpha }{}_\beta ^𝔭\}:YD_\alpha ^𝔭(N)\}`$.
We also use $`𝔭^{}`$ for $`𝔭^{}=𝔭^{𝕍[𝔾_{}]}`$.
The proof of the next lemma is straightforward.
###### Lemma 4.8.
Let $`𝔭,`$ and $`𝔾_{}`$ be as in Definition 4.7.
1. $`𝔭^{𝕍[𝔾_{}]}`$ is a reasonable parameter in $`𝕍[𝔾_{}]`$.
2. If $`𝔭`$ is, in $`𝕍`$, a $`\mathrm{}`$-winner (or non-$`\mathrm{}`$-loser or $`\mathrm{}^{}`$-winner or non-$`\mathrm{}^{}`$-loser), then $`𝔭^{𝕍[𝔾_{}]}`$ is so in $`𝕍[𝔾_{}]`$.
3. If $`𝔭`$ is, in $`𝕍`$, a $`\mathrm{}_{\alpha ,\beta }`$-winner (or non- $`\mathrm{}_{\alpha ,\beta }`$-loser or $`\mathrm{}_{\alpha ,\beta }^{}`$-winner or non-$`\mathrm{}_{\alpha ,\beta }^{}`$-loser), then $`𝔭^{𝕍[𝔾_{}]}`$ is so in $`𝕍[𝔾_p]`$.
###### Proof.
(1). It is easily seen, using the fact that $``$ is an NNR proper forcing notion, that $`𝔭^{𝕍[𝔾_{}]}`$ satisfies items (a)-(i) of Definition 3.2 in $`𝕍[𝔾_{}]`$.
(2). Suppose $`𝔭`$ is a winner in $`𝕍`$. We show that $`𝔭^{}=𝔭^{𝕍[𝔾_{}]}`$ is a winner in $`𝕍[𝔾_{}]`$. Suppose $`\alpha <\mathrm{}g(𝔭)`$ and $`N^{}_\alpha ^𝔭^{}.`$ Then for some $`N_\alpha ^𝔭,N^{}=N[𝔾_{}]`$. Let $`\sigma `$ be a winning strategy for $`𝔭`$ for the game $`\mathrm{}_\alpha (N,𝔭)`$. We define the winning strategy $`\sigma ^{}`$ for the game $`\mathrm{}_\alpha (N^{},𝔭^{})`$ in $`𝕍[𝔾_{}]`$ as follows. Following the notation of Definition 3.6, in the $`n`$-th move, challenger chooses some $`X_n^{}D_\alpha ^𝔭^{}(N^{})`$. So for some $`X_nD_\alpha ^𝔭(N)`$, we have $`X_n^{}=\{M[𝔾_{}]_\alpha ^𝔭^{}:MX_n\underset{\beta <\alpha }{}_\beta ^𝔭\}`$. Let $`M_nX_n`$ and $`Y_nM_nX_n`$ with $`Y_nD^𝔭(M_n)N`$ be the play that chooser does using the strategy $`\sigma `$. Let $`M_n^{}=M_n[𝔾_{}]`$ and $`Y_n^{}=\{M[𝔾_{}]:MY_n\}`$ be what chooser replies via $`\sigma ^{}.`$ Then challenger chooses some $`Z_n^{}Y_n^{}`$ with $`Z_n^{}D^𝔭^{}(M_n^{})`$. Thus for some $`Z_nY_n`$ with $`Z_nD^𝔭(M_n)𝕍,`$ we have $`Z_n^{}=\{M[𝔾_{}]:MZ_n\}.`$ Since $`\sigma `$ is a winning strategy, $`\{\{M_n\}Z_n:n<\omega \}D^𝔭(N).`$ It then follows that
$$\{\{M_n^{}\}Z_n^{}:n<\omega \}D^𝔭^{}(N^{}).$$
Thus $`\sigma ^{}`$ is a winning strategy for chooser for the game $`\mathrm{}_\alpha (N^{},𝔭^{})`$. The proof for non-$`\mathrm{}`$-loser or $`\mathrm{}^{}`$-winner or non-$`\mathrm{}^{}`$-loser is the same.
Clause (3) can be proved similarly. ∎
###### Definition 4.9.
Let $`𝔭`$ be a reasonable parameter and let $``$ be a forcing notion.
1. For $`\alpha \beta <\mathrm{}g(𝔭)`$, we say $``$ is $`(𝔭,\alpha ,\beta )`$-proper if $`𝒫()(\chi _0^𝔭)`$ and:
1. for some $`x(\chi _\beta ^𝔭)`$, if $`N_\beta ^𝔭,\{x,,\alpha \}N,pN`$ and $`YD_\alpha ^𝔭(N)`$, then for some $`q`$ we have:
1. $`pq`$,
2. $`q`$ is $`(N,)`$-generic,
3. for some $`N^{}(_\alpha ^𝔭NY)\{N(\chi _\alpha ^𝔭)\}`$ satisfying $`\alpha =\beta N^{}=N`$ we have q``[𝔾
~
,N,y]YDα𝔭(N)"subscriptforces𝑞``subscriptsubscript𝔾
~
superscript𝑁superscript𝑦𝑌subscriptsuperscript𝐷𝔭𝛼superscript𝑁"q\Vdash_{{\mathbb{Q}}}``{{\mathscr{M}}}_{{\mathbb{Q}}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{Q}}},N^{\prime},y^{*}]\cap Y\in D^{{\mathfrak{p}}}_{\alpha}(N^{\prime})" where $`y^{}=x,p,`$. Note that this implies $`q`$ is $`(N^{},)`$-generic.
We call the above, version 1 of $`(𝔭,\alpha ,\beta )`$-properness. Version 2 of $`(𝔭,\alpha ,\beta )`$-properness is defined similarly, but we demand $`N^{}=N(\chi _\alpha ^𝔭)`$ and $`\alpha _𝔭\beta `$.
2. We say $``$ is $`(𝔭,f)`$-proper, where $`f^𝔭`$ (see Definition 4.4), when for every $`\alpha <\mathrm{}g(𝔭)`$ and $`\beta f(\alpha )`$, the forcing notion $``$ is $`(𝔭,\alpha ,\beta )`$-proper.
3. We say $``$ is $`𝔭`$-proper if $``$ is $`(𝔭,\alpha ,\alpha )`$-proper for $`\alpha <\mathrm{}g(𝔭)`$. We say $``$ is almost $`𝔭`$-proper if $``$ is $`(𝔭,f)`$-proper for some $`f^𝔭`$.
The next lemma shows some relations between the above defined notions.
###### Lemma 4.10.
Assume $`𝔭`$ is a simple reasonable parameter.
1. If $`\alpha ^{}\alpha \beta \beta ^{}<\mathrm{}g(𝔭)`$ (for version 2 we demand $`\alpha ^{}_𝔭\beta ^{}`$ and $`\alpha _𝔭\beta `$) and $``$ is a $`(𝔭,\alpha ,\beta )`$-proper forcing notion, then $``$ is a $`(𝔭,\alpha ^{},\beta ^{})`$-proper forcing notion.
2. Assume $`f,f^{}`$ are in $`^𝔭`$ and $`ff^{}`$. If $``$ is a $`(𝔭,f)`$-proper forcing notion, then $``$ is a $`(𝔭,f^{})`$-proper forcing notion.
###### Proof.
(1) Suppose $``$ is $`(𝔭,\alpha ,\beta )`$-proper as witnessed by $`x(\chi _\beta ^𝔭)(\chi _\beta ^{}^𝔭)`$. We show that $`x`$ witnesses that $``$ is $`(𝔭,\alpha ^{},\beta ^{})`$-proper as well. Thus let $`N_\beta ^{}^𝔭`$ with $`\{x,,\alpha \}N,pN`$ and $`YD_\alpha ^{}^𝔭(N)`$. Then $`N(\chi _\beta ^𝔭)_\beta ^𝔭`$. Let $`qp`$ be as in Definition 4.9 and it witnesses that $``$ is $`(𝔭,\alpha ,\beta )`$-proper with respect to $`N(\chi _\beta ^𝔭),p`$ and $`Y`$. Then $`q`$ witnesses $``$ is $`(𝔭,\alpha ^{},\beta ^{})`$-proper with respect to $`N,p`$ and $`Y`$.
(2) is clear, as for every $`\alpha <\mathrm{}g(𝔭)`$, $`f^{}(\alpha )f(\alpha )`$. ∎
It follows from the above lemma that if $`𝔭`$ is a simple reasonable parameter and $`f^𝔭`$, then
$`𝔭`$-proper $`(𝔭,f)`$-proper $``$ almost $`𝔭`$-proper.
We may like to consider $`(𝔭,f)`$-properness for iterations which may add reals. Then we have to replace $`D_\alpha ^𝔭(N)`$ by a definition which is absolute enough (and the non-loser versions have to be absolute enough as well). In such situations, it is natural to restrict ourselves to those sets $`Y`$ which are $`𝔭`$-closed, see below.
###### Definition 4.11.
Let $`𝔭`$ be a simple reasonable parameter, $`\alpha <\mathrm{}g(𝔭)`$ and $`N_\alpha ^𝔭`$. A subset $`YN`$ is called $`𝔭`$-closed if:
1. $`YN\underset{\beta <\alpha }{}_\beta ^𝔭`$,
2. if $`MN_\beta ^𝔭`$, $`\beta <\alpha `$ (hence $`\beta \alpha M\alpha N`$), $`\gamma M\beta `$ and $`M(\chi _\gamma ^𝔭)_\gamma ^𝔭`$,<sup>7</sup><sup>7</sup>7note that this requirement is redundant if $`𝔭`$ is simple or just $`\gamma _𝔭\beta `$. then
$$M(\chi _\gamma ^𝔭)YMY,$$
3. if $`\beta <\alpha ,M_{\mathrm{}}N_\beta ^𝔭`$ (hence $`\beta \alpha N`$), $`M_{\mathrm{}}M_{\mathrm{}+1}`$ for $`\mathrm{}<\omega `$ and $`M=\underset{\mathrm{}<\omega }{}M_{\mathrm{}}N_\beta ^𝔭`$, and even $`M_{\mathrm{}}:\mathrm{}<\omega N`$, then
$$\left(\underset{\mathrm{}<\omega }{}M_{\mathrm{}}Y\right)MY.$$
###### Definition 4.12.
Assume $`𝔭`$ is a reasonable parameter, $``$ is a forcing notion and $``$ $`\stackrel{~}{}`$ is a $``$-name of a forcing notion.
1. For $`\alpha \beta <\mathrm{}g(𝔭)`$, we say $``$ $`\stackrel{~}{}`$ has $`(\kappa ,\alpha ,\beta )`$-anti w.d. above $``$ (or (,
~
)
~
({\mathbb{P}},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}) has $`(\kappa ,\alpha ,\beta )`$-anti-w.d.), if clause (A) implies clause (B), where:
1. $`(a)N_0_\alpha ^𝔭`$ and $`N_1_\beta ^𝔭`$,
2. $`(b)N_0N_1`$ and {,
~
}N0
~
subscript𝑁0\{{\mathbb{P}},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\}\in N_{0},
3. $`(c)n<1+\kappa `$,
4. $`(d)p_{\mathrm{}}`$ is $`(N_\iota ,)`$-generic for $`\mathrm{}<n`$ and $`\iota =0,1`$,
5. (e)p``𝔾
~
N1=𝔾"forces𝑒subscript𝑝``subscript𝔾
~
subscript𝑁1superscript𝔾"(e)\quad p_{\ell}\Vdash``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}}\cap N_{1}=\mathbb{G}^{\ell}" for $`\mathrm{}<n`$,
6. $`(f)𝔾^{\mathrm{}}N_0=𝔾^{}`$ for $`\mathrm{}<n`$,
7. $`(g)Y=\underset{\mathrm{}<n}{}_{}[𝔾^{\mathrm{}},N_1,y]`$ belongs to $`D_\beta ^𝔭(N_1)`$, where y=N0,,
~
,𝑦subscript𝑁0
~
y=\langle N_{0},{\mathbb{P}},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\rangle,
8. (h)q
~
𝑞
~
(h)\quad\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}} is a $``$-name of a member of $``$ $`\stackrel{~}{}`$ ,
9. (i)q
~
N0𝑖𝑞
~
subscript𝑁0(i)\quad\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\in N_{0},
1. there is a triple (p:<n,q
~
,𝔾)(\langle p^{\prime}_{\ell}:\ell<n\rangle,\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{\prime},\mathbb{G}^{**}) such that:
2. (a)q
~
𝑎superscript𝑞
~
(a)\quad\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{\prime} is a $``$-name of a member of $``$ $`\stackrel{~}{}`$ ,
3. (b)p``q
~
q
~
forces𝑏subscriptsuperscript𝑝``𝑞
~
superscript𝑞
~
(b)\quad p^{\prime}_{\ell}\Vdash``\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\leq\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{\prime}” for $`\mathrm{}<n`$,
4. (c)𝔾Gen(N0,
~
),𝑐superscript𝔾absentGensubscript𝑁0
~
(c)\quad\mathbb{G}^{**}\in{\rm Gen}(N_{0},{\mathbb{P}}*\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}),
5. (d)(p,q
~
)``𝔾
~
N0=𝔾"forces𝑑subscriptsuperscript𝑝superscript𝑞
~
``subscript𝔾
~
subscript𝑁0superscript𝔾absent"(d)\quad(p^{\prime}_{\ell},\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{\prime})\Vdash``\mathbb{G}_{{\mathbb{P}}*\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.60275pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.60275pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.60275pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.60275pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}}\cap N_{0}=\mathbb{G}^{**}" for $`\mathrm{}<n`$.
2. Given a function $`f^𝔭`$, we say $``$ $`\stackrel{~}{}`$ has $`(\kappa ,f)`$-anti w.d. above $``$ (or (,
~
)
~
({\mathbb{P}},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}) has $`(\kappa ,f)`$-anti-w.d.), if for every $`\alpha <\mathrm{}g(𝔭)`$, $``$ $`\stackrel{~}{}`$ has $`(\kappa ,\alpha ,f(\alpha ))`$-anti w.d. above $``$.
###### Remark 4.13.
We may assume that the conditions $`p_{\mathrm{}},\mathrm{}<n,`$ in clause (1)(A)(d) of the above definition are pairwise incompatible. If $`p_{\mathrm{}}`$ and $`p_\iota `$ are compatible, then by clause (1)(A)(e), $`C^{\mathrm{}}=C^\iota ,`$ so we may replace $`p_{\mathrm{}},p_\iota `$ by a common extension $`p_{\mathrm{},\iota }`$ of them and take $`C^{\mathrm{},\iota }=C^{\mathrm{}}.`$
As the sets $`(\chi _\alpha )`$ may change with forcing, we may prefer to use $`_\alpha [\chi _\alpha ]^\mathrm{}_0`$. For this reason, we define the notion of ordinal based parameter, and we will show that any ordinal based parameter gives naturally a reasonable parameter.
###### Definition 4.14.
1. We call $`𝔭`$ an o.b. (ordinal based) parameter if
$$𝔭=(\overline{\chi }^𝔭,\overline{R}^𝔭,\overline{}^𝔭,\overline{D}^𝔭)(=(\overline{\chi },\overline{R},\overline{},\overline{D})),$$
where for some ordinal $`\alpha ^{}`$, called $`\mathrm{}g(𝔭)`$, we have:
1. $`\overline{\chi }=\chi _\alpha :\alpha <\alpha ^{},`$ where $`\chi _\alpha `$ is a regular cardinal and $`((\underset{\beta <\alpha }{}\chi _\beta )^+)(\chi _\alpha )`$,
2. $`\overline{R}=R_\alpha :\alpha <\alpha ^{},`$ where $`R_\alpha `$ is an $`n(R_\alpha )`$-place relation on some bounded subset of $`\chi _\alpha `$, <sup>8</sup><sup>8</sup>8we could have asked “on $`\chi _\alpha `$”, as there is no real difference.
3. $`\overline{}=_\alpha :\alpha <\alpha ^{}`$, where $`_\alpha [\chi _\alpha ]^\mathrm{}_0`$ is stationary,
4. $`\overline{D}=D_\alpha :\alpha <\alpha ^{}`$ and $`D_\alpha `$ is a function with domain $`_\alpha `$ and for each $`a_\alpha ,D_\alpha (a)`$ is a pseudo-filter on $`a`$,
5. let $`𝔭^{[\alpha ]}=\overline{\chi }\alpha ,\overline{R}(\alpha +1),\overline{}\alpha ,\overline{D}\alpha `$,
6. if $`a_\alpha `$ and $`Xa`$, then
$$XD_\alpha (a)X\underset{\beta <\alpha }{}_\beta D_\alpha (a).$$
2. An o.b. parameter $`𝔭`$ is simple if in addition, it satisfies:
1. if $`a_\alpha `$, $`XD_\alpha (a)`$ and $`\beta \alpha a`$, then $`a\chi _\beta _\beta `$.
3. For $`𝔭`$ as above let $`𝔮=𝔭^𝕍`$ be defined by
* $`\mathrm{}g(𝔮)=\mathrm{}g(𝔭),`$
and we define by induction on $`\alpha <\mathrm{}g(𝔭)`$
* $`\chi _\alpha ^𝔮=\chi _\alpha ^𝔭,`$
* $`R_\alpha ^𝔮=R_\alpha ^𝔭,`$
* $`_\alpha ^𝔮=\{N((\chi _\alpha ^𝔮),):N\text{ is countable},N\chi _\alpha ^𝔭_\alpha ^𝔭`$ and $`𝔮^{[\alpha ]}N\}`$. Note that $`𝔮^{[\alpha ]}`$ is well defined by the induction hypothesis.
* $`D_\alpha ^𝔮(N)`$ is defined as
$$\begin{array}{cc}D_\alpha ^𝔮(N)=\{Y^{}:\hfill & Y^{}\underset{\beta <\alpha }{}_\beta ^𝔮\text{ and for some }yN\underset{\beta <\alpha }{}(\chi _\beta ^𝔭)\hfill \\ & \text{ and }YD_\alpha ^𝔭(N\chi _\alpha ^𝔭)\text{ we have }Y^{}\{M:M\hfill \\ & N\underset{\beta <\alpha }{}_\beta ^𝔮\text{ and }yM\text{ and }M\chi _\alpha ^𝔭Y\}\}.\hfill \end{array}$$
4. For an o.b. parameter $`𝔭`$, we say $`\overline{}`$ is an $`NNR_\kappa ^0`$-iteration for $`𝔭`$ if it is an $`NNR_\kappa ^0`$-iteration for $`𝔭^𝕍`$. We say $`𝔭`$ is simple if $`𝔭^𝕍`$ is. Similarly for $`\mathrm{}`$-winner, non-$`\mathrm{}`$-loser, etc.
The next lemma shows that each o.b. parameter leads to a canonical reasonable parameter, namely $`𝔭^𝕍`$, and also shows the advantage of working with o.b. parameters than reasonable parameters.
###### Lemma 4.15.
Assume $`𝔭`$ is an o.b. (simple) parameter in the universe $`𝕍`$.
1. $`𝔭^𝕍`$ is a (simple) reasonable parameter.
2. If $`(\chi _0^𝔭)`$ is a proper forcing notion (or at least it preserves the stationarity of $`_\alpha ^𝔭`$, for each $`\alpha <\mathrm{}g(𝔭)`$), then $`_{}\mathrm{`}\mathrm{`}𝔭`$ is an o.b. (simple) parameter”.
3. If forcing with $``$ adds no reals, then also $`\mathrm{}`$-winner, non-$`\mathrm{}`$-loser, etc., are preserved.
###### Proof.
Straightforward. ∎
###### Definition 4.16.
Let $`𝔭`$ be an o.b. parameter.
1. We say $`𝕍`$ is an $`NNR_\kappa ^0`$-forcing for $`𝔭`$ or is a $`𝔭NNR_\kappa ^0`$-forcing notion, when the following holds:
* if for some transitive class $`𝕍_0`$ with $`𝔭𝕍_0`$ and some $`NNR_\kappa ^0`$-iteration ¯=i,
~
i:i<α𝕍0\bar{{\mathbb{Q}}}=\langle{\mathbb{P}}_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}:i<\alpha\rangle\in\mathbb{V}_{0} we have $`𝕍_0^{\mathrm{Lim}(\overline{})}=𝕍`$, then we can let $`_\alpha =\mathrm{Lim}(\overline{}),`$
~
α=subscript
~
𝛼\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}={\mathbb{Q}} and get an $`NNR_\kappa ^0`$-iteration ¯=i,
~
i:i<α+1\bar{{\mathbb{Q}}}^{\prime}=\langle{\mathbb{P}}_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}:i<\alpha+1\rangle. (i.e. $`𝕍=𝕍_0[𝔾_\alpha ],𝔾_\alpha _\alpha `$ is generic over $`𝕍_0`$ and there is a $`_\alpha `$-name
~
αsubscript
~
𝛼\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha} such that i,
~
i:iαdelimited-⟨⟩:subscript𝑖subscript
~
𝑖𝑖𝛼\langle{\mathbb{P}}_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}:i\leq\alpha\rangle is an $`NNR_\kappa ^0`$-iteration and =
~
α[𝔾α]subscript
~
𝛼delimited-[]subscript𝔾𝛼{\mathbb{Q}}=\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}[\mathbb{G}_{\alpha}]). In particular $``$ is proper and does not add reals.
2. If we omit “for $`𝔭`$” we mean for any $`𝔭`$ which makes sense. Alternatively, we can put a family of $`𝔭`$’s.
3. We add “over $`x`$” if this holds whenever $`x𝕍_0`$. We can use the same definition for other versions of NNR.
In the next section we will present several examples of forcing notions that fit into the above definition, in the sense that they are $`𝔭NNR_\kappa ^0`$ for some $`2\kappa \mathrm{}_0`$ and some reasonable parameter $`𝔭.`$
## 5. Examples: shooting thin clubs
In this section we present some examples that fit into our framework. We already know that $`(<\omega _1)`$-proper forcing notions are $`𝔭`$-proper for standard reasonable parameter $`𝔭`$ of length $`\omega _1`$ (by Lemma 3.17). We first deal with a forcing notion which is $`\text{NNR}_\kappa ^0`$-proper, for some $`\kappa <\mathrm{}_0`$. Second, we deal with shooting clubs of $`\omega _1`$ running away from some $`C_\delta \delta =sup(C_\delta )`$ which are small (see \[6, Ch.XVIII, §1\]). These are the most natural non-$`\omega `$-proper forcing notions which do not add reals.
###### Definition 5.1.
1. Let $`\overline{C}=(C_\delta ,n_\delta ):\delta <\omega _1,\delta \text{ limit}`$, where $`C_\delta `$ is an unbounded subset of $`\delta `$ of order type $`\omega `$ and $`1n_\delta <\omega `$. Let $`\overline{u}=u_\delta :\delta <\omega _1,\delta \text{ limit}`$, where $`u_\delta [2n_\delta +1]^{n_\delta }`$. Then we define $`=_{\overline{C},\overline{u}}`$ by
$$\begin{array}{cccc}_{\overline{C},\overline{u}}=\{f:& \text{for some }\alpha <\omega _1,f\text{ is a function from }\alpha \text{ to }\hfill & & \\ & \omega \text{ such that for every limit ordinal }\delta \alpha ,\hfill & & \\ & \text{ for some }k<2n_\delta +1,ku_\delta \text{ and for every}\hfill & & \\ & iC_\delta \text{ large enough we have }f(i)=k\}.\hfill & & \end{array}$$
$`_{\overline{C},\overline{u}}`$ is ordered by inclusion.
2. Assume $`\overline{C}=C_\delta :\delta <\omega _1,\delta \text{ limit},C_\delta `$ a closed subset of $`\delta `$ of order type less than $`\omega \delta `$ and for $`\delta _1<\delta _2`$ limit ordinals, $`sup(C_{\delta _1}C_{\delta _2})<\delta _1`$, and for limit $`\delta ^{}`$ we have $`\{C_\delta \delta ^{}:\delta <\omega _1\}`$ is countable. Assume further that $`\overline{\kappa }=\kappa _\delta :\delta <\omega _1,\delta \text{ limit}`$ with $`\kappa _\delta \{2,3,\mathrm{},\mathrm{}_0\}`$ and $`\overline{D}=D_\delta :\delta <\omega _1,D_\delta `$ is a family of subsets of $`\mathrm{dom}(D_\delta )`$, such that the intersection of any $`<\kappa _\delta `$ of them is non-empty.
Let $`\overline{f}=f_{\delta ,x}:\delta <\omega _1,x\mathrm{dom}(D_\delta )`$ satisfy $`f_{\delta ,x}:C_\delta \omega `$ and $`\overline{A}=A_\delta :\delta <\omega _1`$ satisfy $`A_\delta D_\delta `$. Then we define $`=_{\overline{C},\overline{D},\overline{\kappa },\overline{f},\overline{A}}`$ as
$$\begin{array}{cccc}_{\overline{C},\overline{D},\overline{\kappa },\overline{f},\overline{A}}=\{f:& \text{ for some }\alpha <\omega _1,f\text{ is a function from }\alpha \text{ to }\omega \hfill & & \\ & \text{ such that for every limit }\delta \alpha \text{ and for some}\hfill & & \\ & xA_\delta \text{ we have }f_{\delta ,x}^{}f\text{ i.e. for every large }\hfill & & \\ & \text{ enough }iC_\delta \text{ we have }f_{\delta ,x}(i)=f(i)\}.\hfill & & \end{array}$$
$`_{\overline{C},\overline{D},\overline{\kappa },\overline{f},\overline{A}}`$ is ordered by inclusion.
###### Lemma 5.2.
1. The forcing notion $`=_{\overline{C},\overline{u}}`$ from Definition 5.1(1) is proper, does not add reals, and is $`(<\omega _1)`$-proper. It is also $`𝔻`$-complete for some simple 2-completeness system, hence
1. if $`\overline{}`$ is a countable support iteration, $`\mathrm{}g(\overline{})=\alpha +1,\overline{}\alpha `$ is NNR$`{}_{2}{}^{}{}_{}{}^{0}`$-iteration and α``
~
α=C¯,u¯",(C¯,u¯)formulae-sequencesubscriptforcessubscript𝛼absent``subscript
~
𝛼subscript¯𝐶¯𝑢"¯𝐶¯𝑢\Vdash_{{\mathbb{P}}_{\alpha}}``\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}={\mathbb{Q}}_{\bar{C},\bar{u}}",\,\,(\bar{C},\bar{u}) as above $`𝕍`$, then $`\overline{}`$ is an NNR$`{}_{2}{}^{}{}_{}{}^{0}`$-iteration.
2. Similarly for the forcing notion $`=_{\overline{C},\overline{D},\overline{\kappa },\overline{f},\overline{A}}`$ from Definition 5.1(2).
###### Proof.
We only prove (1), as (2) can be proved in a similar way. Set $`=_{\overline{C},\overline{u}}`$.
Let us start by showing that $``$ does not add reals. Thus suppose that $`f`$, $`t`$ $`\stackrel{~}{}`$ is a $``$-name and $`f`$t
~
:ω2:𝑡
~
𝜔2\mathchoice{\oalign{$\displaystyle t$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle t$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle t$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle t$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}:\omega\to 2 is a real”. Let $`\chi `$ be a large enough regular cardinal and let $`N_n:n<\omega `$ be a chain of countable elementary submodels of $`(H(\chi ),)`$ such that ,f,t
~
N0𝑓𝑡
~
subscript𝑁0{\mathbb{Q}},f,\mathchoice{\oalign{$\displaystyle t$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle t$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle t$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle t$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\in N_{0} and for each $`n<\omega ,N_nN_{n+1}`$. Set $`N=_{n<\omega }N_n`$ and $`\delta =N\omega _1`$. Also for each $`n<\omega `$ set $`\delta _n=N_n\omega _1.`$ Pick some $`k<2n_\delta +1,ku_\delta .`$ We define by induction on $`n<\omega `$ a sequence $`f_n:n<\omega `$ of conditions in $``$, such that:
1. $`f_nN_n,`$
2. $`f_0=f,`$
3. for some $`\delta _{n1}<\alpha _n<\delta _n,f_n:\alpha _n\omega `$, where $`\delta _1=0,`$
4. $`f_nf_{n+1}`$,
5. $`\alpha _nC_\delta ,`$
6. for all $`iC_\delta \alpha _n(\alpha _0+1),f_n(i)=k`$.
7. $`f_n`$ decides t
~
n𝑡
~
𝑛\mathchoice{\oalign{$\displaystyle t$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle t$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle t$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle t$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\restriction n.
The construction can be done quite easily, noting that for each $`n,C_\delta \delta _n`$ is bounded in $`\delta _n`$. Let $`f^{}=_{n<\omega }f_n.`$ Then $`f^{}`$ extends $`f`$ and it decides $`t`$ $`\stackrel{~}{}`$ .
To show that $``$ is $`<\omega _1`$-proper, fix $`\alpha <\omega _1`$ a limit ordinal and let $`\overline{N}=N_\xi :\xi <\alpha `$ be an increasing and continuous chain of countable elementary submodels of some $`(H(\chi ),)`$ such that $`\alpha ,N_0`$ and $`\overline{N}\xi +1N_{\xi +1}`$. For each $`\xi <\alpha `$ set $`\delta _\xi =N_\xi \omega _1.`$
Let $`fN_0.`$ By essentially the same argument as above, we can find a sequence $`f_\xi :\xi <\alpha `$ of conditions in $``$, with $`f_\xi :\beta _\xi \omega `$ such that:
1. $`f_0=f`$,
2. $`f_\xi f_{\xi +1}`$,
3. if $`\xi `$ is a limit ordinal, then $`f_\xi =_{\zeta <\xi }f_\zeta ,`$
4. $`\delta _\xi <\beta _{\xi +1}<\delta _{\xi +1}`$,
5. $`f_{\xi +1}N_{\xi +1}`$ is $`(,N_\xi )`$-generic,
6. $`f^{}=_{\xi <\alpha }f_\xi `$ is a condition.
Then $`f^{}f`$ is $`(,N_\xi )`$-generic for every $`\xi <\alpha .`$
Let us show that $``$ is $`𝔻`$-complete for some simple 2-completeness system. Let $`\theta `$ be large enough regular and let $`𝔻`$ be a function whose domain consists of those pairs $`(N,f)`$ where $`N(H(\theta ),)`$ is countable, $`N`$ and $`fN.`$ Now suppose that $`(N,f)\mathrm{dom}(𝔻)`$. Set
$$𝔻(N,f)=\{A_x:x\text{ is a finitary relation on }N\},$$
where for any such $`x`$,
$$A_x=\{𝔾\text{Gen}(N,,f):N𝒫(N)\mathrm{\Psi }(x,𝔾,N,,f)\}$$
and the formula $`\mathrm{\Psi }`$ says that “if $`x=(y,k),`$ where $`y`$ is an $`\omega `$-sequence cofinal in $`\delta =N\delta `$ and $`k<2n_\delta +1,ku_\delta `$, then $`𝔾y`$ is eventually equal to $`k`$.
Let us show that $`𝔻`$ is a simple 2-completeness system. Thus suppose that $`x_0,x_1`$ are given, and we have to show that $`A_{x_0}A_{x_1}`$ is non-empty. The only non-trivial case is when $`x_0,x_1`$ satisfy the hypotheses of the formula $`\mathrm{\Psi }.`$ Thus suppose that $`x_0=(y_0,k_0)`$ and $`x_1=(y_1,k_1)`$, where $`y_0,y_1`$ are $`\omega `$-sequences cofinal in $`\delta `$ and $`k_0,k_1<2n_\delta +1,k_0,k_1u_\delta `$. If $`y_0y_1`$ is cofinal in $`\delta `$, then we must have $`k_0=k_1`$ and so by the previous arguments we can find an increasing sequence $`f_n:n<\omega `$ of extensions of $`f`$ in $`N`$ such that $`f^{}=\underset{n<\omega }{}f_n`$ gives rise to a filter $`𝔾\text{Gen}(N,,f)`$ such that $`f^{}(y_0y_1)`$ is eventually equal to $`k_0`$ and we are done. Otherwise, $`y_0y_1`$ is bounded in $`\delta ,`$ so for some $`\eta <\delta ,y_0y_1\eta `$. By enlarging $`\eta `$ we may also assume that $`\mathrm{dom}(f)\eta .`$ Again, by similar arguments as above, it is not difficult to build an increasing sequence $`f_n:n<\omega `$ of extensions of $`f`$ in $`N`$ such that $`f^{}=\underset{n<\omega }{}f_n`$ gives rise to a filter $`𝔾\text{Gen}(N,,f)`$ such that $`f^{}(y_0\eta )`$ is eventually equal to $`k_0`$ and $`f^{}(y_1\eta )`$ is eventually equal to $`k_1`$. Then $`𝔾A_{x_0}A_{x_1}`$ and we are done.
It is now clear that $``$ is $`𝔻`$-complete, which completes the proof. ∎
###### Definition 5.3.
Assume $`\overline{C}=C_\delta :\delta <\omega _1,\delta \text{ limit }`$, where $`C_\delta `$ is an unbounded subset of the limit ordinal $`\delta `$ (think of the case $`C_\delta `$ of order type $`<\delta `$ but not necessarily). Let
$$\begin{array}{cccc}_{\overline{C}}=\{c:& \text{ for some }\alpha <\omega _1,c\text{ is a closed subset of }\alpha \hfill & & \\ & \text{ and for every limit ordinal }\delta \alpha \text{ we have}\hfill & & \\ & \delta =sup(c\delta )cC_\delta \text{ is bounded in }\delta \}.\hfill & & \end{array}$$
Order $`_{\overline{C}}`$ by
$$c_1<c_2c_1\text{ is an initial segment of }c_2.$$
###### Remark 5.4.
1. For more information about the above forcing notion see \[6, Ch.XVIII, §1, 1.9\]. Note that $`_{\overline{C}}`$ may be non-$`\omega `$-proper.
2. Note that $`c_{\overline{C}}`$ is a closed subset of $`\alpha `$, but not necessarily a closed subset of $`\omega _1`$.
In general the forcing notion $`_{\overline{C}}`$ might be trivial, say for example when for every limit ordinal $`\delta ,C_\delta =\delta .`$ We are interested in the cases that this forcing notion is non-trivial, and we first deal with the simple case of $`\mathrm{otp}(C_\delta )=\omega `$.
###### Lemma 5.5.
Assume $`𝔭`$ is a simple reasonable parameter, $`\overline{C}`$ is as in Definition 5.3 and $`\underset{\delta }{}\mathrm{otp}(C_\delta )=\omega `$. Let $`f^𝔭`$ be defined as $`f(0)=0`$ and $`f(\beta )=1+\beta `$ for $`\beta >0`$. Then $`_{\overline{C}}`$ is $`(𝔭,f)`$-proper.
In Lemma 5.6 we prove a stronger result, which includes the above lemma as a very special case. Note that if $`\underset{\delta }{}\mathrm{otp}(C_\delta )<\delta `$, then we can split the analysis by restricting ourselves to $`\{N:N\omega _1S_\gamma \}`$, where $`\gamma <\omega _1`$ is such that $`S_\gamma =\{\delta :\mathrm{otp}(C_\delta )=\gamma \}`$ is stationary.
###### Lemma 5.6.
1. Assume
1. $`𝔭`$ is a simple reasonable parameter,
2. $`f^𝔭`$ and $`𝔭`$ is non-$`\mathrm{}_f`$-loser,
3. $`\gamma ()<\omega _1`$,
4. $`\overline{C}=C_\delta :\delta <\omega _1,C_\delta \delta =sup(C_\delta )`$.
5. $`\mathrm{otp}(C_\delta )\omega ^{\gamma ()}`$ for every $`\delta `$,
Define $`g^𝔭`$ by recursion as
* $`g(0)=0,`$
* $`g(1)=f(1)+\gamma (),`$
* $`g(\alpha +1)=f(g(\alpha ))+\gamma ()+1`$, for $`\alpha >0,`$
* for limit ordinals $`\alpha ,g(\alpha )=\underset{\beta <\alpha }{sup}g(\beta )`$.
Then for every $`\alpha <\mathrm{}g(𝔭)`$, the forcing notion $`=_{\overline{C}}`$ is $`(𝔭,\alpha ,g(\alpha ))`$-proper (version 1).
2. In part (1), we can get “version 2” of $`(𝔭,\alpha ,g(\alpha ))`$-properness when the following is satisfied: if $`g(\delta )=\delta ,N_\alpha ^𝔭,C\omega _1N=sup(C),\mathrm{otp}(C)\omega ^{\gamma ()}`$ and $`YD_\alpha ^𝔭(N)`$, then $`Y^{}=\{MY:M\omega _1C\}D_\alpha ^𝔭(N)`$.
3. If we weaken clause $`(b)`$ to $`(b)_{f,g}`$, where
1. $`(b)_{f,g}:f^𝔭`$, $`f(f(\alpha ))=f(\alpha )`$ for every $`\alpha <\mathrm{}g(𝔭)`$ and $`𝔭`$ is a
2. non-$`\mathrm{}_f^{}`$-loser,
then for $`\alpha <\mathrm{}g(𝔭)`$, the forcing notion $`_{\overline{C}}`$ is $`(𝔭,\alpha ,f(\gamma ()+\alpha ))`$-proper.
###### Proof.
We only prove clause (2). Note that version 2 is harder to prove, and using the extra freedom, we can avoid the need for the extra assumption from (2) to prove (1). First observe that
1. for each $`\alpha <\omega _1`$ the set $`_\alpha ^{}=\{p_{\overline{C}}:\text{ there is }\beta p\text{ which is }\alpha \}`$ is an open dense subset of $`_{\overline{C}}`$.
We prove (2) by induction on $`\alpha `$. Let $`\beta =g(\alpha )`$. Let $`N_\beta ^𝔭`$ be countable with $`_{\overline{C}},\alpha ,\beta ,f,gN`$, $`pN_{\overline{C}}`$, and suppose $`YD_\beta ^𝔭(N)`$ is given. Let $`\delta =\delta _N=N\omega _1`$.
Case 1: $`\alpha =0`$. In this case, we are reduced to show that $`_{\overline{C}}`$ is proper. Let $`_n:n<\omega `$ list the dense open subsets of $`_{\overline{C}}`$ that belong to $`N`$. We shall choose by induction on $`n<\omega `$, a condition $`p_n`$ such that:
1. $`p_0=p,`$
2. for each $`n`$, $`p_nN`$,
3. $`p_np_{n+1}_n`$,
4. the set $`p_{n+1}\{supp_{n+1}\}\backslash (p_n\{supp_n\})`$ is disjoint from $`C_\delta `$.
Set $`p_0=p`$. Now assume $`p_n`$ has been chosen and we shall choose $`p_{n+1}`$ as requested. Let $`F_nN`$ be a function with domain $`_{\overline{C}}`$ such that for all $`q_{\overline{C}},qF_n(q)_n`$.
For $`\alpha <\omega _1`$ let $`q^{[\alpha ]}=q\{sup(q)\}\{sup(q)+1+\alpha \}`$, so clearly $`qq^{[\alpha ]}_{\overline{C}}`$ and the function $`(q,\alpha )q^{[\alpha ]}`$ belongs to $`N`$. Define a function $`H:\omega _1\omega _1`$ by $`H(\alpha )=sup(F_n(p_n^{[\alpha ]}))`$. Clearly it is well defined and belongs to $`N`$. Let
$$C=\{\beta <\omega _1:\beta \text{ a limit ordinal, }\omega \beta =\beta ,(\alpha <\beta )(H(\alpha )<\beta )\text{ and }sup(p_n)<\beta \}.$$
It is easily seen that $`C`$ is a club of $`\omega _1`$ which belongs to $`N`$ and $`\gamma ()N`$, hence we can find $`\beta ^{}C`$ such that $`\mathrm{otp}(\beta ^{}C)`$ is divisible by $`\omega ^{\gamma ()}`$. But $`\mathrm{otp}(C_\delta \beta ^{})<\omega ^{\gamma ()}`$, hence for some $`\beta C`$ we have $`sup(C_\delta \beta )<\beta `$. Let $`p_{n+1}=F_n(p^{[sup(C_\delta \beta )+1]})`$.
Set $`q=\underset{n<\omega }{}p_n`$. Then for each $`\alpha <N\omega _1,_\alpha ^{}\{_n:n<\omega \}`$, hence
$$\beta \left(\beta q\text{ and }\alpha \beta <\omega _1\right).$$
It follows that $`q`$ is $`(N,_{\overline{C}})`$-generic.
Case 2: $`\alpha =1`$. Set $`Y^{}=\{MY:M\omega _1C_\delta \}`$. It is clear that $`g(\delta )=\delta `$ (as $`gN`$ and $`\delta =N\omega _1`$), hence by the hypotheses in clause (2) we have $`Y^{}D_{g(1)}^𝔭(N)`$. Let $`_n:n<\omega `$ list the dense open subsets of $`_{\overline{C}}`$ which belong to $`N`$ and let $`\delta =\underset{n<\omega }{lim}\alpha _n`$ where $`\alpha _n:n<\omega `$ is increasing. We now simulate a strategy for the challenger in the game $`\mathrm{}_{\alpha ,\beta }(N)`$, where in the $`n`$-th move, we let the challenger to choose $`Z_n=\mathrm{}`$ (so the chooser has to use $`Y_n=\mathrm{}`$ as well) and at the end of the $`n`$-th move, the challenger also chooses $`p_{n+1}_{\overline{C}}N`$ such that:
* $`p_0=p,`$
* $`p_np_{n+1}_n`$,
* $`p_{n+1}`$ is $`(M_n,_{\overline{C}})`$-generic and $`sup(p_{n+1})>\alpha _n`$,
* the set $`(p_{n+1}\{supp_{n+1}\})\backslash (p_n\{supp_n\})`$ is disjoint to $`C_\delta `$.
This is possible by Case 1 and its proof, because $`M_n\omega _1C_\delta `$ which holds as $`M_nY^{}`$. As this is a legal strategy for the challenger, it cannot be a winning strategy, hence for some such play the chooser wins, hence $`\{M_n:n<\omega \}D_1^𝔭(N)`$. Now $`q=\underset{n<\omega }{}p_n`$ is well defined, and $`sup(q)=\delta `$ and $`qC_\delta p\{sup(p)\}`$ and qC¯``{Mn:n<ω}[𝔾
~
C¯,N]"subscriptforcessubscript¯𝐶𝑞``conditional-setsubscript𝑀𝑛𝑛𝜔subscript𝔾
~
subscript¯𝐶𝑁"q\Vdash_{{\mathbb{Q}}_{\bar{C}}}``\{M_{n}:n<\omega\}\subseteq{{\mathscr{M}}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{Q}}_{\bar{C}}},N]", so $`q`$ is as required as the chooser has won the play.
Case 3: $`\alpha >1,\alpha `$ successor. The proof is similar to the Case 2, only we use the induction hypothesis instead of using Case 1.
Case 4: $`\alpha `$ a limit ordinal. The proof is again similar to the Case 2.
This completes the induction hypothesis and hence the proof of clause (2) of the lemma. ∎
###### Definition 5.7.
Suppose $`S\omega _1`$ is stationary, $`𝒟_{\omega _1}`$ is the club filter on $`\omega _1`$ and $`f{}_{}{}^{\omega _1}\omega _{1}^{}`$.
1. We say $`f`$ is a $`(𝒟_{\omega _1}+S,\gamma )`$ function, when $`S_{(𝒟_{\omega _1}^+,)}`$ “in 𝕍[𝔾
~
],{x𝕍ω1/𝔾
~
:𝕍ω1/𝔾
~
\mathbb{V}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}],\{x\in\mathbb{V}^{\omega_{1}}/\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}:\mathbb{V}^{\omega_{1}}/\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\models$`x`$ is an ordinal <f/𝔾
~
}<f/\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\} has order type $`\gamma `$”.
2. Assume $`\overline{C}=C_\delta :\delta <\omega _1`$, where $`C_\delta `$ is an unbounded subset of $`\delta `$. We define, by induction on $`\gamma `$, when “$`\overline{C}`$ obeys $`f`$ on $`S`$” for $`f{}_{}{}^{\omega _1}\omega _{1}^{}`$ which is a $`(𝒟_{\omega _1}+S,\gamma )`$ function:
* if $`\gamma <\omega _1`$, this means
$$\{\delta S:\mathrm{otp}(C_\delta )\omega ^{1+f(\delta )}\}=Smod𝒟_{\omega _1}.$$
* if $`\gamma \omega _1`$, it means that for some $`g:\omega _1\omega _1`$ and pressing down function $`h`$ on $`S`$, for every $`\zeta <\omega _1`$ for which $`h^1\{\zeta \}`$ is stationary, for some $`\beta <\gamma `$ and $`f_\beta `$, a $`(𝒟+h^1\{\zeta \},\beta )`$ function, we have $`C_{g(\delta )}\delta :\delta h^1\{\zeta \}`$ obeys $`f_\beta `$.
The next lemma can be proved as in Lemma 5.6.
###### Lemma 5.8.
Assume
1. $`𝔭`$ is a simple reasonable parameter such that $`\mathrm{}g(𝔭)`$ is of uncountable cofinality,
2. $`S𝒟_{\omega _1}^+`$ and $`N\underset{\alpha }{}_\alpha ^𝔭N\omega _1S`$,
3. $`𝔭`$ is a non-$`\mathrm{}_{\alpha ,\alpha }`$-loser (or just non $`\mathrm{}_{\alpha ,\alpha }^{}`$-loser) for all $`\alpha C^{},`$ where $`C^{}`$ is a club of $`\mathrm{}g(𝔭)`$ with $`0<\mathrm{min}(C^{})`$,
4. $`\overline{C}`$ obeys $`f`$ on $`S`$ which is a $`(𝒟_{\omega _1}+S,\gamma )`$-function,
5. for all $`\alpha ,g(\alpha )=\mathrm{min}(C^{}\backslash \alpha )`$.
Then $`_{\overline{C}}`$ is $`(𝔭,g)`$-proper.
Let us now give another example which fits into our general framework (see also \[6, Ch.XVIII\]). Recall that a filter $`𝒟`$ on a countable set is called a $`P`$-filter if it contains all co-finite sets and if $`A_n𝒟`$ for $`n<\omega `$, then for some $`A𝒟`$ and all $`n<\omega `$ we have $`|A\backslash A_n|<\mathrm{}_0`$
###### Definition 5.9.
1. We say $`\overline{𝒟}=𝒟_\delta :\delta <\omega _1,\delta \text{ limit}`$ is an $`\omega _1`$-filter-sequence if:
1. $`𝒟_\delta `$ is a filter on $`\delta `$, containing the co-bounded subsets of $`\delta `$,
2. $`𝒟_\delta `$ is a $`P`$-filter and some $`C_\delta 𝒟_\delta `$ has order type $`\omega `$,
3. for every club $`C\omega _1`$ and $`\alpha <\omega _1`$, the set $`A_C^\alpha [\overline{𝒟}]`$ is stationary, where, by induction on $`\alpha `$, we define $`A_C^\alpha [\overline{𝒟}]`$ by $`A_C^\alpha [\overline{𝒟}]=\{\delta <\omega _1:\delta \text{ is a limit ordinal, }\delta C\text{ and for every }\beta <\alpha \text{ we have}`$ $`\delta =sup(\delta A_C^\beta [\overline{𝒟}])`$, moreover $`\delta A_C^\beta [\overline{𝒟}]𝒟_\delta \}`$.
2. A reasonable parameter $`𝔭`$ obeys $`\overline{𝒟}`$, if for each $`\alpha <\mathrm{}g(𝔭)`$ and $`N_\alpha ^𝔭`$, $`\overline{𝒟}N`$ and we have
$$\begin{array}{cccc}D_\alpha ^𝔭(N)=\{Y:& YN\underset{\beta <\alpha }{}_\beta ^𝔭\text{ is }𝔭\text{-closed and if }\alpha >0,\text{ then there are}\hfill & & \\ & \overline{\beta }=\beta _n:n<\omega \text{ and }\overline{M}=M_n:n<\omega \text{ satisfying: }\hfill & & \\ & (a)\beta _nN\alpha ,\hfill & & \\ & (b)\text{ either for all}n,\alpha =\beta _n+1\hfill & & \\ & \text{ or }\beta _n<\beta _{n+1},sup_{n<\omega }\beta _n=sup(\alpha N),\hfill & & \\ & (c)M_nY_{\beta _n}^𝔭,M_nM_{n+1},\hfill & & \\ & (d)\underset{n<\omega }{}M_n=N\underset{\beta \alpha N}{}(\chi _\beta ^𝔭),\hfill & & \\ & (e)\{M_n\omega _1:n<\omega \}𝒟_{N\omega _1}\}.\hfill & & \end{array}$$
3. A forcing notion $``$ is a $`\overline{𝒟}NNR_\kappa ^0`$-forcing if for every reasonable parameter $`𝔭`$ which obeys $`\overline{𝒟}`$, $``$ is an $`NNR_\kappa ^0`$-forcing over $`𝔭`$ (see Definition 4.16).
4. For a $`P`$-filter $`𝒟`$ on $`\omega `$, we say a reasonable parameter $`𝔭`$ obeys $`𝒟`$ if for every $`N_\alpha ^𝔭`$
$$\begin{array}{cccc}D_\alpha ^𝔭(N)=\{Y:& YN\underset{\beta <\alpha }{}_\beta ^𝔭\text{ is }𝔭\text{-closed and if }\alpha >0,\text{ then there are}\hfill & & \\ & \overline{\beta },\overline{M}\text{ satisfying items }(a),(b)\text{ and }(d)\text{ of clause }(2)\text{ and}\hfill & & \\ & (e)\{n:M_nY\}𝒟\}.\hfill & & \end{array}$$
5. In parts (1) - (4) above, we may replace the word “filter” by “ultrafilter” if the $`𝒟_\alpha `$’s are ultrafilter.
###### Lemma 5.10.
1. If $`\mathrm{}_\mathrm{}_1`$ holds, then there is an $`\omega _1`$-ultrafilter sequence.
2. If $`\overline{𝒟}`$ is an $`\omega _1`$-filter sequence and $`(\chi _\alpha ,_\alpha ):\alpha <\omega _1`$ is as in Definition 3.2, then there is a reasonable parameter $`𝔭`$ of length $`\omega _1`$ obeying $`\overline{𝒟}`$ which is a non-$`\mathrm{}`$-loser. Furthermore, for all $`\alpha <\omega _1`$, $`\chi _\alpha ^𝔭=\chi _\alpha `$ and $`_\alpha ^𝔭=_\alpha `$.
3. If $`\mathrm{}_\mathrm{}_1`$ holds, $`(\chi _\alpha ,_\alpha ):\alpha <\omega _1`$ is as above and $`𝒟`$ is a $`P`$-filter on $`\omega `$, then some reasonable parameter $`𝔭`$ of length $`\omega _1`$ is $`P`$-filter like, non-$`\mathrm{}_{\mathrm{id}}`$-loser with $`\chi _\alpha ^𝔭=\chi _\alpha ,_\alpha ^𝔭=_\alpha `$. Similarly for ultrafilters.
4. Instead of $`\mathrm{}_\mathrm{}_1`$ it is enough to assume CH and that for some $`C_\delta :\delta <\omega _1\text{ limit}`$ and some normal filter $`D`$ on $`\omega _1`$, and for every club $`C`$ of $`\omega _1,\{\delta :\delta >sup(C_\delta \backslash C)\}D`$.
###### Proof.
We only prove (1) and (2), the other parts can be proved similarly.
(1). Let $`S_\delta :\delta <\omega _1`$ be a $`\mathrm{}_\mathrm{}_1`$-sequence, where each $`S_\delta \delta `$ and let $`E_\delta :\delta <\omega _1\text{ limit}`$ be a ladder system, where each $`E_\delta \delta `$ has order type $`\omega `$ and $`E_\delta =S_\delta `$ if $`S_\delta `$ is an $`\omega `$-sequence cofinal in $`\delta .`$ Let $`\overline{𝒟}=𝒟_\delta :\delta <\omega _1\text{ limit}`$, where $`𝒟_\delta `$ is any $`P`$-ultrafiler on $`\delta `$ extending $`𝒟_^\delta ^\text{c}\{E_\delta \}`$, where $`𝒟_\delta ^\text{c}`$ is the filter of co-bounded subsets of $`\delta `$. We show that $`\overline{𝒟}`$ is as required. Items (a) and (b) of Definition 5.9(1) are clearly satisfied. Let us prove clause (3). Thus suppose that $`C\omega _1`$ is a club, $`\alpha <\omega _1`$, and suppose by induction that for all $`\beta <\alpha `$, the set $`A_C^\beta (\overline{𝒟})`$ is stationary. We show that $`A_C^\alpha (\overline{𝒟})`$ is stationary as well. We may assume that $`C`$ only contains limit ordinals. Then
$$A_C^\alpha (\overline{𝒟})=C\underset{\beta <\alpha }{}\{\delta :\delta =sup(\delta A_C^\beta (\overline{𝒟}))\}\{\delta :\beta <\alpha ,\delta A_C^\beta (\overline{𝒟})𝒟_\delta \}.$$
It follows from the induction hypothesis that the set $`C\underset{\beta <\alpha }{}\{\delta :\delta =sup(\delta A_C^\beta (\overline{𝒟}))\}`$ is a club. Suppose by contradiction that the set $`A=\{\delta :\beta <\alpha ,\delta A_C^\beta (\overline{𝒟})𝒟_\delta \}`$ is non-stationary. Thus for some club $`DC`$ and for all $`\delta D`$, there exists some $`\beta _\delta <\alpha `$ such that $`\delta A_C^{\beta _\delta }(\overline{𝒟})𝒟_\delta `$. As $`\alpha <\omega _1,`$ it follows from Födor’s lemma that there are a stationary set $`SD`$ and some fixed $`\beta _{}<\alpha `$ such that
$$\delta S\delta A_C^\beta _{}(\overline{𝒟})𝒟_\delta $$
On the other hand, by the $`\mathrm{}_\mathrm{}_1`$-assumption, the set
$$T=\{\delta S:\delta A_C^\beta _{}(\overline{𝒟})=S_\delta \}$$
is stationary. We may assume without loss of generality that for all $`\delta T,\delta A_C^\beta _{}(\overline{𝒟})`$ has order type $`\omega .`$ But then
$$\delta T\delta A_C^\beta _{}(\overline{𝒟})=S_\delta =E_\delta 𝒟_\delta ,$$
which is a contradiction.
(2) Define $`𝔭`$ of length $`\omega _1`$ such that for all $`\alpha <\omega _1`$,
* $`\chi _\alpha ^𝔭=\chi _\alpha `$,
* $`R_\alpha ^𝔭=[((_{\beta <\alpha }\chi _\beta )^+)]^\mathrm{}_0`$,
* $`_\alpha ^𝔭=_\alpha `$,
* for $`N_\alpha ,`$ $`D_\alpha ^𝔭(N)`$ is defined as in Definition 5.9(2).
Then $`𝔭`$ is as required. ∎
###### Lemma 5.11.
1. If $`𝒟`$ is a $`P`$-filter on $`\omega `$ (or $`P`$-ultrafilter on $`\omega `$) and $`𝔭`$ is a reasonable parameter obeying $`𝒟`$, then for some $`\overline{𝒟},\overline{𝒟}`$ is an $`\omega _1`$-filter-sequence (or $`\omega _1`$-ultrafilter-sequence) and $`𝔭`$ obeys $`\overline{𝒟}`$.
2. If $`𝔭`$ is a $`P`$-point filter (or ultrafilter), then $`𝔭`$ is a non-$`\mathrm{}`$-loser.
###### Proof.
(1) For each limit ordinal $`\delta <\omega _1`$ fix a bijection $`f_\delta :\omega \delta `$ and set $`𝒟_\delta =\{f^{\prime \prime }[X]:X𝒟\}`$. Then $`\overline{𝒟}`$ is as required.
Proof of (2) is essentially similar to the proof of Lemma 3.7. ∎
We now consider the case where the order type of the club sets $`C_\delta `$is higher than $`\omega `$.
###### Lemma 5.12.
1. Assume
1. $`\kappa \omega `$ and $`\overline{C}=C_{\delta ,\mathrm{}}:\mathrm{}<k_\delta ,\delta <\omega _1\text{ limit}`$, where $`1+\kappa k_\delta \omega ,C_{\delta ,\mathrm{}}`$ is a closed unbounded subset of $`\delta `$ and $`\mathrm{}<m<k_\delta C_{\delta ,\mathrm{}}C_{\delta ,m}=\mathrm{}`$,
2. $`=_{\overline{C}}=\{C:C`$ is a closed bounded subset of $`\omega _1`$ such that for every limit $`\delta <sup(C)`$, and for every $`\mathrm{}<k_\delta `$ except $`<1+\kappa `$ many, $`\delta sup(CC_{\delta ,\mathrm{}})\}`$,<sup>9</sup><sup>9</sup>9i.e., $`\{\mathrm{}<k_\delta :\delta >sup(CC_{\delta ,\mathrm{}})\}`$ has size $`<1+\kappa `$.
3. $`𝔭`$ is a reasonable parameter, obeying the $`P`$-ultrafilter $`𝒟`$.
Then $``$ is a $`𝔭`$NNR$`{}_{\kappa }{}^{}{}_{}{}^{0}`$ forcing notion.
2. In part (1), if we add:
1. $`N_\alpha ^𝔭`$ and $`D_\alpha ^𝔭(N)=\{Y:\{n:M_nY\}𝒟_N\},𝒟_N`$ a $`P`$-ultrafilter and $`\mathrm{}<k_\delta \{n<\omega :M_n\omega _1C_{\delta ,\mathrm{}}\}=\mathrm{}mod𝒟_N`$,
then we can allow $`C_{\delta ,0}=C_{\delta ,1}`$.
3. Assume
1. $`D_\delta `$ is a family of subsets of $`\mathrm{dom}(D_\delta )`$, the intersection $`Y`$ of any $`<1+\kappa `$ of them satisfies,
2. $`()n(y_1,\mathrm{},y_nY)[\delta >sup(\underset{\mathrm{}=1}{\overset{n}{}}C_{\delta ,y_{\mathrm{}}})]`$,
3. $`\overline{C}=C_{\delta ,x}:x\mathrm{dom}(D_\delta )`$ and $`\delta `$ is a limit ordinal $`<\omega _1`$,
4. $`C_{\delta ,x}:x\mathrm{dom}(D_\delta )`$ is a sequence of pairwise disjoint subsets of $`\delta `$,
5. $`\overline{X}\underset{\delta <\omega _1}{}\mathrm{dom}(D_\delta )`$,
6. $`_{\overline{C},\overline{X},\overline{D}}=\{C:C`$ is a closed bounded subset of $`\omega _1`$ such that for every limit $`\delta sup(C)`$ we have $`(xX_\delta )(\delta >sup(CC_{\delta ,x}))\}`$ ordered by end extension,
7. $`\overline{𝒟}`$ is a $`P`$-ultrafilter sequence,
8. $`𝔭`$ is a reasonable parameter which obeys $`\overline{𝒟}`$.
Then $``$ is a $`𝔭NNR_\kappa ^0`$ forcing notion.
###### Proof.
We prove clause (1), as other items can be proved in a similar way. So let $`𝕍_0`$ be some transitive class with $`𝔭𝕍_0`$ and suppose that $`\overline{}𝕍_0`$ is an NNR$`{}_{\kappa }{}^{}{}_{}{}^{0}`$-iteration with $`_\alpha =lim(\overline{})`$ such that $`𝕍=𝕍_0^_\alpha `$ and α𝕍0``
~
α=C¯
~
subscriptsuperscriptforcessubscript𝕍0subscript𝛼absent``subscript
~
𝛼subscript¯𝐶
~
\Vdash^{\mathbb{V}_{0}}_{{\mathbb{P}}_{\alpha}}``\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}={\mathbb{Q}}_{\mathchoice{\oalign{$\displaystyle\bar{C}$\crcr\vbox to0.60275pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\bar{C}$\crcr\vbox to0.60275pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\bar{C}$\crcr\vbox to0.60275pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\bar{C}$\crcr\vbox to0.60275pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}} is as above”. Set α+1=α
~
αsubscript𝛼1subscript𝛼subscript
~
𝛼{\mathbb{P}}_{\alpha+1}={\mathbb{P}}_{\alpha}\ast\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}. We have to show that items (a)-(d) of Definition 3.9 (for $`\kappa =\mathrm{}_0`$) or 3.14 (for $`\kappa <\mathrm{}_0`$) are satisfied. We only check clause (d), as other items are easier to prove. Suppose that in $`𝕍_0,`$
* $`N_0\underset{\beta <\alpha }{}_\beta ^𝔭,`$
* $`N_0N_1_\alpha ^𝔭`$,
* $`N_1=\{M_n:n<\omega \}`$,
* $`𝒟_{N_0}`$ is a $`P`$-ultrfiler as in Definition 5.12(2),
* $`𝔾_mN_1_\alpha `$ is generic over $`N_1`$ for $`m<k<1+\kappa ,`$
* $`\underset{m<k}{}[𝔾_mN_0=𝔾^{}]`$,
* $`p_{\alpha +1}N_0`$ is such that $`p\alpha 𝔾^{}`$,
Clearly, without loss of generality $`M_n:n<\omega ,𝒟_{N_0}N_1`$. For $`\iota =0,1`$ set $`\delta _{N_\iota }=N_\iota \omega _1`$. So for each $`\mathrm{}<k_{\delta _{N_0}}`$, we have C
~
δN0,[𝔾m]subscript𝐶
~
subscript𝛿subscript𝑁0delimited-[]subscript𝔾𝑚\mathchoice{\oalign{$\displaystyle C$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle C$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\delta_{N_{0}},\ell}[\mathbb{G}_{m}] is a closed subset of $`\delta `$ and for $`\mathrm{}_1<\mathrm{}_2<k_{\delta _{N_0}}`$ we have C
~
N0ω1,1[𝔾m]C
~
N0ω1,2[𝔾m]=subscript𝐶
~
subscript𝑁0subscript𝜔1subscript1delimited-[]subscript𝔾𝑚subscript𝐶
~
subscript𝑁0subscript𝜔1subscript2delimited-[]subscript𝔾𝑚\mathchoice{\oalign{$\displaystyle C$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle C$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{N_{0}\cap\omega_{1},\ell_{1}}[\mathbb{G}_{m}]\cap\mathchoice{\oalign{$\displaystyle C$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle C$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{N_{0}\cap\omega_{1},\ell_{2}}[\mathbb{G}_{m}]=\emptyset. So for some $`\mathrm{}(m)\{0,1,\mathrm{},k_{\delta _{N_0}}1\}`$ we have
(m)C
~
N0ω1,[𝔾m]=mod𝒟N0.𝑚subscript𝐶
~
subscript𝑁0subscript𝜔1delimited-[]subscript𝔾𝑚modulosubscript𝒟subscript𝑁0\ell\neq\ell(m)\Rightarrow\mathchoice{\oalign{$\displaystyle C$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle C$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{N_{0}\cap\omega_{1},\ell}[\mathbb{G}_{m}]=\emptyset\mod{\mathscr{D}}_{N_{0}}.
Now let
$`B=\{n:`$ if $`\mathrm{}<k_{\delta _{N_0}},\mathrm{}\{\mathrm{}(m):m<k\}`$ and $`\mathrm{}<n`$, then
Mnω1C
~
δN0,[𝔾m]subscript𝑀𝑛subscript𝜔1subscript𝐶
~
subscript𝛿subscript𝑁0delimited-[]subscript𝔾𝑚M_{n}\cap\omega_{1}\notin\mathchoice{\oalign{$\displaystyle C$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle C$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\delta_{N_{0}},\ell}[\mathbb{G}_{m}] for $`m<k`$ and $`pM_n\}`$.
Then $`B`$ belongs to $`N_1𝒟_{N_0}`$. Let $`B=\{n_i:i<\omega \}`$ be an increasing enumeration of $`B`$ and let $`_n:n<\omega `$ list the dense open subsets of $`_{\alpha +1}`$ which belong to $`N_0`$. We choose $`p_i`$, by induction on $`i<\omega `$, such that:
1. $`p_iN_1P_{\alpha +1}`$,
2. $`p_i\alpha \underset{m<k}{}𝔾_m`$,
3. $`p_i\{_n:n<n_i,_nN_1\text{ and }i>0\}`$,
4. $`pp_i`$,
5. $`p_ip_{i+1}`$,
6. pi+1\pi is disjoint to {C
~
δN0,[𝔾m]:<kδN0,<ni and m<k(m)}\subscript𝑝𝑖1subscript𝑝𝑖 is disjoint to conditional-setsubscript𝐶
~
subscript𝛿subscript𝑁0delimited-[]subscript𝔾𝑚formulae-sequencesubscript𝑘subscript𝛿subscript𝑁0subscript𝑛𝑖 and 𝑚𝑘𝑚p_{i+1}\backslash p_{i}\text{ is disjoint to }\bigcup\{\mathchoice{\oalign{$\displaystyle C$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle C$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\delta_{N_{0}},\ell}[\mathbb{G}_{m}]:\ell<k_{\delta_{N_{0}}},\ell<n_{i}\text{ and }m<k\Rightarrow\ell\neq\ell(m)\}.
This is possible as, for each $`i<\omega `$,
{C
~
δN0,[𝔾m]Mniω1:<kδN0,<ni and m<k(m)}conditional-setsubscript𝐶
~
subscript𝛿subscript𝑁0delimited-[]subscript𝔾𝑚subscript𝑀subscript𝑛𝑖subscript𝜔1formulae-sequencesubscript𝑘subscript𝛿subscript𝑁0subscript𝑛𝑖 and 𝑚𝑘𝑚\bigcup\{\mathchoice{\oalign{$\displaystyle C$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle C$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle C$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\delta_{N_{0}},\ell}[\mathbb{G}_{m}]\cap M_{n_{i}}\cap\omega_{1}:\ell<k_{\delta_{N_{0}}},\ell<n_{i}\text{ and }m<k\Rightarrow\ell\neq\ell(m)\}
is a bounded subset of $`M_{n_i}\omega _1`$.
Set
$$𝔾^{}=\{q_{\alpha +1}N_0:i<\omega (p_iq)\}.$$
Then $`𝔾^{}_{\alpha +1}N_0`$ is generic over $`N_0,p𝔾^{}`$ and if we set $`q=\underset{i<\omega }{}p_i,`$ then $`q`$ witnesses that $`𝔾^{}`$ has an upper bound in α+1/G
~
αsubscript𝛼1subscript𝐺
~
subscript𝛼{\mathbb{P}}_{\alpha+1}/\mathchoice{\oalign{$\displaystyle G$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle G$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle G$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle G$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{\alpha}}. ∎
The proof of the next lemma is similar to the above proofs.
###### Lemma 5.13.
1. The forcing notion $`_{\overline{C}}`$ from Lemma 5.12(1) is $`NNR_\mathrm{}_0^0`$-forcing notion for every $`𝔭`$, non-$`\mathrm{}_{\mathrm{id}}`$-loser.
2. If $`\overline{𝒟}`$ is an $`\omega _1`$-filter sequence and $`𝔭`$ is a reasonable parameter obeying $`\overline{𝒟}`$, then any $`(<\omega _1)`$-proper forcing notion is $`𝔭`$-proper.
## 6. Second preservation of not adding reals
In this section we present our second preservation theorem. We shall concentrate on the simple case.
###### Definition 6.1.
Let $`𝔭`$ be a reasonable parameter and let ¯=i,
~
i:i<g(¯)\bar{{\mathbb{Q}}}=\langle{\mathbb{P}}_{i},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}:i<\ell g(\bar{{\mathbb{Q}}})\rangle be an iteration of forcing notions. We say that $`\overline{}`$ is a $`𝔭`$-NNR$`{}_{\kappa }{}^{}{}_{}{}^{1}`$ iteration, where $`2\kappa \mathrm{}_0`$,<sup>10</sup><sup>10</sup>10we omit $`\kappa =\mathrm{}_1`$ for convenience. when for some $`f_i,g_i_{\mathrm{dc}}^𝔭`$, for $`i<\mathrm{}g(\overline{})`$ (see 4.4), we have:
1. $`\mathrm{cf}(\mathrm{}g(𝔭))>\mathrm{}g(\overline{})`$,
2. $`\overline{}`$ is a countable support iteration of proper forcing notions such that for each $`i<\mathrm{}g(\overline{}),_i`$ adds no reals, <sup>11</sup><sup>11</sup>11this follows from other parts (close (d)), even for $`_{i+1},i<\mathrm{}g(\overline{})`$.
3. (long properness) for each $`i<\mathrm{}g(\overline{})`$, we have i``
~
isubscriptforcessubscript𝑖absent``subscript
~
𝑖\Vdash_{{\mathbb{P}}_{i}}``\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} is $`(𝔭^_i,f_i)`$-proper”,
4. ($`\kappa `$-anti w.d.) if $`i<\mathrm{}g(\overline{})`$ and $`\beta g_i(\alpha )`$, then
~
isubscript
~
𝑖\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} has $`(\kappa ,\alpha ,\beta )`$-anti w.d. above $`_i`$.
###### Remark 6.2.
1. $`\kappa `$ is the amount of “$`𝔻`$-completeness”, in other words what versions of weak diamond we kill by our iteration. So the case $`\kappa =\mathrm{}_0`$ is easier, and we first deal with it in Theorem 6.3.
2. Note that we ask for $`f_i𝕍`$ and not a $`_i`$-name f
~
isubscript𝑓
~
𝑖\mathchoice{\oalign{$\displaystyle f$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle f$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle f$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle f$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} of such a function. The reason is that if for $`i<\mathrm{}g(\overline{}),_i`$ satisfies the $`\mathrm{cf}(\mathrm{}g(𝔭))`$-c.c., then we can find fi𝔭,fif
~
iformulae-sequencesubscriptsuperscript𝑓𝑖superscript𝔭subscriptsuperscript𝑓𝑖subscript𝑓
~
𝑖f^{\prime}_{i}\in{{\mathscr{F}}}^{{\mathfrak{p}}},f^{\prime}_{i}\geq\mathchoice{\oalign{$\displaystyle f$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle f$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle f$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle f$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}. As in practice we usually have $`\mathrm{cf}(\mathrm{}g(𝔭))>|_\alpha |,`$ there is no point at present for $`f_i`$ to be a $`_i`$-name.
3. In clause (d) we have implicitly used:
1. if $`\alpha \beta ^{}<\beta `$, then clause (d) for $`(\alpha ,\beta )`$ and $`\kappa `$ implies clause (d) for $`(\alpha ,\beta ^{})`$ and $`\kappa `$.
This holds by clause (i) of Definition 3.2.
4. We could replace $`f_i`$ by a club $`E_i`$ of $`\mathrm{}g(𝔭)`$, letting $`f_i(\alpha )=E_i\backslash \alpha `$.
5. In clause (c), for a club $`C`$ of $`\mathrm{}g(𝔭)`$ we catch our tail, that is $`f_i(\alpha )C=C\backslash \alpha `$ for a club of $`\alpha <\mathrm{}g(𝔭)`$.
6. In clause (d), much of the freedom/variation will be due to the decision how “similar” are $`G^{\mathrm{}}:\mathrm{}<k`$ such that $`𝔾^{}`$ exists. Here we demand
1. $`YD_\beta ^𝔭(N_1)`$.
In \[6, Ch.VIII\], it is essentially required that
1. $`𝔾^0\times 𝔾^1\times \mathrm{}\times 𝔾^{k1}(_i\times \mathrm{}\times _i)N_1`$ ($`k`$ times) is generic over $`N_1`$.
In \[6, Ch.V\], it is required that
1. the common $`Y`$ is a pre-determined increasing sequence of models.
Clause $`(\beta )`$ makes demand (d) in Definition 6.1 easier, but the parallel of (c) is harder compared to clause $`(\alpha )`$.
We now state and prove the main results of this section. We first deal with the $`𝔭`$-NNR$`{}_{\mathrm{}_0}{}^{}{}_{}{}^{1}`$ iterations.
###### Theorem 6.3.
Assume $`\overline{}`$ is a $`𝔭`$-$`\text{NNR}_\mathrm{}_0^1`$ iteration, $`𝔭`$ is a reasonable parameter and $`𝔭`$ is a $`\mathrm{}_f`$-winner for some $`f_{\mathrm{club}}^𝔭`$ (or at least is $`\mathrm{}_f^{}`$-non loser).
1. Forcing with $`_{\mathrm{}g(\overline{})}=\mathrm{Lim}(\overline{})`$ does not add reals (so consequently adds no $`\omega `$-sequences, as we are assuming properness).
2. If $`ij\mathrm{}g(\overline{})`$, then
1. $`_j/_i`$ is proper,
2. $`_j/_i`$ is $`(𝔭,f_{i,j})`$-proper, where $`f_{i,j}_{\mathrm{club}}^𝔭`$ is increasing continuous and is computable from the $`f_\epsilon ^𝔭`$ for $`\epsilon [i,j)`$,
3. we have the parallel of clause (d) in the following sense: if $`i<j<\mathrm{}g(\overline{}),`$ then for some function $`g_{\text{cd}}^𝔭`$ in $`𝕍`$ and for all $`\alpha <\mathrm{}g(𝔭)`$ and $`\beta g(\alpha )`$ we have $`_j/_i`$ is $`(\mathrm{}_0,\alpha ,\beta )`$-anti-w.d above $`_i`$.
###### Proof.
The proof is by induction on $`\mathrm{}g(\overline{})`$. For notational simplicity we assume that:
1. all $`f_i`$’s are also in $`_{\mathrm{nd}}^𝔭`$, so we can consider them as increasing and continuous functions from $`\mathrm{}g(𝔭)`$ to $`\mathrm{}g(𝔭)`$. We also demand that the $`f_{i,j}`$’s are also like that, are increasing continuous and moreover $`f_{i,j}(f_{i,j}(\alpha ))=f_{i,j}(\alpha )`$, and they are $`f^{}`$ where $`f^{}_{\mathrm{nd}}^𝔭`$ is increasing continuous and $`𝔭`$ is $`\mathrm{}_f^{}`$-winner (or at least $`\mathrm{}_f^{}^{}`$-non-loser).
Case 1: $`\mathrm{}g(\overline{})=0`$. This is trivial.
Case 2: $`\mathrm{}g(\overline{})=i()+1`$ is a successor ordinal. We show that items (1) and (2) are satisfied.
Clause (1): $`_{i()}`$ adds no reals by the induction hypothesis and $`_{_{i()}}`$
~
i()subscript
~
𝑖\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i(*)} adds no reals”, by clause (d) in Definition 6.1, hence i()+1=i()
~
i()subscript𝑖1subscript𝑖subscript
~
𝑖{\mathbb{P}}_{i(*)+1}={\mathbb{P}}_{i(*)}*\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i(*)} adds no reals.
Clause (2): We have to show that items $`(b)^{},(c)^{}`$ and $`(d)^{}`$ are satisfied.
1. Clause $`(b)^{}`$: By \[6, Ch. III\], $`_j/_i`$ is proper.
2. Clause $`(c)^{}`$: Given $`ij\mathrm{}g(\overline{Q})`$, if $`j<i()+1`$ the conclusion follows by the induction hypothesis. So assume $`j=i()+1`$. If $`i=j`$, the required demand is trivial, so assume $`i<j`$. If $`i=i()`$, use clause (c) of Definition 6.1 for $`i`$ to get the conclusion. So assume that $`i<i()`$. Let
* $`f_{i,j,0}=f_{i()},`$
* $`f_{i,j,m+1}=f_{i()}f_{i,i()}f_{i,j,m}`$,
* $`f_{i,j}(\alpha )=\underset{m<\omega }{sup}f_{i,j,m}(\alpha )`$.
Then the $`f_{i,j}`$’s are as required in $``$. To prove “$`_j/_i`$ is $`(𝔭,f_{i,j})`$-proper”, assume that
1. $`(a)N((\chi ),)`$ is countable,
2. $`(b)\{\overline{},i,j,\alpha ,\beta ,f_{i,i()},f_{i()},f_{i,j}\}N`$,
3. $`(c)\alpha f_{i,j}(\alpha )\beta <\mathrm{}g(𝔭)`$,
4. $`(d)q_i\text{ is }(N,_i)\text{-generic}`$,
5. $`(e)pN_j,piq`$,
6. $`(f)YD_\beta ^𝔭(N)`$,
7. (g)q``Yi[𝔾
~
i,N]forces𝑔𝑞``𝑌subscriptsubscript𝑖subscript𝔾
~
subscript𝑖𝑁(g)\quad q\Vdash``Y\subseteq{{\mathscr{M}}}_{{\mathbb{P}}_{i}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}},N]”.
First we deal with version 2, and assume that $`𝔭`$ is simple. Choose $`y^{}N`$ which codes enough information. Clearly $`\beta ^{}=f_{i()}(\alpha )`$ belongs to $`N`$. So $`f_{i,i()}(\beta ^{})\beta `$, hence by the induction hypothesis there are $`q^{},Y^{}`$ such that:
* $`qq^{}_{i()},`$
* $`pi()q^{},`$
* $`q^{}`$ is $`(N,P_{i()})`$-generic,
* $`Y^{}Y,Y^{}D_\beta ^{}^𝔭(N)`$,
* q``Yi()[𝔾
~
i(),N,y]"forcessuperscript𝑞``superscript𝑌subscriptsubscript𝑖subscript𝔾
~
subscript𝑖𝑁superscript𝑦"q^{\prime}\Vdash``Y^{\prime}\subseteq{{\mathscr{M}}}_{{\mathbb{P}}_{i(*)}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i(*)}},N,y^{*}]".
Next, we apply clause (c) in the Definition 6.1 for $`i()`$, so there are $`q^{\prime \prime },Y^{\prime \prime }`$ such that
* $`q^{}q^{\prime \prime }_{i()+1}=_j,`$
* $`pq^{\prime \prime },`$
* $`q^{\prime \prime }`$ is $`(N,_j)`$-generic,
* $`Y^{\prime \prime }Y^{},Y^{\prime \prime }D_\alpha ^𝔭(N)`$,
* q′′``Y′′[𝔾
~
j,N,y]"forcessuperscript𝑞′′``superscript𝑌′′subscript𝔾
~
subscript𝑗𝑁superscript𝑦"q^{\prime\prime}\Vdash``Y^{\prime\prime}\subseteq{{\mathscr{M}}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{j}},N,y^{*}]".
The result follows immediately. The proof for version 1 is similar.
3. Clause $`(d)^{}`$: Recall that we have demanded $`f_{i,j}(f_{i,j}(\alpha ))=f_{i,j}(\alpha )`$ (see $``$ at the beginning of the proof).
Let $`N_0,N_1,\alpha ,\beta ,i,j,p,k,q_{\mathrm{}}`$ (for $`\mathrm{}<k`$), $`𝔾^{\mathrm{}}`$ (for $`\mathrm{}<k`$) and $`𝔾^{}`$ be as in the assumptions of Definition 6.1(d) (see Definition 4.12).
Without loss of generality $`i<i()<j=i()+1`$, since the other cases are trivial as in the proof of clause $`(c)^{}`$ . First choose $`𝔾^{}N_1`$ for $`𝔾^{},\alpha ,\beta ,i,i()`$. For each $`\mathrm{}<k`$, if for some $`s_{\mathrm{}}𝔾^{\mathrm{}}`$ we have
1. $`s_{\mathrm{}}__i`$ there is an upper bound for $`𝔾^{}`$ in i()/𝔾
~
isubscript𝑖subscript𝔾
~
subscript𝑖{\mathbb{P}}_{i(*)}/\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}}”,
then, as $`𝔾^{\mathrm{}}`$ is generic over $`N_0`$, by increasing $`s_{\mathrm{}}`$ if necessary, there are $`s_{\mathrm{}}𝔾^{\mathrm{}}`$ and $`r_{\mathrm{}}P_{i()}N_1`$ such that $`s_{\mathrm{}}`$ forces that $`r_{\mathrm{}}`$ is an upper bound for $`𝔾^{}`$, and without loss of generality $`r_{\mathrm{}}is_{\mathrm{}}`$. Now without loss of generality
$`𝔾^\mathrm{}_1=𝔾^\mathrm{}_2s_\mathrm{}_1=s_\mathrm{}_2`$
and
$`𝔾_\mathrm{}_1𝔾_\mathrm{}_2s_\mathrm{}_1,s_\mathrm{}_2`$ are incompatible.<sup>12</sup><sup>12</sup>12see Remark 4.13.
Now choose $`r_{i()}N_1`$ with domain $`i()\backslash i`$ as follows:
* $`\mathrm{dom}(r)=\underset{\mathrm{}<k}{}\mathrm{dom}(r_{\mathrm{}})\backslash i`$,
* $`r(\alpha )=r_{\mathrm{}}(\alpha )`$ if s𝔾
~
i,<kformulae-sequencesubscript𝑠subscript𝔾
~
subscript𝑖𝑘s_{\ell}\in\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}},\ell<k,
* $`r(\alpha )=\mathrm{}__\alpha `$ if this occurs for no $`\mathrm{}`$.
Renaming $`rN_i_{i()},\mathrm{dom}(r)i()\backslash i`$ and $`s_{\mathrm{}}𝔾^{\mathrm{}},r_{\mathrm{}}=s_{\mathrm{}}r`$ is above $`𝔾^{}`$ in $`_{i()}`$. Let $`\beta _{\mathrm{}}=f_{i,j,1+\mathrm{}}(\alpha )`$ for $`\mathrm{}k`$.
We choose, by induction on $`\mathrm{}k`$, the objects $`Y_{\mathrm{}},q_{\mathrm{}}^{},M_{\mathrm{}}`$ such that:
1. $`(a)Y_0=Y`$,
2. $`(b)M_0=N_1`$,
3. $`(c)N_0M_{\mathrm{}+1}`$,
4. $`(d)M_{\mathrm{}+1}M_{\mathrm{}}_{\beta _k\mathrm{}}^𝔭`$,
5. $`(e)Y_{\mathrm{}+1}Y_{\mathrm{}}`$,
6. $`(f)Y_{\mathrm{}}D_\beta _{\mathrm{}}^𝔭(M_{\mathrm{}})`$,
7. $`(g)M_{\mathrm{}+1}Y_{\mathrm{}}`$,
8. $`(h)q_{\mathrm{}}q_{\mathrm{}}^{}P_{i()}`$,
9. $`(i)q_{\mathrm{}}^{}\text{ is }(M_{\mathrm{}+1},_{i()})\text{-generic}`$,
10. (j)q forces a value for
G
~
i()M+1𝑗subscriptsuperscript𝑞subscript forces a value for
G
~
subscript𝑖subscript𝑀1(j)\quad q^{\prime}_{\ell}\text{ forces a value for }\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i(*)}}\cap M_{\ell+1},
11. $`(k)q_{\mathrm{}}^{}\text{ is }(N_0,_{i()})\text{-generic}`$,
12. (l)q``Y+1i()[𝔾
~
i(),M]"forces𝑙subscriptsuperscript𝑞``subscript𝑌1subscriptsubscript𝑖subscript𝔾
~
subscript𝑖subscript𝑀"(l)\quad q^{\prime}_{\ell}\Vdash``Y_{\ell+1}\subseteq{{\mathscr{M}}}_{{\mathbb{P}}_{i(*)}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i(*)}},M_{\ell}]",
13. $`(m)q_{\mathrm{}}^{}iG^{\mathrm{}}`$.
Now apply clause (d) of the definition for $`i(),N_0,M_k,q_{\mathrm{}}^{}:\mathrm{}<k,Y_k`$ and $`𝔾^{}`$ and get $`𝔾^{}`$ as required.
Case 3: $`\delta =\mathrm{}g(\overline{})`$ is a limit ordinal. We show that items (1) and (2) are satisfied in this case as well.
Clause (1): This follows from clause $`(2)(d)^{}`$ proved below.
Clause (2): Again, we have to check items $`(b)^{}`$, $`(c)^{}`$ and $`(d)^{}`$. Let $`f_{i,j}`$ be fast enough functions.
1. Clause $`(b)^{}`$: This is obvious.
2. Clause $`(d)^{}`$: We first prove clause $`(d)^{}`$ and later prove clause $`(c)^{}`$. As before, without loss of generality $`i<j=\delta `$. Let $`N_0,N_1,p,𝔾^{},\alpha ,\beta ,k<\mathrm{}_0`$ and $`𝔾^{\mathrm{}},q_{\mathrm{}}`$ for $`\mathrm{}<k`$ be as in the assumptions of clause (d) of Definition 6.1.
Choose $`\gamma N_0,\alpha <\gamma <\beta `$ such that $`\gamma `$ is large enough, in particular,
$$ii^{}<j^{}<jf_{i^{},j^{}}(\gamma )=\gamma .$$
Let $`i_m:m<\omega N_1`$ be such that
1. $`(a)i_0=i`$,
2. $`(b)i_m<i_{m+1}`$,
3. $`(c)\underset{m<\omega }{sup}i_m=sup(jN_0)`$.
Choose $`y^{}N_1(\chi _\gamma )`$ coding enough information. We choose
$$M_0,M_1,M_2,M_3,M_4N_1_\gamma ^𝔭Y$$
such that
1. $`(a)N_0M_0M_1M_2M_3M_4`$,
2. $`(b)YM_mD_\gamma ^𝔭(M_m)\text{ for }m<5`$.
Choose $`q_{\mathrm{}}^{}𝔾^{\mathrm{}}M_4`$ above $`𝔾^{\mathrm{}}M_3`$ so that $`q_{\mathrm{}}^{}`$ is $`(M_t,_i)`$-generic for $`t<4`$. Let $`_m:m<\omega M_0`$ list the dense open subsets of $`_j`$ from $`N_0`$. Now we shall use the diagonal argument and choose 𝔾
~
N0,pmimN0,rmimformulae-sequencesubscript𝔾
~
subscript𝑁0subscript𝑝𝑚subscriptsubscript𝑖𝑚subscript𝑁0subscript𝑟𝑚subscriptsubscript𝑖𝑚\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}}\cap N_{0},p_{m}\in{\mathbb{P}}_{i_{m}}\cap N_{0},r_{m}\in{\mathbb{P}}_{i_{m}}. We fulfill the above in $`M_4`$, so that at the end can find a solution in $`N_1`$, by using a canonical construction.
But to carry this, we need to have finitely many candidates for 𝔾
~
imM0subscript𝔾
~
subscriptsubscript𝑖𝑚subscript𝑀0\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{0} with a common $`Y_m`$. To get this in the inductive step, we need in step $`m1`$ that for $`M_1`$ we just have finitely many candidates for 𝔾
~
imM1subscript𝔾
~
subscriptsubscript𝑖𝑚subscript𝑀1\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{1}, and in turn to get this in the step $`m1`$, we use that in step $`m2`$ for $`M_2`$ and from every maximal antichain we choose a finite subset. To get this we use that for $`M_3`$ we just ask M3[𝔾
~
im3]𝕍=M3subscript𝑀3delimited-[]subscript𝔾
~
subscriptsubscript𝑖𝑚3𝕍subscript𝑀3M_{3}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m-3}}}]\cap\mathbb{V}=M_{3}. So along the way $`N_0,M_0,M_1,M_2,M_3`$ our induction demands go down, but slowly, so that in each step $`m`$, advancing for say $`M_0`$, we have to preserve less than really knowing 𝔾
~
imM0subscript𝔾
~
subscriptsubscript𝑖𝑚subscript𝑀0\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{0}, and are helped by our demand on $`M_1`$, just like in \[6, Ch. XVIII\]. So compared to \[6, Ch. V\], we have a finite tower.
Thus we choose by induction on $`m<\omega `$ the objects $`r_m,𝔾_m^{},p_m,n_m`$, $`𝔾_m^{\mathrm{}}:\mathrm{}<n_m`$ and $`Y_m`$ such that:
1. $`(a)r_m_{i_m}M_4`$,
2. $`(b)\mathrm{dom}(r_m)[i,i_m)`$,
3. $`(c)r_{m+1}i_m=r_m`$,
4. $`(d)q_{\mathrm{}}^{}r_m_{i_m}`$ is $`(M_t,_{i_m})`$-generic for $`t<4`$,
5. $`(e)`$ if $`\mathrm{}<k,𝒥_{i_m}`$ is dense open and $`𝒥M_2`$, then for some
finite $`𝒥M_2,𝒥`$ is predense above $`q_{\mathrm{}}^{}r_m`$,
6. $`(f)n_m<\omega `$, and for $`\mathrm{}<n_m,𝔾_m^{\mathrm{}}`$ is a subset of $`_{i_m}M_0`$ generic over
$`M_0`$ and $`𝔾_m^{\mathrm{}}M_1`$,
7. $`(g)𝔾_{m+1}^{\mathrm{}}_{i_m}\{𝔾_m^{\mathrm{}}:\mathrm{}<n_m\}`$,
8. $`(h)n_0=k`$ and $`𝔾_0^{\mathrm{}}=𝔾^{\mathrm{}}M_1`$,
9. (i)qrm``𝔾
~
i,mM1{𝔾m:<nm}"forces𝑖subscript𝑞subscript𝑟𝑚``subscript𝔾
~
subscript𝑖𝑚subscript𝑀1conditional-setsubscriptsuperscript𝔾𝑚subscript𝑛𝑚"(i)\quad q_{\ell}\cup r_{m}\Vdash``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i,m}}\cap M_{1}\in\{\mathbb{G}^{\ell}_{m}:\ell<n_{m}\}",
10. $`(j)𝔾_m^{}`$ is a subset of $`_{i_m}N_0`$ generic over $`N_0`$,
11. $`(k)𝔾_m^{}𝔾_m^{\mathrm{}}`$, so that $`𝔾_m^{}𝔾_{m+1}^{}`$ and $`𝔾_0^{}=𝔾^{}`$,
12. $`(l)p_m_jN_1,`$
13. $`(m)p_0=p`$,
14. $`(n)p_mi_m𝔾_m^{}`$,
15. $`(o)p_mp_{m+1}_m`$,
16. $`(p)Y_m_{_{i_m}}[𝔾_m^{\mathrm{}},M_0,y^{}]`$,
17. $`(q)Y_mD_\gamma ^𝔭(M_0)`$.
Let us now explain the induction construction. If $`m=0`$, this is trivial, so suppose that it holds for $`m`$ and we do it for $`m+1`$. This is done in several stages.
Stage A: Choosing $`p_{m+1}`$ is trivial, the demands are: $`p_{m+1}p_m,p_{m+1}i_m𝔾_m^{}`$ and $`p_{m+1}_m`$.
Stage B: To choose $`𝔾_{m+1}^{}`$, apply the induction hypothesis using clause $`(d)^{}`$ of what we are proving with $`i_m,i_{m+1},\gamma ,f_{i_m,i_{m+1}}(\gamma ),N_0,M_1`$ here standing for $`i,j,\alpha ,\beta ,N_0,N_1`$ there.
Stage C: Let $`\{H_m^{\mathrm{}}:\mathrm{}<n_{m+1}\}`$ list the possibilities of 𝔾
~
imM1subscript𝔾
~
subscriptsubscript𝑖𝑚subscript𝑀1\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m}}}\cap M_{1} (by clause (e) this exists). Without loss of generality $`H_m^{\mathrm{}}M_0=G_m^{h(\mathrm{})}`$, for some function $`h=h_m:n_{m+1}n_m`$. We choose $`s_m^{\mathrm{}}_{i_{m+1}}M_1`$ above $`𝔾_{m+1}^{}`$, such that $`s_m^{\mathrm{}}i_m𝔾_m^{h(\mathrm{})}`$. Now we repeat the argument of the successor stage of shrinking $`Y`$, so we can find $`t_m^{\mathrm{}}`$ such that
* $`t_m^{\mathrm{}}_{i_{m+1}}M_1`$, above $`s_m^{\mathrm{}},`$
* $`t_m^{\mathrm{}}i_mH_m^{\mathrm{}},`$
* tn``𝔾
~
im+1M0=:𝔾m+1"t^{\ell}_{n}\Vdash``\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m+1}}}\cap M_{0}=:\mathbb{G}^{\ell}_{m+1}",
and such that
1. $`Y_{m+1}=:\underset{i<n_{m+1}}{}[𝔾_{m+1}^{\mathrm{}},M_0,y^{}]D_\gamma ^𝔭(M_0)`$.
The rest is as in the proof of Theorem 3.11.
Now, without loss of generality the construction belongs to $`N_1`$. So
$$𝔾^{}=\{s_jN_0:\underset{n<\omega }{}sp_m\}$$
is as required, as $`q_{\mathrm{}}^{}=:q_{\mathrm{}}\underset{m}{}r_m_jN_1`$, and is above $`𝔾^{}`$ and $`pq_{\mathrm{}}^{}`$. This finishes proving clause $`(d)^{}`$ in the case $`\mathrm{}g(\overline{})`$ is a limit ordinal.
3. Clause $`(c)^{}`$: Again, without loss of generality $`i<j=\delta `$. So assume $`f_{i,j}(\alpha )\beta ,\{i,j,\alpha ,\beta \}N^{}_\beta ^𝔭,q`$ is $`(N^{},_i)`$-generic, $`Y^{}D_\beta ^𝔭(N^{}),q_\alpha `$ and q``Yi[𝔾
~
i,N,y]"forces𝑞``superscript𝑌subscriptsubscript𝑖subscript𝔾
~
subscript𝑖superscript𝑁superscript𝑦"q\Vdash``Y^{*}\subseteq{{\mathscr{M}}}_{{\mathbb{P}}_{i}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i}},N^{*},y^{*}]" are given. We prove the desired conclusion by induction on $`\alpha `$. For each $`\alpha `$, we would like to simulate a play of $`\mathrm{}_{\alpha ,\beta }(N^{})`$, supplying the challenger with a strategy. For this we apply the proof of clause $`(d)^{}`$. Choose $`N_0,N_1,M_0,\mathrm{},M_4,q_0,𝔾_0,𝔾^{}`$ (and $`k=1`$) as there so that for some $`\alpha ^{}<\gamma ^{}<\beta ^{}`$ as there, $`\beta <\alpha ^{}`$ and $`N^{},q,Y^{}N_0`$.
During the construction, we demand $`p_mN^{}_j`$, so a generic for $`N_0`$ is not necessarily created. But still $`p_mp_{m+1},p_mi_m𝔾_m^{}`$. Now $`p_m`$ will be played by the chooser. Now $`g(1+\alpha )`$ will be a fixed point of $`f_{i_m,i_m+1}`$. So we can add the demand $`N^{}[𝔾_m^{}N^{}]𝕍=N^{}`$, i.e. $`𝔾_m^{}`$ is generic over $`N^{}`$ and
1. $`[𝔾_m^{}N^{},N^{},y^{}]D_{g(1+\alpha )}^𝔭(N^{})`$.
We will define the game so that the following are satisfied:
* The challenger chooses
1. $`X_{m+1}=[𝔾_m^{}N^{},N^{},y^{}]\underset{\xi <g(1+\alpha )}{}_\xi ^𝔭\{M:p_mM\}N_0`$.
* Now the chooser chooses $`\alpha _m,\beta _m^{}`$ and then the challenger chooses $`\beta _m\beta _m^{},f_{i_m,i_{m+1}}(\alpha )`$ in $`N_0^{}j`$ and the chooser chooses $`M_{m+1}^{}`$ such that $`p_mM_{m+1}^{}`$.
* Now the chooser chooses $`Y_mD_{\beta _m}^𝔭(M_0),Y_mX_mM_0,Y_mN_0`$.
* Now we play $`Z_m`$ for the challenger as follows: there is $`p_{m+1}p_m`$ which is $`(M_{m+1}^{},_{i_{m+1}})`$-generic such that $`p_{m+1}i_m𝔾_m^{}`$, $`p_{m+1}^{}`$ decides 𝔾
~
im+1Nsubscript𝔾
~
subscriptsubscript𝑖𝑚1superscript𝑁\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m+1}}}\cap N^{*} and forces Zm=Ymim+1[𝔾
~
im+1,Mm,y]Dαm𝔭(Mm)subscript𝑍𝑚subscript𝑌𝑚subscriptsubscriptsubscript𝑖𝑚1subscript𝔾
~
subscriptsubscript𝑖𝑚1subscriptsuperscript𝑀𝑚superscript𝑦subscriptsuperscript𝐷𝔭subscript𝛼𝑚subscriptsuperscript𝑀𝑚Z_{m}=Y_{m}\cap{{\mathscr{M}}}_{{\mathbb{P}}_{i_{m+1}}}[\mathchoice{\oalign{$\displaystyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{G}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{{\mathbb{P}}_{i_{m+1}}},M^{*}_{m},y^{*}]\in D^{{\mathfrak{p}}}_{\alpha_{m}}(M^{*}_{m}).
Let us now give the details. Choose $`i_m^{}:m<\omega N_0`$ such that $`i_mN^{},i_0=i,i_m<i_{m+1}`$ and $`sup\{i_m:m<\omega \}=sup(N^{}j)`$ and let $`_m^{}:m<\omega `$ list the dense open subsets of $`_j`$ from $`N^{}`$. For $`𝕞<\omega `$ let $`𝒯_𝕞`$ be the set of finite sequences $`𝔵`$ from $`M_4`$ coding
* $`r_{𝔵,m}:m𝕞,`$
* $`𝔾_{𝔵,m}:m𝕞,`$
* $`p_{𝔵,m}:m𝕞,`$
* $`n_{𝔵,m}:m𝕞,`$
* $`𝔾_{𝔵,m}^{\mathrm{}}:\mathrm{}n_{𝔵,m},m𝕞,`$
* $`Y_{𝔵,m}:m𝕞`$,
* $`(X_{𝔵,m},\alpha _{𝔵,m},\beta _{𝔵,m}^{},\beta _{𝔵,m},M_{𝔵,n},y_{𝔵,m}^{},M_{𝔵,m}^{},y_{𝔵,m})`$ for $`m𝕞`$,
* $`Z_{𝔵,m}`$ for $`m<𝕞`$,
satisfying items (a)-(k), (m), (n), (p) and (q) from the proof of $`(d)^{}`$ above and
1. $`(l)^{}p_m_jN^{}`$,
2. $`(o)^{}p_mp_{m+1}_m^{}`$,
3. $`(r)^{}r_{𝔵,m}`$ is $`(N^{},𝔾_{i_m})`$-generic for $`m𝕞`$,
4. $`(s)^{}(X_{𝔵,m},\alpha _{𝔵,m},\beta _{𝔵,m}^{},\beta _{𝔵,m},M_{𝔵,m},Y_{𝔵,m},M_{𝔵,m}^{},Z_{𝔵,m^{}}):m𝕞`$
belongs to $`N`$ and is an initial segment of a play of the game
$`\mathrm{}_{\alpha ,\beta }^{}(N^{},𝔭)`$ or just $`\mathrm{}_{\alpha ,\beta }^{}(N^{},N,𝔭)`$,<sup>13</sup><sup>13</sup>13 note that in the $`𝕞`$-th move the challenger has not yet chose $`Z_{𝔵,𝕞}`$, (see clause (e) of Definition 4.2(1).
5. $`(t)^{}Z_{𝔵,m}Y_{m+1}`$,
6. $`(u)^{}y_{𝔵,m}`$ codes $`p_m,i_m:m<\omega `$,
7. $`(v)^{}f_{i_m,i_{m+1}}(\alpha _m)\beta _m^{}`$ for $`m𝕞`$.
We let $`𝔵𝔶`$ to have the natural meaning for $`𝔵𝒯_{𝕞_1},𝔶𝒯_{𝕞_2},𝕞_1<𝕞_2`$. Note that
1. $`𝒯_𝕞N`$ for $`𝕞<\omega `$,
2. $`𝒯_0\mathrm{}`$,
3. if $`𝕩𝒯_𝕞`$, then $`𝔵`$ is an initial segment of a play of the game $`\mathrm{}_{\alpha ,\beta }(N^{},𝔭)`$ (see clause $`(s)^{}`$ above).
Now we show that $`()_{11}()_{12},`$ where
1. $`(a)`$$`M_𝕞^{}Y_{𝔵,𝕞}_{\alpha _𝕞}^𝔭(M_{𝔵,𝕞}\{M_{𝔵,𝕞}(\chi _𝕞^𝔭)\}`$ satisfies $`y_𝕞,y_𝕞^{}`$
$`M_𝕞^{}`$,
2. $`(b)`$$`Z_𝕞𝒟_{\alpha _{𝔵,m}}(M_m^{}),`$
3. $`(c)`$$`Z_𝕞Y_𝕞`$ (hence $`Z_𝕞X_{𝔵,𝕞})`$,
4. $`(d)`$$`X_{𝕞+1}D_\beta ^𝔭(N^{})X_{𝔵,𝕞}`$ is such that $`Z_𝕞X_{𝕞+1}`$,
5. $`(e)`$$`\alpha _{𝕞+1}\alpha N^{}`$,
6. $`(f)`$$`\beta _{𝕞+1}^{}\beta N^{}\backslash p_{i_{𝕞+1},i_{𝕞_r}}(\alpha _{𝕞+1}),`$
7. $`(g)`$$`y_{𝕞+1}^{}N(\chi _{\alpha _{𝕞+1}}^𝔭)`$ and $`y_{𝕞+1}^{}M_{𝕞+1}`$,
8. $`(h)`$$`\beta _𝕞\beta N\backslash \beta _n^{}\backslash \alpha _n`$ and $`M_{𝕞+1}X_{𝕞+1}_{\beta _{𝕞+1}}^𝔭,`$
9. $`(i)`$$`y_{𝕞+1}M_{𝕞+1}(\chi _{\alpha _𝕞}^𝔭)`$,
10. $`(j)`$$`Y_{𝕞+1}ND_{\beta _{𝕞+1}}^𝔭(M_{𝕞+1})`$,
11. $`(k)`$ any $`M_{𝕞+1}^{}Y_{𝕞+1}_{\alpha _m}^𝔭(M_{𝕞+1}\{M_{𝕞+1}(\chi _{\alpha _m}^𝔭)\}`$
satisfies $`y_{𝕞+1},y_{𝕞+1}^{}M_{𝕞+1}^{}`$.
and
1. there is $`𝔶𝒯_{𝕞+1}`$ such that $`𝔵𝔶`$ and $`(Z_{𝔶,𝕞},X_{𝔶,𝕞+1}`$, $`\alpha _{𝔶,𝕞+1},\beta _{𝔶,m+1}^{},y_{𝔶,𝕞+1}^{}`$, $`\beta _{𝔶,𝕞+1},y_{𝔶,𝕞+1},M_{𝔶,𝕞+1},Y_{𝔶,𝕞+1},M_{𝔶,𝕞+1}^{})`$ is equal to $`(Z_𝕞,X_{𝕞+1},\alpha _{𝕞+1},\beta _{𝕞+1}^{}`$, $`y_{𝕞+1}^{},\beta _{𝕞+1},y_{𝕞+1}`$, $`M_{𝕞+1},Y_{𝕞+1},M_{𝕞+1}^{})`$.
To see this, note that $`f_{i_𝕞,i_{𝕞+1}}(\alpha _𝕞)\beta _𝕞`$, hence $`_{i_{𝕞+1}}/_{i_𝕞}`$ is $`(𝔭,\alpha _𝕞,\beta _𝕞)`$-proper; thus let $`𝔾_{i_𝕞}_{i_𝕞}`$ be generic over $`𝕍,r_𝕞𝔾_{i_𝕞}`$ to the model $`M_{𝔵,𝕞}^{}`$ and the set $`Y_{𝔵,𝕞}`$. Thus we can describe a strategy for the challenger in the game $`\mathrm{}_{\alpha ,\beta }(N^{},𝔭)`$ (or $`\mathrm{}_{\alpha ,\beta }^{}(N^{},N,𝔭)`$) delaying his choice of $`M_𝕞^{},Z_𝕞`$ to the $`(𝕞+1)`$-th move, he just chose on the side $`𝔵_𝕞𝒯_𝕞`$ which “codes” what they played so far and preserve $`𝔵_𝕞𝔵_{𝕞+1}`$.
By $`_3`$ this is possible, all possible choices of the chooser are allowed, that is this gives a well defined strategy for the challenger. Now take $`𝔶𝒯_{𝕞+1}`$ be such that for all $`𝕞`$, $`𝔵_𝕞𝔶`$
As the challenger does not have a winning strategy, there is a play where the chooser wins. This gives us
$$𝔾^{\prime \prime }=\underset{n<\omega }{}𝔾_n^{}N^{}$$
with a bound. Also, there is such a choice $`𝔵_𝕞:𝕞<\omega `$ with $`\{(M_{𝔵_{𝕞+1},𝕞}^{}\}Y_{𝔵_{𝕞+1},𝕞}:𝕞<\omega \}D_\alpha ^𝔭(N)`$ and $`q^{}=\underset{𝕞<\omega }{}r_𝕞`$ is as required.
The proof is complete. ∎
Now we deal with the case of $`𝔭`$-NNR$`{}_{\kappa }{}^{}{}_{}{}^{1}`$ iteration, where $`2\kappa <\mathrm{}_2`$. The adaptation for the proof of Theorem 6.3 when $`2\kappa <\mathrm{}_2`$ should be clear.
###### Theorem 6.4.
Assume $`\overline{}`$ is a $`𝔭`$-NNR$`{}_{\kappa }{}^{}{}_{}{}^{1}`$ iteration where $`2\kappa <\mathrm{}_2`$, $`𝔭`$ is a reasonable parameter and $`𝔭`$ is a $`\mathrm{}_f`$-winner for some $`f_{\mathrm{club}}^𝔭`$ (or at least is $`\mathrm{}_f^{}`$-non loser).
1. Forcing with $`_{\mathrm{}g(\overline{})}=\mathrm{Lim}(\overline{})`$ does not add reals.
2. If $`ij\mathrm{}g(\overline{})`$, then
1. $`_j/_i`$ is proper,
2. $`_j/_i`$ is $`(𝔭,f_{i,j})`$-proper, where $`f_{i,j}_{\mathrm{club}}^𝔭`$ is increasing continuous and is computable from the $`f_\epsilon ^𝔭`$ for $`\epsilon [i,j)`$,
3. if $`i<j<\mathrm{}g(\overline{}),`$ then for some function $`g_{\text{cd}}^𝔭`$ in $`𝕍`$ and for all $`\alpha <\mathrm{}g(𝔭)`$ and $`\beta g(\alpha )`$ we have $`_j/_i`$ is $`(\mathrm{}_0,\alpha ,\beta )`$-anti-w.d above $`_i`$.
###### Proof..
Similar to the proof of Theorem 6.3, with some changes as in the proof of Theorem 3.15. ∎
###### Remark 6.5.
We may be interested in non-proper forcing notions, say semi-proper and UP ones (see \[6, Ch. X, XI, XV\]). Here the change from reasonable parameter $`𝔭=𝔭^V`$ to $`𝔭^{𝕍[𝔾]}`$ is more serious as $`\{N\chi _\alpha :N^{𝔭^{𝕍[𝔾]}}\}`$ is in general not equal to $`\{N\chi _\alpha :N_\alpha ^𝔭\}`$. This is treated in .
## 7. Forcing axioms compatible with CH
As is well known, iteration theorems give us consistency of axioms and in this section we present a few of such examples. We consider $`\kappa \{2,\mathrm{}_0\}`$, but could also have $`\kappa =\mathrm{}_1`$ at some points.
###### Definition 7.1.
Suppose $`𝔭`$ is an o.b. parameter. Then $`\mathrm{Ax}_\lambda ^\alpha (𝔭,\kappa ,0)`$ means: if $``$ is $`\mathrm{}_2`$-e.c.c. ($`\mathrm{}_2`$-pic if $`\lambda =\mathrm{}_2`$) and an $`NNR_\kappa ^0`$-forcing notion for $`𝔭,`$ $`_\beta `$ is a dense open subset of $``$, for $`\beta <\beta ^{}<\lambda `$, and S
~
isubscript𝑆
~
𝑖\mathchoice{\oalign{$\displaystyle S$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle S$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle S$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle S$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} is a $``$-name of a stationary subset of $`\omega _1`$, for $`i<i^{}<\alpha `$, then for some directed $`𝔾`$ we have:
* $`\beta <\beta ^{}𝔾_\beta \mathrm{},`$
* i<iS
~
i[𝔾]={γ<ω1:(r𝔾)(r``γS
~
i")}𝑖superscript𝑖subscript𝑆
~
𝑖delimited-[]𝔾conditional-set𝛾subscript𝜔1𝑟𝔾subscriptforces𝑟``𝛾subscript𝑆
~
𝑖"i<i^{*}\Rightarrow\mathchoice{\oalign{$\displaystyle S$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle S$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle S$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle S$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}[\mathbb{G}]=\{\gamma<\omega_{1}:(\exists r\in\mathbb{G})(r\Vdash_{{\mathbb{Q}}}``\gamma\in\mathchoice{\oalign{$\displaystyle S$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle S$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle S$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle S$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i}")\} is a stationary subset of $`\omega _1`$.
We remove $`\alpha `$ when $`\alpha =0.`$
Now we use our results on preservation of being an $`\text{NNR}_\kappa ^0`$-forcing notion for $`𝔭`$ to get the consistency of $`\mathrm{Ax}_\lambda ^\alpha (𝔭,\kappa ,0)`$.
###### Lemma 7.2.
1. If $`𝔭`$ is an o.b. parameter of length $`\mathrm{}g(𝔭)=\omega _1`$ which is non-$`\mathrm{}_{\mathrm{id}}^{}`$-loser, and if $`\overline{}`$ is a countable support iteration such that for α<g(¯),α``
~
α\alpha<\ell g(\bar{{\mathbb{Q}}}),~{}\Vdash_{{\mathbb{P}}_{\alpha}}``\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha} is an $`NNR_\kappa ^0`$-forcing notion for $`𝔭`$”, then $`\overline{}`$ is an $`NNR_\kappa ^0`$-iteration for $`𝔭`$
2. Assume $`\mathrm{CH}+\mu =\mu ^{<\mu }\lambda `$. If $`𝔭`$ is a non$`\mathrm{}_{\mathrm{id}}^{}`$-loser o.b. parameter, $`\chi _0^𝔭>2^\lambda `$, then for some $`\mathrm{}_2`$-e.c.c. ($`\mathrm{}_2`$-pic, if $`\lambda =\mathrm{}_2`$) $`NNR_\kappa ^0`$-forcing notion $``$ of size $`\mu `$ we have $`_{}\mathrm{`}\mathrm{`}\mathrm{Ax}_\lambda (𝔭,\kappa ,0)`$”.
###### Proof.
(1). Follows from Theorem 3.11,
(2). It follows using a suitable countable support iteration of length $`\mu ,`$ forcing all instances of the axiom $`\mathrm{Ax}_\lambda (𝔭,\kappa ,0)`$ at some stage of the iteration. Clause (1) and Theorem 1.10 guarantee that the iteration is as required.. ∎
###### Definition 7.3.
Assume $`\lambda =\lambda ^{<\lambda }\mathrm{}_1\kappa 2`$ and $`𝔭`$ is a reasonable parameter such that $`\lambda <\chi _0^𝔭`$ and $`\lambda \mathrm{cf}(\mathrm{}g(𝔭))`$. Let also
$`=_{\lambda ,𝔭}=(\{\overline{}:\overline{}(\lambda )`$ is a $`𝔭`$-NNR$`{}_{\kappa }{}^{}{}_{}{}^{0}`$ iteration$`\},_{})`$
where
$$\overline{}^1_{}\overline{}^2\overline{}^1=\overline{}^2\mathrm{}g(\overline{}^1).$$
1. $``$ $`\stackrel{~}{}`$ is absolutely $`(\lambda ,𝔭,)`$-NNR$`{}_{\kappa }{}^{}{}_{}{}^{0}`$ forcing above $`\overline{}`$, when
1. $`\overline{}`$,
2. $``$ $`\stackrel{~}{}`$ is a $`\mathrm{Lim}(\overline{})`$-name of a forcing notion from $`(\lambda )^{𝕍[𝔾_{\mathrm{Lim}(\overline{}}]}`$,
3. if $`\overline{}_{}\overline{}^1`$, then $``$ $`\stackrel{~}{}`$ is $`(\overline{}^1,\overline{},𝔭)`$-NNR$`{}_{\kappa }{}^{}{}_{}{}^{0}`$, which means that there is $`\overline{}^2,\overline{}^1_{}\overline{}^2`$ and ¯g(¯1)2=
~
subscriptsuperscript¯2𝑔superscript¯1
~
\bar{{\mathbb{Q}}}^{2}_{\ell g(\bar{{\mathbb{Q}}}^{1})}=\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}} so $``$ $`\stackrel{~}{}`$ is a $`\mathrm{Lim}(\overline{}^1)`$-name $`(\lambda )`$.
2. (
~
,¯
~
)
~
¯
~
(\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}},\mathchoice{\oalign{$\displaystyle\bar{{\mathscr{I}}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\bar{{\mathscr{I}}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\bar{{\mathscr{I}}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\bar{{\mathscr{I}}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}) is an absolute $`(\lambda ,𝔭,)`$-NNR$`{}_{\kappa }{}^{}{}_{}{}^{0}`$-problem above $`\overline{}`$, when
1. $``$ $`\stackrel{~}{}`$ is a $`\mathrm{Lim}(\overline{})`$-name of a forcing notion from $`(\lambda )^{𝕍[𝔾_{\mathrm{Lim}(\overline{})}]}`$,
2. $`\overline{}`$ $`\stackrel{~}{}`$ is a $`\mathrm{Lim}(\overline{})`$-name for a sequence of $`<\lambda `$ subsets of $``$ $`\stackrel{~}{}`$ ,
3. if $`\overline{}_{}\overline{}^1`$, then there is $`\overline{}^2`$ such that $`\overline{}^1_{}\overline{}^2`$ and Lim(¯2)``(
~
,¯
~
)subscriptforcesLimsuperscript¯2absent``
~
¯
~
\Vdash_{{\rm Lim}(\bar{{\mathbb{Q}}}^{2})}``(\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}},\mathchoice{\oalign{$\displaystyle\bar{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\bar{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\bar{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\bar{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}) is solved”, which means there is a directed 𝔾
~
𝔾
~
\mathbb{G}\subseteq\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}} meeting
~
εsubscript
~
𝜀\mathchoice{\oalign{$\displaystyle{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\varepsilon} for every ε<g(¯
~
)"𝜀𝑔¯
~
"\varepsilon<\ell g(\mathchoice{\oalign{$\displaystyle\bar{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\bar{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\bar{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\bar{\mathscr{I}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}})".
###### Lemma 7.4.
Suppose that $`\mathrm{CH}`$ holds, $`\lambda =\lambda ^{<\lambda }\mathrm{}_1\kappa 2`$ and $`𝔭`$ is a reasonable parameter such that $`\lambda <\chi _0^𝔭`$ and $`\lambda \mathrm{cf}(\mathrm{}g(𝔭))`$. Then there is a proper $`\lambda `$-c.c. forcing notion $`_{}`$of cardinality $`\lambda `$, such that
1. Forcing with $`_{}`$ adds no reals,
2. $`_{}=\mathrm{Lim}(\overline{}_{}),`$ where $`\overline{}_{}`$ is a countable support iteration α,
~
α:α<λ\langle{\mathbb{P}}_{\alpha},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}:\alpha<\lambda\rangle with $`__\alpha \mathrm{`}\mathrm{`}|_\alpha |<\lambda `$”, such that $`\overline{}_{}\alpha (\lambda )`$ for $`\alpha <\lambda `$. In particular, $`_{}=\underset{\alpha <\lambda }{}_\alpha `$,
3. $`\overline{}_{}`$ is $`𝔭`$-NNR$`{}_{\kappa }{}^{}{}_{}{}^{1}`$-iteration,
4. if $`𝕀`$ is a dense open subset of
$$=_{\lambda ,𝔭}=(\{\overline{}:\overline{}(\lambda )\text{ is a }𝔭\text{-NNR}_\kappa ^0\text{ iteration}\},_{}),$$
where $`\overline{}^1_{}\overline{}^2\overline{}^1=\overline{}^2\mathrm{}g(\overline{}^1)`$, and if $`𝕀`$ is definable in $`((\lambda ),)`$ from a parameter, then $`\lambda =sup\{\alpha <\lambda :\overline{}_{}\alpha 𝕀\}`$,
5. if $`𝕍^{_{}}`$ is a forcing notion of cardinality $`\mathrm{}_1`$, so without loss of generality whose set of elements is a subset of $`\omega _1`$, and if $`\alpha <\lambda `$ is such that $`𝕍^_\alpha `$ and $``$ is absolute $`(𝔭\alpha ,\lambda ,)`$-proper, then for some $`\beta (\alpha ,\lambda ),_\beta =`$, hence $`𝕍^{_{}}\mathrm{Ax}_\lambda ()`$.
6. similarly for $`(,\overline{})`$.
###### Proof.
First note that $``$ is non-empty and with no maximal member. Indeed, the iteration of length zero belongs to $``$ and if ¯=β,
~
β:β<α\bar{{\mathbb{Q}}}=\langle{\mathbb{P}}_{\beta},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\beta}:\beta<\alpha\rangle\in{\mathbb{R}}, we can define $`\overline{}^{}`$ above it by letting ¯=β,
~
β:βα,\bar{{\mathbb{Q}}}^{\prime}=\langle{\mathbb{P}}_{\beta},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\beta}:\beta\leq\alpha\rangle, where $`_\alpha =lim(\overline{})`$ and $`__\alpha `$
~
α=(2ω1>,)\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}=({}^{\omega_{1}>}2,\triangleleft)”.
Now fix $`\mathrm{\Phi }:\lambda (\lambda )`$ such that for each $`x(\lambda ),`$ the set $`\mathrm{\Phi }^1\{x\}`$ is stationary in $`\lambda .`$ Let ¯=α,
~
α:α<λ\bar{{\mathbb{Q}}}_{*}=\langle{\mathbb{P}}_{\alpha},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}:\alpha<\lambda\rangle be a countable support iteration of forcing notions, where at stage $`\alpha `$, if $`\mathrm{\Phi }(\alpha )`$ is a $`_\alpha `$-name for a forcing notion as in $`(e)`$, then we let $`__\alpha `$
~
α=Φ(α)subscript
~
𝛼Φ𝛼\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha}=\Phi(\alpha)”. Otherwise,
~
αsubscript
~
𝛼\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\alpha} is forced to be the trivial forcing notion.
Using the preservation theorems we have proved earlier, we can easily show that $`_{}=lim(\overline{}_{})`$ is as required. ∎
It is natural to restrict ourselves to the linear case, but this is not a real difference when we allow to change the $`f`$ a little.
###### Definition 7.5.
Let $`𝔭`$ be a reasonable parameter.
1. $`𝔭`$ is linear, if whenever $`\alpha <\mathrm{}g(𝔭),N_\alpha ^𝔭`$ and $`YD_\alpha ^𝔭(N)`$, then there is $`Z`$ such that:
1. $`ZD_\alpha ^𝔭(N)`$,
2. $`ZY`$,
3. if $`aZ_\gamma ^𝔭`$ then $`NaN(\chi _\gamma )`$,
4. $`Z`$ is linear, which means
5. $`(\alpha )Z`$ is well ordered by $``$ (and by $``$),
6. $`(\beta )`$ if $`aZ`$ then $`ZaN`$ and $`Na:aZ`$ is $``$-increasing continuous.
2. $`𝔭`$ is linearly standard when it is standard and linear.
3. Assume $`f^𝔭`$ and $`g^𝔭`$ is defined as
$$g(\alpha )=\{f(\beta ):\beta f(\alpha )\}.$$
Let $`𝔮=𝔭^{[f]}`$ be defined as in $`𝔭`$, except that for each $`\alpha <\mathrm{}g(𝔭)`$ and $`N`$, $`D_\alpha ^𝔮(N)=\{YD_\alpha ^𝔭(N)`$: for some $`\beta Nf(\alpha )`$ there is $`ZY,Z`$ linear and $`ZN(\chi _\beta ^𝔭)\}`$.
###### Lemma 7.6.
If $`𝔭`$ is a reasonable parameter and $`f,g`$ and $`𝔮`$ are as in Definition 7.5(3), then
1. $`𝔮`$ is a reasonable parameter,
2. if $`f`$ is increasing continuous then so is $`g`$,
3. if $`f(\alpha )=\alpha `$ then $`g(\alpha )=\alpha `$ and $`D_\alpha ^𝔮(N)=D_\alpha ^𝔭(N)`$.
###### Proof.
Straightforward. ∎
###### Lemma 7.7.
Assume $`𝔭`$ is a reasonable parameter, $`f^𝔭,\lambda =\lambda ^{<\lambda }`$ is large enough regular and $`=_{\lambda ,𝔭}`$. Let $`\overline{}`$ and $`=\mathrm{Lim}(\overline{})`$.
1. If $`𝔭`$ is linear and $``$ $`\stackrel{~}{}`$ is a $``$-name of a ($`<^+\omega _1)`$-proper forcing notion from $`(\lambda )`$ and $`f_{\mathrm{dc}}^𝔭`$, then, ``
~
subscriptforcesabsent``
~
\Vdash_{{\mathbb{P}}}``\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}} is $`(𝔭^{},f)`$-proper”.
2. if $``$ $`\stackrel{~}{}`$ satisfies the $`\kappa `$-completeness system $`𝔻𝕍`$ over $``$ and $``$ forces it is an NNR proper forcing notion, then for some $`f_{\mathrm{dc}}^𝔭`$, we have (,
~
)
~
({\mathbb{P}},\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}) is $`(\kappa ,f)`$-anti w.d. (see Definition 4.12),
###### Proof.
(1). The proof is essentially the same as the proof of Lemma 3.17.
(2). Define the function $`f`$ such that for each $`\alpha <\mathrm{}g(𝔭),`$ $`f(\alpha )=[\theta _\alpha ,\mathrm{}g(𝔭))`$, where $`\theta _\alpha \alpha `$ is large enough so that $`(\chi _{\theta _\alpha })`$ contains all the relevant information and the cardinal $`\theta `$ from Definition 1.3 witnessing $`𝔻`$ is a completeness system is below $`\chi _{\theta _\alpha }`$.
Suppose $`\alpha <\mathrm{}g(𝔭)`$, $`\beta f(\alpha )`$ and suppose that $`N_0,N_1,n,p_{\mathrm{}}:\mathrm{}<n,𝔾^{\mathrm{}}:\mathrm{}<n,𝔾^{},Y`$ and $`q`$ $`\stackrel{~}{}`$ are as in Definition 4.12(1)(A). Without loss of generality, we can assume that the $`p_{\mathrm{}}`$’s are pairwise incompatible.
Pick some $`MY,`$ so that for each $`\mathrm{}<n,(𝔾^{\mathrm{}}M)`$ is $`(M)`$-generic over $`M`$. Fix some $`\mathrm{}<n.`$ Consider the pair (M[𝔾],q
~
[𝔾])dom(𝔻)𝑀delimited-[]superscript𝔾𝑞
~
delimited-[]superscript𝔾dom𝔻(M[\mathbb{G}^{\ell}],\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}[\mathbb{G}^{*}])\in{\rm dom}({\mathbb{D}}). By the assumption, there are $`p_{\mathrm{}}^{}p^{\mathrm{}}`$ and $`^{\mathrm{}}`$ $`\stackrel{~}{}`$ such that
p
H
~
is in Gen+(M[𝔾],
~
,q
~
).forcessubscriptsuperscript𝑝
H
~
is in superscriptGen𝑀delimited-[]superscript𝔾
~
𝑞
~
p^{\prime}_{\ell}\Vdash\text{``}\mathchoice{\oalign{$\displaystyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\text{~{}is in~{}}{\rm Gen}^{+}(M[\mathbb{G}^{\ell}],\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}},\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}})\text{''}.
By extending $`p_{\mathrm{}}^{}`$ if necessary, we can assume that for some q
~
subscript𝑞
~
\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\ell} we also have
p
q
~
q
~
is an upper bound for
H
~
,forcessubscriptsuperscript𝑝
q
~
q
~
is an upper bound for
H
~
p^{\prime}_{\ell}\Vdash\text{``}\mathchoice{\oalign{$\displaystyle q_{\ell}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q_{\ell}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q_{\ell}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q_{\ell}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\geq\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\text{~{}is an upper bound for~{}}\mathchoice{\oalign{$\displaystyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}\text{''},
Set 𝕂=𝔾
~
superscript𝕂superscript𝔾fragmentsH
~
\mathbb{K}^{\ell}=\mathbb{G}^{\ell}\ast\mathchoice{\oalign{$\displaystyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\mathbb{H}^{\ell}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}. As the iteration
~
~
{\mathbb{P}}\ast\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}} does not add any new $`\omega `$-sequences of elements of $`𝕍`$, and by our choice of $`𝔾^{}`$, by extending (p,q
~
),subscriptsuperscript𝑝subscript𝑞
~
(p^{\prime}_{\ell},\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\ell}), we may assume that for some fixed $`𝔾^{}`$ and for all $`\mathrm{}<n,`$ we have
(p,q
~
)𝕂N0=𝔾.forcessubscriptsuperscript𝑝subscript𝑞
~
superscript𝕂subscript𝑁0superscript𝔾absent(p^{\prime}_{\ell},\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\ell})\Vdash\text{``}\mathbb{K}^{\ell}\cap N_{0}=\mathbb{G}^{**}\text{''}.
It follows that 𝔾Gen(N0,
~
)superscript𝔾absentGensubscript𝑁0
~
\mathbb{G}^{**}\in{\rm Gen}(N_{0},{\mathbb{P}}\ast\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}). Let q
~
superscript𝑞
~
\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{\prime} be such that for all $`\mathrm{}<n,p_{\mathrm{}}^{}`$q
~
=q
~
superscript𝑞
~
subscript𝑞
~
\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{\prime}=\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\ell}”. It is clear that p:<n,q
~
\langle p^{\prime}_{\ell}:\ell<n\rangle,\mathchoice{\oalign{$\displaystyle q$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle q$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle q$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}^{\prime} and $`𝔾^{}`$ are as required by Definition 4.12(1)(B). ∎
For the rest of this section, assume that $`\mathrm{CH}`$ holds, $`\lambda =\lambda ^{<\lambda }\mathrm{}_1\kappa 2`$ and $`𝔭`$ is a reasonable parameter such that $`\lambda <\chi _0^𝔭`$ and $`\lambda \mathrm{cf}(\mathrm{}g(𝔭))`$. Let $`_{}`$ be as in Lemma 7.4.
###### Lemma 7.8.
Under the above assumptions, if $`𝕍^{_{}}`$ is one of the following forcing notions, then it satisfies clause (e) of Lemma 7.4, in particular $`𝕍^{_{}}\mathrm{Ax}_\lambda ()`$.
1. $`=_{\overline{C},\overline{u}}`$ is as in Definition 5.1(1),
2. ~
=C¯
~
subscript¯𝐶\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}={\mathbb{Q}}_{\bar{C}} is as in Definition 5.3, for $`\overline{C}=C_\delta :\delta <\omega _1`$ limit$`,`$ where for each limit ordinal $`\delta ,\mathrm{otp}(C_\delta )=\omega .`$
3. ~
=C¯
~
subscript¯𝐶\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}={\mathbb{Q}}_{\bar{C}} is as in Definition 5.3, for $`\overline{C}=C_\delta :\delta <\omega _1`$ limit$``$, where for some countable ordinal $`\gamma (),\delta <\omega _1\mathrm{otp}(C_\delta )\omega ^{\gamma ()}`$.
###### Proof.
The $`(𝔭,f)`$-properness of $`_{\overline{C},\overline{u}}`$ follows from Lemmas 5.2 and 7.7. The $`(𝔭,f)`$-properness of $`_{\overline{C}}`$ follows from Lemma 5.5 for $`\overline{C}`$ as in (b) and from Lemma 5.6 for $`\overline{C}`$ as in (c). The $`(𝔭,g)`$-anti w.d. is straightforward. ∎
Given an Aronszajn tree $`T`$, let $`_T`$ be the forcing notion of \[6, Ch. V, Definition 6.5\]. Let also $`\overline{}_T=_{T,\alpha }:\alpha <\omega _1`$ where
$$_{T,\alpha }=\{(f,C,\mathrm{\Psi })_T:T_\alpha \mathrm{dom}(f)\}.$$
###### Lemma 7.9.
Under the above assumption, we have the following:
1. If $`T`$ $`\stackrel{~}{}`$ is a $`_{}`$-name of an Aronszajn tree, then the pair (
~
T
~
,¯T
~
)subscript
~
𝑇
~
fragments¯𝑇
~
(\mathchoice{\oalign{$\displaystyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle{\mathbb{Q}}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\mathchoice{\oalign{$\displaystyle T$\crcr\vbox to0.60275pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle T$\crcr\vbox to0.60275pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle T$\crcr\vbox to0.60275pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle T$\crcr\vbox to0.60275pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}},\mathchoice{\oalign{$\displaystyle\bar{\mathscr{I}}_{T}$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\bar{\mathscr{I}}_{T}$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\bar{\mathscr{I}}_{T}$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\bar{\mathscr{I}}_{T}$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}) is an absolute $`(\lambda ,𝔭,)`$-NNR$`{}_{\mathrm{}_1}{}^{}{}_{}{}^{0}`$ problem over $`_{}`$.
2. If $`\lambda `$ is strongly inaccessible<sup>14</sup><sup>14</sup>14We may avoid this, if we use iterations as in \[6, Ch. VIII\], i.e. $`\overline{}`$ is only a class of $`((\lambda ),)`$, satisfying a strong version of $`\lambda `$-c.c., so $`\lambda =\mathrm{}_2`$ is sufficient, then every Aronszajn tree is special.
###### Proof.
(1) follows from \[6, Ch. V, Theorem 6.1\] and Lemma 7.7.
(2) is clear, as for any Aronszajn tree $`T`$, by (1), the forcing notion $`_T`$ satisfies clause (e) of Lemma 7.4. ∎
## 8. On Moore’s question
In this section we answer a question of Justin Moore about the consistency of strong failure of club guessing sequences with CH.
###### Definition 8.1.
Let cd:$`(\mathrm{}_1)\omega _1`$ be one-to-one. We say $`E`$ solves cd, when $`E`$ is a club of $`\omega _1`$ and for every $`\alpha E`$ we have cd$`(E(\alpha +1))<\mathrm{min}(E\backslash (\alpha +1))`$.
Justin Moore asked the following question.
###### Question 8.2.
Is the following consistent:
1. $`\mathrm{CH}`$,
2. for every one-to-one function cd from $`(\mathrm{}_1)`$ to $`\omega _1`$, there is some $`E`$ which solves it.
3. if $`\overline{C}=C_\delta :\delta <\omega _1`$ limit$``$, where $`C_\delta \delta =sup(C_\delta )`$ and $`\mathrm{otp}(C_\delta )=\omega `$, then for some club $`E`$ of $`\omega _1,(\delta )(\delta >sup(C_\delta E))`$.
We give a positive answer to the above question by proving the following theorem.
###### Theorem 8.3.
Suppose $`\mathrm{CH}`$ holds, $`\lambda =\lambda ^{<\lambda }\mathrm{}_1\kappa 2`$ and $`𝔭`$ is a reasonable parameter such that $`\lambda <\chi _0^𝔭`$ and $`\lambda \mathrm{cf}(\mathrm{}g(𝔭))`$. Let $`_{}`$ be as in Lemma 7.4 and set $`𝕍_1=𝕍^{_{}}`$. Then $`𝕍_1`$ satisfies the requirements of Question 8.2.
###### Proof.
In $`𝕍_1,\mathrm{CH}`$ holds by Lemma 7.4. Clause (c) of 8.1 holds by Lemma 7.8(b). To show that clause (b) of 8.2 is satisfied, let cd:$`(\mathrm{}_1)\omega _1`$ be a one-to-one function. Define $`=_{\text{cd}}`$ as follows:
1. $`p`$ iff $`p`$ is a closed bounded subset of $`\omega _1`$ satisfying
$$(\alpha p)[\alpha \mathrm{max}(p)\text{cd}(p(\alpha +1))<\mathrm{min}(p\backslash (\alpha +1))].$$
2. $`p_{}qp,q`$ and $`p`$ is an initial segment of $`q`$.
It is easily seen that $`_{\text{cd}}`$ is $`(<^+\omega _1)`$-proper and that it satisfies clause (e) of Lemma 7.4. The result follows immediately. ∎ |
warning/0003/hep-ph0003303.html | ar5iv | text | # A model of quark and lepton masses I: The neutrino sector
## I Introduction
There are strong indications- the latest of which came from the SuperKamiokande collaboration \- that neutrinos do have a mass, albeit a very tiny one, and, as a result, “oscillate”. The exact nature of the masses as well as the oscillation angles is an important subject which is under intense investigation . Consequently, there exists many interesting models which, in one way or another, try to accomodate most of the known data. It is perhaps prudent to think that the subject of neutrino masses and oscillation is still a very open one.
It is fair to say that the extreme smallness of neutrino masses suggests something very peculiar about these particles. This peculiarity could come from the way the neutrinos obtain their masses and/or from the very special nature of the neutrinos themselves which distinguish them from all other particles. For example, do right-handed neutrinos (present in most models of neutrino masses) carry quantum numbers which are absent in some or all other (left- or right-handed) fermions? After all, right-handed neutrinos, if present, would be singlets under $`SU(3)SU(2)_LU(1)_Y`$ anyway.
Most efforts on the problem of neutrino masses, at least on the model-building front, are concentrated on the construction of lepton mass matrices based on various ansatzes. There is one common assumption present in many of such models, which is one in which light neutrino masses arise from a see-saw mechanism . The smallness of neutrino masses would come from an expression that goes like $`m_D^2/`$, where $`m_D`$ is a Dirac mass , and $``$ is a Majorana mass which typically is very much larger than $`m_D`$ . In these models, the scale of new physics $``$, as suggested by the lightness of neutrino masses, would be some kind of Grand Unified scale or even the breaking scale of Left-Right symmetry models . (Lepton number is not a conserved quantity in this class of models.) The see-saw mechanism is a very elegant approach which is widely embraced.
However, one could not help but wonder if there might be some other mechanism for obtaining tiny neutrino masses, and if so, how it would fare compared with the see-saw mechanism. Would this new mechanism shed light on other important issues? What would be its scale of new physics? Can one find an experimental distinction between the two mechanisms? This was the topic discussed in .
At the present time, it is not clear that, if neutrinos do have a mass, it would be of the Majorana or Dirac type. As we have mentioned above, with Majorana neutrinos and the see-saw mechanism, one could “easily” obtain small neutrino masses. Now if the mass were to be of the Dirac type, one can straightforwardly write down a gauge-invariant Yukawa coupling in the SM itself (endowed with right-handed neutrinos, of course). But to obtain a small neutrino mass, one has to put in by hand a Yukawa coupling which is incredibly small, of the order of $`10^{11}`$. Such a fine tuning is highly unnatural and that might be the reason why little attention is given to the construction of models based on Dirac neutrino masses. Did we leave something out by ignoring it? What if the mass is truly of the Dirac type? Until this question is settled, it is worthwhile to investigate possible alternatives to the see-saw mechanism. This paper and a previous one propose one of such alternatives by constructing a model of Dirac neutrino masses where the smallness of their values arises dynamically. One of the criteria used in building such a model is the wish to go beyond the mere presentation of a neutrino mass matrix. In particular, we would like to see if there might be other phenomenological consequences which could be testable: New particles, new physics signals, etc.. This is the aim we set about in building our model.
The construction of the model presented in was based on the following questions: If neutrino masses were so small compared to all other known masses, would there be an appearance of a special symmetry when one lets the mass go to zero? Could this special symmetry, if it exists, be a peculiar feature of the right-handed neutral leptons alone? Could there be additional purposes for its existence other than providing a small mass for the neutrinos? In other words, can one learn something more from it? It was found in Ref. that there is indeed an interesting symmetry which acts only on the right-handed neutrinos and which, in addition to providing a reason for the smallness of the neutrino masses, also constrains the nature (even or odd) of the number of generations. Furthermore, the way in which neutrino masses are constructed can be used to build a model for charged lepton and quark masses. In addition, this particular way of constructing masses might even have some bearing on the strong CP problem. Last but not least: Are there additional tests of various neutrino models other than neutrino oscillations? For the see-saw mechanism with Majorana neutrinos, one already sees that one of such additional signals is, for example, the phenomenon of neutrinoless double beta decay. As it will be presented below, the addtional signals of the model presented here will involve a number of very concrete predictions: the absence of neutrinoless double beta decay, the possible presence of “low mass” ( a couple of hundreds of GeV e.g.) vector-like fermions, among other things. In particular, the detection of these vector-like fermions do not in any way involve neutrinos.
One particularly important feature of our model is the following predictions for neutrino oscillations, assuming only the validity of the atmostpheric and solar neutrino data: 1) The three light neutrinos are nearly degenerate; 2) If the light neutrinos have a mass large enough to form a component of the Hot Dark Matter (HDM) then only the MSW solution to the solar neutrino oscillation is favored; 3) If the vacuum solution to the solar neutrino problem turns out to be the correct one, our model will only be able to accomodate tiny neutrino masses, around $`10^3eV`$ or less, ruling out near-degenerate neutrinos as components of HDM. As a result, in our model, one cannot have both vacuum solution and HDM. We will show below the correlation between the masses and the differences of mass squared, $`\mathrm{\Delta }m^2`$, which enter the neutrino oscillation phenomena.
Assuming the existence of the aforementioned symmetry, how can one construct Dirac neutrino masses to be dynamically small? By “dynamically”, it is meant that the mass is zero at tree level and that any non-zero value would have to arise at the one-loop (or more) level. Now, the peculiar (and toughest) thing about neutrinos is the fact that their mass is so small- at least eleven orders of magnitude smaller than the electroweak scale. In constructing our model for Dirac neutrino masses, it is then reasonable to ask under what conditions would the dynamical Dirac mass of the neutrinos obtained at the one loop level be “naturally” small, i.e. devoid of excessive fine tuning. In this paper, we present the following interesting results: In the four-generation model, it is found that the fourth neutrino can be naturally heavy while the other three obtain their masses at one loop, with the result that these masses can be tiny provided some ratios of masses of particles which participate in the loop integration are “large”, regardless of their actual values. This is interesting because, as we shall see below, some of the particles which participate in the loop integration, in particular the lightest ones, can have masses as low as a few hundred GeVs and which could provide a direct test of this model. We will also see that, in order to obtain very small neutrino masses, at least one of the particles needs to be much heavier than the lightest one- a result which is somewhat reminescent of the see-saw mechanism. We will also see that the mass of the light neutrinos is intrinsically tied to the extra global symmetry present in the scalar sector of the model. In fact, the extra Nambu-Goldstone (NG) bosons which are not absorbed by the (family and $`SU(2)_{\nu _R}`$) gauge bosons acquire a mass due to the presence of the gauge-invariant “cross-coupling” terms in the potential which explicitely break the extra global symmetry.
The above brief statement will be made clearer in the discussion of neutrino masses. Notice, in particular, that the result given for light neutrino masses in is only a very special case of the present discussion.
The plan of the paper is as follows. First, the model is presented with a description of the gauge structure along with its particle content. It is shown how a new symmetry prevents neutrinos from obtaining a mass unless it is broken. Next, the special properties of this extra symmetry associated with the right-handed neutrinos are discussed. In particular, if that symmetry is a chiral $`SU(2)`$ as is the case in this paper, nontrivial constraints coming from the nonperturbative Witten anomaly can be applied to the nature of the number of families. This is the extra bonus mentioned above. The paper then proceeds to discuss the generation of light neutrino masses, principally by radiative corrections of the type mentioned above. It is then followed by a discussion of the neutrino mass matrix. In particular, we will present the correlation between the values of the neutrino masses and $`\mathrm{\Delta }m^2`$. Most importantly, we will show how $`\mathrm{\Delta }m^2`$ increases or decreases with the masses themselves, with two resulting implications: either one has HDM and MSW or vacuum solution and no HDM. Either of these solutions will have an important cosmological implication. We end the paper with a brief discussion of the charged lepton mass matrix, the primary purpose of which being the wish to complete the discussion by presenting some examples on what the oscillation angles might look like. A followed-up paper will deal seperately with the charged lepton sector and, as a consequence, with a full discussion of the angles.
We would like to emphasize for the purpose of clarity that the charged lepton sector (which will be dealt with in a separate paper) is different in structure from the neutrino sector, as we shall see below, and does not have the same hierarchical structure. The fact that, in this model, the three light neutrinos are nearly degenerate does not imply that it would be the same in the charged lepton sector. In fact, it is not as we will show in a subsequent paper.
Finally, a section will be devoted to various other phenomenological implications of the model.
We shall assume throughout this paper the existence of right-handed neutrinos.
Since this manuscript is meant to be comprehensive, and hence lengthy, one could skip the three subsections of the next section, after first reading its introduction.(Its reading is nevertheless recommended because the physics motivations are discussed there.)
## II A Model
It is well-known that all that is needed to give neutrinos a mass is to simply add extra right-handed neutrinos to the Standard Model. One can then construct a (Dirac) mass term with an arbitrary Yukawa coupling, $`g_\nu \overline{l}_L\varphi \nu _R+H.c.`$, which can be made to be as small as one wishes. This, of course, is unsatisfactory because, if neutrinos have masses in the $`eV`$ range or less, this would require the Yukawa coupling, $`g_\nu `$, to be of O($`10^{11}`$) or less. Fine tuning to such a precision is normally considered to be unnatural. At this point, one might be tempted to try to explain this fact by simply invoking a fourth generation with a democratic mass matrix, at least for the neutrinos, as has been done by Ref. . The diagonalization of the neutrino mass matrix would then give one heavy eigenstate and three massless states. By adding some arbitrary phases to the mass matrix, one can “provide” a small mass ( depending on the values of those phases) to the three neutrinos. This purely phenomenological ansatz (Ref. ) appears to “fit” the recent data on neutrino oscillations with the appropriate choices of the phases. However, the fourth generation lepton masses came out to be extremely heavy and split, which practically seems to be ruled out by analyses of precision experiments .
In , a model of Dirac neutrino masses was constructed and based on a four generation scenario that was very different from the democratic ansatz made in . One of the reasons for using such a scenario is the fact that, as of the present time, a fourth generation is not ruled out by experiment and, as a consequence, it is interesting to explore its possible implications. A recent review gave a comprehensive discussion of various topics concerning quarks and leptons beyond the third generation, including the present experimental status and future searches.
If a fourth generation were to be used in the investigation of neutrino masses, one should keep in mind various phenomenological constraints concerning not only leptons but also quarks. For instance, constraints on the $`\rho `$ parameter limit the mass splitting within each doublet of extra quarks and leptons: the up and down members of a fourth generation should be very close in mass. They should be long-lived enough to escape present detection. This, in turns, tells us something about the mixing between the fourth generation and the other three. All of these issues have been described in . In the construction of the model presented in Ref. , these phenomenological constraints were kept in mind.
As mentioned briefly in the Introduction, our approach, as described in Ref. , is based on a dynamical justification for for the small value of the neutrino Yukawa couplings. The question that was asked was: Could there be a scenario in which a symmetry appears as one lets the Yukawa coupling go to zero? The tiny Yukawa coupling which would give the neutrino a very small mass would then arise dynamically when that symmetry is broken. These Yukawa couplings then appear as effective couplings which could be small for dynamical reasons and are not fundamental parameters that are put in by hand and which are needed to be fine tuned. What is the nature of that symmetry and how a dynamical Yukawa coupling appears will be the subject of this section and the following two.
It is obvious that an extension of the Standard Model(SM) is needed in addressing the above issues. One simply cannot stay solely within the SM if one wishes to deal with the mass of the neutrinos. What it is that one needs when one goes beyond the SM is a matter of taste, modulo a very obvious requirement: predictability of new phenomena or particles which can be tested.
We first describe the model, presenting its gauge structure and representations. Next, explanations are provided for the reasons behind the choices of the extended gauge group and its particle content. The crucial assumption here is the existence of two new symmetries, one of which will be particular to right-handed neutrinos, as alluded to earlier, and the other one is a family gauge symmetry. As we shall see below, it is the breaking of these new symmetries that will give a mass to the neutrinos.
In this work, the SM is extended in the following way. Generically, it takes the form: $`SU(3)_cSU(2)_LU(1)_Y(Familysymmetry)(righthandedneutrinospecialsymmetry)`$. Why a “Family symmetry”? This is so for two reasons: a) We wish to investigate the family replication problem and the mixing among different generations; b) The special symmetry endowed by the right-handed neutrinos might have some bearing on the family symmetry itself. After all, if one would like to investigate the family problem, some kind of family symmetry-be it discrete or continuous, global or gauge- is needed. Why a special symmetry for the right-handed neutrinos? The reasons were already expounded above: To provide a framework for an understanding of the smallness of neutrino masses. Our next task is then to determine what this special symmetry might be and what form the family symmetry might take.
Our model is described by:
$$SU(3)_cSU(2)_LU(1)_YSO(N_f)SU(2)_{\nu _R}$$
(1)
where $`SO(N_f)`$ and $`SU(2)_{\nu _R}`$ are the family gauge group and the special gauge group for the right-handed neutrinos respectively. The particle content of the model is listed in Table 1. Notice that we have denoted the right-handed neutrinos by $`\eta _R=(\nu _R^\alpha ,\stackrel{~}{\nu }_R^\alpha )`$ because they are assumed to transform as doublets under $`SU(2)_{\nu _R}`$. The two options listed for the right-handed neutrinos as well as the meaning of the non-standard particles will be discussed below. We would first like to explain the choices of the extra gauge groups. Here, the extra symmetries are chosen to be gauge symmetries because, as it is well known, powerful constraints can be obtained from models built on the gauge assumption.
### A Why $`SU(2)_{\nu _R}`$?
Let us first look at Table 1. In this model, all standard (left-handed and right-handed) particles are singlets under $`SU(2)_{\nu _R}`$. Hence the subscript $`\nu _R`$. In this respect, $`SU(2)_{\nu _R}`$ is very different from $`SU(2)_R`$ of the popular Left-Right model . In that model, right-handed quarks and leptons form doublets under $`SU(2)_R`$, for every family. Because of our assignment, all weak interactions among standard particles are pure V-A, in contrast with the Left-Right model. What is the motivation behind our choice that makes it so different from the Left-Right model? To answer that question, let us recall an interesting feature of chiral $`SU(2)`$: the presence or absence of the so-called Witten global anomaly.
If chiral fermions transform as doublets under $`SU(2)`$, there exists a nonperturbative anomaly- the so-called Witten anomaly \- associated with an odd number of doublets. Briefly speaking, this is so because the fermionic determinant $`\sqrt{deti\overline{)}(A_\mu )}`$ changes sign under a “large” gauge transformation $`A_\mu ^U=U^1A_\mu UiU^1_\mu U`$ if the number of chiral doublets is odd. This would make the partition function $`Z`$ vanish and the theory would be ill-defined. This nonperturbative anomaly would then require the number of Weyl doublets to be even in order for the theory to be consistent. (This ambiguity in sign stems from the fact that the fourth homotopy group $`\mathrm{\Pi }_4(SU(2))=Z_2`$.) Other groups that also have similar non-trivial constraints are $`Sp(N)`$ for any $`N`$ and $`O(N)`$ for $`N5`$.
It is amusing to recall a well-known but forgotten fact about the SM. There the chiral gauge group is $`SU(2)_L`$. Each family contains one lepton and three quark doublets and, as such, is free from the global Witten anomaly. (Let us recall that the cancellation of the perturbative triangle anomaly in the SM only relates the lepton charge to that of the quark.) If, instead of three, the number of colors, $`N_c`$, were arbitrary, the freedom from such an anomaly would require $`1+N_c`$ to be even, and hence, $`N_c`$ to be odd, namely $`N_c`$ = 3, 5,… Why nature choses $`N_c=3`$ instead of some other odd number is a question which can only be answered in the context of some deeper theory such as e.g. $`SU(5)`$. Although the Witten anomaly does not fix the size of $`N_c`$, it is nevertheless a powerful constraint in the sense that, once a fermion content is known (e.g. one color singlet (leptons) and one fundamental representation (quarks) in the SM), $`N_c`$ is constrained (e.g. odd in the case of the SM).
The above simple lesson taught us something about the powerful constraint that a chiral $`SU(2)`$ exerts on the number of chiral doublets. This is the reason why it is chosen to be the special symmetry of the right-handed neutrinos. Let us contrast the constraint coming from $`SU(2)_{\nu _R}`$ with that coming from $`SU(2)_R`$ (Left-Right model). For our model, with $`SU(2)_{\nu _R}`$, only $`\eta _R`$ transforms as doublets. Absence from the Witten anomaly then requires the number of such doublets to be even. If $`\eta _R`$ carries, in addition, family indices then the anomaly requirement restricts the number of generations to be even such as in Option 1 as indicated in Table 1. If there exists an $`\eta _R`$ which is a family singlet (denoted by $`\eta _R^{}`$), the number of generations would be odd such as in Option 2 of Table 1. With the Left-Right model, each family contains four doublets of $`SU(2)_R`$: $`(\nu _R,e_R)`$ and $`(u_R,d_R)_i`$ with $`i=1,..,3`$. Therefore, the Witten anomaly requirement is automatically satisfied per family. This is one of the few differences between our model and the Left-Right model.
A final word of caution is in order here. Although the Witten anomaly constraint allows us to make a statement on the evenness or oddness of the number of generations- a subject to which we shall come back in the next subsection, it does not determine that number. This should come from a deeper and as-yet-unknown theory. Our goal is much more modest: Given a fermion content (Option 1 or 2 below), we can say whether or not the number of generations is odd or even, and that is all. We shall however try to constraint that number from a different route which is more phenomenological in nature, and point out the differences between Option 1 and 2.
### B Why $`SO(N_f)`$?
In the construction of any model, there is a time-honored requirement: the absence of the perturbative triangle anomaly. Even if the Witten anomaly were absent, this requirement is a must for any gauge theory. (It just happens that,in the SM, both requirements are simultaneously satisfied.) In our case, if a family index is assigned to all standard fermions and to $`\eta _R`$, the family gauge group that is chosen cannot be a vector-like theory, which is anomaly-free, because $`\eta _R`$ posseses an additional quantum number, that of $`SU(2)_{\nu _R}`$. This is unlike QCD or even the Left-Right model if left and right-handed fermions carry similar family quantum numbers. A safe group and representations have to be chosen.
The choice made in this paper is $`SO(N_f)`$ for the family gauge group, with chiral (left- and right-handed) fermions transforming as (real) vector representations with $`N_f`$ components each. As such, the model is also free of the perturbative triangle anomaly.
Our model based on $`SU(3)_cSU(2)_LU(1)_YSO(N_f)SU(2)_{\nu _R}`$ with an even number of $`SU(2)_{\nu _R}`$\- doublets and chiral fermions transforming as vector representations of $`SO(N_f)`$ is free from both nonperturbative and perturbative anomalies.
### C Constraints on $`N_f`$
As shown in Table 1, there are two options for $`\eta _R`$, each of which should contain an even number of $`SU(2)_{\nu _R}`$ doublets.
a) Option 1:
$`\eta _R^\alpha `$ carries the family index $`\alpha =1..N_f`$ where $`N_f=2,4,6,8,..`$.
b) Option 2:
Here we have $`\eta _R^{}`$ (a family singlet) and $`\eta _R^\alpha `$. The constraint is now $`1+N_f=`$ even, which means that $`N_f=3,5,7,..`$ (excluding the trivial case of 1 family).
Unlike the SM where one knows the fermion content for each family, i.e. quarks and leptons, and hence the nature of $`N_c`$\- it is odd\- our scenario involves incomplete experimental informations, and as such, the nature (odd or even) of $`N_f`$ cannot be completely fixed. Each choice, however, represents a distinct particle content (no family singlets for the even option and one family singlet for the odd option) which implies a possible distinct route for a yet-unknown unification.
Recognizing the fact that there are deep differences between the even and odd options-a point to be discussed below- and in the absence of a deeper theory, one might wonder what can be done to narrow down the choices, not between odd or even, but within each option itself. Below we present an argument that could help in finding a way to further restrict $`N_f`$. This argument is only suggestive, being a combination of “theoretical prejudice” and phenomenological constraint.
One might require that gauge couplings are free from Landau singularities below the Planck scale in such a way that unification of the SM gauge couplings, if it exists, occurs in the perturbative regime . With this criterion, one can see that the even option can only accomodate $`N_f=2,4,6`$, while the odd option can only accomodate $`N_f=3,5`$. This is because for $`N_f7`$, one or more gauge couplings will “blow up” before the Planck scale. There are no reasons, in the absence of a deeper theory, to rule out any of the above choices. This will require other yet-unknown conditions. The only thing one can say, in the context of our model, is that electroweak precision experiments appear to rule out $`N_f5`$ and and that existential facts tell us that $`N_f`$ is at least three. This leaves us with the choice $`N_f=4`$ for the even option and $`N_f=3`$ for the odd option.
If $`N_f4`$ comes from the above argument, what then is the role of the Witten anomaly in all of this? It tells us about the particle content of the right-handed neutrinos. For $`N_f`$ = 4, the right-handed neutrinos are simply $`\eta _R=(1,1,0,4,2)`$ while for $`N_f`$ = 3, one has $`\eta _R=(1,1,0,3,2)`$ plus a family singlet $`\eta _R^{}=(1,1,0,1,2)`$. What observed differences can there be between these two options? The former predicts the existence of a fourth generation whose consequences have been recently discussed in Refs. and . The latter predicts the existence of a neutral family-singlet $`\eta _R^{}`$ (doublet under $`SU(2)_{\nu _R}`$) which could have cosmological consequences of a yet-unknown nature. In addition, as we point out below, it appears that the even option prefers three almost degenerate light neutrinos while the odd option prefers a hierarchical structure for the light neutrinos. If a fourth generation is discovered, which alone does not necessarily imply the even option presented here, and if the light neutrino masses are convincingly “proven” to be nearly degenerate (instead of a hierarchical structure), the even option might be viable. Furthermore, as we shall see below, another possibility for testing this model is to look for signals of some of the lightest particles- the vector-like fermions- which participate in the loop diagram of Fig. 1. As discussed below, the light neutrino masses depend only on the ratios of these masses and not on their magnitudes and these vector-like fermions can be as light as a few hundred GeVs.
## III Neutrino Masses
This section will be devoted to the discussion of how neutrino masses can be generated in our model for Option 1. We shall comment on Option 2 at the end of the manuscript. We shall concentrate only on the lepton sector and, in particular, on the neutrino one, leaving the full discussion of the charged lepton and quark sectors for a subsequent publication.
Since we will be dealing only with Dirac neutrino masses, we shall require that all fermions be endowed with a global $`BL`$ symmetry. Since we are concerned only with leptons in this section, a global $`L`$ symmetry is sufficient for the present purpose. This global $`L`$ symmetry would prevent a Majorana mass term of the type $`\eta _R^{i\alpha }\eta _{i\alpha R}`$, where $`i=1,2`$ and $`\alpha =1,..,4`$. Only Dirac masses will be allowed.
There might be other suggestive reasons as to why Dirac masses for the neutrinos might be attractive. For example, a combined fit of massive neutrinos as components of Hot Dark Matter (HDM) and atmospheric neutrino oscillations seems to prefer a scenario in which two or three light neutrinos are nearly degenerate and have mass in the O(eV) range. Recent data on neutrinoless double beta decay (or absence thereof) appear to rule out Majorana neutrinos heavier than 0.2 eV, at least in the simplest versions. Here it will be shown how, in our scenario, one can obtain three near-degenerate neutrinos whose mass can be of the order of a few eV’s and is of the Dirac type. Consequently, in our model, there will be no neutrinoless double beta decay, and hence no contraint on the Dirac neutrino masses from such a search.
As we have discussed above, Option 1 contains no family-singlet fermion field and freedom from the Witten anomaly dictates that the number of families should be even. Furthermore, we have argued that this even number should be four. As a result, the gauge group for this option is:
$$SU(3)_cSU(2)_LU(1)_YSO(4)SU(2)_{\nu _R}$$
(2)
The reader is referred to Table 1 for a list of particles that participate in this model.
### A Computation of the diagonal elements of the $`4\times 4`$ neutrino mass matrix
Without the extra vector-like fermions, $`F`$, $`M_1`$ and $`M_2`$, the only gauge-invariant Yukawa coupling involving leptons would be $`_Y=g_E\overline{l}_L^\alpha \varphi e_{\alpha R}+H.c.`$, (where $`\alpha =1,..,4`$ is the family index), giving rise to equal masses for the charged leptons. Unbroken $`SU(2)_{\nu R}`$ forbids a similar term for the neutrinos and they remain massless at this level. (Notice that, since we are only interested in Dirac neutrino masses, a gauge-invariant Majorana mass term of the type $`\eta _R^{i\alpha }\eta _{i\alpha R}`$ is forbidden by $`L`$ symmetry.) We know that the charged leptons are not degenerate in mass. We also know that the width of the Z boson constrains the mass of the fourth neutrino to be larger than half the Z mass. This is where the vector-like fermions listed in Table 1 come in. Because of their vector-like nature, they can have arbitrary gauge-invariant bare masses. It is seen below that some of these masses can be as low as a few hundreds GeVs and are thus accessible to future experimental searches.
The Yukawa part of the Lagrangian involving leptons can be written as
$`_{Lepton}^Y`$ $`=`$ $`g_E\overline{l}_L^\alpha \varphi e_{\alpha R}+G_1\overline{l}_L^\alpha \mathrm{\Omega }_\alpha F_R+G_{M_1}\overline{F}_L\varphi _{1R}+G_{M_2}\overline{F}_L\stackrel{~}{\varphi }_{2R}+G_2\overline{}_{1L}\mathrm{\Omega }_\alpha e_R^\alpha +`$ (4)
$`G_3\overline{}_{2L}\rho _m^\alpha \eta _{\alpha R}^m+M_F\overline{F}_LF_R+M_1\overline{}_{1L}_{1R}+M_2\overline{}_{2L}_{2R}+h.c.`$
The assumption of an unbroken $`L`$ symmetry forbids the presence of Majorana mass terms as mentioned above.
Notice that the values of $`M_{F,1,2}`$ are arbitrary. What they might be will be the subject of the discussion presented below. After integrating out the $`F`$, $`_1`$, and $`_2`$ fields, the relevant part of the effective Lagrangian below $`M_{F,1,2}`$ reads
$`_{Lepton}^{Y,eff}`$ $`=`$ $`g_E\overline{l}_L^\alpha \varphi e_{\alpha R}+G_E\overline{l}_L^\alpha (\mathrm{\Omega }_\alpha \varphi \mathrm{\Omega }^\beta )e_{\beta R}+`$ (6)
$`G_N\overline{l}_L^\alpha (\mathrm{\Omega }_\alpha \stackrel{~}{\varphi }\rho _i^\beta )\eta _{\beta R}^i+H.c.,`$
where
$$G_E=\frac{G_1G_{M_1}G_2}{M_FM_1};G_N=\frac{G_1G_{M_2}G_3}{M_FM_2}.$$
(7)
This is a tree-level effective Lagrangian whose consequences are now presented.
Let us discuss the implication of each term on the right-hand side of Eq. (6). As stated in the preceding paragraph, the first term gives rise to equal masses for the charged leptons. The second term would lift the degeneracy of the charged lepton sector once $`\mathrm{\Omega }`$ acquires a vacuum expectation value (VEV). The third term gives rise to a neutrino mass once both $`\mathrm{\Omega }`$ and $`\rho `$ acquire a VEV. It is clear that, in our model, neutrino masses can appear only when both $`SO(4)`$ and $`SU(2)_{\nu _R}`$ are spontaneously broken while the charged lepton masses are non zero (but equal) even if $`SO(4)`$ is unbroken. Only when $`SO(4)`$ is broken will the charged lepton mass degeneracy be lifted.
Let us assume: $`<\mathrm{\Omega }>=(0,0,0,V)`$ and $`<\rho >=(0,0,0,V^{}s_1)`$, where $`s_1=\left(\begin{array}{c}1\\ 0\end{array}\right)`$. Notice that each component (under $`SO(4)`$) of $`\rho `$ transforms as a doublet under $`SU(2)_{\nu _R}`$. If we denote the 4th element of $`\eta _R`$ by $`(N_R,\stackrel{~}{N}_R)`$, one can use the above two VEV’s along with $`<\varphi >=(0,v/\sqrt{2})`$ ( $`v`$ 246 GeV) in Eq.(6) to write down a Dirac mass term for the 4th generation neutrino, namely
$$\stackrel{~}{G}_N\frac{v}{\sqrt{2}}\overline{N}_LN_R+h.c.;\stackrel{~}{G}_N=G_1G_{M_2}G_3\frac{VV^{}}{M_FM_2},$$
(8)
giving
$$m_N=\stackrel{~}{G_N}\frac{v}{\sqrt{2}}.$$
(9)
At tree level, all other neutrinos are massless. Their masses arise at the one-loop level as shown below. The Dirac mass of the fourth neutrino could be rather heavy. In fact, it is not unreasonable to expect $`G_1`$, $`G_{M_2}`$ and $`G_3`$ to be of the order of unity. In consequence, as long as
$$VV^{}/M_FM_2O(1),$$
(10)
one might expect the fourth neutrino to be even as heavy as 175 GeV. Certainly, the LEP bound of $`M_Z/2`$ can easily be satisfied.
Why are the other three neutrinos massless at tree level? Firstly, it is so because, from Eq. (4) and Eq. (6), one can see that, after integrating out the heavy vector-like fermions, there is no (tree-level, dimension 6) operator which contains,as a factor, a term such as $`\overline{l}_L^m\stackrel{~}{\varphi }\eta _{iR}^m`$, where $`m=1,2,3`$ is a family index, which would give rise to a mass term for the three light neutrinos. An effective (dimension 6) operator which contains the aforementioned term would necessarily come from a loop integration such as the one shown in Fig. 1. Just like the various terms which appear in Eq. (6), this effective operator would also contain the scalar fields $`\mathrm{\Omega }`$ and $`\rho `$. It would appear as
$$\overline{l}_L^m\stackrel{~}{\varphi }\eta _{iR}^m(\mathrm{\Omega }_\alpha \rho ^{\alpha i}).$$
(11)
As pointed out in the Appendix, a term such as $`(\mathrm{\Omega }^\alpha \rho _{\alpha i})`$ appears as part of a quartic term in the potential which explicitely breaks the extra global symmetry that the scalar sector posesses. As a result, the extra NG bosons are, in fact, pseudo NG bosons and acquire a mass which is proportional to the coupling $`\lambda _4`$ as shown in Eq. (149) of the Appendix.
In order to compute the one-loop contributions to neutrino masses, let us recall, in this section, the results obtained in the Appendix concerning the relevant mass eigenstates in the scalar sector. We have
$$H_4=\mathrm{cos}\alpha \stackrel{~}{H}_4\mathrm{sin}\alpha \stackrel{~}{h}_4,$$
(13)
$$h_4=\mathrm{sin}\alpha \stackrel{~}{H}_4+\mathrm{cos}\alpha \stackrel{~}{h}_4,$$
(14)
$$\mathrm{\Omega }_i=\mathrm{cos}\beta \stackrel{~}{\mathrm{\Omega }}_i\mathrm{sin}\beta Re\stackrel{~}{\rho }_i,$$
(15)
$$Re\rho _i=\mathrm{sin}\beta \stackrel{~}{\mathrm{\Omega }}_i+\mathrm{cos}\beta Re\stackrel{~}{\rho }_i,$$
(16)
where $`i=1,2,3`$ and where the states with the $`\stackrel{~}{}`$ sign are mass eigenstates. The Yukawa couplings which will be involved in the computation of neutrino masses can now be written in terms of the mass eigenstates. For example, $`G_1\overline{l}_L^\alpha \mathrm{\Omega }_\alpha F_R`$ can be written as
$$G_1\overline{l}_L^4\mathrm{\Omega }_4F_R=G_1\overline{l}_L^4(\mathrm{cos}\alpha \stackrel{~}{H}_4\mathrm{sin}\alpha \stackrel{~}{h}_4)F_R,$$
(17)
$$G_1\overline{l}_L^i\mathrm{\Omega }_iF_R=G_1\overline{l}_L^i(\mathrm{cos}\beta \stackrel{~}{\mathrm{\Omega }}_i\mathrm{sin}\beta Re\stackrel{~}{\rho }_i)F_R.$$
(18)
where $`i=1,2,3`$. Also, $`G_3\overline{M}_{2L}\rho _m^\alpha \eta _{\alpha R}^m`$ ($`m=1,2`$) can be now written as
$$G_3\overline{M}_{2L}\rho _1^4\eta _{4R}^1=G_3\overline{M}_{2L}(\mathrm{sin}\alpha \stackrel{~}{H}_4+\mathrm{cos}\alpha \stackrel{~}{h}_4+iIm\rho _4)_1\eta _{4R}^1,$$
(19)
$$G_3\overline{M}_{2L}\rho _1^i\eta _{iR}^1=G_3\overline{M}_{2L}(\mathrm{sin}\beta \stackrel{~}{\mathrm{\Omega }}_i+\mathrm{cos}\beta Re\stackrel{~}{\rho }_i+iIm\rho _i)_1\eta _R^{1,i}.$$
(20)
The above equations, in addition to $`G_{M_2}\overline{F}_L\stackrel{~}{\varphi }M_{2R}`$, form the basis for constructing the one-loop diagrams as shown in Fig.1. As one can immediately see, the only scalars that participate in the loop integration are $`\stackrel{~}{H}_4`$, $`\stackrel{~}{h}_4`$, $`\stackrel{~}{\mathrm{\Omega }}_i`$, and $`\stackrel{~}{\rho }_i`$. The contributions to the light neutrino masses will contain a factor $`\mathrm{cos}\beta \mathrm{sin}\beta =\mathrm{sin}(2\beta )/2`$ for $`\stackrel{~}{\mathrm{\Omega }}_i`$ and -$`\mathrm{cos}\beta \mathrm{sin}\beta `$ for $`Re\stackrel{~}{\rho }_i`$.
The masses of the physical Higgs scalars, $`H_4`$ and $`h_4`$, and those of the pseudo NG bosons, $`Re\stackrel{~}{\rho }_i`$ ($`i=1,2,3`$), are given by Eqs. (136,149) in the Appendix. Since the one-loop contributions to the 4th neutrino mass are expected to be small compared with its tree-level value, we shall concentrate in this section on the light neutrino masses. There we shall be concerned only with $`\stackrel{~}{\mathrm{\Omega }}_i`$ (NG bosons) and $`Re\stackrel{~}{\rho }_i`$ (pseudo NG bosons) ($`i=1,2,3`$). In the ’tHooft-Feynman gauge, the NG bosons will have a propagator with a mass which is that of the family gauge bosons. We shall denote it by $`M_G`$. We shall call the mass of the pseudo NG bosons, $`M_P`$.
The result obtained from the diagrams as shown in Fig. 1 for the three light neutrinos is
$$m_\nu =\stackrel{~}{G}_\nu \frac{v}{\sqrt{2}},$$
(21)
where
$$\stackrel{~}{G}_\nu =G_1G_{M2}G_3\frac{\mathrm{sin}(2\beta )}{32\pi ^2}(I(\stackrel{~}{\mathrm{\Omega }})I(Re\stackrel{~}{\rho })),$$
(22)
and where
$$I(\stackrel{~}{\mathrm{\Omega }})I(Re\stackrel{~}{\rho })=\frac{1}{M_FM_2}\{\frac{M_F[M_F^2(M_G^2\mathrm{ln}(\frac{M_G^2}{M_F^2})M_P^2\mathrm{ln}(\frac{M_P^2}{M_F^2}))+M_G^2M_P^2\mathrm{ln}(\frac{M_P^2}{M_G^2})]}{(M_G^2M_F^2)(M_P^2M_F^2)}(M_FM_2)\}.$$
(23)
For notational convenience, we shall define:
$$\mathrm{\Delta }I(G,P)I(\stackrel{~}{\mathrm{\Omega }})I(Re\stackrel{~}{\rho }),$$
(24)
It is convenient to express the mass of the light neutrinos by the following ratio:
$$\frac{m_\nu }{m_N}=\frac{M_FM_2}{VV^{}}\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\mathrm{\Delta }I(G,P),$$
(25)
where $`m_N`$ is defined by Eq. (9).
One should mention for completeness the tiny one-loop contribution to the 4th neutrino mass. If we denote by this contribution by $`\delta m_4`$, it is straigthforward to see that it is given precisely by the same formula for the light neutrino mass, Eq. (21), with the following replacements: $`\beta \alpha `$, $`M_GM_{H_4}`$, $`M_PM_{h_4}`$, namely
$$\delta m_4=\stackrel{~}{G}_4\frac{v}{\sqrt{2}},$$
(27)
$$\stackrel{~}{G}_4=G_1G_{M2}G_3\frac{\mathrm{sin}(2\alpha )}{32\pi ^2}\mathrm{\Delta }I(G,P),$$
(28)
where the form of $`I(\stackrel{~}{H}_4)I(\stackrel{~}{h}_4)`$ is identical to Eq. (23) with the replacements as mentioned above. This contribution will play an insignificant role in the mass matrix, but it has to be mentioned for completeness.
The above results were obtained at one loop. One wonders if higher loop contributions might be significant. It turns out that, because of the nature of the interactions, the next correction occurs at the three loop level. It means that the correction to the one-loop light neutrino mass is at the two-loop order. Considering that already the one-loop result is O($`<10^{10}`$), a two-loop correction to that result would most likely be insignificant, even for the mass splitting to be discussed below. Above all, the experimental results are far from being precise enough to even contemplate such a tiny correction. From hereon, we shall assume that these three-loop corrections are insignificant in the computation of the mass splittings.
At this stage, the three light neutrinos are degenerate. A discussion of the lifting of the degeneracy will follow a more general discussion of the implications of Eq. (25). It is clear that the “light family” symmetry would have to be broken in order for the “light” fermions to mix. It is also clear that the neutrino masses (one heavy and three light) derived so far represent only the diagonal elements of a $`4\times 4`$ neutrino mass matrix. If the discussion presented in this section on light neutrino masses is to be at all interesting, it is imperative to assume that the bulk of at least one, if not all, of the light neutrino masses comes from Eq. (21).
At this point, an important remark is in order here. As we have stressed above, the near-degeneracy of the light neutrinos in no way implies that a similar situation will occur in the charged lepton sector. In fact, we will show in a separate paper that this will not be the case.
Under what conditions will $`\stackrel{~}{G}_\nu `$ be of the order of $`10^{11}`$ or less? First of all, as we have seen from Eq. (9), in order to have a “heavy” fourth neutrino, one should have $`G_1G_{M2}G_3\frac{VV^{}}{M_FM_2}O(1)`$. This puts a condition on the angle $`\beta `$ itself, namely ($`\mathrm{tan}\beta V^{}/V`$)
$$\mathrm{tan}\beta \frac{1}{G_1G_{M2}G_3}\frac{M_FM_2}{V^2}.$$
(29)
As we have stated earlier, it is not unreasonable to assume that $`G_1`$, $`G_{M_2}`$ and $`G_3`$ to be of the order of unity. With $`M_G^2g^2V^2`$ (where $`g`$ is the $`SO(4)`$ gauge coupling), Eq. (29) becomes
$$\mathrm{tan}\beta g^2\frac{M_F}{M_G}\frac{M_2}{M_G}.$$
(30)
The above estimate for the constraint on the angle $`\beta `$ will be used in our computation of the light neutrino masses. With this in mind, we can now proceed to make an estimate of the ratio $`m_\nu /m_N`$, where now $`VV^{}/M_FM_2O(1)`$ and Eq. (25) becomes
$$\frac{m_\nu }{m_N}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}(I(\stackrel{~}{\mathrm{\Omega }})I(Re\stackrel{~}{\rho })).$$
(31)
As we have seen above, the result (31) depends only on ratios of masses of the particles in the loop integral and not on their absolute values. Because of that fact, the results will be shown in units of $`M_F`$ which can be as small or as large as one wishes.
Before moving on to discuss the implications of Eqs. (25) and (31), one remark is in order here. From Eq. (23), one can see that the light neutrino mass vanishes when $`M_G=M_P`$. Since there is no reason (as far as the present construction of the model is concerned) for this equality to be valid, we shall dismiss this possibility. We shall concentrate instead on the criteria for having small $`m_\nu `$ for arbitrary $`M_G`$ and $`M_P`$ (and $`M_F`$ and $`M_2`$ as well).
The results are shown in Figs. 2, 3, 4 and 5. A few comments are in order here. First of all, as we have mentioned above, our results depend on ratios of the four masses which enter the loop integral: $`M_F`$, $`M_P`$, $`M_G`$, and $`M_2`$. One can symbolically denote one of the masses as $`M=1`$, and the other three will be multiples of that chosen one. Which one should be chosen is a matter of convenience and phenomenological interest. In particular, we choose $`M_F=1`$ because there is a possibility that the vector-like fermions $`F`$ could be detected if their masses are low enough.
A glance at Figs. 2-5 reveal that it is relatively easy to obtain a very small ratio $`Rm_\nu /m_N`$. In particular, one can see that large values of $`M_2`$, the mass of the singlet fermion field $`_2`$, are sufficient to obtain small values for $`Rm_\nu /m_N`$. For instance, one can see that, roughly speaking, $`Rm_\nu /m_N10^{11}`$ when $`M_210^6`$ (in units of $`M_F`$). Although conceptually quite different, the above fact is very reminescent of the see-saw mechanism in that there is one large scale: Majorana for see-saw, $`M_2`$ for this scenario, and one “small” scale: Dirac mass $`m_D`$ for see-saw, $`M_F`$ for this scenario. The important point that we wish to make is the fact that the general result obtained here, namely the smallness of light neutrino masses, does not depend on one particular combination of masses which would imply fine tuning, a point which was not made quite clear in Ref. , but only on “large” ratio of masses whatever they might be. In this sense, the smallness of neutrino masses in our scenario is no less natural than the ones obtained from the see-saw mechanism.
In Figs. 2-5, we show the results for the case $`M_2>M_G`$. There is, of course, absolutely no reason for this ordering. It is a matter of presentation. We obtain exactly the same results with the roles of $`M_2`$ and $`M_G`$ reversed. As can be inferred from the figures, for a given value of $`M_P`$ ($`M_F=1`$), $`Rm_\nu /m_N10^{11}`$ if the ratio $`M_G/M_2`$ is below a certain value. For example, for $`M_G10^5`$, one has $`M_G/M_210^3`$, while for $`M_G10^7`$, one has $`M_G/M_210^210^1`$. What this says is that the larger the mass is (e.g. $`M_G`$), the less mass splitting is needed in order to have a small $`R`$.
At this stage, we can only say that $`m_\nu `$ can be very small. What we cannot say is exactly what its value should be. This should come from some deeper theory. Instead, we shall use present constraints to restrict the range of values for $`M_{G,P,2}`$.
Having seen how one can obtain very small $`m_\nu `$, the next question would be: How small can one allow $`m_\nu `$ to be if one takes into account the neutrino oscillation data? First of all, atmospheric neutrino oscillation data gives a difference of mass squared $`\mathrm{\Delta }m^210^3eV^2`$ while solar neutrino oscillation data gave $`\mathrm{\Delta }m^210^5eV^2`$ (MSW) or $`10^{10}eV^2`$ (vacuum). In anticipation of new data, the LSND results are not taken into account in our rough estimation of various mass scales. Without any need for a specific model, one can say that the atmospheric data implies that at least one of the three neutrinos should have a mass of at least $`3\times 10^2eV`$, while the solar data implies that at least one of the remaining two should have a mass of at least $`3\times 10^3eV`$ (MSW) or $`10^5eV`$ (vacuum). As we have seen above, the 4th neutrino can be quite heavy. For the sake of argument, let us assume here that its mass is approximately 100 GeV. Since our three light neutrinos are practically degenerate -a lifting of which will be discussed below, the atmospheric data alone constrains $`R`$ to be greater than approximately $`10^{14}`$. This in turn constrains $`M_210^{12}`$ (in units of $`M_F`$) for the case $`M_2>M_G`$, or $`M_G10^{12}`$ for the reverse case. Notice that this rough estimate is only for illustration purpose.
There is however one interesting piece of information which could be quite interesting, phenomenologically speaking: the presence of vector-like fermions which carry weak quatum numbers and which could be relatively “light”. These are the fermions $`F`$ with mass $`M_F`$ as indicated above. Let us recall from the above discussions that $`M_{G,P,2}`$ are all expressed in units of $`M_F`$ which itself could take on any value, even a few hundreds of GeV. The sole restriction will be from experimental constraints, a subject to which we shall come back below. Furthermore, we can see from the results that the mass of the pseudo-NG bosons can also be “low” as well (Fig.1) which could provide a further experimental clue.
We now turn to an important issue: the lifting of the mass degeneracy of the light neutrinos. The analysis presented below will reveal quite interesting implications such as the correlation between the actual values of the masses and $`\mathrm{\Delta }m^2`$, which can have a profound cosmological consequence. For neutrino masses which are large enough to provide part of HDM, the MSW solution of the solar neutrino problem is preferred. If the vacuum solution turns out to be the correct one, the neutrino masses will be much too light in our scenario to play a role in HDM.
We shall divide the discussion presented below into two parts. First we analyze the case when there is no mixing between the 4th neutrino and the lighter three. It will be seen that an interesting feature emerges: $`\mathrm{\Delta }m_{23}^2\mathrm{\Delta }m_{21}^2`$-a quasi-symmetric splitting. ($`\mathrm{\Delta }m_{31}^2`$ is of the same order.) This phenomenon could be called a mass splitting quasi-degeneracy. Of course, solar and atmospheric neutrino data suggest otherwise. Next, we will show how this mass splitting quasi-degeneracy can be lifted, suggesting- at least in our scenario-the presence of a 4th neutrino.
In what follows, we will neglect any possible CP phase in the neutrino mass matrix since we will be concerned only with $`\mathrm{\Delta }m^2`$ and present data on neutrino oscillations are not sensitive to the presence of such a phase. In addition, we shall concentrate in the next two subsections only on $`\mathrm{\Delta }m^2`$. A full comparison with the data will necessitate the inclusion of the leptonic “CKM” angles coming from $`V_L=U_l^{}U_\nu `$. In the two subsections presented below, we shall see what $`U_\nu `$ might look like. To complete the discussion, we shall use a model for $`U_l`$ in order to make some statements about the size of the mixing angles. The subject of the charged lepton mass matrix itself will be dealt with in a subsequent publication.
### B Neutrino mass matrix I: What if there is no mixing between the 4th and the lighter three neutrinos?
The $`4\times 4`$ neutrino mass matrix obtained at this point is purely diagonal. We would like to examine how mass mixing might arise. In particular, we would like to lift the degeneracy of the three light neutrinos. In this section we will concentrate on the scenario where there is mass mixing only among the three light neutrinos. We will show that, in this scenario, $`\mathrm{\Delta }m_{23}^2\mathrm{\Delta }m_{21}^2`$. If this were experimentally the case, it would be hard to detect the influence of the 4th neutrino since it does not mix with the other three. Since the atmostpheric and solar data appear to point to $`\mathrm{\Delta }m_{23}^2\mathrm{\Delta }m_{21}^2`$, we will present in the next section what can be done in order to be in agreement with the data. It turns out that this can be accomplished if one introduces a mixing with the 4th neutrino. This implies that, at least in our model, $`\mathrm{\Delta }m_{23}^2\mathrm{\Delta }m_{21}^2`$ implies the existence of a 4th neutrino, and hence a 4th generation.
The degeneracy of the three light neutrinos at this level comes from the fact that there is a remaining global $`SO(3)`$ symmetry which manifests itself through the equality of the masses of the family gauge bosons ($`M_G`$) as well as those of the pseudo-NG bosons ($`M_P`$). It is then clear that one needs to break that remaining global symmetry in order to remove the degeneracy of the light neutrino masses. We would want to do this in such a way as to preserve the quasi-degeneracy of the light neutrinos. There are probably several ways to achieve this, and we will present one of them here.
Since we have seen how the diagonal elements of the neutrino mass matrix for the three light neutrinos are obtained at the one loop level, it is natural to envision a scenario in which the mixings themselves are obtained at one loop. A look at Figs. 1 reveals that the most “straightforward” way to induce mixings at one loop is for $`\stackrel{~}{\mathrm{\Omega }}_i`$ and/or $`Re\stackrel{~}{\rho }_i`$ to have mixed couplings, i.e. to both $`\nu _{Li}`$ and $`\nu _{Lj}`$ as well as to both $`\eta _{Ri}`$ and $`\eta _{Rj}`$. This could come from mixings among $`\stackrel{~}{\mathrm{\Omega }}_i`$ with different family indices and/or the mixings among $`Re\stackrel{~}{\rho }_i`$. Before getting into the details of what kinds of interactions are needed to break the remaining global $`SO(3)`$ symmetry and hence inducing the mixings, it is instructional to assume that such a mixing among the boson masses occurs and to write down the Yukawa couplings (19,20) in terms of the new boson mass eigenstates.
Let us first look at the states $`\stackrel{~}{\mathrm{\Omega }}_i`$. As we have discussed earlier, these are the NG bosons which are absorbed by the corresponding family gauge bosons. When these NG bosons get mixed, there will be mass mixings among the corresponding family gauge bosons. Let us denote the orthogonal matrix which diagonalizes these family gauge bosons by $`A_\mathrm{\Omega }`$. We shall choose the following representation for $`A_\mathrm{\Omega }`$:
$$A_\mathrm{\Omega }=\left(\begin{array}{ccc}c_2c_3& s_1s_2c_3+c_1s_3& c_1s_2c_3+s_1s_3\\ c_2s_3& c_1c_3+s_1s_2s_3& c_1s_2s_3+s_1c_2\\ s_2& s_1c_2& c_1c_2\end{array}\right)$$
(32)
where $`c`$ and $`s`$ represent the cosine and sine. If we denote by $`\stackrel{~}{\mathrm{\Omega }}_i^{}`$ the logitudinal components of the gauge boson mass eigenstates, its relationship with $`\stackrel{~}{\mathrm{\Omega }}_i`$ in the unmixed case is given by
$$\left(\begin{array}{c}\stackrel{~}{\mathrm{\Omega }}_1\\ \stackrel{~}{\mathrm{\Omega }}_2\\ \stackrel{~}{\mathrm{\Omega }}_3\end{array}\right)=A_\mathrm{\Omega }^T\left(\begin{array}{c}\stackrel{~}{\mathrm{\Omega }}_1^{}\\ \stackrel{~}{\mathrm{\Omega }}_2^{}\\ \stackrel{~}{\mathrm{\Omega }}_3^{}\end{array}\right)$$
(33)
where $`A_\mathrm{\Omega }^T`$ is given by
$$A_\mathrm{\Omega }^T=\left(\begin{array}{ccc}c_2c_3& c_2s_3& s_2\\ s_1s_2c_3+c_1s_3& c_1c_3+s_1s_2s_3& s_1c_2\\ c_1s_2c_3+s_1s_3& c_1s_2s_3+s_1c_2& c_1c_2\end{array}\right)$$
(34)
The masses of the corresponding gauge bosons are now denoted by
$$M_{G_1}^2=M_G^2+\delta _1;M_{G_2}^2=M_G^2+\delta _2;M_{G_3}^2=M_G^2,$$
(35)
where $`\delta _{1,2}`$ can be positive or negative. Notice that $`\delta _{1,2}`$ and the mixing angles shown above are related, i.e. they are all derived from the same boson mass matrix. We will show an example of such fact below.
We can now replace the unprimed states in Eqs.(18,20) by the primed states using Eq. (33). We can then compute the one-loop contributions to the elements of the neutrino mass matrix $`_\nu `$. Let us first look at the contributions to the light neutrino masses and mixings coming from the $`\stackrel{~}{\mathrm{\Omega }}_i`$ states. The two terms which are crucial for this computation are
$$G_1\overline{l}_L^i\mathrm{cos}\beta \stackrel{~}{\mathrm{\Omega }}_iF_R=G_1\overline{l}_L^i\mathrm{cos}\beta A_{\mathrm{\Omega },i}^{T,j}\stackrel{~}{\mathrm{\Omega }}_j^{}F_R$$
(36)
and
$$G_3\overline{M}_{2L}\mathrm{sin}\beta \stackrel{~}{\mathrm{\Omega }}_i\eta _R^{1,i}=G_3\overline{M}_{2L}\mathrm{sin}\beta \stackrel{~}{\mathrm{\Omega }}_j^{}A_{\mathrm{\Omega },i}^j\eta _R^{1,i}$$
(37)
In the loop integrations, one will encounter the following propagators:
$$\frac{1}{k^2M_{G3}^2}=\frac{1}{k^2M_G^2},$$
(39)
$$\frac{1}{k^2M_{G1}^2}=\frac{1}{k^2M_G^2}+\frac{\delta _1}{(k^2M_{G2}^2)(k^2M_G^2)},$$
(40)
$$\frac{1}{k^2M_{G2}^2}=\frac{1}{k^2M_G^2}+\frac{\delta _2}{(k^2M_{G2}^2)(k^2M_G^2)},$$
(41)
With the above remarks in mind, let us proceed to calculate the contributions of $`\stackrel{~}{\mathrm{\Omega }}^{}`$ to the neutrino mass matrix. We shall concentrate first on the $`3\times 3`$ submatrix of the light neutrino sector. As a prelude to the computation of the full submatrix, let us show how two elements are calculated: $`_\nu ^{11}`$ and $`_\nu ^{12}`$. In these computations. we shall use, as an example, the explicit form for $`A_\mathrm{\Omega }`$ shown in Eq. (32). For the complete calculations of the matrix elements, we shall use the notations $`A_{ij}`$ for $`A_\mathrm{\Omega }`$.
a) In the calculation of the contribution of $`\stackrel{~}{\mathrm{\Omega }}^{}`$ to $`_\nu ^{11}`$, one combines Eq. (33) with Eq. (34) to get the following combination of $`\stackrel{~}{\mathrm{\Omega }}^{}`$:
$$(c_2c_3\stackrel{~}{\mathrm{\Omega }}_1^{}c_2s_3\stackrel{~}{\mathrm{\Omega }}_2^{}s_2\stackrel{~}{\mathrm{\Omega }}_3^{})^2,$$
(42)
which gives the following combination of propagators:
$$c_2^2c_3^2\stackrel{~}{\mathrm{\Omega }}_1^{}\stackrel{~}{\mathrm{\Omega }}_1^{}+c_2^2s_3^2\stackrel{~}{\mathrm{\Omega }}_2^{}\stackrel{~}{\mathrm{\Omega }}_2^{}+s_2^2\stackrel{~}{\mathrm{\Omega }}_3^{}\stackrel{~}{\mathrm{\Omega }}_3^{}$$
(43)
Upon using the propagators listed in Eqs.(37) in the one-loop integral (Fig.1), one obtains the following replacement (the reader is referred to Eq. (22) for a comparison):
$$\frac{\mathrm{sin}(2\beta )}{32\pi ^2}I(\stackrel{~}{\mathrm{\Omega }})\frac{\mathrm{sin}(2\beta )}{32\pi ^2}(I(\stackrel{~}{\mathrm{\Omega }})+c_2^2c_3^2\delta _1I(M_G,M_{G1})+c_2^2s_3^2\delta _2I(M_G,M_{G2}))$$
(44)
where ($`i=1,2`$)
$$\delta _iI(M_G,M_{Gi})=\frac{1}{M_FM_2}\{\frac{M_F[M_F^2(M_G^2\mathrm{ln}(\frac{M_G^2}{M_F^2})M_{Gi}^2\mathrm{ln}(\frac{M_{Gi}^2}{M_F^2}))+M_G^2M_{Gi}^2\mathrm{ln}(\frac{M_{Gi}^2}{M_G^2})]}{(M_G^2M_F^2)(M_{Gi}^2M_F^2)}(M_FM_2)\}.$$
(45)
One can see that, in the symmetry limit where $`\delta _i0`$ ($`M_{Gi}M_G`$), $`\delta _iI(M_G,M_{Gi})`$ vanishes identically.
One interesting remark worth mentioning is the following: In (44), the first term $`I(\stackrel{~}{\mathrm{\Omega }})`$ contains no mixing angles. In fact, the coefficient in front of $`I(\stackrel{~}{\mathrm{\Omega }})`$ is $`c_2^2c_3^2+c_2^2s_3^2+s_2^2=1`$, which is the result of $`A_\mathrm{\Omega }`$ being an orthogonal matrix.
We do not give the explicit form for $`I(\stackrel{~}{\mathrm{\Omega }})`$ because, after taking into account the contribution of $`Re\stackrel{~}{\rho }_i`$, one obtains the combination $`I(\stackrel{~}{\mathrm{\Omega }})I(Re\stackrel{~}{\rho })`$ which is already given by Eq. (23).
When the boson mass differences, represented by $`\delta _i`$, are small compared with $`M_G^2`$, another useful form which could be used is given by ($`i=1,2`$)
$$\delta _iI(M_G,M_{Gi})=x_iI(M_G,x_i)$$
(46)
where
$$I(M_G,x_i)=\frac{M_G^2}{M_FM_2}\{\frac{M_F[M_F^2(1+x_i+\mathrm{ln}(\frac{M_G^2}{M_F^2}))+M_G^2(1+x_i)]}{(M_G^2M_F^2)^2(1+x_i(M_G^2/(M_G^2M_F^2)))}(M_FM_2)\},$$
(47)
and where
$$x_i=\frac{\delta _i}{M_G^2},$$
(48)
so that
$$M_{G3}^2=M_G^2;M_{G1}^2=M_G^2(1+x_1);M_{G2}^2=M_G^2(1+x_2).$$
(49)
Here one could explicitely see the vanishing of $`\delta _iI(M_G,M_{Gi})`$ in the symmetry limit because of the explicit appearance of $`\delta _i`$ on the right-hand side of the equation.
The other diagonal elements of the neutrino mass matrix can be analogously calculated. One just needs to replace the combination of angles in (43) with the appropriate ones.
b) For the 1-2 element, the appropriate combination of propagators is given by
$$c_2c_3(s_1s_2c_3+c_1s_3)\stackrel{~}{\mathrm{\Omega }}_1^{}\stackrel{~}{\mathrm{\Omega }}_1^{}c_2s_3(c_1c_3+s_1s_2s_3)\stackrel{~}{\mathrm{\Omega }}_2^{}\stackrel{~}{\mathrm{\Omega }}_2^{}+s_1s_2c_2\stackrel{~}{\mathrm{\Omega }}_3^{}\stackrel{~}{\mathrm{\Omega }}_3^{}$$
(50)
It is now straigthforward to compute $`_\nu ^{12}`$. It is given by
$$_\nu ^{12}(\stackrel{~}{\mathrm{\Omega }})=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}(c_2c_3(s_1s_2c_3+c_1s_3)\delta _1I(M_G,M_{G1})c_2s_3(c_1c_3+s_1s_2s_3)\delta _2I(M_G,M_{G2})),$$
(51)
where we have the appearance of the same $`\delta _iI(M_G,M_{Gi})`$. Notice that $`_\nu ^{12}(\stackrel{~}{\mathrm{\Omega }})`$ denotes the contribution coming from $`\stackrel{~}{\mathrm{\Omega }}`$ only. The full element will also include the contribution coming from $`Re\stackrel{~}{\rho }`$.
Notice that the term $`I(\stackrel{~}{\mathrm{\Omega }})`$ is not present in (51). Again this is due to the orthogonality of $`A_\mathrm{\Omega }`$. The coefficient appearing in front of $`I(\stackrel{~}{\mathrm{\Omega }})`$ is $`c_2c_3(s_1s_2c_3+c_1s_3)c_2s_3(c_1c_3+s_1s_2s_3)+s_1s_2c_2=0`$. The orthogonality of $`A_\mathrm{\Omega }`$ implies that the product of any two columns is equal to zero. As a result we can see that, in the symmetry limit, $`_\nu ^{12}`$ vanishes identically. This applies to all the other off-diagonal elements.
In order to complete the computation of the matrix elements (including the 1-1 and 1-2 elements), one has to say something about the contributions coming from the pseudo-NG bosons themselves. One might imagine that the same mechanism which breaks the global $`SO(3)`$ symmetry also induces mixing among the degenerate pseudo-NG bosons. We will assume that the same matrix $`A_\mathrm{\Omega }`$ diagonalizes the pseudo-NG boson sector so that, instead of the combination of $`\stackrel{~}{\mathrm{\Omega }}_i`$ and $`Re\stackrel{~}{\rho }_i`$ used in Eqs. (18) for the NG and pseudo-NG bosons, we shall use $`A_\mathrm{\Omega }\stackrel{~}{\mathrm{\Omega }}`$ and $`A_\mathrm{\Omega }Re\stackrel{~}{\rho }`$, where $`\stackrel{~}{\mathrm{\Omega }}`$ and $`Re\stackrel{~}{\rho }`$ are now column vectors. With these definitions, one simply gets $`\stackrel{~}{\mathrm{\Omega }}^{}Re\stackrel{~}{\rho }=\stackrel{~}{\mathrm{\Omega }}^{}A_\mathrm{\Omega }^1A_\mathrm{\Omega }Re\stackrel{~}{\rho }`$ . This simple assumption is used for two purposes: 1) To reduce the number of arbitrary parameters; 2) To see how far one can go with it before one needs to modify it. With this assumption, the mass splitting among the pseudo-NG bosons are given as in Eq. (49), namely
$$M_{P3}^2=M_P^2;M_{P1}^2=M_P^2(1+x_1);M_{P2}^2=M_P^2(1+x_2),$$
(52)
with the same $`x_i`$ as for the gauge boson masses. Furthermore, the mixing angles are the same as above. The contributions of the “rotated” pseudo-NG bosons to the neutrino mass matrix elements will therefore be accompanied by a factor $`\frac{\mathrm{sin}(2\beta )}{32\pi ^2}`$, just as in Eq. (44).
As mentioned above, in the full computation of the matrix elements, we shall use, for convenience, $`A_{ij}`$ to denote the matrix elements of $`A_\mathrm{\Omega }`$ instead of the representation of Eq. (32). One should then recall that, because $`A_\mathrm{\Omega }`$ is an orthogonal matrix, one has: $`_jA_{ij}^2=1`$ and $`_kA_{ki}A_{kj}=0`$. The form of the neutrino mass matrix elements will make use of these properties, just as we have done above.
With the above remarks in mind, the full $`4\times 4`$ neutrino mass matrix is now given by:
$$_\nu /m_N=\left(\begin{array}{cccc}m_{11}& m_{12}& m_{13}& 0\\ m_{12}& m_{22}& m_{23}& 0\\ m_{13}& m_{23}& m_{33}& 0\\ 0& 0& 0& 1\end{array}\right)$$
(53)
where $`m_N`$ is the mass of the 4th generation neutrino shown in Eq. (9). In Eq. (53), we have ignored the tiny one-loop contribution to $`m_{44}1`$, in particular when there is no mixing between the 4th neutrino and the lighter three. As we shall see later on, it can be ignored even if there is mixing, the reason being the fact that $`m_{ij}`$, $`i,j=1,2,3`$, are so much smaller than $`m_{44}1`$. A change of $`m_{44}`$ to a value slightly less than or greater than one will not significantly affect the eigenvalues, as we shall see in the numerical examples below.
With
$$\mathrm{\Delta }I(G,P,x_i)I(M_G,x_i)I(M_P,x_i),$$
(54)
where $`I(M_P,x_i)`$ is given by Eq. (47) with the substitution $`M_GM_P`$, one obtains for $`m_{ij}`$:
$$m_{11}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)A_{11}^2x_1\mathrm{\Delta }I(G,P,x_1)A_{12}^2x_2\mathrm{\Delta }I(G,P,x_2)\}$$
(56)
$$m_{22}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)A_{21}^2x_1\mathrm{\Delta }I(G,P,x_1)A_{22}^2x_2\mathrm{\Delta }I(G,P,x_2)\}$$
(57)
$$m_{33}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)A_{31}^2x_1\mathrm{\Delta }I(G,P,x_1)A_{32}^2x_2\mathrm{\Delta }I(G,P,x_2)\}$$
(58)
$$m_{12}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{A_{11}A_{21}x_1\mathrm{\Delta }I(G,P,x_1)+A_{12}A_{22}x_2\mathrm{\Delta }I(G,P,x_2)\}$$
(59)
$$m_{13}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{A_{11}A_{31}x_1\mathrm{\Delta }I(G,P,x_1)+A_{11}A_{32}x_2\mathrm{\Delta }I(G,P,x_2)\}$$
(60)
$$m_{23}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{A_{21}A_{31}x_1\mathrm{\Delta }I(G,P,x_1)+A_{22}A_{32}x_2\mathrm{\Delta }I(G,P,x_2)\}$$
(61)
where $`A_{ij}`$ denote the matrix elements of $`A_\mathrm{\Omega }`$, as mentioned above, and where $`\mathrm{\Delta }I(G,P)`$ was already defined in Eq. (24).
A few remarks are in order here. First, one can see that, in the limit $`x_i0`$, $`_\nu `$ reduces to a diagonal matrix with three equal diagonal elements: $`\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\mathrm{\Delta }I(G,P)`$. Secondly, apart from various mixing angles, the off-diagonal elements depend on results of loop integrals, $`\mathrm{\Delta }I(G,P,x_i)`$ which, in turns, depend on the same parameters as the ones that enter the loop integrals of the diagonal elements in the unbroken case, $`\mathrm{\Delta }I(G,P)`$. The ratio $`R_I\mathrm{\Delta }I(G,P,x_i)/\mathrm{\Delta }I(G,P)`$ is plotted in Figs. (6,7), for two values of the parameter $`x`$, as a function of $`M_2`$ in the similar manner to Fig. 2-5. (The two values of $`x`$ were chosen for the purpose of illustration and to coincide with the two examples given below.) It can be seen that the ratio $`R_I`$ is at most of O($`10^{}2`$), even for $`x`$ as large as 0.5. Therefore, in our model, a small mass splitting in the scalar and gauge sectors results in a scenario with almost degenerate light neutrinos. The difference of the mass squared, $`\mathrm{\Delta }m^2`$, depends, however, on the size of the off-diagonal elements. To see how it actually works, a simple model of mixings will be presented below along with some numerical examples.
We starts out with a very simplistic model of mixing and try to see how far one can go. It is:
$$_{G,P}^2=M_{G,P}^2\left(\begin{array}{ccc}1& b& 0\\ b& 1& 0\\ 0& 0& 1\end{array}\right)$$
(62)
where $`b`$ is a small parameter less than unity. This simple model has the merit of elucidating the points that we have made above. (An extension of this model, showing similar results, will be discussed below.) The above mass mixing (62) could come, for example, from a term in the Lagrangian of the form: $`\lambda _5((\mathrm{\Omega }^\alpha \rho _\alpha ^{})(\mathrm{\Omega }^\beta \rho _\beta ^{\prime \prime })+(\rho ^\alpha \rho _\alpha ^{})(\rho ^\beta \rho _\beta ^{\prime \prime }))`$. Assuming $`<\rho ^{}>=(v^{},0,0,0)`$, $`<\rho ^{\prime \prime }>=(0,v^{\prime \prime },0,0)`$, with $`v^{,\prime \prime }V,V^{}`$, one can obtain the above mass mixing matrix.
It is easy to see that the eigenvalues of (62) are
$$M_{G1,P1}^2=M_{G,P}^2(1+b),M_{G2,P2}^2=M_{G,P}^2(1b),M_{G3,P3}^2=M_{G,P}^2.$$
(63)
$`A_\mathrm{\Omega }`$ as discussed above is now given by
$$A_\mathrm{\Omega }=\left(\begin{array}{ccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0\\ \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0\\ 0& 0& 1\end{array}\right)$$
(64)
Now we can make the following identifications: $`x_1b`$, $`x_2b`$. The various angles are given in $`A_\mathrm{\Omega }`$. The matrix elements of the neutrino mass matrix are now fairly simple:
$$m_{11}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)\frac{1}{2}b(\mathrm{\Delta }I(G,P,b)\mathrm{\Delta }I(G,P,b))\}$$
(66)
$$m_{22}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)\frac{1}{2}b(\mathrm{\Delta }I(G,P,b)\mathrm{\Delta }I(G,P,b))\}$$
(67)
$$m_{33}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)\}$$
(68)
$$m_{12}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\frac{1}{2}b(\mathrm{\Delta }I(G,P,b)+\mathrm{\Delta }I(G,P,b))\}$$
(69)
$$m_{13}=0,$$
(70)
$$m_{23}=0.$$
(71)
The above matrix elements are surprisingly easy to handle. When they are substituted into Eq. (53), one obtains straightforwardly the following mass eigenvalues:
$$m_1=m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)b\mathrm{\Delta }I(G,P,b)\},$$
(73)
$$m_2=m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\mathrm{\Delta }I(G,P),$$
(74)
$$m_3=m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)+b\mathrm{\Delta }I(G,P,b)\},$$
(75)
$$m_4=m_N$$
(76)
The matrix which diagonalizes the above neutrino mass matrix is simply
$$U_\nu =\left(\begin{array}{cccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0& 0\\ 0& 0& 1& 0\\ \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0& 0\\ 0& 0& 0& 1\end{array}\right)$$
(77)
One obtains the following mass splittings:
$$m_3^2m_2^2=(m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2})^2(2b\mathrm{\Delta }I(G,P)\mathrm{\Delta }I(G,P,b)+(b\mathrm{\Delta }I(G,P,b))^2),$$
(79)
$$m_2^2m_1^2=(m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2})^2(2b\mathrm{\Delta }I(G,P)\mathrm{\Delta }I(G,P,b)+(b\mathrm{\Delta }I(G,P,b))^2).$$
(80)
In general, $`\mathrm{\Delta }I(G,P,x_i)\mathrm{\Delta }I(G,P)`$, and combined with the fact that $`b<1`$, one has $`(b\mathrm{\Delta }I(G,P,borb))^22b\mathrm{\Delta }I(G,P)\mathrm{\Delta }I(G,P,borb)`$. One can then neglect the last terms in Eq.(III B). Numerically, one has $`\mathrm{\Delta }I(G,P,b)\mathrm{\Delta }I(G,P,b)`$. This implies that $`m_3^2m_2^2m_2^2m_1^2`$, a quasi-degenerate mass splitting. This holds for any value of $`b`$. Solar and atmospheric data suggest otherwise. This necessitates the lifting of this quasi-degeneracy of the mass splitting. To do this, we need to invoke some kind of mixing between the 4th neutrino and the lighter three. In an indirect way, the disparity between $`\mathrm{\Delta }m_{23}^2`$ and $`\mathrm{\Delta }m_{21}^2`$ indicates- in our model- the influence of a 4th generation. Before discussing this issue which will be presented in the next section, let us illustrate numerically a few examples of the quasi-degenerate case.
First, a few useful points are in order here. Since $`m_2=m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\mathrm{\Delta }I(G,P)`$, one can rewrite the above equations (III B) as (neglecting the last terms on the right-hand side)
$$m_3^2m_2^2=m_2(m_N2b\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\mathrm{\Delta }I(G,P,b)),$$
(82)
$$m_2^2m_1^2=m_2(m_N2b\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\mathrm{\Delta }I(G,P,b)).$$
(83)
For a fixed value of $`m_2`$, the size of the mass splitting, $`\mathrm{\Delta }m^2`$, depends on the size of the factor $`m_N(2b)(\mathrm{sin}(2\beta )/32\pi ^2)\mathrm{\Delta }I(G,P,borb)`$. At first glance, it appears that one can obtain $`\mathrm{\Delta }m^2`$ to be as small as one wants with the appropriate choice of $`b`$. Although it is true that it can be so, we will show that, $`\mathrm{\Delta }m^2`$ can also be very small ($`<10^{10}eV^2`$), even when $`b1`$. This depends on how large the masses of some of the particles participating in the loop diagrams are. As a result, by limiting $`\mathrm{\Delta }m^210^{10}eV^2`$, one puts a constraint on those masses.
In Fig. 8, we present the “median” mass $`m_2`$ as a function of $`M_2`$ and $`M_G`$ for a given $`M_P`$ (as presented in Figs. (2-5)). The mass is given in units of $`(m_N/100GeV)`$. Similarly, we present in Figs.(9,10) $`m_3^2m_2^2`$ and $`m_2^2m_1^2`$ as a function of the same masses, but also for a given value of $`b`$. The results are expressed in units of $`(m_N/100GeV)^2`$. For a more streamlined presentation of the results, we shall limit ourselves to the case $`m_21.67eV`$, coming from the suggestion that the sum of neutrino masses lies between 4 and 5 $`eV`$ in order to form a component of HDM. Similarly, we shall restrict $`\mathrm{\Delta }m^2<1eV^2`$. In our model, for a given value of $`b`$, $`m_2`$ and $`\mathrm{\Delta }m^2`$ are correlated as one can see from Fig. (8,9,10).
Three major remarks are in order here. 1) One can see from Figs. (9, 10) the quasi-degeneracy of the mass splitting in this particular scenario. (In the next section, we shall see how one can lift that degeneracy.) 2) One can also see from Figs. (8, 9, 10) that, were the vacuum solution to the solar neutrino problem favored, i.e. $`\mathrm{\Delta }m^210^{10}eV^2`$, the median value $`m_2`$ will always be less than $`0.1eV`$. (The lifting of the mass splitting degeneracy to satisfy the atmospheric neutrino data will not change this conclusion.) This simply means that, at least in this model, the solar vacuum solution is incompatible with the light neutrinos being significant components of HDM. 3) Also from Figs. (8, 9, 10), it can be seen that the MSW solution, $`\mathrm{\Delta }m^210^5eV^2`$, can correspond to values of $`m_2`$ larger than 1 $`eV`$. (Again, the lifting of the mass splitting degeneracy to satisfy the atmospheric neutrino data will not change this conclusion.) So, in our scenario, the MSW solution is compatible with the light neutrinos being significant components of HDM while the vacuum solution is not. This is a very specific prediction of this model.
The above discussion leaves out the question of the size of the mixing angles. As mentioned above, we have already fixed the neutrino mixing matrix $`U_\nu `$, as given by Eq. (77). To complete the task, one has to model the charged lepton mixing matrix $`U_l`$. This is something that we shall do in the last section. We wish however to reemphasize the main result of this section: the values of $`\mathrm{\Delta }m^2`$ are independent of $`U_l`$. As one can see from Fig. (9, 10), $`\mathrm{\Delta }m^2`$ depends only on the various masses and on the parameter $`b`$, regardless of $`U_l`$. As a consequence, the large angle or small angle solutions as deduced from the data basically constrains, in our scenario, the matrix $`U_l`$ ($`U_\nu `$ being already fixed).
To finish the discussion of this section, we wish to present another form for the boson mass matrix, namely
$$_{G,P}^2=M_{G,P}^2\left(\begin{array}{ccc}1& b& 0\\ b& 1& b\\ 0& b& 1\end{array}\right)$$
(84)
The mass eigenvalues are
$$M_{G1,P1}^2=M_{G,P}^2(1+\sqrt{2}b),M_{G2,P2}^2=M_{G,P}^2(1\sqrt{2}b),M_{G3,P3}^2=M_{G,P}^2.$$
(85)
$`A_\mathrm{\Omega }`$ is now given by
$$A_\mathrm{\Omega }=\left(\begin{array}{ccc}\frac{1}{2}& \frac{1}{\sqrt{2}}& \frac{1}{2}\\ \frac{1}{2}& \frac{1}{\sqrt{2}}& \frac{1}{2}\\ \frac{1}{\sqrt{2}}& 0& \frac{1}{\sqrt{2}}\end{array}\right)$$
(86)
It is now straightforward to see that the neutrino mass matrix elements are
$$m_{11}=m_{33}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)\frac{b}{2\sqrt{2}}(\mathrm{\Delta }I(G,P,b)\mathrm{\Delta }I(G,P,b))\}$$
(87)
$$m_{22}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)\frac{b}{\sqrt{2}}(\mathrm{\Delta }I(G,P,b)\mathrm{\Delta }I(G,P,b))\}$$
(88)
$$m_{12}=m_{23}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\frac{1}{2}b(\mathrm{\Delta }I(G,P,b)+\mathrm{\Delta }I(G,P,b))\}$$
(89)
$$m_{13}=\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\frac{b}{2\sqrt{2}}(\mathrm{\Delta }I(G,P,b)\mathrm{\Delta }I(G,P,b))\},$$
(90)
The eigenvalues are now simply given by
$$m_1=m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)\sqrt{2}b\mathrm{\Delta }I(G,P,b)\},$$
(92)
$$m_2=m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\mathrm{\Delta }I(G,P),$$
(93)
$$m_3=m_N\frac{\mathrm{sin}(2\beta )}{32\pi ^2}\{\mathrm{\Delta }I(G,P)+\sqrt{2}b\mathrm{\Delta }I(G,P,b)\},$$
(94)
$$m_4=m_N$$
(95)
These masses have exactly the same form as those of Eq.(III B), except for the factor of $`\sqrt{2}b`$ instead of $`b`$. The matrix $`U_\nu `$ which diagonalizes the above matrix is exactly the same as in Eq. (77). Furthermore, $`m_3^2m_2^2`$ and $`m_2^2m_1^2`$ are of the same form as Eqs. (III B), with the following replacement in Eqs. (III B): $`bb^{}=\sqrt{2}b`$. The analysis which follows is exactly the same as the one presented above.
One can envision various scenarios for the boson mass matrices, but it is certainly beyond the scope of this paper. To make things more complicated than the simple assumption (62) does not appear to add much to the discussion. Although it might be possible that a more involved ansatz than (62) could lead to the lifting of the mass splitting “quasi-degeneracy”, we have not succeeded in finding it. For this reason, we now turn our attention to the more appealing scenario, at least within our model: the mixing between the 4th neutrino and the rest.
### C Neutrino mass matrix II: Mixing between the 4th and the lighter three neutrinos
We have seen above that the simple ansatz for the boson mass matrices (62) leads to a situation in which the mass splittings are quasi-degenerate. This, of course, is in contradiction with the data. In this model, in order to lift that quasi-degeneracy, one needs a mixing between the 4th neutrino and at least one of the lighter three. To get a feel for what might be needed, we shall first present a few numerical examples. Based on these examples, we shall attempt to give a theoretical basis for these numerical examples.
As an example, we shall choose a specific value for the parameter $`b`$ and for the masses $`M_2`$, $`M_G`$, $`M_P`$ and $`M_F`$ which enter the loop integrals for the neutrino masses. This will fix a definite value for the matrix elements of the neutrino mass matrix. As we have already discussed earlier, the integrals depend only on the ratio of the above masses. We will present two examples for the purpose of comparison. We shall see the reasons why we wish to do so below.
1) First Example:
We shall set (in units of $`M_F`$): $`M_F=1`$, $`M_P=5`$, $`M_G=10^6`$, $`M_2=2.5\times 10^9`$. For $`b`$, we shall choose: $`b=0.035`$. (A smaller value of $`b`$ will give a smaller mass splitting.) The reason for this choice (other choices are equally valid) is the fact that it will give a typical mass of approximately 1.5 eV and a desired mass splitting. Putting these values into the expressions for the integrals as given by Eq.(47), we obtain the following neutrino mass matrix, where $`m_N`$ is assumed to be 100 GeV for convenience:
$$_\nu =(100GeV)\left(\begin{array}{cccc}1.579332216\times 10^{11}& .8697647852\times 10^{17}& 0& 0\\ .8697647852\times 10^{17}& 1.579332216\times 10^{11}& 0& 0\\ 0& 0& 1.579332184\times 10^{11}& 0\\ 0& 0& 0& 1\end{array}\right)$$
(96)
Notice that the above matrix has no mixing between the 4th neutrino and the lighter three. The eigenvalues are just:
$$|m_1|=1.579331346eV;|m_2|=1.579332184eV;|m_3|=1.579333086eV;|m_4|=100GeV.$$
(97)
As we have discussed in the previous section, this gives a quasi-degenerate mass splitting,namely
$$\mathrm{\Delta }m_{32}^2=1.601195367\times 10^6eV^2,$$
(98)
$$\mathrm{\Delta }m_{21}^2=1.535757079\times 10^6eV^2,$$
(99)
where $`\mathrm{\Delta }m_{ji}^2=m_j^2m_i^2`$.
Let us now assume that the mixing with the 4th neutrino is non-zero. We start out with the simplest assumption, namely one in which only the 3rd neutrino mixes with the 4th one. This means that $`m_{34}`$ and $`m_{43}`$ are both non-vanishing. If we wish to have $`m_3^2m_2^210^3eV^2`$ as suggested by the atmospheric neutrino data, it turns out that $`m_{34}`$ and $`m_{43}`$ cannot be too small nor too large, being of order $`10^7m_N`$. Notice that $`m_{34}`$ and $`m_{43}`$ do not have to be equal. We shall see how it might be possible to obtain such a number. Let us first see how it works from a numerical viewpoint.
To guide our understanding of how things work, let us notice that, by adding $`m_{34}`$ and $`m_{43}`$ to $`_\nu `$ above, one changes only one of the three light mass eigenvalues, leaving the other two the same. Now the two unchanged eigenvalues will be the ones that fix one of the two mass splittings, $`\mathrm{\Delta }m^2`$. For convenience, we shall choose the $`\mathrm{\Delta }m^2`$ corresponding to the unmodified mass eigenvalues as the one which corresponds to the solar neutrino problem. As we have learned from the above analysis in Section (III B), if one chooses the MSW solution, then one can find masses which are large enough for HDM, while, if the vacuum solution is chosen, the masses will be too small to form any significant component of HDM. For the numerical example given here, we shall choose the MSW solution as shown above. For $`m_{34}`$ and $`m_{43}`$, we shall first choose a symmetric case (there is no particular reason for this being so) as an example. We have
$$_\nu =(100GeV)\left(\begin{array}{cccc}1.579332216\times 10^{11}& .8697647852\times 10^{17}& 0& 0\\ .8697647852\times 10^{17}& 1.579332216\times 10^{11}& 0& 0\\ 0& 0& 1.579332184\times 10^{11}& .8\times 10^7\\ 0& 0& .8\times 10^7& 1\end{array}\right)$$
(100)
The eigenvalues are
$$|m_1|=1.579331346eV;|m_2|=1.579333086eV;|m_3|=1.579972184eV;m_4=100GeV.$$
(101)
We then get
$$\mathrm{\Delta }m_{32}^2=2.02\times 10^3eV^2,$$
(102)
$$\mathrm{\Delta }m_{21}^2=5.497\times 10^6eV^2.$$
(103)
The matrix which diagonalizes the mass matrix is
$$U_\nu =\left(\begin{array}{cccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0& 0\\ \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0& 0\\ 0& 0& 1& 0.8\times 10^7\\ 0& 0& .8\times 10^7& 1\end{array}\right)$$
(104)
Two remarks are in order here. Firstly, from the values of the light neutrino masses, one obtains $`_{i=1}^3|m_i|4.7eV`$, which is in the range of mass for HDM. Secondly, Eq. (102) corresponds to the best fit for the atmospheric neutrino data, while Eq. (103) corresponds to the best fit for the (small angle) MSW solution to the solar neutrino data. One word of caution: this is not a prediction because we chose the masses ($`M_G`$, $`M_2`$, etc…) in such a way as to “reproduce” the experimental results. It nevertheless shows a dynamical basis for these numbers. Also, for nothing more than a numerical example, the values of $`m_{34,43}`$ were chosen arbitrarily in order to have the desired mass splitting. How to justify these values is the subject to be discussed below.
The next numerical example deals with the case when $`m_{34}m_{43}`$. In doing the analysis, we notice that it does not matter whether $`m_{34}`$ is greater than $`m_{43}`$ or the other way around. One obtains the same result either way. We shall require that $`\mathrm{\Delta }m_{32}^2(eV^2)=10^310^2`$. It turns out that $`m_{34}`$ and $`m_{43}`$ can range (in units of $`m_N`$) only between approximately $`0.4\times 10^6`$ and $`0.8\times 10^8`$. To be explicit, one has
$$_\nu =(100GeV)\left(\begin{array}{cccc}1.579332216\times 10^{11}& .8697647852\times 10^{17}& 0& 0\\ .8697647852\times 10^{17}& 1.579332216\times 10^{11}& 0& 0\\ 0& 0& 1.579332184\times 10^{11}& .4\times 10^6\\ 0& 0& .8\times 10^7& 1\end{array}\right)$$
(105)
gives $`\mathrm{\Delta }m_{32}^2(eV^2)10^2`$, while
$$_\nu =(100GeV)\left(\begin{array}{cccc}1.579332216\times 10^{11}& .8697647852\times 10^{17}& 0& 0\\ .8697647852\times 10^{17}& 1.579332216\times 10^{11}& 0& 0\\ 0& 0& 1.579332184\times 10^{11}& .4\times 10^6\\ 0& 0& .8\times 10^8& 1\end{array}\right)$$
(106)
gives $`\mathrm{\Delta }m_{32}^2(eV^2)10^3`$. Notice that $`\mathrm{\Delta }m_{21}^2`$ stays the same. The above numerical results show that $`m_{34}`$ can differ from $`m_{43}`$ by a large factor (50 in this case) while keeping $`\mathrm{\Delta }m_{32}`$ within the desired range.
2) Second Example:
In this example, we choose (in units of $`M_F`$): $`M_F=1`$, $`M_P=5`$, $`M_G=10^4`$, $`M_2=1.2\times 10^9`$. For $`b`$, we shall choose: $`b=0.000095`$. For simplicity, we shall assume, as we have already done above, the following values for $`m_{34,43}`$, namely $`m_{34}=m_{43}=0.8\times 10^7(100GeV)`$. The mass matrix is now
$$_\nu =(100GeV)\left(\begin{array}{cccc}1.382258467\times 10^{11}& .981382953\times 10^{17}& 0& 0\\ .981382953\times 10^{17}& 1.382258467\times 10^{11}& 0& 0\\ 0& 0& 1.382258467\times 10^{11}& .8\times 10^7\\ 0& 0& .8\times 10^7& 1\end{array}\right)$$
(107)
The eigenvalues are
$$m_1=1.382259448eV;m_2=1.382257486eV;m_3=1.381618467eV;m_4=100GeV,$$
(108)
with the corresponding diagonalization matrix given by
$$U_\nu =\left(\begin{array}{cccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0& 0\\ \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0& 0\\ 0& 0& 1& 0.8\times 10^7\\ 0& 0& .8\times 10^7& 1\end{array}\right)$$
(109)
The mass splittings are
$$|\mathrm{\Delta }m_{32}^2|=1.77\times 10^3eV^2,$$
(110)
$$|\mathrm{\Delta }m_{21}^2|=5.42\times 10^6eV^2.$$
(111)
The above two examples are chosen sololy for illustration. Other values of $`\mathrm{\Delta }m^2`$ are possible with different choices of various masses ($`M_G`$, $`M_2`$, etc..) and/or the parameter $`b`$.
Before turning to the discussion on the possible origins of $`m_{34,43}`$, let us briefly discuss the “tiny” one-loop contribution to $`m_{44}`$, namely $`\delta m_4`$ as given by Eq. (III A). One might wonder how it would affect the light mass eigenvalues. It turns out however that, as long as $`\delta m_41`$ (which is the case in this paper), it does not matter what value it takes. It is easy to see how. A $`2\times 2`$ matrix of the form $`(a,c;c,b)`$, where $`a,cb`$, has as eigenvalues: $`b+c^2/b+(1/4)a^2/b)+O(c^4,a^4)`$ and $`ac^2/b(1/4)a^2/b)+O(c^4,a^4)`$. One can see that, for the smaller eigenvalue, a small change in $`b`$ affects very little its value. As an example, we put 0.99 instead of 1 in Eq. (). We obtain $`\mathrm{\Delta }m_{32}^2(eV^2)1.02\times 10^3`$ instead of $`1.06\times 10^3`$ (for 1). If we put 1.1 instead of 1, we obtain $`\mathrm{\Delta }m_{32}^2(eV^2)0.923\times 10^3`$. Considering the kind of accuracy that one has at the present time, this is completely irrelevant.
There are probably several scenarios for calculating $`m_{34,43}`$. However, considering the fact that the present experimental status is not accurate enough for a detailed model, we will present below a more or less “generic” scenario which will show how one can obtain $`m_{34,43}`$ of the right order of magnitude.
What might be the origin of $`m_{34,43}`$? It might be obvious up until now that the vacuum expectation values of $`\mathrm{\Omega }`$ and $`\rho `$ shown in Subsection (III.A) cannot generate such a mixing. One needs at least one additional scalar with a non-vanishing vacuum expectation value along the 3rd direction. Let that field be $`\mathrm{\Omega }^{}`$ and let us assume that $`<\mathrm{\Omega }^{}>=(0,0,\stackrel{~}{v},0)`$. Let us also assume that there are couplings of the type:
$$\lambda _{34}\mathrm{\Omega }^\alpha \mathrm{\Omega }_\alpha ^{}\rho ^\beta \rho _\beta ;\lambda _{43}\mathrm{\Omega }^\alpha \rho _\alpha \mathrm{\Omega }^{,\beta }\rho _\beta ,$$
(112)
where, for convenience, we have omitted the $`SU(2)_{\nu R}`$ index in $`\rho `$. With the above couplings, one can construct diagrams for $`m_{34}`$ and $`m_{43}`$ as shown in Fig. 11.
We shall denote the masses of $`\stackrel{~}{H}_4`$ and $`\stackrel{~}{h}_4`$ by $`M_{H_4}`$ and $`M_{h_4}`$ respectively. Let us define the following quantities:
$$\mathrm{\Delta }M^2(G,\stackrel{~}{H}_4)=M_G^2M_{H_4}^2,$$
(114)
$$\mathrm{\Delta }M^2(G,\stackrel{~}{h}_4)=M_G^2M_{h_4}^2,$$
(115)
$$\mathrm{\Delta }M^2(P,\stackrel{~}{H}_4)=M_P^2M_{H_4}^2,$$
(116)
$$\mathrm{\Delta }M^2(P,\stackrel{~}{h}_4)=M_P^2M_{h_4}^2.$$
(117)
From Fig. 8, we obtain:
$`m_{34}/m_N`$ $`=`$ $`({\displaystyle \frac{\lambda _{34}}{16\pi ^2}})({\displaystyle \frac{\stackrel{~}{v}M_FM_2}{V}})(c_\beta ^2s_\alpha ^2{\displaystyle \frac{\mathrm{\Delta }I(G,\stackrel{~}{H}_4)}{\mathrm{\Delta }M^2(G,\stackrel{~}{H}_4)}}+c_\beta ^2c_\alpha ^2{\displaystyle \frac{\mathrm{\Delta }I(G,\stackrel{~}{h}_4)}{\mathrm{\Delta }M^2(G,\stackrel{~}{h}_4)}}+s_\beta ^2s_\alpha ^2{\displaystyle \frac{\mathrm{\Delta }I(P,\stackrel{~}{H}_4)}{\mathrm{\Delta }M^2(P,\stackrel{~}{H}_4)}}`$ (120)
$`+s_\beta ^2c_\alpha ^2{\displaystyle \frac{\mathrm{\Delta }I(G,\stackrel{~}{h}_4)}{\mathrm{\Delta }M^2(G,\stackrel{~}{h}_4)}}),`$
$`m_{43}/m_N`$ $`=`$ $`({\displaystyle \frac{\lambda _{34}}{16\pi ^2}})({\displaystyle \frac{\stackrel{~}{v}M_FM_2}{V}})(s_\beta ^2c_\alpha ^2{\displaystyle \frac{\mathrm{\Delta }I(G,\stackrel{~}{H}_4)}{\mathrm{\Delta }M^2(G,\stackrel{~}{H}_4)}}+s_\beta ^2s_\alpha ^2{\displaystyle \frac{\mathrm{\Delta }I(G,\stackrel{~}{h}_4)}{\mathrm{\Delta }M^2(G,\stackrel{~}{h}_4)}}+c_\beta ^2c_\alpha ^2{\displaystyle \frac{\mathrm{\Delta }I(P,\stackrel{~}{H}_4)}{\mathrm{\Delta }M^2(P,\stackrel{~}{H}_4)}}`$ (122)
$`+c_\beta ^2s_\alpha ^2{\displaystyle \frac{\mathrm{\Delta }I(G,\stackrel{~}{h}_4)}{\mathrm{\Delta }M^2(G,\stackrel{~}{h}_4)}}),`$
where $`c`$ and $`s`$ stand for $`\mathrm{cos}`$ and $`\mathrm{sin}`$, and $`\mathrm{\Delta }I(G,\stackrel{~}{H}_4)`$ and the other similar quantities in Eq. (117) are given by Eq. (24), with the substitution of the appropriate masses taken into account.
As one can see from the above equations, the expressions appear rather complicated at first look. However, one can make an estimate as to which term in $`m_{34}`$ and $`m_{43}`$ is the most important. Each term in Eqs. (117) is of the form: $`\lambda (\stackrel{~}{v}/V)(M_F/M_2)(M_2^2/\mathrm{\Delta }M^2)\mathrm{\Delta }I(mixingangles)`$, where $`\lambda `$ stands for $`\lambda _{34,43}`$. First, we have seen from the above numerical analysis that, if we wish to have a mass of O(1-2 eV), then $`M_F/M_210^9`$. It is reasonable to assume that $`\lambda (\stackrel{~}{v}/V)\times (mixingangles)1`$. If one of the terms in Eq. (117) were to be the dominant one and that $`m_{34,43}10^7`$, then one should have $`(M_2^2/\mathrm{\Delta }M^2)\mathrm{\Delta }I10^2`$. Let us first look at the $`(G;\stackrel{~}{H},\stackrel{~}{h})`$ contribution. Assuming that $`M_{H_4,h_4}<M_G`$ so that $`(M_2^2/\mathrm{\Delta }M^2)M_2^2/M_G^2`$, it turns out numerically that $`(M_2^2/M_G^2)\mathrm{\Delta }I`$ is always less than $`10`$. For $`M_{H_4,h_4}>M_G`$, $`\mathrm{\Delta }I`$ is larger in value than the previous case, but then with $`(M_2^2/\mathrm{\Delta }M^2)M_2^2/M_{H_4,h_4}^2`$, one will again have $`(M_2^2/M_{H_4,h_4}^2)\mathrm{\Delta }I`$ less than $`10^2`$. Taking into account the actual calculation of $`m_{34,43}`$ which includes mixing angles and various factors, the $`(G;\stackrel{~}{H},\stackrel{~}{h})`$ would be too small to actually affect the mass splittings. This leaves us with the contribution coming from $`(P;\stackrel{~}{H},\stackrel{~}{h})`$. Here, as we have done above, we will set $`M_P=5`$ in units of $`M_F`$. There are several possibilities that one can explore. We will present here one of such possibilities. The main purpose will be to show that, under reasonable assumptions, one can obtain the desired order of magnitude for $`m_{34,43}`$. In addition, one would like to see phenomenological implications coming from such a scenario- something extra other than just a mass matrix.
Let us assume that, by an appropriate choice of parameters in the Higgs potential, one has $`M_{H_4}`$ to be of O($`M_G`$), and that $`M_{h_4}M_2`$. Furthermore, let us assume that one also has $`\beta \alpha `$. Although it is not really necessary, let us further assume that $`\lambda _{34}\lambda _{43}`$. Now numerically, $`(M_2^2/\mathrm{\Delta }M^2)\mathrm{\Delta }I<10^2`$ when one of the masses in $`\mathrm{\Delta }M^2`$ is much larger than the other one and not too much different from $`M_2`$. This is just the case for $`M_{H_4}=O(M_G)M_P`$. Under these assumptions, we are left with the $`(P;h)`$ contribution. In this case, one has $`m_{34}m_{43}`$. So we get
$$|m_{34}||m_{43}|m_N\lambda _{34}\frac{\stackrel{~}{v}}{V}\frac{M_F}{M_2}|\frac{M_2^2}{M_P^2M_{h_4}^2}|\mathrm{\Delta }M^2(G,\stackrel{~}{h}_4)s_\beta ^2c_\alpha ^2.$$
(123)
Typically, $`\mathrm{\Delta }M^2(G,\stackrel{~}{h}_4)=O(10^710^{11})`$. In most of our examples, $`M_F/M_210^9`$. So one would expect $`(M_F/M_2)\mathrm{\Delta }M^2(G,\stackrel{~}{h}_4)10^{16}10^{20}`$. If we wish $`m_{34}m_{43}m_N.8\times 10^7`$, for example, the other factors have to be sufficiently large. First, the ratio $`|\frac{M_2^2}{M_P^2M_{h_4}^2}|`$ can be rather large if $`M_{h_4}`$ is small compared with $`M_2`$. Secondly, even if the previous ratio can be large, it can still be offset by $`s_\beta ^2c_\alpha ^2`$. Let us recall from Eq. (29) that $`\mathrm{tan}\beta g^2(M_F/M_2)(M_2^2/M_G^2)g^210^9(M_2^2/M_G^2)`$. Therefore the angle can be very small if $`M_G`$ is too “close” in mass to $`M_2`$. A numerical investigation reveals that, if one wants to have a mass of $`O(1eV)`$ and, at the same time, a large enough angle, $`M_G`$ can be relatively “low” ($`10^4`$ in units of $`M_F`$). (This would imply that the scale of family symmetry could be a few thousands of TeV if $`M_F`$ is a few hundred GeV’s.) We now give a couple of numerical estimates. We shall take the Second Example as a prototype. There one can calculate the factor $`s_\beta ^2c_\alpha ^2`$ to be $`0.134`$. 1) For $`M_{h_4}=100`$ with all other masses being the same as those of the Second Example, we obtain
$$m_{34}m_{43}m_N\lambda _{34}\frac{\stackrel{~}{v}}{V}\times 4.7\times 10^7.$$
(124)
If we wish $`m_{34}m_{43}m_N.8\times 10^7`$, then $`\lambda _{34}\frac{\stackrel{~}{v}}{V}0.17`$. So one could either have $`\lambda _{34}.2`$ and $`\stackrel{~}{v}V`$, or some other combination. 2)For $`M_{h_4}=10`$, we have
$$m_{34}m_{43}m_N\lambda _{34}\frac{\stackrel{~}{v}}{V}\times 1.4\times 10^5,$$
(125)
which would imply $`\lambda _{34}\frac{\stackrel{~}{v}}{V}0.006`$\- a reasonable constraint.
It turns out that the cases with $`M_{h_4}1000`$ (in units of $`M_F`$) do not work because then the mass ratios are not large enough to compensate for the smallness of the integrals. It is interesting that one can have scenarios where $`\stackrel{~}{h}_4`$ is light enough (i.e. not too much heavier than $`F`$)- a feature which could have interesting phenomenological implications.
### D Oscillation Angles
To discuss the neutrino oscillation angles, one needs to give the leptonic “CKM” matrix, namely $`V_L=U_l^{}U_\nu `$. It is beyond the scope of this paper to discuss the charged lepton sector, and hence $`U_l`$. This will be the subject of the following publication. However, we can give an example of $`U_l`$ by adopting, at least for this paper, a simple model of charged lepton masses of Ref. (), which is a phenomenological model based on a generalization to the lepton sector of the “democratic mass” ansatz of the quark sector. The reason why we use, as an example, Ref. () is because the matrix which diagonalizes the neutrino mass matrix, $`U_\nu `$, is identical to the $`3\times 3`$ submatrix of our Eq. (77) (apart from a difference in in the overall sign), namely
$$U_\nu ^{(3)}=\left(\begin{array}{ccc}\frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0\\ \frac{1}{\sqrt{2}}& \frac{1}{\sqrt{2}}& 0\\ 0& 0& 1\end{array}\right)$$
(126)
Although Ref. () discussed an ansatz for three generations, we will use it here because the mixing with the 4th generation is not relevant for the oscillation angles we are interested in. (It was relevant for the mass splitting.) So, basically, we will be using only the phenomenological ansatz for the charged lepton mass matrix of Ref. (). In fact, we will only use the matrix which diagonalizes that mass matrix.
The $`3\times 3`$ leptonic “CKM” matrix written down by Ref. () is
$$V_l=(AB_l)^{}U_\nu \left(\begin{array}{ccc}1& (1/\sqrt{3})\sqrt{m_e/m_\mu }& (2/\sqrt{6})\sqrt{m_e/m_\mu }\\ \sqrt{m_e/m_\mu }& 1/\sqrt{3}& 2/\sqrt{6}\\ 0& 2/\sqrt{6}& 1/\sqrt{3}\end{array}\right)$$
(127)
where $`AB_l`$ is the matrix which diagonalizes the charged lepton mass matrix, $`U_\nu `$ is given above, and $`m_e`$ and $`m_\mu `$ are the electron and muon masses respectively. Now, the probability for $`\nu _e\nu _\mu `$ is
$$P(\nu _e\nu _\mu )2(V_{11}^2V_{21}^2+V_{12}^2V_{22}^2V_{13}^2V_{23}^2)\mathrm{sin}^2(1.27\mathrm{\Delta }m_{12}^2L/E),$$
(128)
where the usual notation $`\mathrm{sin}^2(2\theta _{12})`$ is simply the coefficient of $`\mathrm{sin}^2(1.27\mathrm{\Delta }m_{12}^2L/E)`$. Similarly
$$P(\nu _\mu \nu _\tau )4V_{23}^2V_{33}^2\mathrm{sin}^2(1.27\mathrm{\Delta }m_{23}^2L/E),$$
(129)
with $`\mathrm{sin}^2(2\theta _{23})`$ being the coefficient of $`\mathrm{sin}^2(1.27\mathrm{\Delta }m_{23}^2L/E)`$. Putting in the values of $`m_e`$ and $`m_\mu `$ to evaluate the matrix elements of $`V_l`$, one readily obtains
$$\mathrm{sin}^2(2\theta _{12})6.5\times 10^3;\mathrm{sin}^2(2\theta _{23})0.89.$$
(130)
These results correspond to the small angle MSW solution, and to the large angle atmospheric solution respectively. This is consistent with the best fit for the two neutrino oscillation problems.
The above results should be viewed with caution. The small angle MSW solution given above, as well as the large angle solution for the atmospheric oscillation, depends on the charged lepton sector \- the neutrino sector diagonalizatin matrix being already fixed by Eq. (77). One can easily imagine how these angles can drastically change if the charged lepton mass matrix has a different texture. This will be the subject of a subsequent paper where we will examine the charged lepton mass matrix in the context of the present model- the basic interaction Lagrangian being already given by Eq. (4).
## IV Epilogue
The above discussions focused entirely on the atmospheric and solar neutrino data. We have left out the LSND result for two reasons. Firstly, it is because it might be prudent to wait for future experiments, either to confirm or to refute these results. Secondly, it is because it is extremely hard to incorporate all three experiments simultaneously in a “natural” model. In general, one needs to invoke some kind of sterile neutrino that mixes with the lightest neutrino to explain the solar data. If this sterile neutrino were to arise from some kind of model, it is rather hard to invent, in a “natural” way, a scenario to explain why this sterile neutrino is so light and close in mass to one of the three active light neutrinos.
Let us suppose that the LSND result are verified by future experiments. What does the model presented in this paper have to say about a sterile neutrino? Let us remember that $`\eta _R=(\nu _R^\alpha ,\stackrel{~}{\nu }_R^\alpha )`$ is an electroweak singlet. Furthermore we have seen that it is $`\nu _R^\alpha `$ which mixes with $`l_L^\alpha `$ to give masses to the neutrinos. Its $`SU(2)_{\nu _R}`$ partner, $`\stackrel{~}{\nu }_R^\alpha )`$, remains massless, at least within the framework of the preceding sections. Could these be the so-called sterile neutrinos? If so, how would they get a mass? How would they mix with the light neutrinos? These are the questions which are under investigation.
We have concentrated in this manuscript on the even option. One might wonder about the odd option and its implication on neutrino masses. It is beyond the scope of this paper to investigate this issue, however a preliminary investigation of the odd option, with three families and one family singlet $`\eta ^{}`$, appears to indicate that the preferred solution for the neutrino masses is that in which there is a hierarchy $`m_1m_2m_3`$.
There are numerous phenomenological consequences to be worked out in subsequent publications. One can, however, make one rather solid prediction: neutrinos, being of the Dirac nature, will not give rise to the phenomenon of neutrinoless double beta decay. Another interesting consequence is the possible existence of “light” (i.e. 200 GeV or so) vector-like fermions: $`F`$, as well as TeV-scale pseudo NG bosons which carry family and $`SU(2)_{\nu _R}`$ quatum numbers. This will be dealt with in a separate paper.
Several other phenomenological issues remain to be investigated. For instance, what are the consequences of a broken $`SU(2)_{\nu R}`$ and what might the cosmological implications of $`\stackrel{~}{\nu }_R`$’s and $`\eta _R^{}`$ be? When $`SU(2)_{\nu R}`$ is broken by $`\rho _i^\alpha `$, the gauge bosons are expected to acquire a mass of O($`V^{}`$) and can be quite heavy. Since only right-handed neutral leptons participate in $`SU(2)_{\nu R}`$ interactions, a place where the effects of those gauge bosons might show up is in the decays of neutrinos. Without going into detail, it is easy to see that the decay of the light (near-degenerate) neutrinos into each other is completely negligible for lack of phase space and for the fact that neutrino masses are tiny compared with $`V^{}`$ (even if the latter is in the TeV region). This leaves us with the decay of the (heavy) fourth-generation neutral lepton $`N`$ for which we have $`N\stackrel{~}{N}+\nu _i+\stackrel{~}{\nu }_i`$ (1) via the exchange of $`SU(2)_{\nu R}`$ gauge bosons, and $`Nl_i^{}+l_j^++\nu _j`$ (2) if $`m_N<m_W`$ or $`Nl_i^{}+W`$ (3) if $`m_N>m_W`$. In addition, one could have $`NE+l_j^++\nu _j`$ when $`m_N>m_E`$, via the exchange of $`W`$. Whether or not $`m_N`$ is larger or smaller than $`m_E`$, the relevant decays to compare with each other are (1) and (3). To make an estimate, let us assume the the family gauge coupling is about the same size as the electroweak coupling ($`g0.7`$). The ratio of the decay widths for (1) and (3) is approximately $`\mathrm{\Gamma }(1)/\mathrm{\Gamma }(3)7.5\times 10^4(M_W/M_{\stackrel{~}{G}})^2(m_N/M_{\stackrel{~}{G}})^2(1(M_W/m_N)^4)^2x^2`$, where $`M_{\stackrel{~}{G}}`$ represent the mass of the $`SU(2)_{\nu _R}`$ gauge bosons and $`x`$ represents the mixing cofficient between the 4th neutrino and a light charged lepton. Now let us remember that the computation of the neutrino masses does not involve $`M_{\stackrel{~}{G}}`$ and as a result there appears to be no constraint there. However, $`M_{\stackrel{~}{G}}gV^{}`$ and $`M_GgV`$, and as a result $`\mathrm{tan}\beta V^{}/VM_{\stackrel{~}{G}}/M_Gg^210^9(M_2^2/M_G^2)`$. In the second example discussed in the previous example, $`V^{}V`$ (with $`M_G=10^4M_F`$) which implies $`M_{\stackrel{~}{G}}M_G`$. Now $`\mathrm{\Gamma }(1)/\mathrm{\Gamma }(3)`$ can also be appreciable if $`m_N`$ is close to $`m_W`$. For example, if $`M_F200GeV`$ and $`m_N82GeV`$, $`\mathrm{\Gamma }(1)/\mathrm{\Gamma }(3)1`$ provided $`x10^9`$. If this were the case, the signal would be quite interesting: a long-lived massive neutral lepton whose electroweak decay width is not what it should be. It is certainly beyond the scope of this paper to explore numerous phenomenological consequences which might arise from our scenario.
As for the cosmological consequences of $`\stackrel{~}{\nu }_R`$’s and $`\eta _R^{}`$, if they are massless, one should recall our earlier discussion: These particles only have family and $`SU(2)_{\nu R}`$ gauge interactions (both for $`\stackrel{~}{\nu _R}`$’s and the latter only for $`\eta _R^{}`$). Therefore, they cannot influence big-bang nucleosynthesis. One can estimate their decoupling temperatures by comparing the interaction rate $`\mathrm{\Gamma }_{int}G^2T^5`$, where $`G^21/(64V^{()4})`$, with the Hubble rate $`HT^2/m_{pl}`$. Decoupling occurs when $`\mathrm{\Gamma }_{int}<H`$ which gives a temperature of O($`10^6`$) GeV if $`V^{()}10^9`$ GeV for example. After this, their temperature would scale like $`T1/R`$. It is not clear what else they can do except to exist as almost non- interacting relativistic relics with an energy density negligible compared with normal matter. At this stage, it is also not clear if they really do need to have a mass. The cosmology of these objects is probably worth exploring further.
Another interesting cosmological subject to explore is the “heaviest” particle in our scenario: The vector-like neutral fermion $`_2`$ which is singlet under all the listed gauge groups in Eq. (2). $``$ couples to other fermions via 4. The decay modes obtained from (4) are: $`_{2R}\varphi ^\pm F_L^{}`$ (1) and $`_{2L}\rho _\alpha \eta ^\alpha `$ (2). Notice that, in the examples given above for the calculations of the neutrino masses, the mass of this fermion is typically $`M_210^9M_F`$. So, if $`M_F200GeV`$ (or a few hundred GeV), one would then expect the mass of $`_2`$ to be around a few times $`10^{11}`$ GeV. If $`M_F1TeV`$, $`_2`$ would have a mass around $`10^{12}`$ GeV. The questions that we would like to investigate are: (1) How many $`_\mathcal{2}`$ are left in the present universe?;(2) Could the decay of the relic $`_\mathcal{2}`$’s manifest itself as ultra high energy cosmic rays (UHECR)(with energy exceeding $`10^{20}eV=10^{11}GeV`$) whose origins are still unknown? It does appear that the mass of $`_2`$ is in the right energy ballpark. This would be the case of a non-accelerated source of UHECR and is part of the “top-down” approach to UHECR . For example, $`_{2R}`$ would decay into the longitudinal component of $`W`$ ($`\varphi ^\pm `$) and $`F_L^{}`$. $`\varphi ^\pm `$ would in turn decay into extremely high-energy quarks and leptons. The quarks will hadronize into hadrons such as pions which will eventually convert into photons, neutrinos, etc..
Last but not least, in the subsequent series of papers, we shall deal with the charged lepton sector and with the quark sector. In particular, we shall see how the generalization of Eq. (4) to the quark sector might yield interesting results.
I would like to thank Vernon Barger, Paul Fishbane and Paul Frampton for reading the manuscript and for useful comments. This work is supported in parts by the US Department of Energy under grant No. DE-A505-89ER40518.
## Higgs Potential
In this Appendix, we shall discuss a simple form of the Higgs potential for the group $`SO(4)SU(2)_{\nu _R}`$. For simplicity, we shall assume that there is no cross coupling between ($`\mathrm{\Omega }`$, $`\rho `$) and the SM Higgs field $`\varphi `$. (One might wonder about the fact that even if the cross coupling were vanishing, it might still be induced through radiative corrections. This, however, would be very small in our model.)
The potential containing $`\mathrm{\Omega }`$ and $`\rho `$ reads
$`V(\mathrm{\Omega },\rho )`$ $`=`$ $`\lambda _1(\mathrm{\Omega }^\alpha \mathrm{\Omega }_\alpha V^2)^2+\lambda _2(\rho ^\alpha \rho _\alpha V^{\mathrm{\hspace{0.17em}2}})^2+\lambda _3[(\mathrm{\Omega }^\alpha \mathrm{\Omega }_\alpha V^2)(\rho ^\alpha \rho _\alpha V^{\mathrm{\hspace{0.17em}2}})]^2`$ (132)
$`+\lambda _4[(\mathrm{\Omega }^\alpha \mathrm{\Omega }_\alpha )(\rho ^\beta \rho _\beta )(\mathrm{\Omega }^\alpha \rho _\alpha ^{})(\mathrm{\Omega }^\beta \rho _\beta )`$
where $`<\mathrm{\Omega }>=(0,0,0,V)`$ and $`<\rho >=(0,0,0,V^{}s_1)`$, with $`s_1=\left(\begin{array}{c}1\\ 0\end{array}\right)`$. Here, we will assume that $`\mathrm{\Omega }`$ is real and $`\rho `$ is complex. We will be particularly interested in the mass eigenstates resulting from Eq. (132).
With $`\mathrm{\Omega }_4=H_4+V`$ and $`\rho _4=\left(\begin{array}{c}h_4+V^{}+i\varphi _4\\ \rho _4^{}\end{array}\right)`$, Eq. (132) gives rise to the following mass matrix for $`H_4`$ and $`h_4`$:
$$8\left(\begin{array}{cc}(\lambda _1+\lambda _3)V^2& \lambda _3VV^{}\\ \lambda _3VV^{}& (\lambda _2+\lambda _3)V^2\end{array}\right)$$
(133)
The eigenvectors are
$$\stackrel{~}{H}_4=\mathrm{cos}\alpha H_4+\mathrm{sin}\alpha h_4$$
(135)
$$\stackrel{~}{h}_4=\mathrm{sin}\alpha H_4+\mathrm{cos}\alpha h_4$$
(136)
The associated eigenvalues are
$$m_{H_4}^2=4(\lambda _2+\lambda _3)V^2m_1^2,$$
(138)
$$m_{h_4}^2=4(\lambda _2+\lambda _3)V^2m_2^2,$$
(139)
where
$$m_{1,2}^2=\frac{1+a\pm \sqrt{(1a)^2+4b^2}}{2},$$
(141)
$$a=(\frac{\lambda _1+\lambda _3}{\lambda _2+\lambda _3})\mathrm{tan}^2\beta ,$$
(142)
$$b=(\frac{\lambda _3}{\lambda _2+\lambda _3})\mathrm{tan}\beta ,$$
(143)
$$\mathrm{tan}\beta =\frac{V^{}}{V},$$
(144)
$$\mathrm{cos}\alpha =\frac{1}{\sqrt{1+[(1m_1^2)/b]^2}}.$$
(145)
The mass matrix for $`\mathrm{\Omega }_i`$ and $`Re\rho _i`$ with $`i=1,2,3`$, is
$$2\lambda _4\left(\begin{array}{cc}V^2& VV^{}\\ VV^{}& V^2\end{array}\right)$$
(146)
The eigenvectors are
$$\stackrel{~}{\mathrm{\Omega }}_i=\mathrm{cos}\beta \mathrm{\Omega }_i+\mathrm{sin}\beta Re\rho _i,$$
(148)
$$Re\stackrel{~}{\rho }_i=\mathrm{sin}\beta \mathrm{\Omega }_i+\mathrm{cos}\beta Re\rho _i,$$
(149)
The associated eigenvalues are
$$m_{\stackrel{~}{\mathrm{\Omega }}}=0,$$
(151)
$$m_{Re\stackrel{~}{\rho }}=2\lambda _4(V^2+V^2).$$
(152)
Notice that $`\stackrel{~}{\mathrm{\Omega }}_i`$ are NG Goldstone bosons which are absorbed by some of the $`SO(4)`$ gauge bosons.
Since it is not of immediate relevance to the paper, we will simply quote the masses of the other scalars obtained from (132). Scalars (Pseudo-NG bosons) which have a mass $`2\lambda _4V^2`$: $`Im\rho _i`$, $`Re\rho _i^{}`$, $`Im\rho _i^{}`$. Goldstone bosons which are absorbed by some of the $`SO(4)SU(2)_{\nu _R}`$ gauge bosons: $`Re\rho _4^{}`$, $`Im\rho _4^{}`$. Notice that the pseudo-NG boson masses are all proportional to $`\lambda _4`$. As a result, their masses tend to zero as $`\lambda _40`$. |
warning/0003/astro-ph0003377.html | ar5iv | text | # Driven-disk model for binaries with precessing donor star. Three-dimensional simulations
## 1 INTRODUCTION
Observations of binary systems over the past ten years indicate that in a number of close binaries, the best known being Her X-1 (HZ Her) and SS433, long-period variations are detected on characteristic timescales substantially longer than the orbital period. To explain these variations, precession of the accretion disk in the binary is widely assume (see, e.g., \[1–3\] and references therein).
Possible reasons for precession of an accretion disk have been analyzed in a number of studies. As early as in 1972 N.I.Shakura noted that the accretion disk can precess if its plane does not coincide with the orbital plane of a binary. Among various mechanisms that can lead to formation of an accretion disk inclined to the orbital plane, two are usually considered to be most probable: influence of magnetic field of the accretor or violation of the symmetry of the donor-star outflow due to rotation of the star. The question of formation an inclined disk under the action of the accretor’s magnetic field remains open: control of the disk orientation would require a strong magnetic field which, in turn, could inhibit the formation of the disk itself.
An inclined accretion disk is more likely to form due to a change of the position of the stream flowing from the inner Lagrangian point $`L_1`$. In particular W.Roberts and A.M.Cherepashchuk proposed a scenario in which a minor asymmetry of the explosion of a supernova (resulting in the formation of a relativistic object in the binary) can decline the orbital plane of a binary relative to the rotation axis of the normal component of the system. In this case, after the supernova explosion and the formation of a relativistic object, the rotation axis of the normal component of the system might become oriented not perpendicular to the orbital plane of the binary. For systems where mass is transfer from the normal star to the relativistic object, the disturbance of the symmetry of the outflow of matter from the donor-star might result in the formation of an accretion disk not aligned with the orbital plane. Precession of such a disk could be caused either by induced precession of the disk itself under the action of the gravitational attraction of donor-star , or by the stream oscillations due to precession of the rotation vector of the donor-star (the ‘slaved disk’ model, ), or by some other mechanism (see, e.g., ).
The description of the gas dynamics of mass transfer binaries of this type calls for the use of three-dimensional models because the rotation vectors of the donor-star is not perpendicular to the orbital plane and is engaged in precession. This means that this problem cannot be reduced to a two-dimensional one. Further, there is an additional complexity in describing such systems connected with the periodic time dependence of the boundary conditions, that is with the absence of a steady-state flow of matter in the system, thus necessitating the consideration of the structure of flow over long periods of time (few times longer than maximum characteristic period of the system). Until very recently, these circumstances, together with insufficient computational resources, made numerical studies of flow structures in binary systems of this type difficult. a Few attempts to consider the formation of an accretion disk in such systems were made in substantially simplified formulation .
Here we consider for the first time numerical, self-consistent solution for the three-dimensional flow structure of mass transfer in semidetached binary systems in which the donor’s rotation axis precesses. The obtained results support the model of ‘driven accretion disk’, based on the idea that the oscillations of the disk relative to the equatorial plane reflect variations of the stream of matter from the inner Lagrangian point $`L_1`$.
## 2 THE MODEL
In we presented a three-dimensional simulation of mass transfer in semidetached binaries with rotation of the donor-star. We investigated both synchronous (the rotational period of the donor-star equals to the orbital period of the system $`P_{}=P_{orb}`$) and asynchronous ($`P_{}P_{orb}`$) rotation of the donor-star . For the case of asynchronous rotation both axially aligned rotation, when the rotation vector of the donor-star is perpendicular to the orbital plane of a binary, and non-aligned, when this vector is inclined relative to the orbital plane were considered. It was also assumed in that the case of asynchronous non-aligned rotation implies that the rotation vector of the donor-star is in a rest in the laboratory coordinate system (i.e., relative to observer).
The existence of observational evidence for precession of rotational vector of the donor-star requires extension of the model. The model we use here assumes that the rotation vector of the donor-star precesses (in a laboratory coordinate system) about its mean position, which coincides with the vector of orbital rotation of the binary. Note that although this model has a formally more general character (rate of precession of the rotation vector in the laboratory coordinate system $`\stackrel{~}{\mathrm{\Omega }}_{lab}0`$) than the model considered in for the case of asynchronous non-aligned rotation (rate of precession of the rotation vector in the laboratory coordinate system $`\stackrel{~}{\mathrm{\Omega }}_{lab}=0`$), these models differ significantly in the laboratory coordinate system only. In a rotating coordinate system (adopted for our calculations), we can expect that there will be qualitatively similar solutions for both models because in this coordinate system the rotation vector of the donor-star in model of also undergoes precessional motion, i.e. for both models $`\stackrel{~}{\mathrm{\Omega }}_{rot}0`$. The different rates of precession in the models should lead only to quantitatively difference of the solutions leaving the general features of the flow structure in a binary unchanged.
To study in detail whether accretion disk will follow the ‘driven disk’ model, we conducted the calculations over a time exceeding the period of long-period variations of the system; this made it possible to obtain established solutions both in a laboratory and in a rotating coordinate system. To be able to compare the solutions with the results obtained in we adopted the same parameters of binary system: the primary (filling its Roche lobe) parameters are $`M_1=0.28M_{}`$, $`T_1=10^4`$ K; the secondary (compact objects) parameters are $`R_2=0.013R_{}`$, $`M_2=1.4M_{}`$; parameters of binary system $`P_{orb}=5^\text{h}.56`$; $`A=1.97R_{}`$. We assumed that the rotation velocity of the donor-star in the laboratory coordinate system exceeds twice the angular velocity of the system. The angle between the rotation vector of the donor star $`𝛀_{}`$ and the ‘$`z`$’-axis was assumed to be $`\vartheta =15^{}`$ (in the laboratory coordinate system).<sup>1</sup><sup>1</sup>1According to the vector transformation rule $`𝛀_{rot}=𝛀_{lab}𝛀`$ the inclination angle of this vector in the rotating coordinate system is $`\vartheta ^{}=30^{}`$. The rate of precession of rotation vector of the donor-star was equal to $`\stackrel{~}{\mathrm{\Omega }}_{lab}={}_{}{}^{1}/_{6}^{}\mathrm{\Omega }`$ in the laboratory coordinate system or $`\stackrel{~}{\mathrm{\Omega }}_{rot}={}_{}{}^{5}/_{6}^{}\mathrm{\Omega }`$ – in the rotating coordinate system (the minus sign indicates that the precession of the rotation vector is retrograde with respect to the orbital rotation).
Following the procedure of , we determined the boundary conditions on the surface of this star by solving the Riemann problem between gas parameters ($`\rho _1,𝒗_1,p_1`$) on the stellar surface and the parameters in the computation gridpoint closest to the given point on the star surface. The asynchronism of rotation of the donor-star, as well as the precession of the vector of its rotation were taken into account when setting the boundary conditions for the gas velocity vector on the Roche lobe of the donor-star. All other parameters of the mathematical model including the details of numerical realization were taken the same as in , namely:
* We adopted the computational domain as a parallepipedon $`(A\mathrm{}2A)\times (A\mathrm{}A)\times (A\mathrm{}A)`$;
* A non-uniform (more fine in the zone near accretor) finite-difference grid with $`91\times 81\times 55`$ gridpoints was used;
* The shape of the donor-star was determined, taking into account the asynchronicity of its rotation.
## 3 RESULTS OF CALCULATIONS
The results of calculations fully confirmed our assumption that, given a rotating coordinate system, the qualitative features of the flow structure are independent of the rate of precession of the rotation vector of the donor-star. Fig. 1 depicts the iso-surfaces of density at the level $`\rho =0.002\rho _{L_1}`$ for 10 moments of time which cover a full period of variation of the boundary conditions in a rotating coordinate system. Note that, in line with the assumed parameters of the system, the period of precession of the rotation vector of the donor-star in a rotating coordinate system $`\stackrel{~}{P}_{rot}`$ is equal to $`{}_{}{}^{6}/_{5}^{}`$ of the orbital period $`P_{orb}`$. Owing to this the results are given in Fig. 1 over the interval $`{}_{}{}^{6}/_{5}^{}P_{orb}`$.
Analysis of the results presented in Fig. 1 shows that the behavior of the disk $`c`$ and of the near-disk matter $`d`$ reflects the variations of the stream of matter $`a`$ flowing from $`L_1`$ (i.e. boundary conditions on the surface of the donor-star $`a`$). This points to the realization of the ‘slaved disk’ model in the calculations. The results show also that the unique morphology of the flow, as well as the effect of circumbinary envelope gas lead to a shock-free interaction between the stream and the outer edge of the accretion disk and, as a consequence, to the absence of a ‘hot spot’ in the disk. In the model discussed here the zone of increased energy release is located beyond the accretion disk – at the region where the gas of circumbinary envelope interacts with the stream and where the extended shock wave is formed. Similarly to the case of non-aligned asynchronous rotation described in , the interaction between the rarefied gas of the circumbinary envelope and the stream of matter results in the formation of gaseous ‘clouds’ located ahead of the front edge of the stream, beyond the zone of disk formation (marker $`d`$ in Fig. 1). The period of variation of these gaseous formations – ‘clouds’ – reflects the period of variation of the boundary conditions at the mass-losing star.
In the laboratory coordinate system the characteristic timescale of the long-period variation (precession of the rotation vector of the donor-star with period $`\stackrel{~}{P}_{lab}`$) is equal to $`6P_{orb}`$. The calculations were made over period of time exceeding $`\stackrel{~}{P}_{lab}`$ starting from the time when the flow was established. As expected, the comparison of results of calculations in the laboratory coordinate system confirms the periodic character of solution with the period $`\stackrel{~}{P}_{lab}`$. This is illustrated in Fig. 2 and 3 where iso-surfaces of density at the level $`\rho =0.002\rho _{L_1}`$ are shown in the laboratory coordinate system. These plots show the 3D view of density distribution in the vicinity of $`L_1`$ and in the near-disk region, respectively, for six moments of time covering the period of precession in the laboratory coordinate system. Analysis of Fig. 2 indicates that in accordance with the variations of boundary conditions, the position of the stream flowing from $`L_1`$ varies with respect to the equatorial plane, and, after a time $`\stackrel{~}{P}_{lab}`$, returns to its initial position. Figure 3 shows that the accretion disk $`c`$ and surrounding gaseous formation (clouds $`d`$ and $`d^{}`$) follow the position of the stream (see Fig. 2), and their shape varies with the same period.
The presence of two characteristic periods – the precessional period of the rotation vector of the donor star in the laboratory frame $`\stackrel{~}{P}_{lab}`$ and orbital period of the system $`P_{orb}`$ – can explain the appearance of other periods of observation evidences, apart from $`\stackrel{~}{P}_{lab}`$. In particular, the interaction of $`\stackrel{~}{P}_{lab}`$ with $`P_{orb}`$ leads to formation of short-period beating oscillations. For example, as concerns the system SS433 ($`P_{orb}=13^\text{d}.1`$, $`\stackrel{~}{P}_{rot}=12^\text{d}.1`$), in additions to the period $`\stackrel{~}{P}_{lab}=162^\text{d}.5`$, there should exist variations with the period of $`6^\text{d}.28`$. These short-period oscillations, as well as the following beating harmonic with period $`5^\text{d}.83`$ are observed in SS433 (see, e.g., and references therein).
## 4 CONCLUSIONS
Our three-dimensional simulations of mass transfer in semidetached binaries with precession of rotation vector of donor star reveal the ‘driven’ nature of forming accretion disk. For typical parameters of numerical viscosity adopted for the numerical model ($`\alpha 0.1÷0.5`$ in terms of $`\alpha `$-disk), the change of the flow pattern, the parameters of accretion disk, and parameters of the near-disk regions reflect the variations of boundary conditions on the donor star. In turn, the periodicity of boundary conditions is determined by the precessional velocity. Analysis of the flow pattern indicates that the basic features of the solution are qualitatively similar to those for calculations previously obtained for the cases of synchronous, aligned asynchronous, and non-aligned asynchronous rotation of the donor-star , and, in turn, indicates to the universal character of the model without a ‘hot spot’ proposed in \[15–18\].
The ‘driven’ character of the solution implies that the emission properties of the accretion disk and intercomponent gaseous structures recur with the precessional period of the rotation axis of donor-star. In binary systems where observed long-period variations can be explained by the precession of the donor-star the periodicity of the solution obtained here can be used to interpret the observational data.
## ACKNOWLEDGMENTS
This work was supported by the Russian Foundation for Basic Research (grant 99-02-17619), the INTAS Foundation (grant 93-93-EXT), and Russian Federation Presedential Grant N 99-15-96022.
## REFERENCES
1. Roberts, W.J. 1974, ApJ, 187, 575
2. Cherepashchuk, A.M. 1988, Itogi Nauki i Techniki. Astronomiya, 38, 60 (in Russian)
3. Margon, B. 1984, ARA&A, 22, 507
4. Shakura, N.I. 1972, Astron. Zh., 49, 921 (Sov. Astron., 16, 756)
5. Cherepashchuk, A.M. 1981, Pis’ma Astron. Zh., 7, 201 (Sov. Astron. Lett., 7, 111)
6. Cherepashchuk, A.M. 1981, Pis’ma Astron. Zh., 7, 726 (Sov. Astron. Lett., 7, 401)
7. Katz, J.I. 1973, Nature Phys. Sci., 246, 87
8. Petterson, J.A. 1975, ApJ, 201, L61
9. Shakura, N.I., Prokhorov, M.E., Postnov, K.A. & Ketsaris, N.A. 1999, A&A, 348, 917
10. Belvedere, G., Lanzafame, G., & Molteni, D. 1993, A&A, 280, 525
11. Lanzafame, G., Belvedere, G., & Molteni, D. 1994, MNRAS, 267, 312
12. Bisikalo, D.V., Boyarchuk, A.A., Kuznetsov, O.A., & Chechetkin, V.M. 1999, Astron. Zh., 76, 270 (Astron. Reports, 43, 229; preprint astro-ph/9812484)
13. Katz, J.I., Anderson, S.F., Margon, B., & Grandi, S.A. 1982, ApJ, 260, 780
14. Margon, B., & Anderson, S.F. 1989, ApJ, 347, 448
15. Bisikalo, D.V., Boyarchuk, A.A., Kuznetsov, O.A., & Chechetkin, V.M. 1997, Astron. Zh., 74, 880 (Astron. Reports, 41, 786; preprint astro-ph/9802004)
16. Bisikalo, D.V., Boyarchuk, A.A., Kuznetsov, O.A., & Chechetkin, V.M. 1997, Astron. Zh., 74, 889 (Astron. Reports, 41, 794; preprint astro-ph/9802039)
17. Bisikalo, D.V., Boyarchuk, A.A., Chechetkin, V.M., Molteni, D., & Kuznetsov, O.A. 1998, MNRAS, 300, 39
18. Bisikalo, D.V., Boyarchuk, A.A., Kuznetsov, O.A., & Chechetkin, V.M. 1998, Astron. Zh., 75, 706 (Astron. Reports, 42, 621; preprint astro-ph/9806013) |
warning/0003/math0003093.html | ar5iv | text | # 1 Statement of results in terms of flat connections
## 1 Statement of results in terms of flat connections
In one respect the summary given in the introduction is slightly inaccurate. The space we study is not exactly the moduli space of flat connections on a compact surface. For that space is generally singular due to the presence of reducible connections. This problem is circumvented by shifting attention to connections of constant central curvature whose degrees are coprime to their ranks. The bait-and-switch is perhaps regrettable, but it is standard practice in the subject.
So let $`C`$ be a surface of genus $`g`$. The fundamental group of $`C`$ has presentation
$$\pi _1(C)=\frac{a_1,b_1,\mathrm{},a_g,b_g}{_{j=1}^ga_j^{}b_j^{}a_j^1b_j^1}.$$
Let $`G`$ be the non-compact group $`\mathrm{GL}(r,)`$. A flat $`G`$-connection on $`C`$ is determined by a representation $`\pi _1(C)G`$. So if $`\mu :G^{2g}G`$ is defined by
$$\mu (A_1,B_1,\mathrm{},A_g,B_g)=\underset{j=1}{\overset{g}{}}A_j^{}B_j^{}A_j^1B_j^1,$$
then any element of $`\mu ^1(I)`$ defines a flat $`G`$-connection. The quotient $`\mu ^1(I)/G`$, where $`G`$ acts by conjugation on all factors, therefore parametrizes all flat $`G`$-connections modulo gauge equivalence. However, as mentioned above, this is generally a singular space. An exception is when $`r=1`$, but then it is nothing but the complex torus $`𝒯=(^\times )^{2g}`$.
Instead of struggling with the singularities, choose any integer $`d`$ coprime to $`r`$, and consider the space $`=\mu ^1(e^{2\pi id/r}I)/G`$. This is a non-compact complex manifold — indeed, a smooth affine variety — and will be our main object of study. It parametrizes gauge equivalence classes of $`G`$-connections on $`C`$ of constant central curvature $`di\omega I`$, where $`\omega `$ is a 2-form on $`C`$ chosen so that $`_C\omega =2\pi /r`$, and $`I`$ is the $`r\times r`$ identity matrix. Indeed, such a connection is again determined by its holonomies $`(A_j,B_j)`$ around $`a_j`$ and $`b_j`$, subject to the constraint that
$$\underset{j=1}{\overset{g}{}}A_j^{}B_j^{}A_j^1B_j^1=\mathrm{exp}_C𝑑i\omega I=e^{2\pi di/r}I.$$
Alternatively, $``$ may be regarded as the space of flat $`G`$-connections on $`C\backslash p`$ having holonomy $`e^{2\pi di/r}I`$ around $`p`$, modulo gauge. This has the advantage that no choice of $`\omega `$ is necessary, but it is less compatible with the Higgs bundle interpretation coming up.
The subject of this paper is the ring structure of the rational cohomology of $``$. To begin this study, some sources of cohomology classes on $``$ are needed.
The simplest thing to do is pull back the generators of $`H^{}(𝒯)`$ by the obvious determinant map $`det:𝒯`$. This produces classes $`\epsilon _1,\mathrm{},\epsilon _{2g}H^1()`$, but they are not very interesting. In fact, the subring they generate can be split off in the following way.
Let $``$ be the space constructed in the same way as $``$, but with $`\mathrm{SL}(r,)`$ substituted for $`\mathrm{GL}(r,)`$. Certainly $``$ is a subspace of $``$; but also, scalar multiplication induces a map $`𝒯\times `$ which is easily seen to be a free quotient by the abelian group $`\mathrm{\Sigma }=_r^{2g}`$. According to a theorem of Grothendieck , the rational cohomology of such a quotient satisfies
$$H^{}()=H^{}(𝒯\times )^\mathrm{\Sigma },$$
where the right-hand side denotes the $`\mathrm{\Sigma }`$-invariant part. Now $`\mathrm{\Sigma }`$ acts on $`𝒯`$ by scalar multiplications, so it acts trivially on cohomology and hence as rings
$$H^{}()=H^{}(𝒯)H^{}()^\mathrm{\Sigma }.$$
Furthermore the composition of the $`\mathrm{\Sigma }`$-quotient with the determinant is the map $`𝒯\times 𝒯`$ given simply by projecting to $`𝒯/\mathrm{\Sigma }=𝒯`$. The subring of $`H^{}()`$ generated by $`\epsilon _1,\mathrm{},\epsilon _{2g}`$ is therefore nothing but the first factor of the tensor product.
To define more interesting cohomology classes on $``$, construct a principal bundle over $`\times C`$ as follows. Let $`\overline{G}=\mathrm{PGL}(r,)`$. Any $`\rho \mu ^1(e^{2\pi di/r}I)`$ induces a well-defined homomorphism $`\pi _1(C)\overline{G}`$. Let $`\stackrel{~}{C}`$ be the universal cover of $`C`$, which is acted on by $`\pi _1(C)`$ via deck transformations. There is then a free action of $`\pi _1(C)\times G`$ on $`\overline{G}\times \mu ^1(e^{2\pi di/r}I)\times \stackrel{~}{C}`$ given by
$$(p,g)(h,\rho ,x)=(\overline{g}\rho (p)h,\overline{g}\rho \overline{g}^1,px),$$
where $`\overline{g}`$ denotes the image of $`g`$ in $`\overline{G}`$. The quotient is the desired principal $`\overline{G}`$-bundle. Like any principal $`\overline{G}`$-bundle, it has characteristic classes $`\overline{c}_2,\mathrm{},\overline{c}_r`$, where $`\overline{c}_iH^{2i}(\times C`$). In terms of formal Chern roots $`\xi _k`$, $`\overline{c}_i`$ can be described as the $`i`$th elementary symmetric polynomial in the $`\xi _k\zeta `$, where $`\zeta `$ is the average of the $`\xi _k`$.
Now let $`\sigma H^2(C)`$ be the fundamental cohomology class, and let $`e_1,\mathrm{},e_{2g}`$ be the basis of $`H^1(C)`$ Poincaré dual to $`a_1,\mathrm{},a_g,b_1,\mathrm{},b_g`$. In terms of these, each of the characteristic classes has a Künneth decomposition
$$c_i=\alpha _i\sigma +\beta _i+\underset{j=1}{\overset{2g}{}}\psi _{i,j}e_j,$$
defining classes $`\alpha _iH^{2i2}()`$, $`\beta _iH^{2i}()`$, and $`\psi _{i,j}H^{2i1}()`$. The pullback of these classes to $`𝒯\times `$, by the way, is easily seen to come entirely from $`H^{}()^\mathrm{\Sigma }`$.
It is convenient to refer to the entire collection of classes $`\alpha _i`$, $`\beta _i`$, $`\psi _{i,j}`$, and $`\epsilon _j`$ as the universal classes.
Now specialize, for the remainder of this section, to the case $`r=2`$. Then $`d`$ must be odd, so that $`=\mu ^1(I)/G`$. The main result of this paper is then the following.
(1.1) When $`r=2`$, the rational cohomology ring of $``$ is generated by the universal classes.
Equivalently, $`^{}()^\mathrm{\Sigma }`$ is generated by the classes $`\alpha =\frac{1}{2}\alpha _2H^2()`$, $`\beta =\frac{1}{4}\beta _2H^4()`$, and $`\psi _j=\psi _{2,j}H^3()`$ for $`j=1,\mathrm{},2g`$. (These normalizations are by now standard in the literature.) It is worth mentioning that $`H^{}()`$ is generally not entirely $`\mathrm{\Sigma }`$-invariant \[24, 7.6\], hence is not generated by the universal classes. However, by the aforementioned theorem of Grothendieck, $`H^{}()^\mathrm{\Sigma }`$ is the rational cohomology of $`/\mathrm{\Sigma }`$, which is a component of the moduli space of flat $`\overline{G}`$-connections on $`C`$.
For completeness, we recount here the main result of our companion paper giving all the relations between these generators. In light of the discussion earlier it suffices to work with $`H^{}()^\mathrm{\Sigma }`$. Let $`\mathrm{\Gamma }=\mathrm{Sp}(2g,)`$. The action of diffeomorphisms on $`C`$ will be shown to induce an action of $`\mathrm{\Gamma }`$ on $`H^{}()^\mathrm{\Sigma }`$, fixing $`\alpha `$ and $`\beta `$ and acting as the standard representation on the span $`V=H^3()`$ of the $`\psi _j`$. Thus $`\gamma =2_{j=1}^g\psi _j\psi _{j+g}`$ is a $`\mathrm{\Gamma }`$-invariant element of $`H^6()`$. Let $`\mathrm{\Lambda }_0^n(\psi )`$ be the kernel of the natural map $`\mathrm{\Lambda }^nV\mathrm{\Lambda }^{2g+2n}V`$ given by the wedge product with $`\gamma ^{g+1n}`$, or equivalently, the $`\mathrm{\Gamma }`$-invariant complement of $`\gamma \mathrm{\Lambda }^{n2}V`$ in $`\mathrm{\Lambda }^nV`$.
For any $`g,n0`$, let $`I_n^g`$ be the ideal within the polynomial ring $`[\alpha ,\beta ,\gamma ]`$ generated by $`\gamma ^{g+1}`$ and the polynomials
$$\rho _{r,s,t}^c=\underset{i=0}{\overset{\mathrm{min}(c,r,s)}{}}(ci)!\frac{\alpha ^{ri}}{(ri)!}\frac{\beta ^{si}}{(si)!}\frac{(2\gamma )^{t+i}}{i!},$$
where $`c=r+3s+2t2g+2n`$, for all $`r,s,t0`$ such that $`r+3s+3t>3g3+n`$ and $`r+2s+2t2g2+n`$.
The main result of our companion paper is then the following.
(1.2) If the rank $`r=2`$, then as an algebra acted on by $`\mathrm{\Gamma }`$,
$$H^{}()^\mathrm{\Sigma }=\underset{n=0}{\overset{g}{}}\mathrm{\Lambda }_0^n(\psi )[\alpha ,\beta ,\gamma ]/I_n^{gn}.$$
Together, the two main theorems completely describe the ring structure of $`H^{}()`$ when $`r=2`$. They do not completely describe $`H^{}()`$ because of the classes not invariant under $`\mathrm{\Sigma }`$. However, these form a relatively minor and simple part of the cohomology, and can be dealt with by hand; this will be carried out in a forthcoming paper .
The main theorems will be proved in the language not of flat connections but rather of Higgs bundles. Indeed, it proved most convenient to deduce them from more general results applying to an infinite sequence of spaces of Higgs bundles, of which $``$ is only the first. We shall next review the definition of Higgs bundles, and the correspondence theorem relating them to flat connections.
## 2 Higgs bundles
A major advance in the study of these representation spaces was made by Hitchin and Simpson , who discovered that they can alternatively be viewed as moduli spaces of holomorphic objects. So now, and for the remainder of the paper, let $`C`$ be a smooth complex projective curve of genus $`g`$.
A Higgs bundle or Higgs pair on $`C`$ with values in a holomorphic line bundle $`L`$ is a pair $`(E,\varphi )`$, where $`E`$ is a holomorphic vector bundle over $`C`$, and $`\varphi `$, called a Higgs field, is any element of $`H^0(\text{End}EL)`$. Its slope is the rational number $`\mathrm{deg}E/\text{rk}E`$. A holomorphic subbundle $`FE`$ is $`\varphi `$-invariant if $`\varphi (F)FL`$. A Higgs bundle is semistable if $`\text{slope}F\text{slope}E`$ for all proper $`\varphi `$-invariant holomorphic subbundles $`FE`$, and stable if this inequality is always strict.
For example, a pair of the form $`(E,0)`$ is stable if and only if the bundle $`E`$ is stable.
We will be concerned entirely with the case when the line bundle $`L`$ is $`K(n)=K𝒪(np)`$, where $`K`$ is the canonical bundle of $`C`$, $`p`$ is a distinguished point in $`C`$, and $`n0`$.
The moduli space of Higgs bundles with values in $`K`$ was constructed by Hitchin and Simpson , and generalized to an arbitrary line bundle by Nitsure . Their work implies the following.
(2.1) For fixed rank $`r`$, degree $`d`$ coprime to $`r`$, and $`n0`$, there exists a moduli space $`_n`$ of Higgs bundles with values in $`K(n)`$, which is a smooth quasi-projective variety of dimension $`r^2(2g2+n)+2`$. For a fixed holomorphic line bundle $`\mathrm{\Xi }`$ of degree $`d`$, the locus $`_n`$ where $`\mathrm{\Lambda }^rE\mathrm{\Xi }`$ and $`\text{tr}\varphi =0`$ is a smooth subvariety of dimension $`(r^21)(2g2+n)`$.
In the case $`n=0`$, Higgs bundles are related to connections of constant central curvature in the following way. Suppose that $`C`$ is equipped with a Kähler metric, and let $`\omega `$ be the Kähler form, again normalized so that $`_C\omega =2\pi /r`$. Then Hitchin showed the following.
(2.2) Suppose that $`r`$ and $`d`$ are coprime and that $`n=0`$. Then a Higgs bundle $`(E,\varphi )`$ is stable if and only if it admits a Hermitian metric so that the metric connection $`A`$ satisfies the equation $`F_A+[\varphi ,\varphi ^{}]=di\omega I`$. This metric is unique up to rescaling and depends smoothly on $`(E,\varphi )`$.
Here $`F_A\mathrm{\Omega }^2(\text{End}E)`$ is the curvature, the Higgs field $`\varphi `$ is regarded as a section of $`\mathrm{\Omega }^{0,1}(\text{End}E)`$, and $`I`$ is the identity in $`\text{End}E`$. For such a connection $`A`$, an easy calculation shows that the $`\mathrm{GL}(r,)`$ connection $`A+\varphi +\varphi ^{}`$ has constant central curvature $`di\omega I`$. Hence there is a natural smooth map from the space $`_0`$ of Higgs bundles to the space $``$ discussed in the previous section.
In fact, this map is a diffeomorphism, as is the restriction $`_0`$. The inverse map is provided by a result of Corlette and Donaldson .
(2.3) Any $`\mathrm{GL}(r,)`$ connection on $`C`$ with constant central curvature $`di\omega I`$ is gauge equivalent to one of the form $`A+\varphi +\varphi ^{}`$, where $`\overline{}_A\varphi =0`$ and $`F_A+[\varphi ,\varphi ^{}]=di\omega I`$.
Both $`_0`$ and $``$ carry natural complex structures, but these are not identified by the diffeomorphism. Rather, they are different members of the family of complex structures which comprises a hyperkähler structure on the moduli space.
Our approach does not use this hyperkähler structure. Indeed, it will be shown in (5) that the cohomology classes defined above in terms of flat connections can also be obtained from the universal family of Higgs bundles. From then on the flat connection point of view will vanish, and the moduli space will be regarded exclusively as a Higgs space.
As pointed out by Simpson , the moduli space of Higgs bundles actually retracts onto a highly singular Lagrangian subvariety, the nilpotent cone. Therefore our results could be viewed as describing the cohomology ring of the nilpotent cone. However, this seems to be only a curiosity and is not relevant to our approach.
The advantage of the Higgs moduli space is that it admits a holomorphic action of the group $`T=^\times `$, given simply by $`\lambda (E,\varphi )=(E,\lambda \varphi )`$. This of course fixes all stable pairs of the form $`(E,0)`$, which are parametrized by the moduli space of stable bundles of rank $`r`$ and degree $`d`$. But the fixed-point set has other components as well, and they will play a crucial role in what follows.
## 3 Deformation theory of Higgs pairs
This section and the next summarize, mostly without proof, some basic facts about Higgs pairs that will be needed later on. The omitted proofs are entirely straightforward, along the lines of Markman \[31, 7.3\], Welters or the second author \[48, 2.1\]. In the rank 2 case, some details are worked out in the first author’s Ph.D. thesis .
The deformation space of a holomorphic bundle $`E`$ is well known to be $`H^1\text{End}E`$; that of a Higgs pair $`(E,\varphi )`$ is similar, but involves hypercohomology.
Let $`(E,\varphi )`$ be a Higgs pair, and let $`𝐄𝐧𝐝(E,\varphi )`$ denote the two-term complex on $`C`$
$$\text{End}E\stackrel{[,\varphi ]}{}\text{End}EK(n).$$
(3.1) The space of infinitesimal deformations of $`(E,\varphi )`$ is the first hypercohomology group $`𝐇^1𝐄𝐧𝐝(E,\varphi )`$. The space of endomorphisms of $`E`$ preserving $`\varphi `$ is $`𝐇^0𝐄𝐧𝐝(E,\varphi )`$. . $`\mathrm{}`$
Similarly, let $`𝐇𝐨𝐦\text{(}(E^{},\varphi ^{}),(E,\varphi )\text{)}`$ denote the complex
$$\text{Hom}(E^{},E)\text{Hom}(E^{},E)K(n)$$
given by $`\psi \psi \varphi ^{}\varphi \psi `$.
(3.2) The space of homomorphisms $`E^{}E`$ intertwining $`\varphi `$ with $`\varphi ^{}`$ is the zeroth hypercohomology group $`𝐇^0𝐇𝐨𝐦\text{(}(E^{},\varphi ^{}),(E,\varphi )\text{)}`$. The space of extensions of $`(E^{},\varphi ^{})`$ by $`(E,\varphi )`$ is $`𝐇^1𝐇𝐨𝐦\text{(}(E^{},\varphi ^{}),(E,\varphi )\text{)}`$. . $`\mathrm{}`$
Here an extension of one Higgs pair by another is a Higgs pair $`(E^{\prime \prime },\varphi ^{\prime \prime })`$ and a short exact sequence
$$0EE^{\prime \prime }E^{}0$$
such that $`\varphi ^{\prime \prime }`$ restricts to $`\varphi `$ on $`E`$ and projects to $`\varphi ^{}`$ on $`E^{}`$.
One more variation on the theme will be needed in §7. Let $`(E,\varphi )`$ be a Higgs pair containing a flag of $`\varphi `$-invariant subbundles. (In practice this flag will always be the Harder-Narasimhan filtration defined in §7.) Let $`\text{End}^{}E`$ be the subbundle of $`\text{End}E`$ consisting of endomorphisms fixing the flag, and let $`𝐄𝐧𝐝^{}(E,\varphi )`$ be the two-term complex
$$\text{End}^{}E\stackrel{[,\varphi ]}{}\text{End}^{}EK(n).$$
(3.3) The space of infinitesimal deformations of the Higgs pair $`(E,\varphi )`$ together with the $`\varphi `$-invariant flag is $`𝐇^1𝐄𝐧𝐝^{}(E,\varphi )`$. . $`\mathrm{}`$
## 4 Universal families of stable Higgs pairs
(4.1) Let $`(E,\varphi )`$ and $`(E^{},\varphi ^{})`$ be stable Higgs pairs with $`\text{slope}E^{}\text{slope}E`$. Then the dimension of $`𝐇^0𝐇𝐨𝐦\text{(}(E^{},\varphi ^{}),(E,\varphi )\text{)}`$ is $`1`$ if $`(E^{},\varphi ^{})`$ and $`(E,\varphi )`$ are isomorphic, and $`0`$ otherwise. . $`\mathrm{}`$
In particular, the space of endomorphisms $`𝐇^0𝐄𝐧𝐝(E,\varphi )`$ consists only of scalar multiplications.
(4.2) Let $`(𝐄,𝚽)`$ and $`(𝐄^{},𝚽^{})X\times C`$ be families of stable Higgs pairs parametrized by $`X`$ such that for all $`xX`$, $`(𝐄,𝚽)_x(𝐄^{},𝚽^{})_x`$. Then there exists a line bundle $`LX`$ such that $`(𝐄,𝚽)(𝐄^{}\pi _1^{}L,𝚽)`$. In particular, $`𝐄`$ and $`\text{End}𝐄`$ are canonical.
Proof. By (4), $`𝐇^0𝐇𝐨𝐦\text{(}(𝐄^{},𝚽^{})_x,(𝐄,𝚽)_x\text{)}`$ is 1-dimensional for all $`x`$. Hence the hyperdirect image $`(𝐑^0\pi _1)_{}𝐇𝐨𝐦\text{(}(𝐄^{},𝚽^{}),(𝐄,𝚽)\text{)}`$ is a line bundle $`LX`$. It is then easy to construct the desired isomorphism. . $`\mathrm{}`$
It is clear from the proof that the above proposition holds true not only for algebraic families of Higgs pairs, but even for smooth families, that is, $`C^{\mathrm{}}`$ bundles $`(𝐄,𝚽)X\times C`$ for any smooth parameter space $`X`$, endowed with a partial holomorphic structure in the $`C`$-directions.
(4.3) Let $`(𝐄,𝚽)`$ be a family of stable Higgs pairs parametrized by $`X`$, and let $`^\times `$ act on $`X`$. If there are two liftings of the action to $`𝐄`$ so that the induced action on $`𝚽`$ is $`\text{Ad}(\lambda )𝚽=\lambda ^1𝚽`$, then one is the tensor product of the other with an action of $`^\times `$ on a trivial line bundle. In particular, there are canonical $`^\times `$-actions on $`𝐄`$ and $`\text{End}𝐄`$.
Proof. Compose one lifting with the inverse of the other. This gives a lifting of the trivial action on $`_n`$ to $`𝐄`$ which preserves $`𝚽`$. By (4), this acts on each fiber via scalar multiplications. . $`\mathrm{}`$
(4.4) There exists a universal family $`(𝐄,𝚽)`$ over $`_n\times C`$, and a lifting of the $`^\times `$-action on $`_n`$ to $`𝐄`$ whose induced action on $`𝚽`$ is $`\text{Ad}(\lambda )𝚽=\lambda ^1𝚽`$.
That is, $`_n`$ is a fine moduli space for the Higgs bundles of degree $`d`$ and rank $`r`$ with values in $`K(n)`$.
Proof. This follows in a standard way, cf. Newstead , from the geometric invariant theory construction of $`_n`$ due to Nitsure . Alternatively, the universal pair can be constructed gauge-theoretically just as in Atiyah-Bott \[2, §9\]. In the rank 2 case, both methods are explained in detail by the first author \[21, 5.3\] \[19, 5.2.3\]. . $`\mathrm{}`$
## 5 Equivalence of the two sets of universal classes
The main results were stated in §1 for the moduli space of flat connections $``$. But what is actually proved is similar results for the Higgs moduli spaces $`_n`$. To show that these imply the statements of §1 in the case $`n=0`$, it suffices to check that the relevant cohomology classes correspond under the diffeomorphism $`_0`$.
Let $`(𝐄,𝚽)`$ be a universal family on $`_n\times C`$. There is a morphism $`_n\text{Jac}^dC`$ given by $`(E,\varphi )\mathrm{\Lambda }^rE`$, so the generators of $`H^{}(\text{Jac}^dC)`$ pull back to classes $`\epsilon _1,\mathrm{},\epsilon _{2g}H^1(_n)`$. Also, let $`c_2,\mathrm{},c_r`$ be the characteristic classes of $`𝐄`$. These are elements of rational cohomology; they can be regarded as the Chern classes of the tensor product of $`𝐄`$ with a formal $`r`$th root of $`\mathrm{\Lambda }^r𝐄^{}`$.
Each of these classes has a Künneth decomposition
$$c_i=\alpha _i\sigma +\beta _i+\underset{j=1}{\overset{2g}{}}\psi _{i,j}e_j,$$
defining classes $`\alpha _iH^{2i2}()`$, $`\beta _iH^{2i}()`$, and $`\psi _{i,j}H^{2i1}()`$. The entire collection of classes $`\alpha _i`$, $`\beta _i`$, $`\psi _{i,j}`$, and $`\epsilon _j`$ will be referred to as the universal classes. What requires proof is then the following.
(5.1) When $`n=0`$, these classes correspond under the diffeomorphism $`_0`$ to their counterparts defined in §1.
Proof. First, there is a diagram
$$\begin{array}{ccc}& \stackrel{det}{}& 𝒯\\ & & \\ _0& \stackrel{\mathrm{\Lambda }^r}{}& \text{Jac}^dC\end{array}$$
relating the map $`det`$ of §1 to the map $`(E,\varphi )\mathrm{\Lambda }^rE`$ mentioned above. It is easy to see from (2) that this commutes, and hence that the classes $`\epsilon _j`$ correspond under the diffeomorphism.
To show that the higher-degree classes $`\alpha _i`$, $`\beta _i`$, and $`\psi _{i,j}`$ correspond, it suffices to show that the principal $`\mathrm{PGL}(r,)`$-bundle associated to $`𝐄`$ corresponds under the diffeomorphism $`_0`$ to the principal bundle
$$\frac{\overline{G}\times S\times \stackrel{~}{C}}{\pi _1(C)\times G}$$
of §1, where $`S=\mu ^1(e^{2\pi id/r}I)`$. (Recall that $`G=\mathrm{GL}(r,)`$, $`\overline{G}=\mathrm{PGL}(r,)`$, and $`\stackrel{~}{C}`$ is the universal cover of $`C`$.) In fact, we will construct a principal $`G`$-bundle $`R`$ over $`_0`$ and show that the principal $`\overline{G}`$-bundle $`U`$ associated to the pullback of $`𝐄`$ to $`R\times C`$ is $`G`$-equivariantly isomorphic to
$$V=\frac{\overline{G}\times S\times \stackrel{~}{C}}{\pi _1(C)}.$$
Let $`R`$ be the total space of the principal $`\overline{G}`$-bundle over $`_0`$ associated to $`𝐄|_{_0\times \{p\}}`$. Then $`R`$ parametrizes stable pairs $`(E,\varphi )`$ equipped with a frame for the fiber $`E_p`$, up to rescaling. On the other hand, $`S`$ parametrizes connections of constant central curvature, together with a frame for the fiber at a base point, up to rescaling. The diffeomorphism $`_0`$ therefore lifts to a $`G`$-equivariant diffeomorphism $`RS`$.
Let $`𝐅`$ be the pullback to $`R\times C`$ of $`𝐄`$. Then $`𝐅`$ admits a natural $`G`$-action lifting that on $`R`$, and it is canonically trivialized on $`R\times \{p\}`$. Moreover, by (2) $`𝐅`$ admits a Hermitian metric so that the restriction of the metric connection $`𝐀`$ to each slice $`\{(E,\varphi )\}\times C`$ satisfies the self-duality equation. Hence $`𝐀+𝚽+𝚽^{}`$ determines a $`G`$-connection on $`𝐅`$ whose restriction to each slice has constant central curvature. In particular, the associated $`\overline{G}`$-connection on the associated $`\overline{G}`$-bundle $`U`$ is flat on each slice.
On the other hand, the bundle $`V`$ over $`S\times C`$ defined above is trivial on $`S\times \{p\}`$, and carries a flat connection on each slice $`\{r\}\times C`$, which has the same holonomy as the one just mentioned and is preserved by the action of $`G`$. These flat connections can be used to extend the isomorphism $`U|_{R\times \{p\}}V|_{S\times \{p\}}`$ of trivial bundles with $`G`$-action to a $`G`$-equivariant isomorphism $`UV`$ lifting the $`G`$-equivariant diffeomorphism $`RS`$. . $`\mathrm{}`$
This marks the last appearance of flat connections in our story. From now on it is all about Higgs bundles.
## 6 Statement of the generation theorem
Let $`\epsilon _j`$, $`\alpha _i`$, $`\beta _i`$, $`\psi _{i,j}`$ be the universal classes, defined in §5, on the Higgs moduli space $`_n`$. The goal of the paper will be to prove this, its main result.
(6.1) The rational cohomology ring of $`_n`$ is generated by the universal classes.
The proof of this generation theorem has several parts, with quite different flavors.
First, we study the stratification of families of Higgs bundles according to their Harder-Narasimhan type. §7 is devoted to finite-dimensional families, and §8 to an infinite-dimensional family analogous to that of Atiyah-Bott . The aim is to show that the strata are smooth of the expected dimension, but this turns out to be true only in a stable sense. We need to consider not only a single $`_n`$, but the chain of inclusions $`_n_{n+1}`$.
We are therefore led to consider in §9 the direct limit $`_{\mathrm{}}`$ of the $`_n`$, and to show that its cohomology is generated by universal classes. Indeed, topological arguments show that it has the homotopy type of the classifying space of the gauge group, and the generation then follows from a theorem of Atiyah-Bott .
Having done this, we then show in §10 that, when $`r=2`$, the cohomology of $`_{\mathrm{}}`$ surjects on that of $`_n`$ for every $`n`$, and hence in particular on that of $``$ itself. This part of the proof is algebro-geometric in nature.
## 7 The finite-dimensional stratification
We wish to adopt the point of view taken by Atiyah-Bott , in which the objects of interest — for us, Higgs pairs — are parametrized by an infinite-dimensional, contractible space. The whole space will be divided into strata on which the level of instability is in some sense constant. So we will first study the analogue of this stratification in finite-dimensional algebraic families, then transfer it to our infinite-dimensional setting. Except for some subtleties surrounding smoothness, the results of §§7 and 8 are mostly analogous to those of Shatz and Atiyah-Bott ; readers familiar with those papers may be willing to skip directly to §9.
Let $`(E,\varphi )`$ be a Higgs pair with values in a line bundle $`L`$. A filtration by $`\varphi `$-invariant subbundles
$$0=E^0E^1\mathrm{}E^l=E$$
is said to be a Harder-Narasimhan filtration (hereinafter HN filtration) if the pairs $`(F^i,\varphi ^i)`$ are semistable with slope strictly decreasing in $`i`$, where $`F^i=E^i/E^{i1}`$ and $`\varphi ^i`$ is induced by $`\varphi `$.
(7.1) Any Higgs pair defined over any field of characteristic 0 possesses a unique HN filtration.
Proof. The analogous statement for bundles without a Higgs field is proved by Harder-Narasimhan and Shatz . The proof for Higgs pairs is entirely parallel: just substitute $`\varphi `$-invariant subbundles for ordinary subbundles everywhere in either proof. Shatz assumes that the ground field is algebraically closed, but his proof of this theorem does not require it. . $`\mathrm{}`$
For a given $`(E,\varphi )`$, the type $`\mu `$ is the $`l`$-tuple $`(r_1,d_1),\mathrm{},(r_l,d_l)`$ of ranks and degrees of the $`F^i`$ appearing in its HN filtration. For example, the type of a semistable pair is the 1-tuple $`(r,d)`$.
Since the slope $`d_i/r_i`$ is strictly decreasing, the pairs $`(0,0),(r_1,d_1),(r_1+r_2,d_1+d_2),\mathrm{},(r,d)`$ consisting of partial sums form the vertices of a convex polygon $`\text{Pol}(\mu )`$ in $`𝐑^2`$, as shown.
Let $`S`$ be a scheme of finite type over $``$, $`(𝐄,𝚽)S\times C`$ a family, parametrized by $`S`$, of Higgs pairs on $`C`$ with values in $`L`$.
(7.2) The set $`\{sS|(𝐄,𝚽)_s\text{ is semistable}\}`$ is open in $`S`$.
Proof. See Nitsure . . $`\mathrm{}`$
For any $`\mu `$, let $`S^\mu `$ be the set of those $`sS`$ such that $`(𝐄,𝚽)_s`$ has type $`\mu `$. Recall that a constructible subset is a finite union of locally closed sets in the Zariski topology.
(7.3) For any family of Higgs pairs over $`S`$ and any type $`\mu `$, $`S^\mu `$ is a constructible set. Moreover, $`S^\mu \mathrm{}`$ for only finitely many $`\mu `$, and each $`S^\mu `$ is covered by constructible sets where the HN filtration varies algebraically, that is, it determines a filtration of $`𝐄`$ by $`𝚽`$-invariant subbundles.
Proof. Without loss of generality assume $`S`$ is irreducible. Given a family $`(𝐄,𝚽)S\times C`$, let $`(𝐄,𝚽)_\xi `$ be the fiber over the generic point $`\xi S`$. This is a Higgs pair defined over the function field $`(S)`$. It therefore has a HN filtration, of some type $`\mu `$, by $`𝚽_\xi `$-invariant subbundles. By Lemma 5 of Shatz there exists an open $`US`$ such that this filtration extends to a filtration of $`𝐄|_{U\times C}`$ by subbundles $`𝐄^i`$. They are $`𝚽`$-invariant, since this is a closed condition and the closure of $`\xi `$ is all of $`S`$.
On the other hand, (7) implies that, since the quotient pairs $`(𝐅^i,𝚽^i)_\xi `$ are semistable, after restricting to a smaller $`U`$ if necessary, $`(𝐅^i,𝚽^i)_s`$ are also semistable for all $`sU`$. Hence this is the HN filtration at every $`sU`$, so $`US^\mu `$.
Now pass to $`S\backslash U`$ and proceed by induction on the dimension of the parameter space. . $`\mathrm{}`$
Following Shatz , define a partial ordering on the set of types by declaring $`\mu \nu `$ if $`\text{Pol}(\mu )\text{Pol}(\nu )`$. Then let $`S^\mu =_{\nu \mu }S^\nu `$.
(7.4) For $`S`$ and $`\mu `$ as above, $`S^\mu S`$ is closed.
The proof requires the following lemma.
(7.5) Let $`(E,\varphi )`$ be a Higgs pair of type $`\mu `$, and let $`FE`$ be a $`\varphi `$-invariant subbundle. Then $`(\text{rk}F,\mathrm{deg}F)\text{Pol}(\mu )`$.
Proof. The analogous statement without a Higgs field is Theorem 2 of Shatz , and the proof of this is entirely parallel. One simply has to note that since the filtration and the subbundle $`F`$ are $`\varphi `$-invariant, so are the subsheaves $`E^iF`$ and $`E^iF`$ considered by Shatz. . $`\mathrm{}`$
Proof of (7). If $`FE`$ is any inclusion of torsion-free sheaves, define $`\widehat{F}E`$ to be the inverse image under the projection $`EE/F`$ of the torsion subsheaf. Then $`\widehat{F}`$ and $`E/\widehat{F}`$ are torsion-free, $`F=\widehat{F}`$ on the locus where $`E/F`$ is torsion-free, and $`F\widehat{F}`$ preserves inclusions of subsheaves of $`E`$.
By (7) $`S^\mu `$ can be regarded as a reduced subscheme of $`S`$. To show that it is closed, by the valuative criterion \[18, II 4.7\] it suffices to show that, if $`X`$ is any smooth curve, and $`f:XS`$ any morphism taking a nonempty open set into $`S^\mu `$, then $`f(X)S^\mu `$.
The generic point of $`X`$ maps to one of the constructible sets named in (7), where the HN filtrations are parametrized by subbundles. Hence there is a non-empty open set $`VX`$ such that the restriction of $`(f\times 1)^{}𝐄`$ to $`V\times C`$ is filtered by subbundles restricting over every point in $`V`$ to the HN filtration.
Like any coherent subsheaf defined on an open set \[18, II Ex. 5.15(d)\], these bundles extend to coherent subsheaves $`𝐄^i`$ of $`(f\times 1)^{}𝐄`$ over all of $`X\times C`$. These can be chosen to remain nested, and as subsheaves of $`𝐄`$ they are of course torsion-free. Furthermore, they can be chosen so that $`𝐄/𝐄^i`$ are torsion-free also, by replacing $`𝐄^i`$ with $`\widehat{𝐄}^i`$.
Since torsion-free sheaves on a smooth surface such as $`X\times C`$ are locally free except on a set of codimension 2 \[13, Cor. 2.38\], it follows that the $`𝐄^i`$ are subbundles except at finitely many points in the fibers over $`X\backslash V`$.
Now on a smooth curve such as one of these fibers, torsion-free sheaves are locally free, and so the procedure of the first paragraph implies the following: every subsheaf of a locally free sheaf is contained in a subbundle having the same rank and no less degree, with equality if and only if it was a subbundle to begin with.
When restricted to $`\{x\}\times C`$ for any $`xX\backslash V`$, then, the nested subsheaves $`𝐄_x^i`$ determine a sequence of subbundles, which only differ from $`𝐄_x^i`$ at finitely many points and hence remain nested and $`𝚽_x`$-invariant, and have the same rank. Since the degrees may have risen, they determine a polygon which contains $`\text{Pol}(\mu )`$. By (7), if the type of $`(𝐄,𝚽)_x`$ is $`\nu _x`$, then $`\text{Pol}(\nu _x)`$ contains this polygon. Hence $`\nu _x\mu `$, so $`f(x)S^\mu `$. . $`\mathrm{}`$
As in Atiyah-Bott, the last statement of (7) can be refined.
(7.6) The HN filtration varies algebraically on all of $`S^\mu `$; that is, there exists a filtration of $`𝐄|_{S^\mu \times C}`$ by subbundles restricting to the HN filtration on each fiber.
The proof again requires a lemma.
(7.7) If $`E^1`$ is the first term in the HN filtration of $`(E,\varphi )`$ and $`FE`$ is another $`\varphi `$-invariant subbundle of the same rank and degree, then $`F=E^1`$.
Proof. The corresponding statement for ordinary bundles is a special case of Lemma 3 of Shatz . The proof of this is again entirely parallel. . $`\mathrm{}`$
Proof of (7). By (7), the HN filtrations determine a constructible subset of the product of Grassmannian bundles $`\times _i\text{Grass}_{r_i}𝐄|_{S^\mu \times C}`$. It must be shown that it is closed. By the valuative criterion, it suffices to show that for any morphism $`f:XS^\mu `$, where $`X`$ is a smooth curve, the HN filtrations determine a filtration of $`(f\times 1)^{}𝐄`$ by subbundles.
As in the proof of (7), a filtration by subbundles does exist over an open $`VX`$, and the subbundles extend to nested, $`𝚽`$-invariant torsion-free sheaves $`𝐄^i`$ over $`X\times C`$.
The restrictions of these to the fibers over $`xX\backslash V`$ generate nested, $`𝚽`$-invariant subbundles whose ranks and degrees span a polygon containing $`\text{Pol}(\mu )`$. But now since $`(𝐄,𝚽)_x`$ also has type $`\mu `$, by (7) this polygon is contained in $`\text{Pol}(\mu )`$ as well. Hence it equals $`\text{Pol}(\mu )`$, so the subbundles have degrees equal to those of the subsheaves which generated them. They therefore coincide with these subsheaves.
Consequently, the sections of the subsheaf $`𝐄^i`$ span an $`r_i`$-dimensional subspace of the fiber of $`(f\times 1)^{}𝐄`$ over every point in $`X\times C`$. It follows that $`𝐄^i`$ is a subbundle.
Finally, we claim that for any $`xX\backslash V`$, the HN filtration of $`𝐄_x`$ is the restriction of the $`𝐄^i`$. First, $`𝐄_x^1`$ is a $`𝚽_x`$-invariant subbundle of rank and degree equal to that of the first term in the HN filtration. By (7) these two subbundles must coincide. Now pass to $`𝐄/𝐄^1`$ and use induction on the length of the HN filtration to do the rest. . $`\mathrm{}`$
Recall that $`𝐄𝐧𝐝^{}`$ refers to the two-term complex, defined in §3, involving endomorphisms fixing a flag of subbundles. In what follows, this flag will always be the HN filtration.
(7.8) There are deformation maps $`T_sS𝐇^1𝐄𝐧𝐝(𝐄,𝚽)_s`$ and $`T_sS^\mu 𝐇^1𝐄𝐧𝐝^{}(𝐄,𝚽)_s`$ so that the following diagram commutes:
$$\begin{array}{ccc}T_sS^\mu & & 𝐇^1𝐄𝐧𝐝^{}(𝐄,𝚽)_s\\ & & \\ T_sS& & 𝐇^1𝐄𝐧𝐝(𝐄,𝚽)_s.\end{array}$$
Proof. Follows immediately from (3), (3) and (7). . $`\mathrm{}`$
(7.9) Let $`(E,\varphi )`$ be a Higgs pair with values in $`L`$. For $`m`$ large enough, $`(E,\varphi (m))`$ belongs to a family, parametrized by a smooth base $`X`$, of Higgs pairs with values in $`L(m)`$ such that the deformation map $`T_{(E,\varphi (m))}X𝐇^1𝐄𝐧𝐝(E,\varphi (m))`$ is an isomorphism.
Proof. It suffices to find a smooth $`X`$ so that the deformation map is surjective, since then one may restrict to a smooth subvariety transverse to the kernel.
Choose $`k`$ large enough that $`E(k)`$ is generated by its sections and has $`H^1=0`$. There is then a surjection $`𝒪^\chi E(k)`$, where $`\chi `$ is the Euler characteristic of $`E(k)`$. This represents a point $`q`$ in the Quot scheme parametrizing quotients of $`𝒪^\chi `$ with fixed rank and degree. Let $`F`$ be the kernel of the map $`𝒪^\chi E(k)`$. The tangent space to the Quot scheme at $`q`$ is then $`H^0\text{Hom}(F,E(k))`$, and the natural map to the deformation space of $`E(k)`$ is the connecting homomorphism in the long exact sequence of
$$0\text{End}E\text{Hom}\text{(}𝒪^\chi ,E(k)\text{)}\text{Hom}\text{(}F,E(k)\text{)}0.$$
This is surjective since $`H^1\text{Hom}\text{(}𝒪^\chi ,E(k)\text{)}=^\chi H^1(E(k))=0`$. For the same reason, $`H^1\text{Hom}\text{(}F,E(k)\text{)}=0`$, so the Quot scheme is smooth at $`q`$.
Now choose $`m`$ large enough that $`H^1\text{(}\text{End}EL(m)\text{)}=0`$, and let $`X`$ be the total space of $`\pi _{}\text{(}\text{End}𝐄L(m)\text{)}`$, where $`𝐄`$ is the tautological quotient on $`\text{Quot}\times C`$, and $`\pi `$ is projection to $`\text{Quot}`$. This is smooth near $`q`$ since the push-forward is locally free there. Moreover, there is a tautological section $`𝚽H^0\text{(}X\times C,\text{End}𝐄L(m)\text{)}`$. That the deformation map is surjective is easily seen from the diagram
$$\begin{array}{ccccccc}H^0\text{(}\text{End}EL(m)\text{)}& & T_{(E,\varphi (m))}X& & H^0\text{Hom}\text{(}F,E(k)\text{)}& & 0\\ =& & & & & & \\ H^0\text{(}\text{End}EL(m)\text{)}& & 𝐇^1𝐄𝐧𝐝(E,\varphi (m))& & H^1\text{End}E& & 0.\end{array}$$
. $`\mathrm{}`$
Now fix a type $`\mu `$, and let $`(E,\varphi )`$ be a pair of this type, with values in $`L`$.
(7.10) For $`m`$ large enough, $`(E,\varphi (m))`$ belongs to a family, parametrized by a smooth base $`Y`$, of Higgs pairs of type $`\mu `$ with values in $`L(m)`$ such that the deformation map $`T_{(E,\varphi (m))}Y𝐇^1𝐄𝐧𝐝^{}(E,\varphi (m))`$ is an isomorphism.
Proof. The proof is parallel to that of the previous theorem. Let $`0=E^0\mathrm{}E^l=E`$ be the HN filtration of $`(E,\varphi )`$, as usual, and let $`E_i=E/E^i`$. Choose $`k`$ large enough that every $`E_i(k)`$ is generated by its sections and every $`H^1(E^i(k))=0`$. Then $`H^1(E_i(k))=0`$ as well, and $`H^0(E(k))H^0(E_i(k))`$ is surjective. If $`\chi =dimH^0(E(k))`$, a choice of basis for $`H^0(E(k))`$ then determines a sequence of quotients
$$𝒪^\chi E(k)\stackrel{\psi _1}{}E_1(k)\stackrel{\psi _2}{}E_2(k)\stackrel{\psi _3}{}\mathrm{}\stackrel{\psi _l}{}E_l(k)$$
determining a point $`q`$ in the product of $`l`$ different Quot schemes. Let $`R`$ be the subspace of this product parametrizing flags of bundles; then $`T_qR`$ is $`𝐇^0(C^{})`$, where $`C^{}`$ is the third row in an exact sequence of two-term complexes
$$\begin{array}{ccc}0& & 0\\ & & \\ \underset{i}{}\text{Hom}(E_i,E_i)& & \underset{i}{}\text{Hom}(E_i,E_{i+1})\\ & & \\ \underset{i}{}\text{Hom}(𝒪^\chi ,E_i)& & \underset{i}{}\text{Hom}(𝒪^\chi ,E_{i+1})\\ & & \\ \underset{i}{}\text{Hom}(F_i,E_i)& & \underset{i}{}\text{Hom}(F_i,E_{i+1}),\\ & & \\ 0& & 0\end{array}$$
and $`F_i`$ is the kernel of the map $`𝒪^\chi E_i`$. Now the second row $`B^{}`$ is isomorphic to $`\chi `$ copies of $`_iE_i_iE_{i+1}`$, with the map given by $`(b_i)(\psi _{i+1}b_ib_{i+1})`$. Using the long exact sequence
$$\begin{array}{c}0𝐇^0(B^{})\underset{i}{}H^0(E_i)\underset{i}{}H^0(E_{i+1})\hfill \\ 𝐇^1(B^{})\underset{i}{}H^1(E_i)\underset{i}{}H^1(E_{i+1})\hfill \\ 𝐇^2(B^{})0\hfill \end{array}$$
together with $`H^1(E_i)=0`$ and the surjectivity of $`H^0(E_i)H^0(E_{i+1})`$, a descending induction on $`i`$ shows that $`𝐇^1(B^{})=𝐇^2(B^{})=0`$. Finally, if $`A^{}`$ is the first row, from the short exact sequence
$$0\text{End}^{}E\underset{i}{}\text{Hom}(E_i,E_i)\underset{i}{}\text{Hom}(E_i,E_{i+1})0$$
we conclude that $`𝐇^{}(A^{})=H^{}(\text{End}^{}E)`$ and hence that $`𝐇^0(C^{})=T_qR`$ surjects on $`𝐇^1(A^{})`$ $`=`$ $`H^1(\text{End}^{}E)`$. Moreover, $`𝐇^2(A^{})=H^2(\text{End}^{}E)=0`$. Hence $`𝐇^1(C^{})=𝐇^2(C^{})=0`$, so $`R`$ is smooth at $`q`$.
Now choose $`m`$ large enough that $`H^1\text{(}\text{End}^{}EL(m)\text{)}=0`$, and let $`Y`$ be the total space of $`\pi _{}\text{(}\text{End}^{}𝐄L(m)\text{)}`$, where $`𝐄`$ is the obvious tautological quotient on $`R`$, $`\text{End}^{}𝐄\text{End}𝐄`$ is the subbundle of endomorphisms preserving the flag, and $`\pi :R\times CR`$ is the projection. Then $`Y`$ is smooth in a neighborhood over $`q`$, contains a point representing $`(E,\varphi (m))`$, and has a tautological Higgs field $`𝚽H^0\text{(}Y\times C,\text{End}^{}𝐄L(m)\text{)}`$. The deformation map is surjective, as may be seen from the diagram
$$\begin{array}{ccccccc}H^0\text{(}\text{End}^{}EL(m)\text{)}& & T_{(E,\varphi (m))}Y& & T_qR& & 0\\ =& & & & & & \\ H^0\text{(}\text{End}^{}EL(m)\text{)}& & 𝐇^1𝐄𝐧𝐝^{}(E,\varphi (m))& & H^1\text{End}^{}E& & 0.\end{array}$$
. $`\mathrm{}`$
## 8 The infinite-dimensional stratification
Let $``$ be a Hermitian vector bundle over $`C`$ of rank $`r`$ and degree $`d`$. A rigorous construction of $`_n`$ as an infinite-dimensional quotient involves connections and sections in Sobolev spaces associated to $``$. So choose any $`k2`$; Atiyah and Bott prefer $`k=2`$, but for us as for them, any greater $`k`$ will also do. Then, for any Hermitian bundle $`𝒱`$ over $`C`$, denote by $`\mathrm{\Omega }^{p,q}(𝒱)`$ the Banach space consisting of sections, of Sobolev class $`L_{kpq}^2`$, of the bundle of differential forms of type $`p,q`$ with values in $`𝒱`$. Also let $`𝒜`$ be the space of holomorphic structures on $``$ differing from a fixed $`C^{\mathrm{}}`$ one by an element of the Sobolev space $`\mathrm{\Omega }^{0,1}(\text{End})`$. We hope the reader will pardon the unorthodox use of $`\mathrm{\Omega }`$ to refer to a Sobolev completion, rather than just the space of smooth forms.
Define a map
$$\overline{}:𝒜\times \mathrm{\Omega }^{1,0}(\text{End}K(n))\mathrm{\Omega }^{1,1}(\text{End}K(n))$$
by $`\overline{}(E,\varphi )=\overline{}_E\varphi `$, and let $`_n=\overline{}^1(0)`$. This parametrizes all pairs where $`\varphi `$ is holomorphic.
Let $`𝒢`$ be the complex gauge group consisting of all complex automorphisms of $``$ of Sobolev class $`L_k^2`$. Then $`𝒢`$ acts naturally and smoothly on $`𝒜`$ as shown by Atiyah-Bott, and likewise on $`\mathrm{\Omega }^{1,0}(\text{End}K(n))`$ since $`L_{k1}^2`$ is a topological $`L_k^2`$-module. The $`𝒢`$-action on the product of these spaces preserves $`_n`$.
(8.1) Every $`𝒢`$-orbit in $`_n`$ has a $`C^{\mathrm{}}`$ representative $`(E,\varphi )`$, and any two are interchanged by a $`C^{\mathrm{}}`$ gauge transformation. The stabilizer of $`(E,\varphi )`$ is the group of holomorphic automorphisms of $`E`$ preserving $`\varphi `$.
Proof. According to Lemma 14.8 of Atiyah-Bott, every $`𝒢`$-orbit in $`𝒜`$ contains a $`C^{\mathrm{}}`$ representative. If $`(E,\varphi )_n`$, so that $`\varphi `$ satisfies the elliptic equation $`\overline{}_E\varphi =0`$, it follows from elliptic regularity that $`\varphi `$ is also $`C^{\mathrm{}}`$. If $`(E,\varphi )`$ and $`(E^{},\varphi ^{})`$ are pairs in the same $`𝒢`$-orbit, then by Lemma 14.9 of Atiyah-Bott, any gauge transformation interchanging $`E`$ and $`E^{}`$ is $`C^{\mathrm{}}`$; hence the same is true for the pairs. Finally, if an element of $`𝒢`$ preserves $`(E,\varphi )`$, this means precisely that it preserves $`\overline{}_E`$, hence is a holomorphic automorphism, and fixes $`\varphi `$. . $`\mathrm{}`$
(8.2) Let $`(E,\varphi )_n`$ be a $`C^{\mathrm{}}`$ pair. Then the normal space to the $`𝒢`$-orbit at $`(E,\varphi )`$ is canonically isomorphic to $`𝐇^1𝐄𝐧𝐝(E,\varphi )`$, and the cokernel of the derivative of $`\overline{}`$ at $`(E,\varphi )`$ is canonically isomorphic to $`𝐇^2𝐄𝐧𝐝(E,\varphi )`$.
Proof. The infinitesimal action of the Lie algebra of $`𝒢`$ is the map $`f`$, and the derivative of $`\overline{}`$ is the map $`g`$, in the complex
$$\mathrm{\Omega }^{0,0}(\text{End})\stackrel{f}{}\mathrm{\Omega }^{0,1}(\text{End})\mathrm{\Omega }^{0,0}(\text{End}K(n))\stackrel{g}{}\mathrm{\Omega }^{0,1}(\text{End}K(n)),$$
where $`f(a)=(\overline{}_E(a),[a,\varphi ])`$ and $`g(b,c)=\overline{}_E(c)+[b,\varphi ]`$. The symbol sequence of this complex is the direct sum of those of the Dolbeault complexes of $`\text{End}E`$ and $`\text{End}EK(n)`$, so it is elliptic. By elliptic regularity the cohomology of the complex is then the same as its counterpart where the Sobolev spaces are replaced by spaces of smooth forms; this is precisely the Dolbeault hypercohomology of $`𝐄𝐧𝐝(E,\varphi )`$. . $`\mathrm{}`$
(8.3) For each $`(E,\varphi )_n`$, $`(E,\varphi (m))`$ is a smooth point of $`_{m+n}`$ for sufficiently large $`m`$.
Proof. Choose $`m`$ large enough that $`H^1(\text{End}EK(m+n))=0`$. Then from the long exact sequence
$$\mathrm{}H^1(\text{End}E)H^1(\text{End}EK(m+n))𝐇^2𝐄𝐧𝐝(E,\varphi (m))0$$
associated to the hypercohomology of a two-term complex, $`𝐇^2𝐄𝐧𝐝(E,\varphi (m))=0`$. Hence $`\overline{}`$ is a submersion at $`(E,\varphi (m))`$, and the implicit function theorem for Banach manifolds \[9, A3\] implies that $`_{m+n}=\overline{}^1(0)`$ is a smooth embedded Banach submanifold in a neighborhood of $`(E,\varphi (m))`$. . $`\mathrm{}`$
Let us now find a slice for the $`𝒢`$-action, using the results of the previous section.
(8.4) Let $`(E,\varphi )`$ be a $`C^{\mathrm{}}`$ pair in $`_n`$. Then for $`m`$ large enough, there is a $`𝒢`$-equivariant submersion $`𝒢\times U_{m+n}`$ onto a neighborhood of $`(E,\varphi (m))`$, where $`U`$ is an open neighborhood of $`x`$ in the algebraic family of (7).
Proof. (7) provides a family $`(𝐄,𝚽)`$ of pairs over some smooth $`Xx`$ such that $`(𝐄,𝚽)_x=(E,\varphi (m))`$ and the natural map $`T_xX𝐇^1𝐄𝐧𝐝(E,\varphi (m))`$ is an isomorphism. Choose a Hermitian metric on $`𝐄`$ extending the given one on $`E`$. This determines a smooth map $`X_{m+n}`$, which by (8) is transverse to the $`𝒢`$-orbit. It extends to a $`𝒢`$-equivariant map $`𝒢\times X_{m+n}`$ whose derivative is a surjection at $`𝒢\times \{x\}`$, and hence in some neighborhood $`𝒢\times U`$, thanks to the diagram
$$\begin{array}{ccccccc}T_e𝒢& & T_{(e,x)}(𝒢\times X)& & T_xX& & 0\\ =& & & & & & \\ \mathrm{\Omega }^{0,0}(\text{End}E)& & T_n& & 𝐇^1𝐄𝐧𝐝(E,\varphi )& & 0.\end{array}$$
. $`\mathrm{}`$
Let $`_n^\mu `$ denote the union of all $`𝒢`$-orbits in $`_n`$ whose $`C^{\mathrm{}}`$ representatives have type $`\mu `$ in the sense of §7. In particular, let $`_n^s`$ denote the stable orbits.
(8.5) For any $`\mu `$, $`_{\nu \mu }_n^\nu `$ is closed in $`_n`$.
Proof. Follows immediately from (8) and the corresponding fact for $`X`$, (7). . $`\mathrm{}`$
Let $`\overline{𝒢}`$ be the quotient of $`𝒢`$ by the central subgroup $`^\times `$. Since by (4) the latter is the stabilizer of all stable pairs, each stable orbit is isomorphic to $`\overline{𝒢}`$.
(8.6) The natural map $`_n^s_n`$ is a principal $`\overline{𝒢}`$-bundle.
Proof. The submersions of (8) descend to maps $`\overline{𝒢}\times U_n`$, whose derivatives are isomorphisms on $`\overline{𝒢}\times \{x\}`$, and which can be made injective by shrinking $`U`$ if necessary. By the inverse function theorem for Banach manifolds \[9, A1\] these are $`\overline{𝒢}`$-equivariant diffeomorphisms onto their images. They therefore constitute an atlas of local trivializations. . $`\mathrm{}`$
Given a $`C^{\mathrm{}}`$ pair $`(E,\varphi )_n`$, define $`\text{End}^{\prime \prime }E`$ by the short exact sequence
$$0\text{End}^{}E\text{End}E\text{End}^{\prime \prime }E0,$$
where $`\text{End}^{}E`$, as before, is the subsheaf of $`\text{End}E`$ preserving the HN filtration of $`(E,\varphi )`$. Also let $`𝐄𝐧𝐝^{\prime \prime }(E,\varphi )`$ be the two-term complex on $`C`$ defined analogously to $`𝐄𝐧𝐝(E,\varphi )`$ and $`𝐄𝐧𝐝^{}(E,\varphi )`$. There is then a short exact sequence of two-term complexes
$$\mathrm{𝟎}𝐄𝐧𝐝^{}(E,\varphi )𝐄𝐧𝐝(E,\varphi )𝐄𝐧𝐝^{\prime \prime }(E,\varphi )\mathrm{𝟎}.$$
(8.7) For any $`(E,\varphi )_n`$ of type $`\mu `$, and for $`m`$ large enough, $`_{m+n}^\mu `$ is an embedded submanifold of $`_{m+n}`$ near $`(E,\varphi (m))`$ with normal space canonically isomorphic to $`𝐇^1𝐄𝐧𝐝^{\prime \prime }(E,\varphi (m))`$.
Proof. By acting with an element of $`𝒢`$ if necessary we may assume that $`(E,\varphi )`$ is $`C^{\mathrm{}}`$.
For $`m`$ large, there was constructed in (7) a family of pairs $`(𝐄,𝚽)`$ of type $`\mu `$ over a smooth base $`Yy`$, having $`(𝐄,𝚽)_y=(E,\varphi (m))`$ and $`T_yY𝐇^1𝐄𝐧𝐝^{}(E,\varphi (m))`$ an isomorphism. Choose a metric on $`𝐄`$ extending the given one on $`E`$. This determines a smooth map $`Y_{m+n}`$, which by (8) is transverse to the $`𝒢`$-orbit.
On an open neighborhood $`V`$ of $`y`$ in $`Y`$, choose a lifting of this map to the domain of the submersion $`𝒢\times U_{m+n}`$ of (8). Projecting this lifting to $`U`$ gives a map $`VU`$ whose image consists of pairs of type $`\mu `$, and hence is contained in $`X^\mu `$. Its derivative at $`y`$ is the natural map from $`T_yY=𝐇^1𝐄𝐧𝐝^{}(E,\varphi (m))`$ to $`T_xX=𝐇^1𝐄𝐧𝐝(E,\varphi (m))`$.
This derivative is injective. Indeed, the kernel is the image of $`𝐇^0𝐄𝐧𝐝^{\prime \prime }(E,\varphi (m))`$. But if $`0=E^0\mathrm{}E^l=E`$ is the HN filtration of $`(E,\varphi )`$ as usual, then there is a short exact sequence
$$0\text{End}^{\prime \prime }E^{l1}\text{End}^{\prime \prime }E\text{Hom}(E^{l1},E/E^{l1})0$$
and hence a short exact sequence of the corresponding two-term complexes. The first hypercohomology $`𝐇^0`$ of both of the outer complexes vanishes by an induction on $`l`$ using (4), so $`𝐇^0𝐄𝐧𝐝^{\prime \prime }(E,\varphi (m))=0`$ too.
Hence, in a neighborhood of $`x`$, $`X^\mu `$ contains an embedded submanifold with tangent space $`𝐇^1𝐄𝐧𝐝^{}(E,\varphi (m))𝐇^1𝐄𝐧𝐝(E,\varphi (m))`$. But by (7), $`X^\mu `$ is a subscheme of $`X`$ with Zariski tangent space contained in $`𝐇^1𝐄𝐧𝐝^{}(E,\varphi (m))`$. It therefore must be smooth near $`x`$.
The inverse image of $`_{m+n}^\mu `$ under the submersion $`𝒢\times U_{m+n}`$ is $`𝒢\times (UX^\mu )`$, so this immediately implies that $`_{m+n}^\mu `$ is a smoothly embedded submanifold in a neighborhood of the orbit of $`(E,\varphi (m))`$. Its normal space at $`(E,\varphi (m))`$ is the quotient of $`𝐇^1𝐄𝐧𝐝(E,\varphi (m))`$ by $`𝐇^1𝐄𝐧𝐝^{}(E,\varphi (m))`$; by choosing $`m`$ large enough we may arrange as in (8) that $`𝐇^2𝐄𝐧𝐝^{}(E,\varphi (m))=0`$, so that this quotient is nothing but $`𝐇^1𝐄𝐧𝐝^{\prime \prime }(E,\varphi (m))`$. . $`\mathrm{}`$
## 9 The direct limit of Higgs spaces
The inclusions $`_n_{n+1}`$ make the set of all $`_n`$ into a directed set. Let $`_{\mathrm{}}`$ be the direct limit. It may be regarded as a set of pairs $`(E,\varphi )`$ as before, but where $`\varphi `$ may now have a pole of arbitrary finite order at $`p`$. Note that for each type $`\mu `$, the direct limit of $`_n^\mu `$ is a subset $`_{\mathrm{}}^\mu `$ of $`_{\mathrm{}}`$. Note also that $`𝒢`$ acts naturally on $`_{\mathrm{}}`$ and that $`_{\mathrm{}}^s/𝒢`$ is $`_{\mathrm{}}`$, the direct limit of the $`_n`$. In another context, $`_n`$ has appeared in the work of Donagi-Markman .
Our aim in this section is to show that $`_{\mathrm{}}^s`$ is contractible. Essentially, the reason is that $`_{\mathrm{}}`$ is contractible, and the complement of the stable set has infinite codimension.
Recall that a subspace of a topological space is a deformation neighborhood retract (hereinafter DNR) if it is the image of a map defined on some open neighborhood of itself and homotopic to the identity. It is equivariant if the homotopy is equivariant for the action of some group.
(9.1) As a subspace of $`_{\mathrm{}}`$, each $`_n`$ is a DNR, and the open sets which retract can be chosen to be nested.
Proof. Since $`_n_{n+1}`$ is an embedding of finite-dimensional manifolds, we may choose a tubular neighborhood $`U_n^1`$ and a projection $`U_n^1_n`$. This tubular neighborhood in turn has a tubular neighborhood $`U_n^2`$ in $`_{n+2}`$, namely its inverse image in $`U_{n+1}^1`$, and so on. The direct limits $`U_n^{\mathrm{}}`$ of these tubular neighborhoods are nested open subsets of $`_{\mathrm{}}`$. Each is homeomorphic to a vector bundle over $`_n`$ and hence deformation retracts onto it. Indeed, the deformation retraction preserves each $`U_n^i`$. . $`\mathrm{}`$
(9.2) As a subspace of $`_{\mathrm{}}`$, each $`_n^s`$ is a $`𝒢`$-equivariant DNR, and the $`𝒢`$-invariant open sets which retract can be chosen to be nested.
Proof. Let $`\pi :_{\mathrm{}}^s_{\mathrm{}}`$ be the quotient map, and let $`U_n^i`$ be the tubular neighborhoods of the previous proof. By the definition of direct limit, it suffices to construct a sequence of $`𝒢`$-equivariant deformation retractions of $`\pi ^1(U_n^i)`$ onto $`_n^s=\pi ^1(_n)`$ compatible with the inclusions $`\pi ^1(U_n^i)\pi ^1(U_n^{i+1})`$. Obviously we would like to lift the deformation retracts of the previous proof to the principal $`𝒢`$-bundle.
These liftings are guaranteed to exist by the first covering homotopy theorem. This asserts that if $`F:Y\times [0,1]Z`$ is a homotopy with any reasonable domain (including manifolds, but unfortunately not their direct limits), and if $`E`$ is a fiber bundle over $`Z`$, then $`F^{}E`$ is isomorphic as a fiber bundle to $`F^{}E|_{Y\times 0}\times [0,1]`$. Actually, a slight refinement of this result is needed, namely that the isomorphism can be chosen so as to extend a given one over a closed DNR $`XY`$. We then apply this refined result to the case where $`X=U_n^i`$, $`Y=U_n^{i+1}`$, and the homotopy is the retraction of $`U_n^i`$ on $`_n`$ described above. It is easy to construct the desired deformation retraction of $`\pi ^1(U_n^i)`$ from the resulting isomorphism.
The slight refinement of the first covering homotopy theorem can be proved by the same argument as the theorem itself, given by Steenrod \[46, §11.3\]. Just choose the atlas for $`F^{}E`$ so that its restriction to $`X\times [0,1]`$ is pulled back from an atlas on $`X`$ using the given isomorphism. The existence of such an atlas follows easily from the fact that $`X`$ is a closed DNR and the ordinary version of the theorem. . $`\mathrm{}`$
(9.3) The quotient map $`\pi :_{\mathrm{}}^s_{\mathrm{}}`$ is a principal $`\overline{𝒢}`$-bundle.
Proof. Immediate from (8) and (9). . $`\mathrm{}`$
(9.4) For all $`k0`$, $`\pi _k(_{\mathrm{}}^s)=1`$.
Proof. The open subsets of $`_{\mathrm{}}`$ provided by (9), which retract onto $`_n^s`$, form a nested open cover of $`_{\mathrm{}}^s`$. By compactness, any map $`S^k_{\mathrm{}}^s`$, and any homotopy of such maps, has image contained in one such neighborhood. Hence $`\pi _k(_{\mathrm{}}^s)=lim_n\mathrm{}\pi _k(_n^s)`$.
So let $`f:S^k_n^s`$ be any map. Now $`_n`$ is contractible just by retracting it first on $`𝒜\times \{0\}`$, so $`f`$ certainly extends to a continuous map $`f:D^{k+1}_n`$. Our task is to perturb this so that it misses the unstable locus.
For each $`xD^{k+1}`$, by (8) and (8) there is some integer $`m_xn`$ such that for all $`mm_x`$, $`f(x)`$ is a smooth point of $`_m`$, and the stratum $`_m^\mu `$ containing $`f(x)`$ is an embedded submanifold at $`f(x)`$. Passing to a finite subcover and taking $`m=\mathrm{max}m_x`$, we find that $`f(D^{k+1})`$ maps entirely into the smooth locus of $`_m`$, and that near its image each stratum $`_m^\mu `$ is an embedded submanifold of finite codimension.
Yet another application of compactness shows that $`f(D^{k+1})`$ intersects only a finite number of strata $`_m^\mu `$. By increasing $`m`$ again if necessary we may assume by (8) that each of these strata has codimension $`>k+1`$. Then, starting with the stratum of highest codimension and working our way up, we may perturb $`f`$ so that it no longer touches that stratum, but so that its value on $`S^k`$ remains unchanged. After these perturbations, $`f`$ will have image entirely within $`_m^s`$.
Thus for any $`f:S^k_n^s`$, the homotopy class of $`f`$ is killed by the inclusion in $`_m^s`$ for some $`mn`$, so $`lim_n\mathrm{}\pi _k(_n^s)=1`$. . $`\mathrm{}`$
For an alternate proof in the rank 2 case, see the first author’s thesis \[19, 7.5.1\].
(9.5) The space $`_{\mathrm{}}^s`$ is contractible.
Proof. A theorem of Whitehead \[47, 10.28\] asserts that if $`X`$ is a CW-space — that is, a space homotopy equivalent to a CW-complex — whose homotopy groups all vanish, then it is contractible. In light of (9) above, it therefore suffices to show that $`_{\mathrm{}}^s`$ is a CW-space.
Consider the fiber bundle
$$_{\mathrm{}}^s\frac{_{\mathrm{}}^s\times E𝒢}{𝒢}B𝒢.$$
(9.6)
We will show that this is a fibration whose total space and base space are CW-spaces. It then follows from Corollary 13 of Stasheff that the fiber is a CW-space.
Note that $`𝒢`$ acts on $`_{\mathrm{}}^s`$ with stabilizer $`^\times `$ and quotient $`_{\mathrm{}}`$. There is therefore a fiber bundle
$$B^\times \frac{_{\mathrm{}}^s\times E𝒢}{𝒢}_{\mathrm{}},$$
which is just the associated bundle to the principal $`\overline{𝒢}`$-bundle $`_{\mathrm{}}^s_{\mathrm{}}`$. The base of this fiber bundle, being a direct limit of manifolds, is metrizable and hence paracompact by Stone’s theorem \[35, 6-4.3\]; hence the fiber bundle is a fibration \[51, I 7.13\]. Moreover, the base is a CW-space, as is the fiber. Proposition 0 of Stasheff then implies that the total space is a CW-space.
According to Proposition 2.4 of Atiyah-Bott , $`B𝒢`$ is a component of the space of maps from $`C`$ to $`BG`$, with the compact-open topology. Since the domain is a compact metric space and the range is a CW-complex, by Corollary 2 of Milnor the space of maps is a CW-space. Moreover, since $`BG`$ is metrizable and $`C`$ is compact, the space of maps $`B𝒢`$ is metrizable, hence paracompact. The fiber bundle (9.6) is therefore a fibration. . $`\mathrm{}`$
(9.7) The space $`_{\mathrm{}}`$ is homotopy equivalent to $`B\overline{𝒢}`$.
Proof. By (9) and (9), there is a principal $`\overline{𝒢}`$-bundle on $`_{\mathrm{}}`$ with contractible total space. . $`\mathrm{}`$
Those who dislike the appearance of infinite-dimensional, gauge-theoretic methods in the last two sections may wish to reflect that it is no doubt possible to replace every infinite-dimensional construction by a finite-dimensional, algebraic approximation, in the style of Bifet et al. or Kirwan . In any case, algebraic geometry will reappear on the scene shortly.
## 10 Surjectivity of the restriction on cohomology
Let $``$ be as in §8. The pull-back of $`C`$ to the product $`_{\mathrm{}}^s\times C`$ is acted on by $`𝒢`$, so the projective bundle $``$ is acted on by $`\overline{𝒢}`$. It therefore descends to a $`^r`$-bundle $`𝐄`$ over $`_{\mathrm{}}\times C`$, whose characteristic classes can be decomposed as usual into Künneth components:
$$\overline{c}_i=\alpha _i\sigma +\underset{j=1}{\overset{2g}{}}\psi _{i,j}e_j+\beta _i.$$
Likewise, the natural determinant maps $`_n\text{Jac}^dC`$ given by $`(E,\varphi )\mathrm{\Lambda }^rE`$ extend to $`_{\mathrm{}}\text{Jac}^dC`$. Let $`\epsilon _1,\mathrm{},\epsilon _{2g}`$ be the pull-backs of the standard generators of $`H^1(\text{Jac}^dC)`$.
It is straightforward to check that the universal classes $`\alpha _i`$, $`\beta _i`$, $`\psi _{i,j}`$, $`\epsilon _j`$ thus defined restrict to their counterparts on $`_n`$ for $`n0`$.
(10.1) The rational cohomology ring $`H^{}(_{\mathrm{}})`$ is generated by these universal classes.
Proof. According to Atiyah-Bott, $`H^{}(B𝒢)`$ is generated by universal classes, which means the following. First of all, $`B𝒢`$ can be identified with the component of the space of maps $`CB\mathrm{U}(r)`$ such that the pull-back of the universal bundle over $`B\mathrm{U}(r)`$ is isomorphic to $``$. Atiyah-Bott call this component $`\text{Map}_{}(C,B\mathrm{U}(r))`$. Then, the pull-back of the universal bundle by the canonical map $`\text{Map}_{}(C,B\mathrm{U}(r))\times CB\mathrm{U}(r)`$ is a bundle whose Chern classes can be decomposed into Künneth components as usual. Atiyah-Bott prove \[2, 2.20\] that these generate the ring $`H^{}(B𝒢)`$.
On the other hand, by (9) $`B\overline{𝒢}`$ can also be identified with $`_{\mathrm{}}`$. Hence $`B𝒢`$ is a bundle over $`_{\mathrm{}}`$ with fiber $`B^\times =^{\mathrm{}}`$. As explained by Atiyah-Bott, the rational cohomology of this bundle splits: $`H^{}(B𝒢)=H^{}(B\overline{𝒢})[h]`$. By restricting to a single $`^{\mathrm{}}`$ fiber, it can be checked that $`\beta _1=rh`$ modulo elements of $`H^2(B\overline{𝒢})`$; it may therefore be discarded since we seek only generators of $`H^{}(B\overline{𝒢})`$. Also $`\alpha _1H^0(B𝒢)`$ may be discarded since by (9) $`B𝒢`$ is connected. Finally it can be checked that $`\psi _{1,j}=\epsilon _j`$, and that for $`i>1`$, the classes $`\alpha _i`$, $`\beta _i`$ and $`\psi _{i,j}`$ of Atiyah-Bott agree with those defined above. (Strictly speaking, they may differ by some lower order terms, since the characteristic classes of a projective bundle are evaluated by formally twisting so that $`c_1=0`$.) . $`\mathrm{}`$
Now by (9), $`H_{}(_{\mathrm{}})`$ is the direct limit of $`H_{}(_n)`$, and hence $`H^{}(_{\mathrm{}})`$ is the inverse limit of $`H^{}(_n)`$. Consequently, the surjectivity of the restriction map $`H^{}(_{\mathrm{}})H^{}(_n)`$ for all $`k`$, and hence the generation theorem (6), is implied by the following result, whose proof occupies the remainder of this section.
(10.2) When $`r=2`$, the restriction $`H^{}(_{n+1})H^{}(_n)`$ is surjective.
Let $`T=^\times `$ act on each $`_n`$ by $`\lambda (E,\varphi )=(E,\lambda \varphi )`$. This action is compatible with the inclusion $`_n_{n+1}`$. Furthermore, since this is an algebraic action on a smooth quasi-projective variety, the $`\mathrm{U}(1)`$-part of the action is Hamiltonian, and the $`𝐑^\times `$ part of the action is the Morse flow of the moment map.
(10.3) For any $`(E,\varphi )_n`$, there exists a limit $`lim_{\lambda 0}(E,\lambda \varphi )_n`$.
Proof. We may regard this as a limit of the downward Morse flow in $`_n`$. Note that it need not be simply $`(E,0)`$ as this may be unstable. Nevertheless, a stable limit always exists; this may be seen in two ways.
First, one can regard $`_n`$ as a space of solutions $`(A,\varphi )`$ to the self-duality equations, as Hitchin regards $``$; the moment map is then $`(A,\varphi )\varphi ^2`$, and what we need to know is that this is proper and bounded below. The boundedness is obvious, and the properness is proved following Hitchin’s argument for $``$ \[24, 7.1(i)\].
Alternatively and more algebraically, one can observe, as does Simpson \[42, Prop. 3\], that the Hitchin map defined by Nitsure \[38, 6.1\], taking $`_n`$ holomorphically to a vector space, is proper and intertwines the $`T`$-action on $``$ with a linear action on the vector space having positive weights. A limit must therefore exist in the zero fiber of the Hitchin map. . $`\mathrm{}`$
Now any $`\mathrm{U}(1)`$ moment map whose Morse flows have lower limits is a perfect Bott-Morse function: see for example Kirwan \[28, 9.1\]. This means that its Morse inequalities are equalities. More explicitly, it means the following. Let $`y_0,\mathrm{},y_k`$ be the critical values of the moment map $`\mu :X𝐑`$, and $`F_i`$ the corresponding critical submanifolds. Choose real numbers $`x_i`$ so that $`x_0<y_0<x_1<y_1<\mathrm{}<y_k<x_{k+1}`$. If $`X_i=\mu ^1(x_0,x_i)`$, then, as for any Bott-Morse function, there is a homotopy equivalence of pairs $`(X_{i+1},X_i)(D_i,S_i)`$, where $`D_i`$ is the disc bundle, and $`S_i`$ the sphere bundle, associated to the negative normal bundle of $`F_i`$, that is, the bundle of downward Morse flows, cf. Milnor . For the function to be perfect means that moreover the connecting homomorphism vanishes in each long exact sequence
$$\mathrm{}H^{}(X_{i+1},X_i)H^{}(X_{i+1})H^{}(X_i)\mathrm{},$$
breaking it up into short exact sequences.
Suppose now that $`X`$ contains a $`T`$-invariant submanifold $`Y`$ on which the moment map is again perfect. Then by induction on $`i`$, $`H^{}(X)`$ surjects on $`H^{}(Y)`$ if and only if $`H^{}(X_{i+1},X_i)`$ surjects on $`H^{}(Y_{i+1},Y_i)`$ for all $`i`$.
We find ourselves in this situation, with $`X=_{n+1}`$ and $`Y=_n`$. To prove (10), it therefore suffices to show that the relative cohomology of the disc bundle for the downward flow from each critical submanifold in $`_{n+1}`$ surjects on that of its intersection with $`_n`$. This will be true in the case $`r=2`$. Indeed, the Thom isomorphism identifies this relative cohomology with the ordinary cohomology of the critical submanifold itself. We will prove, first, that this identification is compatible with the restriction to $`_n`$, and second, that the latter restriction is surjective. Both will follow from the description below of the critical set (cf. Hitchin \[24, 7.1\]).
(10.4) The critical submanifolds of the moment map on $`_n`$ are a disjoint union
$$_n=\underset{j=0}{\overset{g+\left[\frac{n1}{2}\right]}{}}F_n^j,$$
where:
1. for $`j=0`$, the absolute minimum $`F_n^0`$ of the moment map is the moduli space of stable bundles of rank $`2`$ and degree $`d`$, parametrizing Higgs bundles $`(E,\varphi )`$ with $`\varphi =0`$;
2. for $`j>0`$, $`F_n^j=\text{Jac}^{\frac{d+1}{2}j}C\times \text{Sym}^{2g+n12j}C`$, parametrizing Higgs bundles $`(E,\varphi )`$ with $`E=LM`$, $`\mathrm{deg}L=\frac{d+1}{2}j`$, and
$$\varphi =\left(\begin{array}{cc}0& 0\\ s& 0\end{array}\right),$$
where $`sH^0(KML^1(n))`$ vanishes on an effective divisor of degree $`2g+n12j`$.
Proof. The critical points for the moment map of the action of $`\mathrm{U}(1)T`$ are exactly the fixed points of $`T`$. For any $`(E,\varphi )_n`$ fixed by $`T`$, by (4) $`T`$ acts by automorphisms on the universal bundle restricted to $`\{(E,\varphi )\}\times C`$, which is nothing but $`E`$ itself, and $`\lambda T`$ takes $`\varphi `$ to $`\lambda \varphi `$. If the weights of the action are distinct, this splits $`E`$ as a sum of line bundles $`LM`$, and $`\varphi `$ is forced to be of the stated form. If the weights are not distinct, then $`T`$ acts by scalars, so $`\varphi `$ is invariant and hence must be $`0`$. . $`\mathrm{}`$
(10.5) The downward flow from $`F_n^j=F_{n+1}^j_n`$ in $`_{n+1}`$ is wholly contained in $`_n`$.
Proof. The statement is vacuous for $`j=0`$, since at the absolute minimum there is no downward flow. Consider then a point $`(E,\varphi )F_n^j`$ for as described in (10)(b). According to (3), the tangent space to $`_{n+1}`$ at $`(E,\varphi )`$ is the hypercohomology $`𝐇^1𝐄𝐧𝐝(E,\varphi )`$, where $`𝐄𝐧𝐝(E,\varphi )`$ is the two-term complex
$$\text{End}E\stackrel{[,\varphi ]}{}\text{End}EK(n+1).$$
Since $`E=LM`$ and $`\varphi `$ is strictly lower-triangular, this breaks up as a direct sum of complexes, which are the weight spaces for the $`T`$-action. The downward flow corresponds to $`𝐇^1`$ of the complex
$$\text{Hom}(L,M)0,$$
which is of course just $`H^1(ML^1)`$. This is independent of $`n`$, and so the downward flow in $`_{n+1}`$ is wholly contained in $`_n`$, as desired. . $`\mathrm{}`$
Consequently, we may use the Thom isomorphisms to identify the map of relative cohomology on the disc bundles with the ordinary restriction map $`H^{}(_{n+1})H^{}(_n)`$. To prove (10), then, it remains to prove the following statement.
(10.6) The restriction $`H^{}(_{n+1})H^{}(_n)`$ is surjective.
Proof. Again, this is vacuous for $`j=0`$, since $`F_{n+1}^0=F_n^0`$. It is not much harder for $`j>0`$, for $`F_n^j`$ is isomorphic to a product of a Jacobian and a symmetric product, and the embedding $`F_n^jF_{n+1}^j`$ corresponds to the identity on the first factor and a map of effective divisors of the form $`DD+p`$ on the second. The latter map is easily seen (for example, from the description of Macdonald ) to induce a surjection on cohomology. . $`\mathrm{}`$
This completes the proof of (10), and hence of the generation theorem (6).
Note, by the way, that the theorem is false for the moduli space $`_0`$ consisting of pairs with fixed determinant line bundle $`\mathrm{\Lambda }^nE`$ and trace-free $`\varphi `$. One can see this already from Hitchin’s description \[24, 7.6\] of its cohomology: the universal classes are all invariant under the natural action of $`\mathrm{\Sigma }=_2^{2g}`$, but there also exist classes which are not $`\mathrm{\Sigma }`$-invariant. (See also our companion paper \[23, §4\].)
In the cases of ranks 2 and 3, the first author has found an alternative proof of the generation theorem, which will appear in a forthcoming paper . It uses the vector bundles over $`_n`$ whose Chern classes above their ranks furnish the so-called “Mumford relations”; these are well-known to have the appropriate dimension in the $`\mathrm{GL}(r)`$ case, but not in the $`\mathrm{SL}(r)`$ case.
The generation theorem does, however, tell us the following about the cohomology ring of the fixed-determinant moduli space.
(10.7) Let $`\mathrm{\Xi }`$ be a fixed line bundle of odd degree, and let $`_n`$ be the moduli space of Higgs bundles $`(E,\varphi )_n`$ with $`\mathrm{\Lambda }^2E\mathrm{\Xi }`$ and $`\text{tr}\varphi =0`$. Then $`\mathrm{\Sigma }_2^{2g}`$ acts naturally on $`_n`$, and
$$H^{}(_n)=H^{}(\text{Jac}C)H^{}(_n)^\mathrm{\Sigma },$$
where $`H^{}(\text{Jac}C)`$ is generated by the $`\epsilon _1,\mathrm{},\epsilon _{2g}`$, and $`H^{}(_n)^\mathrm{\Sigma }`$ by the remaining universal classes.
Proof. In fact $`_n`$ is the quotient of $`T^{}\text{Jac}C\times _n`$ by the free action of $`\mathrm{\Sigma }`$; the quotient map is $`(L,\psi )\times (E,\varphi )(LE,\psi \text{id}+\varphi )`$. A theorem of Grothendieck referred to earlier asserts that the rational cohomology of a quotient by a finite group is the invariant part of the rational cohomology. Hence
$`H^{}(_n)`$ $`=`$ $`H^{}(T^{}\text{Jac}C\times _n)^\mathrm{\Sigma }`$
$`=`$ $`H^{}(\text{Jac}C)H^{}(_n)^\mathrm{\Sigma },`$
since $`\mathrm{\Sigma }`$ acts on $`T^{}\text{Jac}C`$ only by translations. The determinant map $`_n\text{Jac}^dC`$ lifts to the map $`T^{}\text{Jac}C\times _n\text{Jac}^dC`$ given by projection on the first factor followed by an isogeny of order $`r`$; hence $`\epsilon _1,\mathrm{},\epsilon _{2g}`$ generate the rational cohomology of the first factor. A universal pair on $`_n`$ pulls back to the tensor product of the Poincaré line bundle on $`\text{Jac}C\times C`$ with a universal pair on $`_n\times C`$; hence its projectivization, and consequently the remaining universal classes, are pulled back from the second factor. . $`\mathrm{}`$ |
warning/0003/cond-mat0003134.html | ar5iv | text | # Ab initio calculation of the lattice distortions induced by substitutional Ag- and Cu- impurities in alkali halide crystals.
## I Introduction
Most of the luminescent materials presently used in several technological applications involve the doping of a pure ionic crystal, that is substitution of some of the ions by other ions with specific absorption-emission characteristics. The fine details of the absorption-emission spectra, as well as the efficiency and resolution of the scintillator, are determined by the system-specific embedding potential acting on the impurity, which is in turn sensitive to the distortion induced by the impurity on the crystal lattice. Thus, a theoretical understanding and accurate determination of those distortions is of paramount importance, moreover if we realize that their experimental measurement is a difficult task.
Two main methods are applied nowadays to model impurity systems: supercell techniques, that exploit the convenience of the Bloch theorem by periodically duplicating a finite region of the crystal around the impurity ; and the cluster approach, in which the doped crystal is modeled by a finite cluster centered on the impurity and embedded in a field representing the rest of the host lattice. This cluster approach is the one chosen in the present study, and has been used in the past to study the geometrical and optical properties of doped crystals. The cluster (active space) can be studied by using standard quantum-mechanical methods. The rest of the crystal (environment) can be described in several ways. In the simplest and most frequently used approach, the environment is simulated by placing point charges on the lattice sites, but this procedure has to be improved in order to obtain a realistic description of the lattice distortions around the impurity. Model potentials have been developed to represent the effects of the environment on the active cluster, that include attractive and repulsive quantum-mechanical terms aside from the classical Madelung term, but a problem still remains: the large computational cost of conventional molecular orbital (MO) calculations prevents from performing an exhaustive geometrical relaxation of the lattice around the impurity. In the most accurate MO calculations, only the positions of the ions in the first shell around the impurity are allowed to relax. However, geometrical relaxations far beyond the first shell of neighbors can be expected. In fact, recent semiempirical simulations of solids, performed employing phenomenological potentials, have shown the importance of considering appropriate large-scale lattice relaxations in the study of a variety of intrinsic and extrinsic defects in ionic crystals. As we will show below, the systems under study in this paper can not be properly described by simply considering an expansion of the first shell of neighbors around the impurity.
In this contribution we report theoretical calculations of the lattice distortions induced by Ag<sup>-</sup> and Cu<sup>-</sup> substitutional impurities in 16 alkali halide crystals with the rock-salt structure, namely all those noncontaining cesium. For this purpose we use the ab initio Perturbed Ion (PI) model, which circunvents the problems mentioned above: (a) The active cluster is embedded in an environment represented by the ab initio model potentials of Huzinaga et al. ; (b) The computational simplicity of the PI model allows for the geometrical relaxation of several coordination shells around the impurity. Moreover, it allows us to study a whole family of systems in order to look for systematic trends that might be useful in later theoretical studies of doped crystals similar to those here considered.
The remainder of this paper is organized as follows: In Section II we describe the active cluster which has been used to model the doped systems. In Section III we present and discuss the results of the calculations, and Section IV summarizes the main conclusions.
## II Cluster model
The ab initio Perturbed Ion model is a particular application of the theory of electronic separability of Huzinaga and coworkers to ionic solids, in which the basic building blocks are reduced to single ions. The PI model was first developed for perfect crystals. Its application to the study of impurity centers in ionic crystals has been described in refs. , and we refer to those papers for a full account of the method. In brief, an active cluster containing the impurity is considered, and the Hartree-Fock-Roothaan (HFR) equations for each ion in the active cluster are solved in the field of the other ions. The Fock operator includes, apart from the usual intra-atomic terms, an accurate quantum-mechanical crystal potential and a lattice projection operator which accounts for the energy contribution due to the overlap between the wave functions of the ions. The atomic-like HFR solutions are used to describe the ions in the active cluster in an iterative stepwise procedure. The wave functions of the lattice ions external to the active cluster are taken from a PI calculation for the perfect crystal and are kept frozen during the embedded-cluster calculation. Those wave functions are explicitely considered for ions up to a distance $`d`$ from the center of the active cluster such that the quantal contribution from the most distant frozen shell to the effective cluster energy is less than 10<sup>-6</sup> hartree. Ions at distances beyond $`d`$ contribute to the effective energy of the active cluster just through the long-range Madelung interaction, so they are represented by point charges. At the end of the calculation, the ionic wave functions are selfconsistent within the active cluster and consistent with the frozen description of the rest of the lattice. The intraatomic Coulomb correlation, which is neglected at the Hartree-Fock level, is computed as a correction by using the Coulomb-Hartree-Fock (CHF) model of Clementi.
In a previous work we employed several active clusters of increasing size and with different embedding schemes to describe the scintillator system Tl<sup>+</sup>:NaI. That study was undertaken in order to find the necessary requirements that a cluster model has to fulfill in order to describe properly a doped crystal. Here we just describe the best cluster model between those studied in ref. . This active cluster, shown in Figure 1, has 179 ions which correspond to the central impurity (Ag<sup>-</sup> or Cu<sup>-</sup>) plus twelve coordination shells. Those ions are further split up into two subsets. One is formed by the central impurity plus the first four coordination shells, having a total of 33 ions, and both the wave functions and positions of the ions in this subset are allowed to relax. The lattice positions of the other 146 ions of the active cluster are held fixed during the calculations but their wave functions have been selfconsistently optimized. This is done so that the connection between the region where distortions are relevant and the rest of the crystal is as smooth as possible. In our previous study we showed how an unphysically abrupt connection between those two regions fails in describing properly the lattice distortions induced by the impurity. The ions in the interface region can respond to those distortions by selfconsistently adapting their wave functions to the new potential, and thus contribute to build a more realistic (selfconsistent) environment. The geometrical relaxation around the impurity has been performed by allowing for the independent breathing displacements of each shell of ions, and minimizing the total energy with respect to those displacements until the effective cluster energies are converged up to 1 meV. A downhill simplex algorithm was used. For the ions we have used large STO basis sets, all taken from Clementi-Roetti tables.
The cluster used in this work has been shown to be self-embedding consistent for NaI in our previous work. By this we mean that if the pure crystal is represented by this cluster model (that is, if the central impurity is replaced by the halogen ion corresponding to the pure crystal), the results of the cluster model calculations closely agree with those from a PI calculation for the pure crystal, where all cations (or anions) are equivalent by translational symmetry. The same is true for all the family of alkali halide crystals considered here. Nevertheless, the self-embedding consistency is never complete. In order to supress systematic errors from the distortions calculated with the cluster method, the radial displacements of each shell have been calculated using the following formula:
$$\mathrm{\Delta }R_i=R_i(Imp^{}:AX)R_i(X^{}:AX),$$
(1)
where R<sub>i</sub> (i=1, 2, 3, 4) refer to the radii of the first, second, third, and fourth shells around the impurity in the AX crystal, A = Li, Na, K, Rb, X = F, Cl, Br, I, and Imp<sup>-</sup> = Ag<sup>-</sup>, Cu<sup>-</sup>. Thus both systems (pure and doped crystals) are treated in eq. (1) on equal foot with the cluster model, and not with different methodologies, and the calculated distortions are free from that potential source of error. Also, in order to have the correct Madelung potential at the impurity site, the calculations have been performed by employing the experimental lattice constants to describe the geometrically frozen part of the crystals.
The only terms omitted in our description are the dispersion terms (coming from interatomic correlation) and relativistic effects for the heavy ions. Although the importance of both effects increase with atomic number, they are not crucial for the structural properties of the systems studied here. Specifically, Martín Pendás et al. have shown that the PI method gives lattice constants and bulk moduli in close agreement with experimental results for all alkali halides. The properties of these crystals under the influence of an applied external pressure, a situation where the importance of interatomic correlation effects increases, are also properly reproduced.
## III Results and discussion
The calculated distortions, collected in Table I, are the main quantitative result from our study. For visualization of the trends, however, it is better to display the results in a figure, and this is done in Fig. 2, where we have plotted the distortion of each of the four shells in terms of the empirical cationic radii extracted from ref. . Those points corresponding to the same anion have been joined with a line to guide the eye. The figure contains only the results for Ag<sup>-</sup> because the trends are the same in the case of the Cu<sup>-</sup> impurity. Next we describe, shell by shell, the general trends in Fig. 2:
First shell. This shell is formed by 6 cations in ($`\frac{1}{2}`$,0,0) crystallographic sites, and undergoes an expansion, as might be expected from the larger size of Ag<sup>-</sup> compared to the halogen anions. The impurity anion pushes the neighbor cations to make room for itself in the lattice. The expansion is substantial, with percentage values between 9 and 15 %. In the F salts that expansion is larger the larger the cation size, but this trend is violated in the Cl, Br, and I crystals. If we fix the cation, for K and Rb salts the expansion is larger the smaller the anion (notice that the anion size increases in the order F<sup>-</sup>, Cl<sup>-</sup>, Br<sup>-</sup>, I<sup>-</sup>). This rule is inverted in the case of Li salts, whereas Na salts constitute an intermediate case. It should be recognised, nevertheless, that the expansion is almost independent of the halogen element in the Li and Na salts.
Second shell. The displacement of the second shell, formed by 12 anions at ($`\frac{1}{2}`$,$`\frac{1}{2}`$,0) positions, is always a small contraction. If we fix the anion, the contraction is larger (absolute value) the larger is cation size. If the cation is fixed, for Na, K and Rb salts the contraction is larger the smaller the anion size. Again this trend is inverted for Li salts.
Third shell. This shell, formed by 8 cations at ($`\frac{1}{2}`$,$`\frac{1}{2}`$,$`\frac{1}{2}`$) sites, experiences a small expansion. If we fix the anion, the expansion is smaller the larger the cation size. If the cation is fixed, there is not a definite trend. In the case of Li and Na salts the expansion increases with the anion size (LiI is an exception). This trend is partially inverted in the case of Rb salts, and for K salts all the expansions are almost identical.
Fourth sell. This shell, formed by 6 anions at (1,0,0) positions, experiences an expansion. That expansion increases with cation size if we fix the anion. If the cation is fixed, in K and Rb salts the expansion is smaller the larger the anion size. In Na salts, NaI is again an exception to this general rule, whereas in Li salts the expansion is almost constant.
In the following we try to find some rationalization for the calculated trends. The working rule still in use nowadays stating that the expansion of the first coordination shell can be approximated by the difference between the ionic radii of the impurity and the substituted ion is somewhat misleading. First of all, the ions in a crystal are not hard spheres, but weakly overlapping soft spheres, so one cannot use values for the ionic radii of ions in vacuum in order to predict lattice distortions accurately. In particular, the size of an anion may vary in a nonnegligible way from crystal to crystal. To investigate this, we show $`(<r^2>)^{1/2}`$ for F<sup>-</sup> in fluoride crystals in Table II, where the expectation value is taken over the outermost orbital of the anion (2p), and is calculated from the crystal-consistent ionic wave functions obtained through a PI calculation on the pure crystals. r<sub>X</sub> = $`(<r^2>)^{1/2}`$ can be taken as a rough measure of the anion size. We also show the analogous quantity for the 5s orbital of the Ag<sup>-</sup> impurity in fluorides. The size of the F<sup>-</sup> anion varies by a maximum of 2 %, small compared with the size variation of the Ag<sup>-</sup> anion (7 %). Ag<sup>-</sup> is more compressible than the halogen anions because its outer electronic shell is an s-shell, while it is a p-shell for the halogens. The same can be said of all the halogens. As the size of the cation decreases, Ag<sup>-</sup> is more compressed by the crystal environment. This shows that the standard ionic radii cannot be used for predicting distortions, because their values are a genuine output of the selfconsistent process. Nevertheless, though they are not useful for accurate predictions, physical insight tells us that the distortions should be correlated with the size of the ions, in the sense that one always expect that larger impurities induce larger distortions. These should be useful at least at a qualitative level.
In Table III we show the differences
$$\delta =r(Ag^{}:AX)r_X(X^{}:AX),$$
(2)
where r(Ag<sup>-</sup>:AX) is the radius of the Ag<sup>-</sup> impurity in the AX crystal, and r<sub>X</sub>(X<sup>-</sup>:AX) is the radius of the halogen anion. The differences are expected to be related with the first-shell expansions $`\mathrm{\Delta }R_1`$. The values of $`\delta `$ do not show quantitative agreement with those of $`\mathrm{\Delta }R_1`$. If the anion is fixed, $`\delta `$ increases with the cation size, which is consistent with the main trend in $`\mathrm{\Delta }R_1`$ of fluorides, but does not explain the behavior of other halides. The main trends discussed when the cation is fixed are reproduced for the K and Rb crystals but not for the others. From Figure 1 we can see that the exceptional crystals, concerning the trends in $`\mathrm{\Delta }R_1`$, are LiCl, LiBr, LiI, and NaI. The peculiar feature of those four cases is that anions are much larger than cations: the ratios r<sub>A</sub>/r<sub>X</sub> are the smallest in the family of alkali halides. In a study of the structures of small alkali halide clusters, it was found that materials with small r<sub>A</sub>/r<sub>X</sub> have a cluster growing pattern different from the rest. Specifically, those systems showed a marked tendency towards ring-like structures, whereas the others adopt fragments of the rocksalt lattice as their minimum energy structures. The main reason is that anion-anion repulsions are much more important when the ratio r<sub>A</sub>/r<sub>X</sub> is small. In a recent study, A. Martín Pendás et al. have applied the atoms–in–molecules (AIM) theory of Bader to study the topology of the electron density in crystals. They found that the whole group of alkali halides with rocksalt structure can be divided up into three topological families, called R<sub>1</sub>, B<sub>1</sub>, and B<sub>2</sub> in that paper, and that the ionic radii are the topological organizers. The R<sub>1</sub> family contains KCl, NaF, KF, RbF, RbBr, RbCl and all the cesium halides. The B<sub>1</sub> family contains KI, KBr, LiF, RbI, NaBr, NaCl and NaI, and the B<sub>2</sub> family contains LiCl, LiBr and LiI. There are constant r<sub>A</sub>/r<sub>X</sub> lines that isolate each family. The largest values of r<sub>A</sub>/r<sub>X</sub> are found for family R<sub>1</sub> and the smallest for family B<sub>2</sub>, with intermediate values for B<sub>1</sub>. NaI is so near the B<sub>2</sub> region that it is not surprising that it behaves in Fig. 2 like the elements of the B<sub>2</sub> family. In the AIM theory the critical points of the electron density scalar field are classified as nuclei, bond points, ring points and cage points. When a bond point is found between two nuclei, a bond is established between the corresponding atoms. In the R<sub>1</sub> family there are just anion-cation bonds. We have found that for the (undoped) crystals of this family, the anion-anion overlap is at least one order of magnitude smaller than the anion-cation overlap. In the crystals of the B<sub>1</sub> and B<sub>2</sub> families there are bond critical points between anions, so that the effective local coordination is 6 for cations and 18 for anions (6 anion-cation and 12 anion-anion bonds). In the B<sub>1</sub> crystals (excepting NaI) we have found that the anion-anion overlap is smaller but of the same order of magnitude than cation-anion overlap. In the B<sub>2</sub> crystals, and also in NaI, anion-anion overlap is the largest contribution to the repulsive interactions, and thus anion-cation contacts are less important. The cations of the B<sub>2</sub> family occupy the interstitial holes left in the anionic fcc sublattice. It is then not surprising that when the cation-anion overlap is not so important, the expansion of the cation shell is an exception to the general trends; it is, in fact, nearly constant for the B<sub>2</sub> family.
Let us turn to discuss the distortion of the second and third shells. In all cases, the second shell suffers a contraction of a small magnitude compared to the large expansion of the first shell. In Figure 1 one can see that the radial outward motion of the first cation shell is not going to affect much the positions of the twelve anions of shell 2, so it becomes understandable that the anions of that shell move little. The small contraction of the second shell optimizes the Madelung energy around the impurity and also serves to pack more efficiently the ions in response to the outward motion of the cations. The quantitative trend of that contraction is understood with reference to the three topological families discussed in the previous paragraph and their relation to the anion-anion overlap: the contraction is largest for those crystals where the anion-anion overlap is small (R<sub>1</sub> crystals), intermediate when that overlap begins to count (B<sub>1</sub> crystals), and finally, it is lowest for the B<sub>2</sub> crystals, where anion-anion contacts are important. This explains the trends observed: if the anion is fixed, the contraction is larger the larger the cation size, because the Ag<sup>-</sup>-X<sup>-</sup> overlap decreases with increasing cation size. On the other hand, if the cation is fixed, in Rb, K, and Na salts the contraction decreases with increasing anion size, because the Ag<sup>-</sup>-X<sup>-</sup> overlap increases with anion size. But in Li salts the trend is inverted because Ag<sup>-</sup>-X<sup>-</sup> overlaps decrease with anion size. The distortion of the third shell is not directly related to the introduction of the impurity, as the overlap between the cations of that shell and Ag<sup>-</sup> is very small (see Fig. 1). The expansion of this shell appears to be again of a purely electrostatic origin. The relative values of the displacements (R<sub>3</sub> \- R$`{}_{}{}^{crystal}{}_{3}{}^{}`$)/R$`{}_{}{}^{crystal}{}_{3}{}^{}`$ are very small, less than 1 % except in LiF.
The $`\mathrm{\Delta }R_4`$ displacements, always an expansion, proceed along the same crystallographic direction as the $`\mathrm{\Delta }R_1`$ displacements. Thus, the expansion is clearly induced by the expansion of the first shell. The relative displacements (R<sub>4</sub>-R$`{}_{}{}^{crystal}{}_{4}{}^{}`$)/R$`{}_{}{}^{crystal}{}_{4}{}^{}`$ adopt values between 1 % and 5%, compared to values of 9–15 % for (R<sub>1</sub>-R$`{}_{}{}^{crystal}{}_{1}{}^{}`$)/R$`{}_{}{}^{crystal}{}_{1}{}^{}`$. These numbers indicate that the expansion of the (ninth) shell formed by six cations at ($`\frac{3}{2}`$,0,0) is not expected to be higher than 1 %. In the K and Rb salts $`\mathrm{\Delta }R_4`$ is larger when $`\mathrm{\Delta }R_1`$ is larger, while in Li and Na salts $`\mathrm{\Delta }R_4`$ and $`\mathrm{\Delta }R_1`$ are both almost independent of the anion, so the displacements of the first and fourth shells are correlated. If the anion is fixed, the expansion increases with the cation size, and no special behaviour is observed in the cases of LiCl, LiBr, LiI and NaI.
We conclude that the lattice relaxation around the substitutional impurity in the alkali halides involves the concerted movement of several coordination shells. However, it is not yet clear from the results presented up to this point whether the lattice relaxations of the second, third and fourth shells have a substantial influence on the energy of formation of the defect. At low pressure and temperature conditions, the formation of the impurity centers should be discussed in terms of the internal energy difference for the exchange reaction
$$(X^{}:AX)_s+Imp_g^{}(Imp^{}:AX)_s+X_g^{},$$
(3)
where the $`s`$ and $`g`$ subindexes refer to solid and gas phases, respectively. In order to establish the importance of the relaxation of the lattice beyond the first coordination shell, we have calculated the formation energy $`\mathrm{\Delta }`$H of the defects by employing two different models for the active cluster, shown in Fig. 1: one of them is that formed by 179 ions described in section II; the other includes just four coordination shells around the central ion (a total of 33 ions), and only the positions of the six ions in the first coordination shell are allowed to relax. The results are shown in Table IV. We see that according to the small cluster model the energy of formation of the defects is always positive, that is none of the impurity centers are stable centers. Enlarging the cluster size to include a selfconsistent treatment of 179 ions and extending the lattice relaxation up to the fourth coordination shell induces a huge stabilization of all the impurity centers. The trend of $`\mathrm{\Delta }`$H is simple. For a given impurity (Ag<sup>-</sup> or Cu<sup>-</sup>) the heat of formation decreases by increasing the atomic number of the cation (alkali) or of the anion (halogen). The positive value of $`\mathrm{\Delta }`$H in the calculation for the small cluster is understandable: the impurity simply pushes its neighbor cations producing a high elastic strain energy since the rest of the lattice is not allowed to respond. In the second model three more shells are allowed to move in response to that initial stress and the relaxation lowers the elastic energy so much that the electronic contributions turn $`\mathrm{\Delta }`$H negative in all cases except Ag<sup>-</sup>:LiF and Ag<sup>-</sup>:NaF ($`\mathrm{\Delta }`$H is nearly zero in Ag<sup>-</sup>:LiCl). It is useful to notice that $`\mathrm{\Delta }`$R<sub>1</sub>, the displacement of the first shell, is lower for the first cluster model compared to the second. This means that due to the constraints imposed by the first model the atoms of the first shell are unable to reach their preferred equilibrium positions in the presence of the impurity, a fact that is consistent with the large calculated $`\mathrm{\Delta }`$H. The change of sign in $`\mathrm{\Delta }`$H can be interpreted as suggesting that most of the elastic relaxation of the lattice has been accounted for and that allowing for the elastic relaxation of more shells will have a minor effect. The two cases with a positive $`\mathrm{\Delta }`$H, Ag<sup>-</sup>:LiF and Ag<sup>-</sup>:NaF, are still intriguing. These two crystals have the smallest lattice parameters within the whole family studied here, and it is conceivable that the elastic effects will be largest. The question if the relaxation of more coordination shells is able to stabilize those two systems deserves further investigation.
As indicated above the difference in the heats of formation given by the two models, $`\mathrm{\Delta }`$H(model 2) - $`\mathrm{\Delta }`$H(model 1), gives a measure of the lowering of elastic strain when more coordination shells are allowed to relax. From Table IV one can verify that, for a given crystal, this energy is essentially independent of the impurity, while both $`\mathrm{\Delta }`$H(model 1) and $`\mathrm{\Delta }`$H(model 2) depend on the impurity. This confirms our interpretation of the effect of allowing for the elastic relaxation of several shells: that relaxation is mainly a host effect.
## IV Summary
We have reported a study of the local lattice distortions induced by substitutional Ag<sup>-</sup> and Cu<sup>-</sup> impurities in the family of alkali halide crystals excepting those containing cesium. For this purpose, the ab initio Perturbed Ion (PI) model has been used. A large active cluster of 179 ions, embedded in an accurate quantum environment representing the rest of the crystal, has been studied. The local distortions obtained extend beyond the first shell of neighbors in all cases. Thus, the assumptions frequently employed in impurity calculations, which consider the active space as formed by the central impurity plus its first coordination shell only, should be taken with some care. Distortion trends have been identified and discussed. The first coordination shell (cations) around the impurity experiences an expansion as a consequence of the larger size of the impurity anion compared to the halogens. That expansion is larger for the Ag<sup>-</sup> than for the Cu<sup>-</sup> impurity, also because the first anion is larger than the second. The trends can be qualitatively explained by considering the difference in size between the impurity and the substituted anion in all cases except in those crystals with a very small size ratio between cation and anion. In those cases, that is for LiCl, LiBr, LiI and NaI, anion-anion contacts are important. Those four materials have been found to exhibit special behavior in a number of previous studies involving crystals and clusters. The contraction of the second shell as well as the expansion of the third shell are small and arise from a combination of electrostatic and packing origins. The fourth shell experiences a substantial expansion as a consequence of the direct pushing induced by the expansion of the first shell. The analysis of the energies of formation of the defects clearly shows that elastic relaxation of several coordination shells around the impurity is necessary in the modeling of these materials since this affects even the sign of the energy of formation.
Acknowledgements: Work supported by DGES (PB98-0345) and Junta de Castilla y León (VA28/99). A. Aguado is grateful to University of Valladolid for financial support. We thank the suggestions of one referee.
Captions of Tables
Table I. Radial displacements $`\mathrm{\Delta }R_i`$ (see eq. (1)), in Å, of the first four shells of ions around the silver and copper impurities.
Table II. $`(<r^2>)^{1/2}`$, where the expectation values are taken over the outermost orbital of the F<sup>-</sup> anion in pure alkali fluorides and of the silver anion in Ag<sup>-</sup>-doped alkali fluorides. All quantities in Å.
Table III. Difference of radii between Ag<sup>-</sup> and the substituted anion (see eq. (2)), in Å.
Table IV. Formation energies (in eV) of copper and silver substitutional centers in different alkali halide host lattices, calculated employing two different models for the active cluster. First row: the active cluster contains 33 ions, and only the positions of the ions in the first coordination shell are allowed to relax. Second row: the active cluster contains 179 ions, and the positions of the ions in the first four coordination shells are allowed to relax.
Captions of Figures
Figure 1. The active cluster (ImpA<sub>92</sub>X<sub>86</sub>)<sup>5+</sup> employed to represent the region around the impurity, where Imp=Ag, Cu; light spheres are cations and dark spheres anions. The core of the cluster, formed by the four first coordination shells, which are allowed to breath, is also indicated separately.
Figure 2. Shell distortions $`\mathrm{\Delta }`$R<sub>i</sub> (i=1,2,3,4) around Ag<sup>-</sup>, plotted as a function of cation size.
\[ |
warning/0003/math0003179.html | ar5iv | text | # Remarks on plane maximal curves
## 1. Introduction
A $`𝐅_{q^2}`$-maximal curve of genus $`g`$ is a projective, geometrically irreducible, non-singular, algebraic curve defined over a finite field $`𝐅_{q^2}`$ of order $`q^2`$ such that the number of its $`𝐅_{q^2}`$-rational points attains the Hasse-Weil upper bound
$$1+q^2+2qg.$$
Maximal curves, especially those having large genus with respect to $`q`$, are known to be very useful in Coding theory . Also, there are various ways of employing them in Cryptography, and it is expected that this interesting connection will be be explored more fully, see \[34, Chapter 8\]. Other motivation for the study of maximal curves comes from Correlations of Shift Register Sequences , Exponentials Sums over Finite Fields , and Finite Geometry . Recent papers on maximal curves which also contain background and expository accounts are , , , , , , , , , , , and .
A relevant result on $`𝐅_{q^2}`$-maximal curves $`𝒳`$ with genus $`g`$ states that either $`g=q(q1)/2`$ and $`𝒳`$ is $`𝐅_{q^2}`$-isomorphic to the Hermitian curve $``$ of equation
(1.1)
$$X^{q+1}+Y^{q+1}+Z^{q+1}=0,$$
or $`g(q1)^2/4`$; see , , and . One expects that the bound $`(q1)^2/4`$ can be substantially lowered apart from a certain number of exceptional values of $`g`$. Finding such values is one of the problems of current interest in the study of maximal curves; see \[9, Section 3\], \[11, Proposition 2.5\], \[7, Section 3\], and .
In this paper we investigate plane maximal curves. In Section 2 we prove the non-existence of a plane $`𝐅_{q^2}`$-maximal curve whose genus belongs to the interval $`(q(q2)/8,q(q2)/4]`$, for $`q`$ even, and $`((q1)(q3)/8,(q1)^2/4]`$ for $`q`$ odd; see Corollary 2.3. The curves studied in Section 3 show that these bounds are sharp in some cases. In contrast, a few examples of (non planar) $`𝐅_{q^2}`$-maximal curves with genera in these intervals are known to exist; see \[9, Section 3\], \[7, pp. 74–75\], , , and \[8, Theorem 2.1\].
In the course of our investigation we point out that the Hermitian curve $``$ is the unique $`𝐅_{q^2}`$-maximal curve (up to $`𝐅_{q^2}`$-isomorphism) which is $`𝐅_{q^2}`$-Frobenius non-classical with respect to the linear series $`\mathrm{\Sigma }_1`$ cut out by lines; see Proposition 2.2. Also, the order of contact $`ϵ_2`$ of a non-classical (with respect to $`\mathrm{\Sigma }_1`$) $`𝐅_{q^2}`$-maximal curve with the tangent at a general point satisfies $`ϵ_2^2q/p`$, where $`p:=\mathrm{char}(𝐅_{\mathrm{q}^2})`$; see Corollary 2.8. In particular, plane $`𝐅_{q^2}`$-maximal curves with $`q=p`$ and $`q=p^2`$ are classical with respect to $`\mathrm{\Sigma }_1`$.
According to \[27, Prop. 6\], every curve which is $`𝐅_{q^2}`$-covered by the Hermitian curve is $`𝐅_{q^2}`$-maximal. An open problem of considerable interest is to decide whether the converse of this statement also holds. In Section 3 we solve this problem for the family of the so-called Hurwitz curves. Recall that a Hurwitz curve of degree $`n+1`$ is defined as a non–singular plane curve of equation
(1.2)
$$X^nY+Y^nZ+Z^nX=0,$$
where $`p=\mathrm{char}(𝐅_{\mathrm{q}^2})`$ does not divide $`n^2n+1`$. Theorem 3.1 together with Corollary 3.3 states indeed that the Hurwitz curve is $`𝐅_{q^2}`$-covered by the Hermitian curve if and only if
(1.3)
$$q+10(mod(n^2n+1)).$$
It should be noted on the other hand that for certain $`n`$ and $`p`$, the Hurwitz curve is not $`𝐅_{q^2}`$-maximal for any power $`q`$ of $`p`$; this occurs, for instance, for $`n=3`$ and $`p1(mod7)`$. One can then ask for conditions in terms of $`n`$ and $`p`$ which assure that the Hurwitz curve is $`𝐅_{q^2}`$-maximal for some power $`q`$ of $`p`$. Our results in this direction are given in Remarks 3.6 and 3.10, and Corollaries 3.7, 3.8. They generalise some previous results obtained in \[4, Lemmes 3.3, 3.6\]. Another feature of the Hurwitz curve is that it is non-classical provided that $`p^e`$ divides $`n`$ with $`p^e3`$; see Remark 3.11. So if both (1.3) and $`p^e|n`$ hold then the Hurwitz curve turns out to be a non-classical plane $`𝐅_{q^2}`$-maximal curve. As far as we know, these Hurwitz curves together with the Hermitian curves and the Fermat curves of degree $`n^2n+1`$ (see Corollary 3.3), are the only known examples of non-classical plane $`𝐅_{q^2}`$-maximal curves. As mentioned before, these curves show the sharpness of some of the results obtained in Section 2.
Hurwitz curves as well as their generalizations have been investigated for several reasons by many authors; see \[3, Section 1\] and . This gives a motivation to the final Section 4 where we show that the main results of Section 3 extend to (the non–singular model of) the curve with equation
$$X^nY^{\mathrm{}}+Y^nZ^{\mathrm{}}+Z^nX^{\mathrm{}}=0,$$
where $`n\mathrm{}2`$ and $`p=\mathrm{char}(𝐅_{q^2})`$ does not divide $`Q(n,\mathrm{}):=n^2n\mathrm{}+\mathrm{}^2`$.
Our investigation uses some concepts, such as non-classicity, from Stöhr-Voloch’s paper where an alternative proof to the Hasse-Weil bound was given among other things. We also refer to that paper for terminology and background results on orders and Frobenius orders of linear series on curves.
## 2. The degree of a plane maximal curve
Let $`𝒳`$ be a plane $`𝐅_{q^2}`$-maximal curve of degree $`d2`$. Since the genus of $`𝒳`$ is equal to $`(d1)(d2)/2`$, the upper bound for $`g`$ quoted in Sec. 1 can be rephrased in terms of $`d`$:
(2.1)
$$dd_1(q):=\frac{3+\sqrt{2(q3)(q+1)+9}}{2}\text{or}d=q+1.$$
The main result in this section is the improvement of (2.1) given in Theorem 2.12: Apart from small $`q`$’s, either $`d=q+1`$, or $`d=(q+2)/2`$, or $`d`$ is upper bounded by a certain function $`d_5(q)`$ such that $`d_5(q)/q2/5`$. Our first step consists in lowering $`d_1(q)`$ to $`d_2(q)`$ with $`d_2(q)/q1/2`$.
Let $`\mathrm{\Sigma }_1`$ be the linear series cut out by lines of $`𝐏^2(\overline{𝐅}_{q^2})`$ on $`𝒳`$. For $`P𝒳`$, let $`j_0(P)=0<j_1(P)=1<j_2(P)`$ be the $`(\mathrm{\Sigma }_1,P)`$-orders, and $`ϵ_0=0<ϵ_1=1<ϵ_2`$ (resp. $`\nu _0=0<\nu _1`$) the orders (resp. $`𝐅_{q^2}`$-Frobenius orders) of $`\mathrm{\Sigma }_1`$. We let $`p`$ be the characteristic of $`𝐅_{q^2}`$.
###### Lemma 2.1.
1. $`\nu _1\{1,ϵ_2\};`$
2. $`ϵ_2q;`$
3. $`ϵ_2`$ is a power of $`p`$ whenever $`ϵ_2>2.`$
###### Proof.
For (1), see \[36, Prop. 2.1\]. For (2), suppose that $`ϵ_2>q`$, then $`ϵ_2=q+1`$ as $`ϵ_2d`$ and $`dq+1`$ by (2.1). Then, by the $`p`$-adic criterion \[36, Cor. 1.9\], $`q`$ would be a $`\mathrm{\Sigma }_1`$-order, a contradiction. For (3), see \[16, Prop. 2\]. ∎
The following result is a complement to \[30, Prop. 3.7\], \[22, Thm. 6.1\], and \[21, Prop. 6\].
###### Proposition 2.2.
For a plane $`𝐅_{q^2}`$-maximal curve $`𝒳`$ of degree $`d3`$, the following conditions are equivalent:
1. $`d=q+1;`$
2. $`𝒳`$ is $`𝐅_{q^2}`$-isomorphic to the Hermitian of equation (1.1);
3. $`ϵ_2=q;`$
4. $`\nu _1=q;`$
5. $`j_2(P)=q+1`$ for each $`P𝒳(𝐅_{q^2});`$
6. $`\nu _1>1`$; i.e, $`\mathrm{\Sigma }_1`$ is $`𝐅_{q^2}`$-Frobenius non-classical.
###### Proof.
$`\text{(1)}\text{(2)}:`$ Since the genus of a non–singular plane curve of degree $`d`$ is $`q(q1)/2`$, part (2) follows from .
$`\text{(2)}\text{(3)}:`$ This is well known property of the Hermitian curve; see e.g. \[10, p. 105\] or .
$`\text{(3)}\text{(4)}:`$ If $`q=2`$, then from $`dϵ_2=q`$ and (2.1), either $`d=2`$ or $`d=3`$. By hypothesis, $`d=3`$ can only occur, and so, by parts (1) and (2), $`𝒳`$ is $`𝐅_4`$-isomorphic to the Hermitian curve $`X^3+Y^3+Z^3=0`$. Then $`\nu _1=ϵ_2=2`$; see loc. cit.
Let $`q3`$. By Lemma 2.1(1), $`\nu _1\{1,q\}`$. Suppose that $`\nu _1=1`$ and let $`S_1`$ be the $`𝐅_{q^2}`$-Frobenius divisor associated to $`\mathrm{\Sigma }_1`$. Then \[36, Thm. 2.13\]
$$\mathrm{deg}(S_1)=(2g2)+(q^2+2)d2\mathrm{\#}𝒳(𝐅_{q^2})=2(q+1)^2+2q(2g2)$$
so that $`((2q1)d(q^2+2q+1))(d2)0`$, and hence
(2.2)
$$dF(q):=(q^2+2q+1)/(2q1).$$
Thus, as $`dϵ_2=q`$, we would have $`q^23q10`$ and hence $`q3`$. If $`q=3`$, from (2.2) we have that $`d=3`$; this contradicts \[30, Cor. 2.2\] (cf. Remark 2.5(ii)).
$`\text{(4)}\text{(5)}:`$ By \[36, Cor. 2.6\], $`\nu _1j_2(P)1`$ for any $`P𝒳(𝐅_{q^2})`$. Then part (5) follows as $`j_2(P)d`$ and $`dq+1`$ by (2.1).
$`\text{(5)}\text{(6)}:`$ Suppose that $`\nu _1=1`$. Then, by \[36, Prop. 2.4(a)\], $`v_P(S_1)q+1`$ for any $`P𝒳(𝐅_{q^2})`$. Therefore
$$\mathrm{deg}(S_1)=(2g2)+(q^2+2)d(q+1)\mathrm{\#}𝒳(𝐅_{q^2})=(q+1)^3+(q+1)q(2g2),$$
a contradiction as $`3dq+1`$.
$`\text{(6)}\text{(1)}:`$ From \[21, Thm. 1\] and the $`𝐅_{q^2}`$-maximality of $`𝒳`$ we have
$$\mathrm{\#}𝒳(𝐅_{q^2})=d(q^21)(2g2)=(1+q)^2+q(2g2).$$
Since $`2g2=d(d3)`$ and $`d>1`$, part (1) follows. ∎
###### Corollary 2.3.
Let $`d3`$ be the degree of a plane $`𝐅_{q^2}`$-maximal curve. Then either $`d=q+1`$ or
$$dd_2(q):=\{\begin{array}{cc}(q+2)/2\hfill & \text{if }q4\text{ and }q3,5\text{,}\hfill \\ 3\hfill & \text{if }q=3\text{,}\hfill \\ 4\hfill & \text{if }q=5\text{.}\hfill \end{array}$$
In particular, for $`q3,5`$, a $`𝐅_{q^2}`$-maximal curve has no non-singular plane model if its genus is assumed to belong to the interval $`(q(q2)/8,q(q2)/4]`$, for $`q`$ even, and $`((q1)(q3)/8,(q1)^2/4]`$, for $`q`$ odd.
###### Proof.
The statement on the genus follows immediately from the upper bound on $`d`$. By (2.1) we have that $`dq+1`$. If $`d<q+1`$, then $`q3`$ and from Proposition 2.2 $`\mathrm{\Sigma }_1`$ is $`𝐅_{q^2}`$-Frobenius classical. In particular, (2.2) holds true; i.e., we have $`dF(q)`$. It is easy to see that $`F(q)<(q+3)/2`$ for $`q>5`$ and that $`F(4)=25/7`$. Moreover, $`F(3)=16/5`$ and $`F(5)=4`$, and the result follows. ∎
###### Remark 2.4.
Let $`d`$ be the degree of a plane $`𝐅_{q^2}`$-maximal curve of degree $`d`$ and assume that $`3dd_2(q)`$.
(i) If $`q`$ is odd, then the $`𝐅_{q^2}`$-maximal curve of equation
$$X^{(q+1)/2}+Y^{(q+1)/2}+Z^{(q+1)/2}=0,$$
shows that the upper bound $`d_2(q)=(q+1)/2`$ in Corollary 2.3 is the best possible as far as $`q3,5`$. We notice that this curve is the unique $`𝐅_{q^2}`$-maximal plane curve (up to $`𝐅_{q^2}`$-isomorphism) of degree $`(q+1)/2`$ provided that $`q11`$; see .
(ii) From results of Deuring, Tate and Watherhouse (see e.g. \[37, Thm. 4\]), there exist elliptic $`𝐅_{q^2}`$-maximal curves for any $`q`$. In particular, $`d_2(q)=3`$ is sharp for $`q=3`$.
(iii) From \[33, Sec. 4\], there exists a plane quartic $`𝐅_{25}`$-maximal; so $`d_2(q)=4`$ is sharp for $`q=5`$.
(iv) By part (ii), $`d_2(q)=3`$ is sharp for $`q=4`$. For $`q8`$, $`q`$ even, no information is currently available to asses how good the bound $`d_2(q)=(q+2)/2`$ is.
We go on to look for an upper bound for the degree $`d`$ of a $`𝐅_{q^2}`$-maximal curve satisfying the condition $`d<(q+2)/2`$. Our approach is inspired on \[6, Sec. 3\] where the $`𝐅_{q^2}`$-Frobenius divisor $`S_2`$ associated with the linear series $`\mathrm{\Sigma }_2`$ cut out on $`𝒳`$ by conics was employed to obtain upper bounds for the number of $`𝐅_{q^2}`$-rational points of plane curves. In fact, if we use $`\mathrm{\Sigma }_2`$ instead of $`\mathrm{\Sigma }_1`$, we can get better results for values of $`d`$ ranging in certain intervals depending on $`q`$. This was pointed out at the first time in .
In order to compute the $`\mathrm{\Sigma }_2`$-orders of a plane $`𝐅_{q^2}`$-maximal curve $`𝒳`$, one needs to know whether $`𝒳`$ is classical or not with respect to $`\mathrm{\Sigma }_1`$. This gives the motivation to Proposition 2.6. The following remark will be useful in the proof.
###### Remark 2.5.
(i) If a projective, geometrically irreducible, non-singular, algebraic curve defined over a field of characteristic $`p>0`$ admits a linear series $`\mathrm{\Sigma }`$ of degree $`D`$, then $`\mathrm{\Sigma }`$ is classical provided that $`p>D`$; see \[36, Cor. 1.8\].
(ii) If a non-singular plane curve of degree $`D`$ defined over a field of characteristic $`p>`$ is non-classical with respect to the linear series cut out by lines, then $`D1(modp)`$; see \[30, Cor. 2.2\], and \[23, Cor. 2.4\].
###### Proposition 2.6.
Let $`𝒳`$ be a plane $`𝐅_{q^2}`$-maximal curve of degree $`d`$ such that $`3dd_2(q)`$, where $`d_2(q)`$ is as in Corollary 2.3. Then the linear series $`\mathrm{\Sigma }_1`$ on $`𝒳`$ is classical provided that one of the following conditions holds:
1. $`pd`$ or $`d1(modp);`$
2. $`q=4,8,16,32;`$
3. $`p3`$ and either $`q=p`$ or $`q=p^2;`$
4. $`p=2`$, $`q64`$, and either $`d4`$, or $`dd_3(q):=q/41`$ for $`q=64,128,256`$, or $`dd_3(q):=q/4`$ for $`q512;`$
5. $`p3`$, $`q=p^v`$ with $`v3`$, and $`dd_3(q):=q/pp+2.`$
###### Proof.
If (i) holds, then $`\mathrm{\Sigma }_1`$ is classical by Remark 2.5. For $`q=p`$, the hypothesis on $`d`$ yields $`p3`$ and hence $`d(p+1)/2<p`$. Thus $`\mathrm{\Sigma }_1`$ is classical by Remark 2.5(i). Note that the following computations will provide another proof of this fact.
For the rest of the proof we assume $`\mathrm{\Sigma }_1`$ to be non-classical, and we show that no one of the conditions (i),…,(v) holds. From Lemma 2.1(3), $`ϵ_2M`$, where $`M=4`$ for $`p=2`$, and $`M=p`$ for $`p3`$. Also, $`\nu _1=1`$ by Proposition 2.2. Therefore, as $`j_2(P)ϵ_2`$ for each $`P𝒳`$ \[36, p. 5\]. From \[36, Prop. 2.4(a)\] we have that $`v_P(S_1)M`$ for each $`P𝒳(𝐅_{q^2})`$, where as before $`S_1`$ denotes the $`𝐅_{q^2}`$-Frobenius divisor associated to $`\mathrm{\Sigma }_1`$. Thus,
$$\mathrm{deg}(S_1)=(2g2)+(q^2+2)dM\mathrm{\#}𝒳(𝐅_{q^2})=M(q+1)^2+Mq(2g2),$$
or, equivalently,
$$(Mq1)d^2(q^2+3Mq1)d+M(q+1)^20.$$
On the other hand, the discriminant of the above quadratic polynomial in $`d`$ is
$$\mathrm{\Delta }_M(q):=q^4(4M^26M)q^3+(M^2+4M2)q^2(4M^22M)q+4M+1,$$
and hence $`\mathrm{\Delta }_M(q)<0`$ if and only if either $`q=4,8,16,32`$ and $`M=4`$, or $`q=p,p^2`$ and $`M=p3`$. For these $`q`$’s, the above inequality cannot actually hold, and hence $`\mathrm{\Sigma }_1`$ must be classical. Furthermore, if $`\mathrm{\Delta }_M(q)0`$, then
$$F^{}(M,q):=\frac{q^2+3Mq1\sqrt{\mathrm{\Delta }_M(q)}}{2(Mq1)}dF(M,q):=\frac{q^2+3Mq1+\sqrt{\mathrm{\Delta }_M(q)}}{2(Mq1)}.$$
It is easy to check that $`F^{}(4,q)>4`$, $`F(4,q)<q/41`$ for $`q=64,128,256`$, and that $`F(4,q)<q/4`$ for $`q512`$; hence if (iv) holds, then $`\mathrm{\Sigma }_1`$ must be classical. Let $`p3`$. If $`q/pp+2dq/p`$, then $`\mathrm{\Sigma }_1`$ must be classical by (i). So we can suppose that $`dq/p+1`$. It turns out that $`F(p,q)<q/p+1`$ and hence the result follows when (v) is assumed to be true. ∎
###### Remark 2.7.
For $`q=p^3`$, $`p3`$, the bound $`d_3(q)`$ in Proposition 2.6 is sharp. Indeed, there exists a plane $`𝐅_{p^6}`$-maximal curve of degree $`p^2p+1`$ which is non-classical for $`\mathrm{\Sigma }_1`$; see Corollary 3.3 and Remark 3.11.
###### Corollary 2.8.
Let $`𝒳`$ be a plane $`𝐅_{q^2}`$-maximal curve of degree $`d`$ as in Proposition 2.6. Assume that $`𝒳`$ is non-classical for $`\mathrm{\Sigma }_1`$ and let $`ϵ_2`$ be the order of contact of $`𝒳`$ with the tangent at a general point. Then
1. $`q64`$ if $`p=2`$, and $`qp^3`$ for $`p3;`$
2. $`ϵ_2^2q/p.`$
###### Proof.
Part (1) follows from Proposition 2.6(ii)(iii). To prove (2), we first note that $`ϵ_2<q`$ (cf. Proposition 2.2), and that $`ϵ_2`$ is a power of $`p`$ (see Lemma 2.1(3)). Now, with the same notation as in the proof of the previous proposition, we get $`dF(M,q)`$ with $`M=ϵ_2`$. So $`dq/ϵ_2`$. Furthermore, $`dϵ_2`$ and so $`dϵ_2+1`$ by Remark 2.5(ii). Hence $`ϵ_2+1q/ϵ_2`$ and part (2) follows. ∎
###### Remark 2.9.
The example in Remark 2.7 shows that Corollary 2.8(1) is sharp for $`p3`$.
Our next step is to show that every plane $`𝐅_{q^2}`$-maximal curve which is classical for $`\mathrm{\Sigma }_1`$ contains an $`𝐅_{q^2}`$-rational point different from its inflexions.
###### Lemma 2.10.
Let $`𝒳`$ be a $`𝐅_{q^2}`$-maximal curve of degree $`d3`$ which is classical with respect to $`\mathrm{\Sigma }_1`$. Then there exists $`P_0𝒳(𝐅_{q^2})`$ whose $`(\mathrm{\Sigma }_1,P_0)`$-orders are $`0,1,2.`$
###### Proof.
Let $`R_1`$ be the ramification divisor associated to $`\mathrm{\Sigma }_1`$ and suppose that $`j_2(P)3`$ for each $`P𝒳(𝐅_{q^2})`$. Then from \[36, p. 12\],
$$\mathrm{deg}(R_1)=3(2g2)+3d\mathrm{\#}𝒳(𝐅_{q^2})=(q+1)^2+q(2g2)$$
which is a contradiction as $`g1`$ and $`3d<q+1`$. ∎
It should be noticed that Lemma 2.10 improves a previous result, see \[6, Cor. 3.2\].
We are in a position to establish some useful properties of the linear series $`\mathrm{\Sigma }_2`$ cut out by conics of $`𝐏^2(\overline{𝐅}_{q^2})`$ on plane $`𝐅_{q^2}`$-maximal curve $`𝒳`$ of degree $`d3`$. Since $`𝒳`$ is non-singular, $`\mathrm{\Sigma }_2=2\mathrm{\Sigma }_1`$. Taking into account $`d3`$, we see that $`\mathrm{\Sigma }_2`$ is a 5-dimensional linear series of degree $`2d`$.
###### Lemma 2.11.
Let $`d`$ be the degree of a plane $`𝐅_{q^2}`$-maximal curve $`𝒳`$. Let $`q=8`$ or $`q11`$, and suppose that
$$d_4(q):=\frac{2q^2+15q20+\sqrt{4q^440q^3+145q^2300q+600}}{10(q2)}<dd_2(q),$$
where $`d_2(q)`$ is as in Corollary 2.3. Then the orders (resp. $`𝐅_{q^2}`$-Frobenius orders) of $`\mathrm{\Sigma }_2`$ are $`0,1,2,3,4,ϵ`$ (resp. $`0,1,2,3,ϵ`$) with $`5ϵq`$. Furthermore, $`p`$ divides $`ϵ.`$
###### Proof.
By some computations we obtain that $`d_4(q)`$ is bigger than $`d_3(q)`$ in Proposition 2.6. So the curve $`𝒳`$ is classical for $`\mathrm{\Sigma }_1`$. Let $`P_0𝒳(𝐅_{q^2})`$ be as in Lemma 2.10. Then the $`(\mathrm{\Sigma }_2,P_0)`$-orders are $`0,1,2,3,4`$ and $`j_0`$ with $`5j_02d`$ (cf. \[16, p. 464\]). Therefore, the $`\mathrm{\Sigma }_2`$-orders are $`0,1,2,3,4`$ and $`ϵ`$ with $`5ϵj_0`$. Since $`j_02d`$, from Corollary 2.3, $`ϵq+2`$, and hence $`ϵq`$ by the $`p`$-adic criterion \[36, Cor. 1.9\]. Also, the $`𝐅_{q^2}`$-Frobenius orders of $`\mathrm{\Sigma }_2`$ are $`0,1,2,3`$ and $`\nu `$ with $`\nu \{4,ϵ\}`$; see \[36, Prop. 2.1, Cor. 2.6\]. Suppose that $`\nu =4`$ and keep up $`S_2`$ to denote the $`𝐅_{q^2}`$-Frobenius divisor associated to $`\mathrm{\Sigma }_2`$. Then \[36, Thm. 2.13\]
$$\mathrm{deg}(S_2)=10(2g2)+(q^2+5)2d5\mathrm{\#}𝒳(𝐅_{q^2})=5(q+1)^2+5q(2g2)$$
or equivalently
$$(5q10)d^2(2q^2+15q20)d+5(q+1)^20.$$
The discriminant of this equation is $`4q^240q^3+145q^2300q+600`$ and it is positive for any $`q`$. Since $`d_4(q)`$ is the biggest root of the quadratic polynomial in $`d`$ above, $`dd_4(q)`$, a contradiction. Finally, $`p`$ divides $`ϵ`$ by \[12, Cor. 3\]. ∎
Let $`d_4(q)`$ be as in Lemma 2.11 and for $`q=p^v`$, $`v2`$, let $`d_4(p,q)`$ denote the function
$$\frac{2q^2+3(5\frac{1}{p})q8+\sqrt{4q^48(5\frac{1}{p})q^3+(113\frac{50}{p}+\frac{9}{p^2})q^24(25\frac{17}{p})q+184}}{2(5\frac{1}{p})q12}.$$
###### Theorem 2.12.
Let $`d`$ be the degree of a plane $`𝐅_{q^2}`$-maximal curve $`𝒳`$. Suppose that $`3d<q+1`$ and that $`q=8`$ or $`q11`$. Then
$$dd_5(q):=\{\begin{array}{cc}d_4(q)\hfill & \text{if }q=p\text{,}\hfill \\ d_4(p,q)\hfill & \text{if }q=p^v\text{}v2\text{,}\hfill \end{array}\text{or}d=(q+2)/2.$$
###### Proof.
Suppose that $`d>d_5(q)`$. By means of some computations, $`d_5(p,q)>d_4(q)`$ and hence Lemma 2.11 holds true. With the same notation as in the proof of that lemma, we can then use the following two facts: $`ϵ=\nu q`$, and $`p|ϵ`$. Actually, we will improve the latter one.
###### Claim 1.
$`ϵ`$ is a power of $`p.`$
Indeed, by $`p|ϵ`$ and the $`p`$-adic criterion \[36, Cor. 1.9\], a necessary and sufficient condition for $`ϵ`$ not to be a power of $`p`$ is that $`p\{2,3\}`$ and $`ϵ=6`$. If this occurs, one can argue as in the previous proof and obtain the following result:
$$(5q2)d^2(q^2+15q31)d+5(q+1)^20.$$
From this,
$$dG(q):=\frac{q^2+15q31+\sqrt{q^470q^3+203q^2550q+1201}}{2(5q12)},$$
which is a contradiction as $`G(q)<d_5(q)`$.
###### Claim 2.
$`ϵ=q.`$
The claim is certainly true for $`q=p`$. So, $`q=p^v`$, with $`v2`$. If $`ϵ<q`$, by Claim 1 we have $`ϵq/p`$. Thus, this fact together with
$$\mathrm{deg}(S_2)=(6+\nu )(2g2)+(q^2+5)2d5\mathrm{\#}𝒳(𝐅_{q^2})=5(q+1)^2+5q(2g2),$$
would yield
$$(5qq/p6)d^2(2q^2+15q3q/p8)d+5(q+1)^20,$$
and hence $`dd_4(p,q)`$, a contradiction.
Now from Claim 2 and \[36, Cor. 2.6\], we have
$$q=ϵ=\nu j_5(P_0)12d1,$$
and Theorem 2.12 follows from Corollary 2.3. ∎
## 3. Maximal Hurwitz’s curves
In this section we give a necessary and sufficient condition for $`q`$ in order that the Hurwitz curve $`𝒳_n`$ defined by Eq. (1.2) be $`𝐅_{q^2}`$-maximal.
###### Theorem 3.1.
The curve $`𝒳_n`$ is $`𝐅_{q^2}`$-maximal if and only if (1.3) holds.
We first prove two lemmas.
###### Lemma 3.2.
(\[4, p. 210\]) The Hurwitz curve $`𝒳_n`$ is $`𝐅_p`$-covered by the Fermat curve $`_{n^2n+1}`$
$$U^{n^2n+1}+V^{n^2n+1}+W^{n^2n+1}=0.$$
###### Proof.
Let $`u=U/W`$ and $`v:=V/W`$. Then the image of the morphism $`(u:v:1)(x:y:1)=(u^nv^1:uv^{n1}:1)`$ is the curve defined by $`x^ny+y^n+x=0`$. This proves the lemma. ∎
###### Corollary 3.3.
Suppose that (1.3) holds. Then both curves $`𝒳_n`$ and $`_{n^2n+1}`$ are $`𝐅_{q^2}`$-covered by the Hermitian curve of equation (1.1). In particular, both are $`𝐅_{q^2}`$-maximal.
###### Proof.
If (1.3) holds, it is clear that $`_{n^2n+1}`$ is $`𝐅_{q^2}`$-covered by the Hermitian curve. This property extends to $`𝒳_n`$ via the previous lemma. For both curves, the $`𝐅_{q^2}`$-maximality now follows from \[27, Prop. 6\]. ∎
###### Lemma 3.4.
(\[5, p. 5249\]) The Weierstrass semigroup of $`𝒳_n`$ at the point $`(0:1:0)`$ is generated by the set $`S:=\{s(n1)+1:s=1,\mathrm{},n\}`$.
###### Proof.
Let $`P_0:=(1:0:0)`$, $`P_1=(0:1:0)`$, and $`P_2=(0:0:1)`$. Then $`\mathrm{div}(x)=nP_2(n1)P_1P_0`$ and $`\mathrm{div}(y)=(n1)P_0+P_2nP_1`$ so that
$$\mathrm{div}(x^{s1}y)=((n(s1)+1)P_2+(ns)P_0(s(n1)+1)P_1.$$
This shows that $`S`$ is contained in the Weierstrass semigroup $`H(P_1)`$ at $`P_1`$. In particular, $`H(P_1)S`$. Since $`\mathrm{\#}(𝐍_0S)=n(n1)/2`$ (see ), the result follows. ∎
Proof of Theorem 3.1. If (1.3) holds, then $`𝒳_n`$ is $`𝐅_{q^2}`$-maximal by Corollary 3.3. Conversely, assume that $`𝒳_n`$ is $`𝐅_{q^2}`$-maximal. Then $`(q+1)P_1(q+1)P_2`$ \[32, Lemma 1\], and the case $`s=n`$ in the proof of Lemma 3.4 gives $`(n^2n+1)P_1(n^2n+1)P_2`$. Therefore $`d:=\mathrm{gcd}(n^2n+1,q+1)`$ belongs to $`H(P_1)`$. According to Lemma 3.4 we have that $`d=A(n1)+B`$ with $`AB1`$. Now, there exists $`C1`$ such that $`(A(n1)+B)C=n^2n+1`$ and so $`BC=D(n1)+1`$ for some $`D0`$. Therefore, $`AD(n1)+A+BD=Bn`$. We claim that $`D=0`$, otherwise the left side of the last equality would be bigger than $`Bn`$. Then $`B=C=1`$ and so $`A=n`$; i.e., $`d=n^2n+1`$ and the proof is complete.
###### Corollary 3.5.
The curve $`_{n^2n+1}`$ in Lemma 3.2 is $`𝐅_{q^2}`$-maximal if and only if (1.3) holds.
###### Proof.
If (1.3) is satisfied, the result follows from Corollary 3.3. Now if $`_{n^2n+1}`$ is $`𝐅_{q^2}`$-maximal, then $`𝒳_n`$ is also $`𝐅_{q^2}`$-maximal by Lemma 3.2 and \[27, Prop. 6\]. Then the corollary follows from Theorem 3.1. ∎
###### Remark 3.6.
For a given positive integer $`n`$, we are led to look for a power $`q`$ of a prime $`p`$ such that $`q+10(modm)`$ with $`m=n^2n+1`$. Since $`m0(modp)`$, and $`p0(modm)`$, a necessary and sufficient condition for $`q`$ to have the requested property (1.3) is $`px(modm)`$, where $`x`$ is a solution of the congruence $`X^w+10(modm)`$, and $`w`$ is defined by $`q=p^{\varphi (m)v+w}`$, $`w\{1,2,\mathrm{},\varphi (m)1\}`$; here $`\varphi `$ denotes the Euler function.
Regarding specific examples, we notice that Carbonne and Henocq \[4, Lemmes 3.3, 3.6\] pointed out that $`𝒳_n`$ is $`𝐅_{q^2}`$-maximal in the following cases:
1. $`n=3`$, $`q=p^{6v+3}`$ and $`p3,5(mod7)`$;
2. $`n=4`$, $`q=p^{12v+6}`$ and $`p2,6,7,11(mod13)`$.
By using Theorem 3.1 and Remark 3.6 we have the following result.
###### Corollary 3.7.
1. The curve $`𝒳_2`$ is $`𝐅_{q^2}`$-maximal if and only if $`q=p^{2v+1}`$ and $`p2(mod3);`$
2. The curve $`𝒳_3`$ is $`𝐅_{q^2}`$-maximal if and only if either $`q=p^{6v+1}`$ and $`p6(mod7)`$, or $`q=p^{6v+3}`$ and $`p3,5,6(mod7)`$, or $`q=p^{6v+5}`$ and $`p6(mod7);`$
3. The curve $`𝒳_4`$ is $`𝐅_{q^2}`$-maximal if and only if either $`q=p^{12v+1}`$ and $`p12(mod13)`$, or $`q=p^{12v+2}`$ and $`p5,8`$, or $`q=p^{12v+3}`$ and $`p4,10,12(mod13)`$, or $`q=p^{12v+5}`$ and $`p12(mod13)`$, or $`q=p^{12v+6}`$ and $`p2,5,6,7,8,11(mod13)`$, or $`q=p^{12v+7}`$ and $`p12(mod13)`$, or $`q=p^{12v+9}`$ and $`p4,10,12(mod13)`$, or $`q=p^{12v+11}`$ and $`p12(mod13).`$
###### Corollary 3.8.
Let $`n`$ be a positive integer, $`m:=n^2n+1`$ and $`p`$ a prime.
1. If $`n=p^e`$ with $`e1`$, then the curve $`𝒳_n`$ is $`𝐅_{q^2}`$-maximal with $`q=p^{\varphi (m)v+3e}`$.
2. Let $`p3(mod4)`$ and $`n0,1(modp)`$ such that $`m`$ is prime and that $`m3(mod4)`$. Then $`𝒳_n`$ is $`𝐅_{q^2}`$-maximal with $`q=p^{(m1)v+(m1)/2}`$.
###### Proof.
Part (1) follows from the identity $`p^{3e}+1=(p^e+1)(p^{2e}p^e+1)`$ and Theorem 3.1.
To show (2), it is enough to check that $`p^{(m1)/2}+10(modm)`$. Recall that the Legendre symbol $`(a/p)`$ is defined by:
$$(a/p)=\{\begin{array}{cc}1\hfill & \text{if }x^2a(modp)\text{ has two solutions in }𝐙_p\text{ ,}\hfill \\ 1\hfill & \text{if }x^2a(modp)\text{ has no solution in }𝐙_p\text{ .}\hfill \end{array}$$
In our case, since $`m1(modp)`$, $`(m/p)=1`$. By the quadratic reciprocity low
$$(m/p)(p/m)=(1)^{((m1)/2)((p1)/2)},$$
from $`(m/p)=1`$ and $`m3(mod4)`$ we get $`(p/m)=(1)^{(p1)/2}`$. Now, as $`p3(mod4)`$, we have that $`(p/m)=1`$. In other words, $`p`$ viewed as an element in $`𝐅_m`$ is a non-square in $`𝐅_m`$. Since $`1`$ is as well a non-square in $`𝐅_m`$, it follows then that $`p(1)u^2(modm)`$ with $`u𝐙`$ such that $`u0(modm)`$. Hence $`p^{(m1)/2}(1)(modm)`$ as, in particular, $`m`$ is odd and as $`u^{m1}1(modm)`$. ∎
###### Remark 3.9.
The hypothesis $`m3(mod4)`$ in the above corollary cannot be relaxed. In fact, for $`n=4`$ we have $`m=13`$ but, according with Corollary 3.7, $`𝒳_4`$ is no $`𝐅_{3^6}`$-maximal.
###### Remark 3.10.
Let us assume the hypothesis in Corollary 3.8(2) with $`m`$ not necessarily prime. In this case, to study the congruence in (1.3) we have to consider the multiplicative group $`\mathrm{\Phi }_m`$ of the units in $`𝐙_m`$. This group has order $`\varphi (m)`$, and $`p\mathrm{\Phi }_m`$ since $`m1(modp)`$. Now suppose that $`p`$, as an element of $`\mathrm{\Phi }_m`$, has even order $`2i`$. Then $`p^{2i}1(modm)`$ and hence $`(p^i+1)(p^i1)0(modm)`$. Since $`p`$ has order greater than $`i`$, we have that $`p^i10(modm)`$ unless both $`p^i+1`$ and $`p^i1`$ are zero divisors in $`𝐙_m`$. If we assume that this does not happen, then equivalence (1.3) follows for $`q=p^{\varphi (m)v+i}`$.
###### Remark 3.11.
Let $`p`$ be a prime, $`n:=p^eu`$ with $`e1`$ and $`\mathrm{gcd}(p,u)=1`$. Assume $`e2`$ if $`p=2`$. Then the Hurwitz curve $`𝒳_n`$ as well as the curve $`_{n^2n+1}`$ are non-classical with respect to $`\mathrm{\Sigma }_1`$. It is easy to see that $`0,1`$ and $`p^e`$ are their $`\mathrm{\Sigma }_1`$-orders.
## 4. On the maximality of generalized Hurwitz curves
In this section we investigate the $`𝐅_{q^2}`$-maximality of the non-singular model of the so-called generalized Hurwitz curve $`𝒳_{n,\mathrm{}}`$ of equation
$$X^nY^{\mathrm{}}+Y^nZ^{\mathrm{}}+Z^nX^{\mathrm{}}=0,$$
where $`n\mathrm{}2`$ and $`p=\mathrm{char}(𝐅_{q^2})`$ does not divide $`Q(n,\mathrm{}):=n^2n\mathrm{}+\mathrm{}^2`$. The singular points of $`𝒳_{n,\mathrm{}}`$ are $`P_0:=(1:0:0)`$, $`P_1=(0:1:0)`$, and $`P_2=(0:0:1)`$; each of them is unibranched with $`\delta `$-invariant equal to $`(n\mathrm{}n\mathrm{}+\mathrm{gcd}(n,\mathrm{}))/2`$. Therefore its genus $`g`$ (cf. \[3, Sec. 4\] and \[2, Example 4.5\]) is equal to
$$g=\frac{n^2n\mathrm{}+\mathrm{}^2+23\mathrm{g}\mathrm{c}\mathrm{d}(n,\mathrm{})}{2}.$$
First we generalize Lemma 3.2.
###### Lemma 4.1.
The curve $`𝒳_{n,\mathrm{}}`$ is $`𝐅_{q^2}`$-covered by the Fermat curve $`_{n^2n\mathrm{}+\mathrm{}^2}`$
$$U^{n^2n\mathrm{}+\mathrm{}^2}+V^{n^2n\mathrm{}+\mathrm{}^2}+W^{n^2n\mathrm{}+\mathrm{}^2}=0.$$
###### Proof.
The curve $`𝒳_{n,\mathrm{}}`$ is $`𝐅_{q^2}`$-covered by $`_{n^2n\mathrm{}+\mathrm{}^2}`$ via the morphism $`(u:v:1)(x:y:1):=(u^nv^m:u^mv^{nm}:1)`$, where $`u:=U/W`$ and $`v:=V/W`$. ∎
From this lemma and \[27, Prop. 6\] we have the following.
###### Corollary 4.2.
The curve $`_{n^2n\mathrm{}+\mathrm{}^2}`$ in the above lemma and the $`𝐅_{q^2}`$-non-singular model of $`𝒳_{n,\mathrm{}}`$ are $`𝐅_{q^2}`$-maximal provided that
(4.1)
$$n^2n\mathrm{}+\mathrm{}^20(mod(q+1)).$$
Now, we generalize Lemma 3.4 for any two coprime $`n`$ and $`\mathrm{}`$. For $`0i2`$, let $`Q_i`$ be the unique point in the non-singular model of $`𝒳_{n,\mathrm{}}`$ lying over $`P_i`$.
###### Lemma 4.3.
Suppose that $`\mathrm{gcd}(n,\mathrm{})=1`$. Then the Weierstrass semigroup $`H(Q_1)`$ at $`Q_1`$ is given by
(4.2)
$$\{(n\mathrm{})s+nt:s,t𝐙;t0\frac{\mathrm{}}{n}ts\frac{n\mathrm{}}{\mathrm{}}t\}.$$
###### Proof.
Let $`x:=X/Z,y:=Y/Z`$. It is not difficult to see that $`\mathrm{div}(x)=nQ_2(n\mathrm{})Q_1\mathrm{}Q_0`$ and $`\mathrm{div}(y)=(n\mathrm{})Q_0+\mathrm{}Q_2nQ_1`$. Hence, for $`s,t𝐙`$,
$$\mathrm{div}(x^sy^t)=(ns+\mathrm{}t)Q_2+(\mathrm{}s+(n\mathrm{})t)Q_0((n\mathrm{})s+nt)Q_1,$$
and hence $`(n\mathrm{})s+ntH(Q_1)`$ provided that $`ns+\mathrm{}t0`$ and $`\mathrm{}s+(n\mathrm{})t0`$. Let $`H`$ denote the set introduced in (4.2). Then $`HH(Q_1)`$, and it is easily checked that $`H`$ is a semigroup. By means of some computations we see that $`\mathrm{\#}(𝐍H)=(n^2n\mathrm{}+\mathrm{}^21)/2`$, whence $`H=H(Q_1)`$ follows. ∎
###### Remark 4.4.
The above Weierstrass semigroup $`H(Q_1)`$ was computed for $`\mathrm{}=n1`$, and $`(n,\mathrm{})=(5,2)`$ in .
We are able to generalize Theorem 3.1 for certain curves $`𝒳_{n,\mathrm{}}`$.
###### Theorem 4.5.
Assume that $`\mathrm{gcd}(n,\mathrm{})=1`$ and that $`Q:=Q(n,\mathrm{})=n^2n\mathrm{}+\mathrm{}^2`$ is prime. Then $`𝒳_{n,\mathrm{}}`$ is $`𝐅_{q^2}`$-maximal if and only if $`(\text{4.1})`$ holds.
###### Proof.
The “if” part follows from Corollary 4.2 and here we do not use the hypothesis that $`Q`$ is prime. For the “only if” part, we first notice that each $`Q_i`$ is $`𝐅_{q^2}`$-rational. Now the case $`s=nm`$ and $`t=m`$ in the proof of Lemma 4.3 gives $`QQ_2QQ_1`$. Therefore $`d=\mathrm{gcd}(Q,q+1)H(Q_1)`$ because $`(q+1)Q_1(q+1)Q_2`$ \[32, lemma1\]. As $`1H(Q_1)`$ and $`Q`$ is prime, the result follows. ∎
###### Corollary 4.6.
Let $`n`$, $`\mathrm{}`$ and $`Q`$ be as in Theorem 4.5. Then the curve $`_{n^2n\mathrm{}+\mathrm{}^2}`$ in Lemma 4.1 is $`𝐅_{q^2}`$-maximal if and only if (4.1) holds.
###### Proof.
Similar to the proof of Corollary 3.5. ∎
###### Remark 4.7.
There are infinitely many $`n,\mathrm{}`$ with $`n>\mathrm{}1`$ such that $`Q(n,\mathrm{})`$ is prime. In fact, for a prime $`p^{}`$ such that $`p^{}1(mod6)`$, there exists such $`n`$ and $`\mathrm{}`$ so that $`p^{}=Q(n,\mathrm{})`$; see \[3, Remarque 4\]. |
warning/0003/hep-ph0003099.html | ar5iv | text | # Neutrino scattering on polarized electron target as a test of neutrino magnetic moment
## Figure captions
Figure 1. Kinematics of the neutrino scattering off the polarized electron.
Figure 2. The ratio of the total recoil electron spectrum and the weak one (Eq. 10) for different values of the polarization parameter $`|\stackrel{}{\xi }_e|\mathrm{cos}\theta _\xi =0,0.9,1`$ and for the fixed neutrino magnetic moment $`\mu _\nu =310^{13}\mu _\mathrm{B}`$. Tritium antineutrino emitter.
Figure 3. The ratio of the total recoil electron spectrum and the weak one (Eq. 10) for different values of the neutrino magnetic moment $`\mu _\nu =10^{12}\mu _\mathrm{B}`$, $`310^{13}\mu _\mathrm{B}`$, $`10^{13}\mu _\mathrm{B}`$ and the maximal electron polarization ($`|\stackrel{}{\xi }_e|\mathrm{cos}\theta _\xi =1`$). Tritium antineutrino emitter. |
warning/0003/gr-qc0003085.html | ar5iv | text | # 1 Introduction
## 1 Introduction
Fluids with negative pressures have become of utmost importance in modern cosmology. They were first considered in the context of the inflationary scenario in which the early universe has a very short period of accelerating expansion . This inflationary phase solves many problems of the standard model which are connected to specific choices of initial conditions, like the flatness and horizon problem. At the same time, the origin of the seeds of the large scale structures observed today in the universe has a natural explanation in the inflationary scenario, that considers them as quantum fluctuations of a scalar field in a de Sitter background. The inflationary phase must end up with a transition to the radiative phase of the standard model. In many aspects, the inflationary scenario is plagued with problems of arbitrary choice of parameters of the microphysics, but its great success with the above mentioned problems led the community to consider it as part of the standard scenario.
More recently, the observation of the supernova of high redshift led to the preliminary conclusion that the Universe today is in an accelerating expansion . This is a surprising result since there was little doubts that the Universe was desacelerating and that the desaceleration parameter $`q`$ was positive. If the results of the supernova observations are confirmed, the energy of the Universe today must be dominated by a fluid with negative pressure. It could be a cosmological constant, but other cases are not excluded, like a fluid of cosmic string or domain walls or some scalar field, with a peculiar kind of potential, called quintessence .
In many pratical cases, these fluids with negative pressure may be represented by a perfect fluid with a barotropic equation of state $`p=\alpha \rho `$, with $`\alpha <0`$. To be more precise, in order to have an accelerating universe the strong energy condition must be violated and $`\alpha <\frac{1}{3}`$. A fluid of cosmic string leads to $`\alpha =\frac{1}{3}`$, representing the limiting case between an accelerating and desaccelerating universe; $`\alpha =\frac{2}{3}`$ and $`\alpha =1`$ represent respectively a fluid of domain wall and a cosmological constant. The quintessence fluid only approximativelly can be expressed by a barotropic fluid with $`\alpha <0`$.
In perturbations in fluids with negative pressure were studied in the hydrodynamical representation. It was found that when the strong energy condition is violated, there are instabilities in the small wavelength limit. This result can be understood by remembering that in this limit the expansion of the universe plays no important role: the negative pressure at this limit acts in the same way as gravity and nothing can prevent the collapse. This can be easily seen in the case of a cosmic string fluid for which the density contrast behaves as $`\mathrm{\Delta }t^{1\pm \sqrt{1+\frac{n^2}{3}}}`$. When $`n\mathrm{}`$ divergences appear, leading to the instability of the background model. In the cosmological perturbation theory, we are generally interested on the unstable modes, but that are not divergent. From here on, instabilities will refer to these undesirable divergent modes.
This study of density perturbations for a fluid of cosmic string has been extended in , where a two-fluid model was considered, one of them being the string fluid and the other ordinary matter with positive pressure. Special attention was payed to the case of a closed spatial section. From the point of view of the background, such models have many interesting features, mainly conected to the horizon problem and to the age of the universe. However, the study of scalar fluctuations around this background shows the presence of instability in the small scale limit, as in the case of the one fluid model. On the other hand, the study of gravitational waves for these models reveals a very regular behaviour because gravitational waves are mainly connected with the scale factor behaviour, being quite insensitive to the matter representation. This indicates that the instabilities detected in could be due to the hydrodynamical representation, which could disappear in a more fundamental approach.
In fact, the hydrodynamic representation for fluids with negative pressure is frequently a very poor approximation. Negative pressures appear in situations involving phase transitions in a primordial universe (topological defects) or a fundamental self-interacting scalar field. The exact formulation involves consequently fields, and a representation using a fluid with a barotropic equation of state only in very special situations may be employed. The employment of a perfect fluid representation, mainly when fluids of negative pressure are involved, can be viewed as a practical simplification which in many situations gives the same results as those that could be obtained by employing a more fundamental field representation.
The main point is that the equivalence of a hydrodynamical representation with a more fundamental one is very restricted and it can lead to complete misleading results depending on the problem treated. The instability in the small wavelength limit quoted above is an example. Fluids with negative pressure should have a field representation where their main features could be retained. The representation we will investigate here involves the more reasonable coupling of gravity with a scalar field with a self interacting term.
The aim of this paper is to show that the scalar field representation can more conveniently retain the features we could expect from fluids with negative pressure, mainly for those that are interesting for cosmology, as the objects resulting from phase transitions, like cosmic string. We will concern mainly with the consequences of these different representations for an analysis of perturbations around a homogenous and isotropic expanding universe. We verify that when we use a field representation, with the same behaviour for the scale factor as the corresponding hydrodynamical representation, the instabilities present in the latter case are absent in the former one. This is due essentialy to the fact that when we pass from a hydrodynamical representation to a field representation of the fluid we also pass from an Euler’s type equation to a Klein-Gordon’s type equation, and there is no correspondence when the pressure is negative. On the other hand, for large scale perturbations, both approachs lead to the same results.
We begin by analysing a stiff matter fluid which can mimic a scalar field in a very simple way. As it is well known, a free scalar field minimally coupled to gravity reproduces the stiff matter equation of state. We will show in section $`2`$ that for the stiff matter and free scalar field models, the agreement between the results occurs not only at the background level but also at the perturbative level. In section $`3`$ we extend this analysis to a perfect fluid with arbitrary equation of state $`p=\alpha \rho `$: we determine the corresponding field representation, showing that, at perturbative level, the equivelence between these two approachs remains only when $`\alpha >0`$. In section $`4`$ we review briefly the results obtained in and we discuss the possibility of a field representation for this two fluid model. In section $`5`$, we review a model of variable cosmological constant that leads, from the point of view of the evolution of the scalar factor, to the same behaviour as a cosmic string fluid. In section $`6`$, we perform a perturbative analysis of the variable constant model, and show explicitly that they are free of instabilities, both for scalar perturbations and tensor perturbations. In section $`7`$, we discuss the results.
## 2 Free scalar field model
The most simple gravity model with a scalar field is described by the action
$$𝒮=\sqrt{g}[R\varphi ,_\alpha \varphi ,^\alpha ]d^4x.$$
(1)
It represents a free scalar field minimally coupled to gravity. The field equations obtained in accordance with the principle of least action by varying $`S`$ with respect to the dynamical variables are:
$$R_{\mu \nu }\frac{1}{2}g_{\mu \nu }R=\varphi _{;\mu }\varphi _{;\nu }\frac{1}{2}g_{\mu \nu }\varphi _{;\rho }\varphi ^{;\rho },$$
(2)
$$\mathrm{}\varphi =0.$$
(3)
In a Friedmann-Robertson-Walker flat space-time (FRW), the metric describing the four dimensional geometry is
$$ds^2=dt^2a^2(t)(dx^2+dy^2+dz^2),$$
(4)
and the field equations take the form
$`3({\displaystyle \frac{\dot{a}}{a}})^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\dot{\varphi }^2,`$ (5)
$`2{\displaystyle \frac{\ddot{a}}{a}}+({\displaystyle \frac{\dot{a}}{a}})^2`$ $`=`$ $`{\displaystyle \frac{\dot{\varphi }^2}{2}},`$ (6)
$`\ddot{\varphi }+3{\displaystyle \frac{\dot{a}}{a}}\dot{\varphi }`$ $`=`$ $`0,`$ (7)
where the overdot denotes the derivative with respect to the time coordinate $`t`$. The equations (5,6,7) are not independent due to the Bianchi identities.
On the other hand, in an universe filled with a perfect fluid we have the field equations
$`R_{\mu \nu }{\displaystyle \frac{1}{2}}g_{\mu \nu }R`$ $`=`$ $`8\pi GT_{\mu \nu },`$ (8)
$`T_{;\mu }^{\mu \nu }`$ $`=`$ $`0,`$ (9)
where
$$T_{\mu \nu }=(\rho +p)u_\mu u_\nu pg_{\mu \nu },$$
(10)
with $`p=\alpha \rho `$, $`\alpha `$ being a constant. The most common values of $`\alpha `$ of cosmological interest are $`0`$ (pressurelless matter), $`\frac{1}{3}`$ (radiation) and $`1`$ (stiff matter). Negative values of $`\alpha `$ has acquired increasing importance due to the inflationary paradigm and the new results coming from the supernova type Ia observations, as it was discussed before. Topological defects also require a negative equation of state. Again, equations (8,9) are connected by the Bianchi identities.
In the same FRW flat background, (8,9) take the form
$`3({\displaystyle \frac{\dot{a}}{a}})^2`$ $`=`$ $`8\pi G\rho ,`$ (11)
$`2{\displaystyle \frac{\ddot{a}}{a}}+({\displaystyle \frac{\dot{a}}{a}})^2`$ $`=`$ $`8\pi Gp,`$ (12)
$`\dot{\rho }+3(1+\alpha ){\displaystyle \frac{\dot{a}}{a}}\rho `$ $`=`$ $`0.`$ (13)
It is a straightforward to check that the equations (5,6,7) and equations (11,12,13) permit the identification
$$\rho _\varphi =p_\varphi =\frac{\dot{\varphi }^2}{2},$$
(14)
i.e., the ”scalar field fluid” behaves like a stiff matter in the hydrodynamical approach. The scalar field sound velocity, in this case, is $`c_\varphi ^2=\dot{p}_\varphi /\dot{\rho }_\varphi =1`$. The scale factor both in the free scalar field and hydrodynamical stiff matter cases behaves as $`at^{1/3}`$.
The evaluation of the perturbed quantities follows the well known approach of Lifshitz and Khalatnikov . It can be treated either with the synchronous gauge or the gauge-invariant formalism. Here we choose to work in the synchronous gauge formalism, where $`h_{\mu 0}=0`$.
We study first density perturbations and then gravitational waves for this free field model.
### 2.1 Density perturbations
Introducing small perturbations around the background solutions, the perturbed equations for the scalar-tensor model read
$$\ddot{h}+2\frac{\dot{a}}{a}\dot{h}=4\dot{\varphi }\dot{\delta \varphi },$$
(15)
$$\ddot{\delta \varphi }+3\frac{\dot{a}}{a}\dot{\delta \varphi }+\frac{n^2}{a^2}\delta \varphi \frac{1}{2}\dot{h}\dot{\varphi }=0,$$
(16)
where $`h=h_{kk}/a^2`$ and $`n^2`$ comes from the Helmholtz equation $`^2𝒬+n^2𝒬=0`$: the scalar functions $`𝒬(x^k)`$ are the eingefunctions of the three-dimensional Laplacian operator.
In order, to solve equations (15,16) it is more convenient to work in the conformal time, $`dt=ad\eta `$. The scale factor behaves as $`a\eta ^{1/2}`$. In terms of this new time parameter the solution of the perturbed equations is given by
$$\delta \varphi =\eta ^{3/2}\eta ^{3/2}(c_1(n)J_1(n\eta )+c_2(n)N_1(n\eta ))d\eta ,$$
(17)
where $`J_1`$ and $`N_1`$ are, respectively, the Bessel and Neumann functions of the first order, and $`c_1(n)`$, $`c_2(n)`$ are two arbitrary constants.
We need to verify if the evolution of the perturbations are specified by the equation $`\delta p_\varphi =\alpha \delta \rho _\varphi `$ as it happens with the background evolution showed previously. This should imply that the adiabatic approximation is verified here. To do this, we consider the perturbation of the equation (2) and (10)
$$\delta G^{\mu \nu }=\delta \varphi ,^\mu \varphi ,^\nu +\varphi ,^\mu \delta \varphi ,^\nu +\frac{1}{2}h^{\mu \nu }g^{\alpha \beta }\varphi ,_\beta \varphi ,_\alpha $$
$$+\frac{1}{2}g^{\mu \nu }h^{\alpha \beta }\varphi ,_\beta \varphi ,_\alpha \frac{1}{2}g^{\mu \nu }g^{\alpha \beta }(\delta \varphi ,_\beta \varphi ,_\alpha +\varphi ,_\beta \delta \varphi ,_\alpha ),$$
(18)
$$\delta T^{\mu \nu }=(\delta \rho _\varphi +\delta p_\varphi )U^\mu U^\nu +(\rho _\varphi +p_\varphi )(\delta U^\mu U^\nu +U^\mu \delta U^\nu )\delta p_\varphi g^{\mu \nu }+p_\varphi h^{\mu \nu }.$$
(19)
Using the synchronous gauge condition, we have
$$\delta G_{00}=\dot{\varphi }\delta \dot{\varphi },$$
(20)
$$\delta G_{ij}=\frac{1}{2}h_{ij}\dot{\varphi }^2+a^2\delta _{ij}\dot{\varphi }\delta \dot{\varphi },$$
(21)
$$\delta T_{00}=\delta \rho _\varphi ,$$
(22)
$$\delta T_{ij}=p_\varphi h_{ij}+a^2\delta _{ij}\delta p_\varphi .$$
(23)
If we consider the perturbed Einstein equations $`\delta G_{\mu \nu }=8\pi G\delta T_{\mu \nu }`$ and the values of $`\rho _\varphi `$ and $`p_\varphi `$ obtained from equations (11) and (12), we have that
$$\delta \rho _\varphi =\delta p_\varphi .$$
(24)
In the hydrodynamical approach, the solution for the Einstein perturbed equations with $`p=\rho `$ and $`\delta p=\delta \rho `$, leads to the expression
$$\mathrm{\Delta }=\frac{\delta \rho }{\rho }=\eta ^{3/2}\eta ^{5/2}(d_1(n)J_0(n\eta )+d_2(n)N_0(n\eta ))d\eta ,$$
(25)
where $`d_1(n)`$ and $`d_2(n)`$ are the integration constants. Remembering that $`\rho _\varphi =\frac{\varphi _{}^{}{}_{}{}^{2}}{2a^2}`$ and $`\delta \rho _\varphi =2\frac{\varphi ^{}\delta \varphi ^{}}{a^2}`$, the quantity $`\mathrm{\Delta }_\varphi =\frac{\delta \varphi ^{}}{\varphi ^{}}`$, computed from (17) reduces to (25), using simple recurrence relations for Bessel’s functions.
Hence, in this simple model where the matter is described by a scalar field, the background and perturbed equations can be put in a barotropic form. The ”velocity of sound” $`\delta p_\varphi /\delta \rho _\varphi `$ is the same as the one defined by $`c_\varphi ^2=\dot{p}_\varphi /\dot{\rho }_\varphi `$ in agreement with , where it is shown that this result corresponds to the low-frequency regime of the scalar field oscillations.
### 2.2 Gravitational waves
Here, the perturbed equation of gravitational waves is:
$$h^{\prime \prime }2\frac{a^{}}{a}h^{}+\left[n^22\frac{a^{\prime \prime }}{a}+2\frac{a^2}{a^2}\right]h=0,$$
(26)
where $`h_{ij}=h(\eta )Q_{ij}`$, $`Q_{ij}`$ being a tracelless transverse eigenfunction on the spatial section, and primes denote derivative with respect to conformal time $`d\eta =adt`$.
After inserting the backgroung solutions, we obtain the solution of the equation (26), as follow:
$$h=\eta \left(e_1(n)J_1(n\eta )+e_2N_1(n\eta )\right),$$
(27)
where $`e_1(n)`$ and $`e_2(n)`$ are the integration constants. This solution is valid for both representations.
It is easy to verify that the above solution is well-behaved and stable.
## 3 Field representation for an arbitrary barotropic equation of state
Let us return to the Einstein’s equations coupled to a perfect fluid, with a barotropic equation of state $`p=\alpha \rho `$. Solving the Einstein’s equation for a flat spatial section, we obtain for the scale factor $`a=a_0t^{\frac{2}{3(1+\alpha )}}`$. Let us now consider a self interacting scalar field coupled to gravity. The Lagrangian has the form,
$$L=\sqrt{g}[R\varphi _{;\rho }\varphi ^{;\rho }+2V(\varphi )]$$
(28)
where the potential $`V(\varphi )`$ represents the self interaction term. The field equations are
$$R_{\mu \nu }\frac{1}{2}g_{\mu \nu }R=\varphi _{;\mu }\varphi _{;\nu }\frac{1}{2}g_{\mu \nu }\varphi _{;\rho }\varphi ^{;\rho }+g_{\mu \nu }V(\varphi ),\mathrm{}\varphi =V_\varphi (\varphi ),$$
(29)
where $`V_\varphi `$ means partial derivative of the potential with respect to $`\varphi `$.
This scalar-tensor model may lead to the same behaviour of the scale factor as in the perfect fluid model provided that the potential takes the form,
$$V(\varphi )=\frac{2}{3}\frac{(1\alpha )}{(1+\alpha )^2}\mathrm{exp}(\sqrt{3(1+\alpha )}\varphi ).$$
(30)
Indeed, for a FRW background, this potential leads to the solutions
$$a=a_0t^{\frac{2}{3(1+\alpha )}},\varphi =\pm \frac{2}{\sqrt{3(1+\alpha )}}\mathrm{ln}t.$$
(31)
For the case of a cosmic string fluid, $`at`$, the potential is $`V(\varphi )=2e^{\sqrt{2}\varphi }`$.
We now turn to the perturbative level. The perturbation of the scalar-tensor model leads to
$`h^{\prime \prime }+{\displaystyle \frac{a^{}}{a}}h^{}`$ $`=`$ $`4\varphi ^{}\delta \varphi ^{}+2(\varphi ^{\prime \prime }+2{\displaystyle \frac{a^{}}{a}}\varphi ^{})\delta \varphi ,`$ (32)
$`\delta \varphi ^{\prime \prime }+2{\displaystyle \frac{a^{}}{a}}\delta ^{}+(n^2+V_{\varphi \varphi }a^2)\delta \varphi `$ $`=`$ $`{\displaystyle \frac{1}{2}}\varphi ^{}h^{},`$ (33)
where we follow the same definitions as before and we have employed the conformal time.
Combining equations (32,33), and inserting the background solutions, we obtain the third order differential equation
$`\delta \varphi ^{\prime \prime \prime }+{\displaystyle \frac{7+3\alpha }{1+3\alpha }}{\displaystyle \frac{\delta \varphi ^{\prime \prime }}{\eta }}+\{n^2+\left[{\displaystyle \frac{224\alpha 18\alpha ^2}{(1+3\alpha )^2\eta ^2}}\right]\}\delta \varphi ^{}`$
$`+\{3{\displaystyle \frac{1+\alpha }{1+3\alpha }}{\displaystyle \frac{n^2}{\eta }}18{\displaystyle \frac{1\alpha ^2}{(1+3\alpha )^2}}{\displaystyle \frac{1}{\eta ^3}}\}\delta \varphi =0.`$ (34)
One solution for this equation is $`\delta \varphi =\eta ^{3\frac{1+\alpha }{1+3\alpha }}`$, which is given by the residual coordinate transformation freedom characteristic of the synchronous coordinate condition. Using this known solution and some suitable transformation, we can reduce the third order equation to a second order Bessel equation. The final solution for $`\delta \varphi `$ is
$$\delta \varphi =\eta ^{3\frac{1+\alpha }{1+3\alpha }}c_\pm \eta ^{\frac{3}{2}}J_{\pm \nu }(n\eta )𝑑\eta ,$$
(35)
where $`c_\pm `$ are integration constants, that in general depend on $`n`$, and $`J_{\pm \nu }`$ is a Bessel function of order $`\nu =\frac{5+3\alpha }{2(1+3\alpha )}`$.
The solution for the density contrast in the case of the hydrodynamical representation, with $`p=\alpha \rho `$ is well known. It can be written as
$`\mathrm{\Delta }`$ $`=`$ $`\eta ^{3\frac{1+\alpha }{1+3\alpha }}{\displaystyle }\eta ^{\frac{5}{2}}(c_+J_\mu (\sqrt{\alpha }n\eta )+c_{}J_\mu (\sqrt{\alpha }n\eta ))d\eta ,\alpha >0,`$ (36)
$`\mathrm{\Delta }`$ $`=`$ $`\eta ^{3\frac{1+\alpha }{1+3\alpha }}{\displaystyle }\eta ^{\frac{5}{2}}(c_+I_\mu (\sqrt{\alpha }n\eta )+c_{}K_\mu (\sqrt{\alpha }n\eta ))d\eta ,\alpha <0,`$ (37)
where now $`I_\mu `$ and $`K_\mu `$ are the modified Bessel functions, and $`\mu =\frac{3}{2}\left(\frac{1\alpha }{1+\alpha }\right)`$. In order to connect both results we must remember that $`\mathrm{\Delta }_\varphi =\frac{\delta \rho _\varphi }{\rho _\varphi }`$, where $`\rho _\varphi =\frac{\dot{\varphi }^2}{2}+V(\varphi )`$. Using some Bessel’s recurrence relations, as in the previous section we find
$$\mathrm{\Delta }_\varphi =C_\pm \eta ^{3\frac{1+\alpha }{1+3\alpha }}\{(1\frac{1}{\alpha })\eta ^{\frac{5}{2}}J_{\pm \mu }(n\eta )+\eta ^{\frac{5}{2}}J_{\pm \mu }(n\eta )d\eta \}.$$
(38)
This expression differs from (36) by the factor $`\alpha `$ in the argument of the Bessel function, and by the first term in (38). However, when $`\alpha >0`$ solutions (36,38) have the same behaviour in the large and small wavelength limit. For $`\alpha =1`$, the equivalence between the two approaches is complete, as can be easily checked by comparing (38) with (36) for this special case. For $`\alpha <0`$, the correspondence between solutions (37,38) occurs only for $`n0`$ (large scale perturbations); for $`n\mathrm{}`$ (small scale perturbations) the hydrodynamical representation exhibits strong instabilities while the scalar field representation displays accoustic oscillation.
Finally we remark that both representations give the same behaviour for gravitational waves.
## 4 Perturbations in cosmic string fluid
An important particular case of the scalar-tensor model develloped in the previous section concerns the case of cosmic string. A cosmic string fluid may mimic a perfect fluid with an equation of state $`p=\frac{\rho }{3}`$. In the fate of density perturbation in fluids with negative pressure has been studied. The main conclusion was that, in the long wavelength approximation, there are only decreasing modes when the equation of state is such that the strong energy condition is violated; for small wavelengths, instabilities can arise due to the fact that the pressure contributes in the same sense as gravity. For the equation of state of a cosmic string fluid displayed above, density perturbations behave as
$$\mathrm{\Delta }=t^{1\pm \sqrt{1+\frac{n^2}{3}}}.$$
(39)
Hence, in the small wavelength limit, $`n\mathrm{}`$, instabilities come to scene.
A scalar-tensor representation of the cosmic string fluid is given by a scalar tensor model with $`V(\varphi )=2\mathrm{exp}(\pm \sqrt{2}\varphi )`$. A perturbation analysis as it was performed in section $`3`$ leads to the following expression for the perturbation in the scalar field (the integration procedure follows very closely to that sketched in the previous section):
$$\mathrm{\Delta }_\varphi t^{1\pm \sqrt{1n^2}}.$$
(40)
In the large wavelength limit both representations give the same results; in the small wavelength limit the hydrodynamical representation display instabilities while the scalar-tensor model exhibits accoustic oscillations.
In it was considered the case of two non interacting fluid, one of them represented by an equation of state like $`p=(1/3)\rho `$ while the second fluid is the ordinary matter with a barotropic equation of state $`p=\alpha \rho `$. This may represent in more realistic way the presence of fluids with negative pressure in the universe. The equations of motion are
$$3(\frac{\dot{a}}{a})^2+3\frac{k}{a^2}=8\pi G(\rho _m+\rho _s),$$
(41)
$$2\frac{\ddot{a}}{a}+(\frac{\dot{a}}{a})^2+\frac{k}{a^2}=\frac{8\pi G}{3}(\rho _s3\alpha \rho _m),$$
(42)
$$\dot{\rho }_m+3\frac{\dot{a}}{a}(1+\alpha )\rho _m=0,$$
(43)
$$\dot{\rho }_s+2\frac{\dot{a}}{a}\rho _s=0.$$
(44)
In this equations $`\rho _m`$ and $`\rho _s`$ denote the ordinary fluid and stringlike fluid densities respectivelly.
The background solutions are:
$$a=a_0\mathrm{sinh}^2(\frac{\sqrt{\gamma }}{2}\eta ),\text{when}\alpha =0;$$
(45)
$$a=a_0\mathrm{sinh}(\sqrt{\gamma }\eta ),\text{when}\alpha =1/3;$$
(46)
$$a=a_0\sqrt{\mathrm{sinh}(2\sqrt{\gamma }\eta )},\text{when}\alpha =1,$$
(47)
where $`\gamma =|\frac{8}{3}\pi G\rho _{0s}k|`$, $`\rho _{0s}`$ is defined as the ratio $`\rho _s=\frac{\rho _{0s}}{a^2}`$, $`\eta `$ being the conformal time.
### 4.1 Density perturbations
Perturbing the Einstein’s equations and imposing the syncrhonous coordinate condition, a set of equations for the ordinary matter and the cosmic fluid perturbations is obtained:
$$\ddot{h}+2\frac{\dot{a}}{a}\dot{h}=6\frac{\ddot{a}}{a}\mathrm{\Delta }_m,$$
(48)
$$\dot{\mathrm{\Delta }}_m+(1+\alpha )(\mathrm{\Psi }\frac{\dot{h}}{2})=0,$$
(49)
$$(1+\alpha )\dot{\mathrm{\Psi }}+(1+\alpha )(23\alpha )\frac{\dot{a}}{a}\mathrm{\Psi }\frac{n^2}{a^2}\alpha \mathrm{\Delta }_m=0,$$
(50)
$$\dot{\mathrm{\Delta }}_s+\frac{2}{3}(\mathrm{\Theta }\frac{\dot{h}}{2})=0,$$
(51)
$$\dot{\mathrm{\Theta }}+3\frac{\dot{a}}{a}\mathrm{\Theta }+\frac{n^2}{2a^2}\mathrm{\Delta }_s=0.$$
(52)
The integration procedure is standard , so we will just present the final results:
| model | perturbations | material phase $`(t0)`$ | string phase$`(t\mathrm{})`$ |
| --- | --- | --- | --- |
| $`\alpha =0`$ | $`\mathrm{\Delta }_m`$ | $`\eta ^2`$ | constant |
| | $`\mathrm{\Delta }_s`$ | $`\frac{1}{\eta ^{\frac{2}{3}}}(a_1K_{\frac{3}{2}}(n\eta )+a_2I_{\frac{3}{2}}(n\eta ))`$ | $`t^{1\pm \sqrt{1+\frac{n^2}{3}}}`$ |
| $`\alpha =1/3`$ | $`\mathrm{\Delta }_m`$ | $`\frac{1}{\eta ^{\frac{1}{2}}}\eta ^{\frac{5}{2}}[c_1J_{\frac{1}{2}}(n\eta )+c_2J_{\frac{1}{2}}(n\eta )]d\eta `$ | $`\mathrm{cos}(\frac{n}{\sqrt{3}}\mathrm{ln}t)`$ |
| | $`\mathrm{\Delta }_s`$ | $`\frac{1}{\eta ^{\frac{1}{2}}}(d_1I_{\frac{1}{2}}(n\eta )+d_2K_{\frac{1}{2}}(n\eta ))`$ | $`t^{1\pm \sqrt{1+\frac{n^2}{3}}}`$ |
| $`\alpha =1`$ | $`\mathrm{\Delta }_m`$ | $`\frac{1}{\eta ^{\frac{3}{2}}}\eta ^{\frac{5}{2}}[c_1J_0(n\eta )+c_2J_0(n\eta )]d\eta `$ | $`\mathrm{cos}(n\mathrm{ln}t)`$ |
| | $`\mathrm{\Delta }_s`$ | $`d_1I_0(n\eta )+d_2K_0(n\eta ))`$ | $`t^{1\pm \sqrt{1+\frac{n^2}{3}}}`$ |
We must exhibit the behaviour of the perturbative model for $`t0`$ and $`t\mathrm{}`$ to find an instability in the density perturbations of cosmic string fluid. Indeed, for small wavelengths, i.e. $`n\mathrm{}`$, the modified Bessel functions $`I_\nu (x)`$ and $`K_\nu (x)`$ go as $`e^{\pm x}/\sqrt{x}`$. However, the ordinary matter has a very regular behaviour for $`t0`$ and $`t\mathrm{}`$.
The instabilities presented above can be attributed to the matter content of the universe and, mainly, to the approach used to describe the density perturbations. We will see that these instabilities do not appear in gravitational waves because the matter content of the universe plays a less important role, being only sensitive to the behaviour of scale factor.
In fact, the same behaviour for the scale factor described above can be obtained by a scalar-tensor model, with a suitable potential, coupled to ordinary matter. For example, in the case of pressurelles ordinary matter ($`p=0`$), the results for the background can be recovered if the potential reads
$$V(\varphi )=V_0\mathrm{sinh}^4(C\varphi ),V_0=\frac{8\mathrm{\Omega }^2+2}{a_{0}^{}{}_{}{}^{2}},C=\frac{1}{\sqrt{8\mathrm{\Omega }^2+2}},\mathrm{\Omega }=\frac{2\pi G\rho _0}{3},$$
(53)
where $`\rho _0`$ is defined by the first integral of the conservation of the energy-moment tensor for the pressurelless fluid, $`\rho =\frac{\rho _0}{a^3}`$. We note that the scalar field representation for the two fluid model (pressurelles matter plus string fluid) requires a different potential with respect to the case of the one fluid model (only string fluid).
At the perturbative level, the effect of replacing the barotropic fluid with negative pressure by a scalar-field model is quite the same as that described in section $`3`$. With respect with the preceding table of solutions, the modified Bessel functions must be replaced by ordinary Bessel functions for the perturbation in the exotic fluid. Hence the instability present in the small wavelength limit disappear again; the behaviour in the long wavelength limit is not changed.
### 4.2 Gravitational waves
Following the work , we have the perturbed equations that govern the evolution of gravitational waves
$$h^{\prime \prime }2\frac{a^{}}{a}h^{}+\left[\stackrel{~}{n}^22\left(\frac{a^{\prime \prime }}{a}\frac{a^2}{a^2}\right)\right]=0,$$
(54)
where $`\stackrel{~}{n}^2=(1/r^2)(n^2+2k)`$ and prime here denotes derivative with respect to $`\theta =r\eta `$.
After inserting the background solutions (45,46,47), equation (54) can be generally rewritten in terms of a hypergeometric equation. Its final solutions for different phases of the evolution of the universe read as follows:
| model | $`h_1`$ | $`h_2`$ |
| --- | --- | --- |
| $`\alpha =1`$ | $`\sqrt{x^21}\left[\frac{1+x}{2}\right]^{2+\sqrt{1\stackrel{~}{n}^2}}\times `$ | $`\sqrt{x^21}\left[\frac{1+x}{2}\right]^{2\sqrt{1\stackrel{~}{n}^2}}\times `$ |
| | $`{}_{2}{}^{}F_{1}^{}(2\sqrt{1\stackrel{~}{n}^2},\frac{1}{2}\sqrt{1\stackrel{~}{n}^2},`$ | $`{}_{2}{}^{}F_{1}^{}(\frac{1}{2}+\sqrt{1\stackrel{~}{n}^2},2+\sqrt{1\stackrel{~}{n}^2},`$ |
| | $`12\sqrt{1\stackrel{~}{n}^2},\frac{2}{1+x})`$ | $`1+2\sqrt{1\stackrel{~}{n}^2},\frac{2}{1+x})`$ |
| $`\alpha =1/3`$ | $`\mathrm{exp}((\sqrt{1\stackrel{~}{n}^2})\eta )\mathrm{sinh}\eta `$ | $`\mathrm{exp}((\sqrt{1\stackrel{~}{n}^2})\eta )\mathrm{sinh}\eta `$ |
| $`\alpha =0`$ | $`\sqrt{x^21}\left[\frac{1+x}{2}\right]^{1+\sqrt{4\stackrel{~}{n}^2}}\times `$ | $`\sqrt{x^21}\left[\frac{1+x}{2}\right]^{1\sqrt{4\stackrel{~}{n}^2}}\times `$ |
| | $`{}_{2}{}^{}F_{1}^{}(1\sqrt{4\stackrel{~}{n}^2},\frac{1}{2}\sqrt{4\stackrel{~}{n}^2},`$ | $`{}_{2}{}^{}F_{1}^{}(\frac{1}{2}+\sqrt{4\stackrel{~}{n}^2},1+\sqrt{4\stackrel{~}{n}^2},`$ |
| | $`12\sqrt{4\stackrel{~}{n}^2},\frac{2}{1+x})`$ | $`1+2\sqrt{4\stackrel{~}{n}^2},\frac{2}{1+x})`$ |
| $`\alpha =1`$ | $`\sqrt{x^21}\left[\frac{1+x}{2}\right]^{(1+\sqrt{14\stackrel{~}{n}^2})/2}\times `$ | $`\sqrt{x^21}\left[\frac{1+x}{2}\right]^{(1\sqrt{14\stackrel{~}{n}^2})/2}\times `$ |
| | $`{}_{2}{}^{}F_{1}^{}(\frac{1\sqrt{14\stackrel{~}{n}^2}}{2},\frac{1\sqrt{14\stackrel{~}{n}^2}}{2},`$ | $`\times _2F_1(\frac{1+\sqrt{14\stackrel{~}{n}^2}}{2},\frac{1+\sqrt{14\stackrel{~}{n}^2}}{2},`$ |
| | $`1\sqrt{14\stackrel{~}{n}^2},\frac{2}{1+x})`$ | $`1+\sqrt{14\stackrel{~}{n}^2},\frac{2}{1+x})`$ |
where in the above expressions, $`{}_{2}{}^{}F_{1}^{}`$ is a hypergeometric function and $`x=\mathrm{cos}(r\theta )`$, $`r`$ being a constant.
In this two-fluid model, the behaviour of gravitational waves in a closed universe exhibits, in what concerns the behaviour of the scale factor, the dynamic of an open universe. It would cause, also, distorsion in the spectrum of the CMBR. The determination of the evolution of perturbations and its connection with this distorsion is technically difficult to be evaluated. But, for the proposal of the present work, the fundamental feature to be retained in the above solutions is that they do not exhibit instabilities.
## 5 Time dependent cosmological constant model
We present for completeness the background equations based on the reference which is the traditional scalar-tensor theory with a potential that is equivalent to a time dependent cosmological constant model. This is one of the cases which can be represented phenomenologically by a fluid with an equation of state $`p=\frac{\rho }{3}`$ in what concerns the behaviour of the scale factor. The action is given by
$$𝒮=\frac{1}{16\pi G}\sqrt{g}[\varphi R\varphi ^1\omega g^{\mu \nu }_\mu _\nu \varphi +2\varphi \mathrm{\Lambda }(\varphi )]d^4x+𝒮_{ng}.$$
(55)
We remark, however, that in the present case we have a non-minimally coupled scalar field, in opposition to the models described before.
The field equations are
$$G_{\mu \nu }=\frac{8\pi T_{\mu \nu }}{\varphi }+\mathrm{\Lambda }(\varphi )g_{\mu \nu }+\omega \varphi ^2(\varphi ,_\mu \varphi ,_\nu \frac{1}{2}g_{\mu \nu }\varphi ,_\alpha \varphi ,^\alpha )+\varphi ^1(\varphi ;_{\mu \nu }g_{\mu \nu }\mathrm{}\varphi ),$$
(56)
$$\mathrm{}\varphi +\frac{2\varphi ^2d\mathrm{\Lambda }/d\varphi 2\varphi \mathrm{\Lambda }(\varphi )}{3+2\omega }=\frac{1}{3+2\omega }(8\pi T\frac{d\omega }{d\varphi }\varphi ,_\alpha \varphi ,^\alpha ).$$
(57)
We shall consider the case where $`\omega =`$ constant, $`\mathrm{\Lambda }(\varphi )=c\varphi ^m`$ and $`\varphi =\varphi _1t^q`$, with $`c`$, $`m`$ and $`q`$ constants. The ansatz allow us to obtain analytical solutions in the form of power law, which leads to closed expressions for the perturbative equations.
The energy-momentum tensor describes an ordinary perfect fluid, such that
$$T^{\mu \nu };_\nu =0.$$
(58)
The equations of motion are
$$3\frac{\dot{a}^2}{a^2}+3\frac{k}{a^2}c\varphi ^m=\frac{8\pi \rho }{\varphi }+\frac{\omega }{2}\frac{\dot{\varphi }^2}{\varphi ^2}3\frac{\dot{a}}{a}\frac{\dot{\varphi }}{\varphi },$$
(59)
$$2\frac{\ddot{a}}{a}\frac{\dot{a}^2}{a^2}\frac{k}{a^2}+c\varphi ^m=\frac{8\pi p}{\varphi }+\frac{\omega }{2}\frac{\dot{\varphi }^2}{\varphi ^2}+\frac{\ddot{\varphi }}{\varphi }+2\frac{\dot{a}}{a}\frac{\dot{\varphi }}{\varphi },$$
(60)
$$\frac{\ddot{\varphi }}{\varphi }+3\frac{\dot{a}}{a}\frac{\dot{\varphi }}{\varphi }=\frac{2c(1m)\varphi ^m}{3+2\omega }+\frac{8\pi (\rho 3p)}{\varphi (3+2\omega )}.$$
(61)
The background solutions are
| model | curvature | $`a(t)`$ | $`\varphi (t)`$ | $`\rho `$ | $`\mathrm{\Lambda }(t)`$ |
| --- | --- | --- | --- | --- | --- |
| $`\rho =0`$ | $`k0`$ | $`a_1t`$ | $`\varphi _1t^{\frac{2}{m}}`$ | - | $`\mathrm{\Lambda }_1/t^2`$ |
| | $`k=0`$ | $`a_1t`$ | $`\varphi _1t^{\frac{2}{1\pm \sqrt{3+2\omega }}}`$ | - | $`\mathrm{\Lambda }_1/t^2`$ |
| $`p=\alpha \rho `$ | any $`k`$ | $`a_1t`$ | $`\varphi _1t^{(1+3\alpha )}`$ | $`s/a^{3(1+\alpha )}`$ | $`\mathrm{\Lambda }_1/t^2`$ |
| $`p=\rho `$ | any $`k`$ | $`a_1t`$ | $`\varphi _1t^2`$ | $`\rho _0=\text{const.}`$ | $`\mathrm{\Lambda }_1/t^2`$ |
where $`a_1`$, $`\varphi _1`$, $`\mathrm{\Lambda }_1`$, and $`s`$ are constants.
The solutions for the scale factor above are characteristic of an equation of state $`p_\varphi =\frac{1}{3}\rho _\varphi `$.
We remark that in all above solutions the scale-factor behaves as $`at`$ corresponding to a ”coasting” universe. From the point of view of the background behaviour, the above solutions exhibit the same characteristics as the perfect fluid formulation.
## 6 Perturbations in a time dependent cosmological constant model
We extend our perturbative analysis for this model, first for density perturbations and then for gravitational waves.
### 6.1 Density perturbations
The perturbed equations for the time dependent cosmological constant model are:
$$\ddot{h}+2\frac{\dot{a}}{a}\dot{h}=\frac{16\pi }{\varphi }(\mathrm{\Delta }\lambda )\left(\frac{3\alpha \omega +3\alpha +\omega +2}{3+2\omega }\right)\rho $$
$$2cm\varphi ^m\left(\frac{m+2\omega +2}{3+2\omega }\right)\lambda +2\ddot{\lambda }+4\frac{\dot{\varphi }}{\varphi }(1+\omega )\dot{\lambda },$$
(62)
$$\ddot{\lambda }+(2\frac{\dot{\varphi }}{\varphi }+3\frac{\dot{a}}{a})\dot{\lambda }+(\frac{\ddot{\varphi }}{\varphi }+3\frac{\dot{a}}{a}\frac{\dot{\varphi }}{\varphi })\lambda \frac{1}{2}\frac{\dot{\varphi }}{\varphi }\dot{h}$$
$$+\frac{n^2}{a^2}\lambda +\frac{2c(m^21)\varphi ^m}{3+2\omega }\lambda =\frac{8\pi }{(3+2\omega )\varphi }\mathrm{\Delta }(13\alpha )\rho ,$$
(63)
$$\dot{\mathrm{\Delta }}=(1+\alpha )(\frac{1}{2}\dot{h}\delta U^k,_k),$$
(64)
$$\frac{}{t}(a^5\delta U^k(1+\alpha )\rho )=a^3\alpha \rho \mathrm{\Delta },^k,$$
(65)
where $`h=h_{kk}a^2`$, $`\lambda =\delta \varphi /\varphi `$ and $`\mathrm{\Delta }=\delta \rho /\rho `$. All functions are spatially expanded into spherical harmonics $`f(x,t)=f(t)𝒬`$, with $`^2𝒬=n^2𝒬`$.
Next, we solve the above equations for the vacuum fluid ($`\alpha =1`$), the empty universe ($`\rho =0`$), radiation phase ($`\alpha =1/3`$), and dust phase ($`\alpha =0`$). We use the background relations in order to simplify the final system of equations.
#### 6.1.1 Vacuum fluid phase ($`\alpha =1`$)
In this case $`\mathrm{\Delta }=0`$ and we have the following equations:
$$\ddot{h}+\frac{2}{t}\dot{h}=(1+2\omega )\frac{\lambda }{t^2}+8(1+\omega )\frac{\dot{\lambda }}{t}+2\ddot{\lambda },$$
(66)
$$\ddot{\lambda }+7\frac{\dot{\lambda }}{t}+[8+\frac{n^2}{a_1}]\frac{\lambda }{t^2}\frac{1}{t}\dot{h}=0,$$
(67)
with the solution
$$\lambda =t^p\text{and}ht^p,\text{where}p=2\pm \sqrt{8\omega \frac{n^2}{a_1^2}}.$$
(68)
There is no divergent behaviour in the small wavelength limit.
#### 6.1.2 Empty universe ($`\rho =0`$)
Here, the system of second order differential equations is
$$\ddot{h}+\frac{2}{t}\dot{h}=\frac{4}{m}\left[\frac{m+2\omega +2}{3+2\omega }\right]\frac{\lambda }{t^2}\frac{8(1+\omega )}{m}\frac{\dot{\lambda }}{t}+2\ddot{\lambda },$$
(69)
$$\ddot{\lambda }+\left[\frac{3m4}{m}\right]\dot{\lambda }+[\frac{n^2}{a_{1}^{}{}_{}{}^{2}}4(2+m)\frac{1m}{m^2}]\frac{\lambda }{t^2}+\frac{1}{m}\frac{\dot{h}}{t}=0,$$
(70)
whose solution is
$$\lambda =t^p\text{and}ht^p\text{where}p=\frac{1}{m}\left(1m\pm \sqrt{9+6m3m^2+8\omega m^2\frac{n^2}{a_1^2}}\right),$$
(71)
where $`m`$ is arbitrary for $`k0`$ and $`m=1\pm \sqrt{3+2\omega }`$ for $`k=0`$. As in the preceding case, the solutions are stable.
#### 6.1.3 Radiation phase ($`\alpha =1/3`$)
Combining conveniently the perturbed equations, we find a fifth order Euler’s type equation for $`\lambda `$:
$`\lambda ^v+11{\displaystyle \frac{\lambda ^{iv}}{t}}+[{\displaystyle \frac{4}{3}}{\displaystyle \frac{n^2}{a_{1}^{}{}_{}{}^{2}}}+312d2{\displaystyle \frac{k}{a_1^2}}]{\displaystyle \frac{\stackrel{\mathrm{}}{\lambda }}{t^2}}+[{\displaystyle \frac{16}{3}}{\displaystyle \frac{n^2}{a_{1}^{}{}_{}{}^{2}}}+228d8{\displaystyle \frac{k}{a_1^2}}]{\displaystyle \frac{\ddot{\lambda }}{t^3}}`$
$`+[{\displaystyle \frac{1}{3}}{\displaystyle \frac{n^4}{a_{1}^{}{}_{}{}^{4}}}+{\displaystyle \frac{n^2}{a_{1}^{}{}_{}{}^{2}}}({\displaystyle \frac{8}{3}}+{\displaystyle \frac{2}{3}}d2{\displaystyle \frac{k}{a_1^2}})+24d4{\displaystyle \frac{k}{a_1^2}}]{\displaystyle \frac{\dot{\lambda }}{t^4}}+[{\displaystyle \frac{1}{3}}{\displaystyle \frac{n^4}{a_{1}^{}{}_{}{}^{4}}}+{\displaystyle \frac{n^2}{a_{1}^{}{}_{}{}^{2}}}({\displaystyle \frac{2}{3}}d2{\displaystyle \frac{k}{a_1^2}})]{\displaystyle \frac{\lambda }{t^5}}=0,`$ (72)
with $`d=3+2\omega `$. Due to the residual coordinate transformation freedom characteristic of the synchronous coordinate condition, $`\lambda t^1`$ is a solution of this equation. Hence, we can reduce it to a fourth order equation. Supposing a solution of the type $`\lambda t^r`$, we find the polynomial equation for $`r`$
$$r^4+(\frac{4}{3}\frac{n^2}{a_{1}^{}{}_{}{}^{2}}2d2\frac{k}{a_1^2})r^2+(\frac{1}{3}\frac{n^4}{a_{1}^{}{}_{}{}^{4}}\frac{n^2}{a_{1}^{}{}_{}{}^{2}}(\frac{2}{3}d+2\frac{k}{a_1^2}))=0$$
(73)
whose solutions are
$$r_{\pm }^{}{}_{}{}^{2}=\frac{2}{3}\frac{n^2}{a_{1}^{}{}_{}{}^{2}}+d+\frac{k}{a_1^2}\pm \sqrt{\frac{n^4}{9a_{1}^{}{}_{}{}^{4}}+(\frac{2}{3}\frac{k}{a_1^2}2d)\frac{n^2}{a_{1}^{}{}_{}{}^{2}}+(d+\frac{k}{a_1^2})^2}.$$
(74)
In the small wavelength limit, these solutions display an oscillatory behaviour, hence stability. In the longwavelength limit, this expression reduces to $`r_\pm =\pm \sqrt{2(d+\frac{k}{a_1^2})}`$.
#### 6.1.4 Dust Phase ($`\alpha =0`$)
In this case, the equations governing the evolution of density perturbations are
$`\ddot{\mathrm{\Delta }}+2{\displaystyle \frac{\dot{\mathrm{\Delta }}}{t}}(2+\omega )R{\displaystyle \frac{\mathrm{\Delta }}{t^2}}`$ $`=`$ $`\ddot{\lambda }2(1+\omega ){\displaystyle \frac{\dot{\lambda }}{t}}(2+\omega )S{\displaystyle \frac{\lambda }{t^2}},`$ (75)
$`\ddot{\lambda }+{\displaystyle \frac{\dot{\lambda }}{t}}+[{\displaystyle \frac{n^2}{a_{1}^{}{}_{}{}^{2}}}+S]{\displaystyle \frac{\lambda }{t^2}}`$ $`=`$ $`{\displaystyle \frac{\dot{\mathrm{\Delta }}}{t}}+R{\displaystyle \frac{\mathrm{\Delta }}{t^2}},`$ (76)
where $`R=\frac{2\frac{k}{a_{1}^{}{}_{}{}^{2}}\omega 1}{3+2\omega }`$ and $`S=\frac{6\frac{k}{a_{1}^{}{}_{}{}^{2}}+3+\omega }{3+2\omega }`$. These equations may be solved supposing $`\mathrm{\Delta }=\mathrm{\Delta }_0t^r`$ and $`\lambda =\lambda _0t^r`$, $`r`$ obeying the polynomial equation
$`r^4+2r^3+({\displaystyle \frac{n^2}{a_{1}^{}{}_{}{}^{2}}}+S(3+\omega )R(3+2\omega ))r^2`$
$`+({\displaystyle \frac{n^2}{a_{1}^{}{}_{}{}^{2}}}(1+\omega )S+(3+2\omega )R)r(2+\omega )R{\displaystyle \frac{n^2}{a_{1}^{}{}_{}{}^{2}}}=0,`$ (77)
This polynomial equation has no simple closed form solution. But, numerical integration reveals the stability of the model in the small wavelength limit. For example, fixing $`a_1=1`$, choosing $`\omega =1`$, $`k=0`$ and $`n=1`$ we find the roots $`r_{1}^{}{}_{}{}^{\pm }1.16\pm 2.77i`$, $`r_{2}^{}{}_{}{}^{\pm }0.16\pm 0.33i`$, while for $`n=100`$, keeping unchanged the other parameters, we find $`r_{1}^{}{}_{}{}^{\pm }0.50\pm 100i`$, $`r_{2}^{}{}_{}{}^{\pm }0.50\pm 0.97i`$.
### 6.2 Gravitational waves
In this case, the perturbed equation that govern the gravitational waves is
$`\ddot{h}+({\displaystyle \frac{\dot{\varphi }}{\varphi }}{\displaystyle \frac{\dot{a}}{a}})\dot{h}+[{\displaystyle \frac{n^2}{a^2}}+4{\displaystyle \frac{\dot{a}^2}{a^2}}+2\left({\displaystyle \frac{\omega (\alpha 1)1}{3+2\omega }}\right)(3{\displaystyle \frac{\dot{a}^2}{a^2}}+3{\displaystyle \frac{k}{a^2}}{\displaystyle \frac{\omega }{2}}{\displaystyle \frac{\dot{\varphi }^2}{\varphi ^2}}+3{\displaystyle \frac{\dot{a}}{a}}{\displaystyle \frac{\dot{\varphi }}{\varphi }})`$
$`2\left({\displaystyle \frac{1+m+\omega (1+\alpha )}{3+2\omega }}\right)c\varphi ^m]h=0.`$ (78)
The solutions of the above equation are
$`\rho =0`$ $``$ $`h=C_\pm t^{\frac{1}{2}(1+A_1\pm \sqrt{(1+A)^24B4\frac{n^2}{a_{1}^{}{}_{}{}^{2}}}},`$
$`A_1`$ $`=`$ $`{\displaystyle \frac{2+m}{m}},B={\displaystyle \frac{4}{m^2}}(m^2m22\omega ),`$ (79)
$`\alpha =1`$ $``$ $`h=C_\pm t^{\pm \sqrt{(\frac{n^2}{a_{1}^{}{}_{}{}^{2}}+A_2)}},`$
$`A_2`$ $`=`$ $`{\displaystyle \frac{1}{3+2\omega }}[38\omega +4\omega ^23{\displaystyle \frac{k}{a_{1}^{}{}_{}{}^{2}}}(1+2\omega )],`$ (80)
$`\alpha =0`$ $``$ $`h=C_\pm t^{\frac{3\pm \sqrt{94A_34\frac{n^2}{a_{1}^{}{}_{}{}^{2}}}}{2}},A_3=2{\displaystyle \frac{4k}{a_{1}^{}{}_{}{}^{2}}}`$ (81)
$`\alpha ={\displaystyle \frac{1}{3}}`$ $``$ $`h=C_\pm t^{2\pm \sqrt{4\frac{k}{a_{1}^{}{}_{}{}^{2}}\frac{n^2}{a_{1}^{}{}_{}{}^{2}}}},`$ (82)
where $`C_\pm `$ are integration constants.
Here, the solution for the gravitational waves are also well-behaved and stable.
### 6.3 The General Relativity limit
In general, the Brans-Dicke theory reduces to the General Relativity theory when $`\omega \mathrm{}`$, except in some special cases, for example, when the trace of the momentum-energy tensor is zero . The solutions described above do not have a well-behaved limit when $`\omega \mathrm{}`$. In fact, an inspection of the background equations shows that all solutions become trivial in that limit. We could expect that in this case, the Brans-Dicke field could become constant and the model reduces itself to General Relativity with a cosmological constant. But, the imposition that $`at`$ breaks this equivalence.
Concerning the perturbed solutions, they become divergent when $`\omega \mathrm{}`$. This only express the fact that the background scenarios have no sense in this limit. For finite $`\omega `$, the perturbed solutions exhibit growing and decreasing modes as usual, for scalar and tensorial perturbations. One important feature of the solutions found before is that when $`n0`$, all dependence of the solutions on the wavelength of the perturbations is carried out by the integration constants, which must be fitted correctly in order to reproduce the power spectrum observed today.
## 7 Conclusions
In spite of the fact that fluids of negative pressure have become crucial to understand many theoretical and observational aspects of modern cosmology, they are plagued with instabilities in the small wavelength limit. These instabilities appear mainly when the barotropic equation of state is such that the strong energy condition is violated. In this paper we have exploited the possibility that these instabilities are due to the hydrodynamical representation. This possibility was suggested by the fact that, while density perturbations are bad behaved in the small wavelength limit, gravitational waves are well behaved in the same limit. Since gravitational waves depend closely on the behaviour of the scale factor but are quite insensitive to the matter representation, the instabilities should be due to the fact that in the hydrodynamical representation, the equation of state is fixed for all wavelength, while in a more fundamental approach it could depend on the scale of the perturbation.
As a matter of fact, this possibility was first suggested in , and it has been studied under certains conditions in . In reference , this problem has been briefly treated in the realm of minimal scalar-tensor model which was intended to cope with the dark matter problem through the introduction of a scalar field with a convenient potential. However, if we are interested in a field that can drive the accelerated expansion of the Universe, as it is the case in this work, the energy of the scalar field should have a smooth distribution, since it should not be present in the local clusters. Such smooth distribution can be obtained considering that the pressure associated to this field is negative, such that it does not clumpsy in large scale; but in order to avoid instabilities at small scales, the effective equation of state should become positive in this limit, and small fluctuations in this field should oscillate like an accoustic wave.
In the present work, we have verified that, when a fluid of positive pressure can mimic a scalar-tensor model, both formulations are essentially similar in the background and perturbative level. However, when we consider a fluid of negative pressure, the equivalence exists only at the background level: at perturbative level, the models behave in a complete different way. In particular, there is agreement between both representations only in the large wavelength limit: for small wavelengths, the hydrodynamical model is unstable, while the corresponding scalar field representation exhibits accoustic oscillations. Hence, in situations where negative pressures are concerned, a field representation leads to a much more complete scenario, being closer to a realistic model.
This can be understood by remembering that when we pass from a hydrodynamical representation to a fundamental one based for example on scalar fields, we change a relativistic Euler’s type equation to a Klein-Gordon equation: the sign of the laplacian operator in these equations are the same only when the hydrodynamical pressure is positive; otherwise, it changes sign and, in the perfect fluid model, accoustic modes give place to exponential modes, resulting in the presence of instabilities. In the long wavelength limit the laplacian operator can be neglected and the results agree in both representation.
We must notice, however, that the scalar-tensor model that corresponds to a given perfect fluid results is quite model dependent. For example, the one fluid model gives a potential that is different from that of a two-fluid models. It would be interesting to employ in a two fluid model, which is closer to a realistic situation, the potential obtained in the one fluid model, in spite of the great complexity of the equations. We can speculate if the resulting effective equation of state evolves in a quite similar way as the usual quintessence models.
We have extended this study for the case where the cosmic string fluid mimic a variable cosmological constant model, derived from a non-minimal coupled scalar field with a potential. The conclusions are basically the same as before. But, we must stress that, in this case, there is a Brans-Dicke type coupling parameter $`\omega `$ that does not recover the known General Relativity solutions when $`\omega \mathrm{}`$. Moreover, the perturbative behaviour may become unstable in this limit, for any scale. But, in general, for finite values of $`\omega `$ the perturbations do not exhibit either the instabilities that are present in the corresponding hydrodynamical model.
Finally, the fact that both approaches give the same behaviour for the long wavelength limit even if the pressure is negative, implies that the predictions for the power spectrum of the anisotropy of the CMBR for small values of the multipolar expansion parameter $`l`$, that is, for very large structures, is not spoiled by the employment of the hydrodynamical representation. However, for large values of $`l`$, that is, small angular separation, the employment of a field representation, mainly when negative pressures are involved, seems crucial.
## Acknowledgements
We thank Nelson Pinto Neto for his suggestions and for many enlightfull discussions. We acknowledge also the financial support from CNPq and CAPES (Brazil). |
warning/0003/cond-mat0003165.html | ar5iv | text | # Mesoscopic phase separation in La2CuO4.02 - a 139La NQR study
## Abstract
In crystals of $`\mathrm{La}_2\mathrm{CuO}_{4.02}`$ oxygen diffusion can be limited to such small length scales, that the resulting phase separation is invisible for neutrons. Decomposition of the $`{}_{}{}^{139}\mathrm{La}`$ NQR spectra shows the existence of three different regions, of which one orders antiferromagnetically below 17 K concomitantly with the onset of a weak superconductivity in the crystal. These regions are compared to the macroscopic phases seen previously in the title compound and the cluster-glass and striped phases reported for the underdoped Sr-doped cuprates.
Inhomogeneous carrier and spin distributions in high-$`T_c`$ compounds might be related to phase separation and stripes, and are intensively studied especially in hole doped $`\mathrm{La}_2\mathrm{CuO}_4`$. In underdoped $`\mathrm{La}_{2\mathrm{x}}\mathrm{Sr}_\mathrm{x}\mathrm{CuO}_4`$ for $`x>0.06`$ doping leads to spin-density wave order or stripe formation, as seen by neutron scattering at various $`x`$-values. Recent NQR and NMR studies have revealed new interesting features, like line intensity suppression caused by spin/charge fluctuations, stripe condensation at low temperatures under favorable pinning conditions, and the presence of inequivalent copper sites attributed to a stripe formation. At very low $`\mathrm{Sr}`$ content ($`x<0.02`$) the magnetic properties are explained in terms of hole segregation, and the formation of a cluster spin glass (x=0.06).
For $`\mathrm{La}_2\mathrm{CuO}_{4+\mathrm{x}}`$ the presence of the mobile oxygen dopants leads to a macroscopic structural phase separation in antiferromagnetic (AF) and superconducting (SC) regions in the concentration range $`0.01<x<0.06`$ (the so called miscibility gap). The oxygen mobility is linked to lattice imperfections, e.g. it is strongly increased by the presence of planar defects. In $`\mathrm{La}_2\mathrm{CuO}_{4+\mathrm{x}}`$ single crystals prepared by the molten solution method oxygen mobility is very low due to the small number of defects and hence the scale of the structural separation can be minimized. The single crystal with $`x=0.02`$ prepared by this way appears homogeneous below 200 K in X-ray and neutron studies, although the composition is inside the miscibility gap. In the crystal a superconducting transition is observed around 15–17 K with a very weak diamagnetic signal. According to $`\mu `$SR part of the sample becomes magnetically ordered below the same temperature of 15 K, but neutron diffraction does not see any long range magnetic structure. Apparently phase separation happens, but the scale is too small to be visible by standard structural methods.
By performing NQR on the same $`\mathrm{La}_2\mathrm{CuO}_{4.02}`$ single crystal, we will demonstrate the existence of three regions with different oxygen concentrations and show only one (oxygen-poor) phase to have an antiferromagnetic transition. We evaluate the size of the antiferromagnetic regions from the reduction of staggered magnetization. By comparison with the macroscopic phases seen previously in the title compound and the cluster-glass and striped phases seen in $`\mathrm{La}_{1.94}\mathrm{Sr}_{0.06}\mathrm{CuO}_4`$ and La<sub>2-x-y</sub>Sr<sub>x</sub>Eu<sub>y</sub>CuO<sub>4</sub> we will discuss the influence of the character of the local magnetic order on the NQR spectra.
The $`{}_{}{}^{139}\mathrm{La}`$ NQR spectra (Figs.1,2) were taken for the $`\pm 7/2\pm 5/2`$ (referred to as $`m=7/2`$) transition by sweeping the frequency. The nuclear spin-lattice relaxation rate $`T_1^1`$ (Fig.4) was measured by monitoring the recovery of magnetization after a single $`\pi `$ pulse. For the nuclear spin-spin relaxation (inset of Fig.4) the standard spin-echo decay method was applied. Fig.1 shows the evolution of $`{}_{}{}^{139}\mathrm{La}`$ spectra with temperature in the $`\mathrm{La}_2\mathrm{CuO}_{4.02}`$ single crystal. A lower S/N ratio around 15 K is due to the fast relaxation and partial wipe-out of the signal in this region (see below). Differences in S/N in some adjacent spectra are due to a different number of averages. The central part of the spectrum broadens below 30 K. The splitting of the central line (see also Fig.2b) below 17 K is a manifestation of magnetic ordering of the Cu moments generating an internal magnetic field at the La site. Above the magnetic transition the spectra are well described by the sum of three Gaussian lines (labeled 1, 2 and 3 in sequence of increasing width). Below 17 K the spectra show the same structure but with a splitting of the single narrow line 1 into two lines of equal intensity (1a and 1b) as a result of the AF ordering. Figs.2a,b illustrate the results of a spectral decomposition at 73 K and 4.2 K.
As already mentioned above, previous experiments showed the possibility of the phase separation. Our NQR spectra are in agreement with this assumption. In addition our data show that separation is present at temperatures much higher than that of the AF transition. We assign the narrow line 1 (Fig.2a) to oxygen-poor (OP) regions, i.e. a phase with low hole doping. This phase is antiferromagnetically ordered below 17 K (two narrow lines in Fig.2b). The broad line (line 3) has to be associated with oxygen-rich (OR) regions because of its absence in samples with $`x=0`$. The large width (1 MHz) of this line is due to the distortion of the electric field gradient at the La site because of the presence of holes in the $`\mathrm{CuO}_2`$ planes leading to the distribution of the oxygen octahedral tilts and the bond lengths. The OR regions are likely responsible for the superconductivity. The hole concentration in the CuO<sub>2</sub> planes does not depend on the method of doping, which allows a comparison to the line position data for La in La<sub>2-x</sub>Sr<sub>x</sub>CuO<sub>4</sub> (to see the similarity with the copper data in ref.\[\], one has to realize that extra holes shift the Cu resonance up, while lowering the La frequency). In La<sub>2-x</sub>Sr<sub>x</sub>CuO<sub>4</sub> the line positions at 4.2 K obey the relation $`\nu =(19.155x)`$ MHz. Using this expression and a hole to excess oxygen ratio of 2:1, the line positions at 4.2 K of 19.13 MHz (midpoint of lines 1a and 1b), 19.02 MHz and 18.84 MHz give oxygen concentrations of 0, 0.01 and 0.035 for lines 1, 2 and 3 resp. The $`x`$ dependence of $`T_c`$ in La<sub>2-x</sub>SrCuO<sub>4</sub> gives an independent way to estimate the hole concentration in the OR-regions. In the phase diagram superconductivity is suggested to start around 0.06. $`T_c`$ = 15 K gives a hole concentration in the OR phase near 0.07 or excess oxygen concentration of 0.035, in agreement with the value found above. In the macroscopically phase separated sample, the oxygen concentration in the OR phase is 0.06. This concentration difference is not surprising since due to the restricted oxygen mobility the local oxygen concentrations will be far from their equilibrium values. According to the above given evaluation, line 2 belongs to regions with rather low oxygen content, although it shows no sign of antiferromagnetic ordering. This line possibly arises from the interface between OP and OR regions and might be compared to the relatively narrow and weak line observed in macroscopically phase separated $`\mathrm{La}_2\mathrm{CuO}_{4.035}`$ besides the broad metallic and splitted narrow antiferromagnetic lines. In our case the weight in the spectrum will be enhanced due to the small sizes of the correlated regions.
The splitting of the line 1 is almost $`T`$ independent from 14 K to 4.2 K (see Fig.3) and just below the ordering temperature $`T_{AF}=17.0\pm 0.5`$ K seems to increase somewhat faster with lowering $`T`$ than the mean field behavior of the staggered magnetization for $`S`$=1/2 (drawn line on Fig.3). The line splitting (200 kHz) and hence the internal field at the La-site at 4.2 K is 20% lower than the 250 kHz splitting in antiferromagnetic undoped $`\mathrm{La}_2\mathrm{CuO}_4`$. This means a 20% reduction of the staggered magnetization and an ordered magnetic moment, that is 0.43 $`\mu _B/\mathrm{Cu}`$ instead of $`\mu =`$0.48 $`\mu _B/\mathrm{Cu}`$ in the undoped compound. In the SDW–like AF ordered striped phase of the Eu and Sr doped compound this value is $`0.29\mu _B/\mathrm{Cu}`$. A temperature dependent reduction of the staggered magnetization in the system with low Sr doping has been explained by finite size effects caused by microsegregation of the holes in domain walls of flat domains. At low temperatures this effect vanishes due to the hole localization. In our case local field reduction is observed even at 4.2 K and may be explained as a finite size effect caused by a small grain dimension. At $`T`$ = 0 K the staggered magnetization of a domain of size $`L`$ (infinite in the two other directions) drops by 20% for $`L`$ about 20 lattice spacings (8 nm). For a quasi-two-dimensional domain (due to the large anisotropy) the reduction should increase and hence this value has to be considered as a lower limit. The neutron data place an upper limit of the same order of magnitude. Thus an average grain size of the order of 8 nm looks reasonable and confirms phase separation on a mesoscopic rather than macroscopic scale.
The $`T`$ dependencies of the spin-lattice ($`T_1^1`$) and spin-spin ($`T_2^1`$) relaxation rates are illustrated in Fig.4 and its inset. Both rates show a large peak at low temperatures. Near the peak temperature, the nuclear magnetization recoveries after the initial magnetization reversal ($`M_z(t)`$) in case of $`T_1^1`$, or the loss of transversal magnetization ($`M_x(t)`$) in case of $`T_2^1`$, are characterized by stretched exponential time dependencies $`\mathrm{exp}[(t/T_{1,2}]^\alpha )`$. Very close to the maximum, fits of $`M_z(t)`$ require a slow and fast time constant. For both $`T_1`$’s $`\alpha `$ is about 0.5, while the quality of the $`M_z(t)`$-fit is found to depend only weakly on the precise value for $`T_1`$ (slow) (of the order of 0.1 s). Closed circles in Fig.4 show the temperature dependence of the fast rate. Below 17 K there is a sharp decrease of this contribution (open circles in Fig.4).
The strong growth of the $`{}_{}{}^{139}\mathrm{La}`$ relaxation rates around 15 K is explained as a result of hole localization and slowing down of hole mediated spin fluctuations. Here in the same temperature region the AF and SC transitions occur. The coincidence of these phenomena likely leads to even more complicated behavior of nuclear relaxation rates, e.g. the above mentioned decrease of the fast contribution to the longitudinal relaxation just below 17K (Fig.4). The occurrence of slow and fast contributions to $`M_z(t)`$ is typical for a multiphase material. Further analysis is not pursued, as its precise character might not only depend on the AF transition alone, but on the coexisting SC transition as well. The other feature caused by the low temperature relaxation behavior is a partial wipe-out of the La NQR signal in the same temperature region. The total intensity multiplied by $`T`$ and corrected for the echo decay factor (Fig.5) shows a decrease around 15 K by about a factor 3 indicating part of the nuclei to have dephased before the echo signal could be measured. The relative decrease of the intensity of the narrow line (1) starts below 70 K (inset of Fig.5) and becomes very pronounced below 30 K. It demonstrates a difference in relaxation behavior in the OP and OR phases. The imprint on other NQR features, if any, might be limited to the changes in the stretched exponential recoveries (see above).
How do these results compare to those in the low doped Sr-cuprates? In $`\mathrm{La}_{1.94}\mathrm{Sr}_{0.06}\mathrm{CuO}_4`$ the La spectra (NMR and NQR) show the existence of at least two sites below 200 K, of which one site is ascribed to AF clusters. The change in the La NMR spectrum around 20 K seems to signal glass formation. Wipe-out effects are clearly seen for Cu-NMR but no La-NQR wipe-out data are shown. In La<sub>1.48</sub>Sr<sub>0.12</sub>Nd<sub>0.4</sub>CuO<sub>4</sub> La wipe-out is found to be almost complete. In La<sub>2-x-y</sub>Sr<sub>x</sub>Eu<sub>y</sub>CuO<sub>4</sub> with $`0.08<x<0.18`$, Cu-NQR shows the presence of three inequivalent sites, which is also reminiscent to our case. At low doping at 1.3 K the Cu-intensity is strongly reduced compared to a $`x=1/8`$ sample due to wipe-out effects without affecting the relative strengths of the three lines. This comparison shows at least three major differences of the NQR spectra: in La<sub>2</sub>CuO<sub>4.02</sub> wipe-out effects on the La-site are less severe, and mainly linked to the relaxation peak and the antiferromagnetic transition below 17 K, line features are much sharper than in the Sr-doped cuprates, and the line splitting (hence internal field) in the AF state is almost the same as in the undoped case. The large wipe-out values over very extended temperature regions of the Sr-doped samples seem to be typical for mobile striped or glassy phases, which are not expected in our compound.
The most intriguing feature is the coincidence of the AF and superconducting transitions in the investigated single crystal. The AF ordering temperature is very low in comparison with $`T_N`$’s of macroscopically phase separated $`\mathrm{La}_2\mathrm{CuO}_{4+\mathrm{x}}`$ samples, which are usually higher than 200 K. The transition is sharp (Fig.3) and it looks as the AF state is strongly depressed at high temperatures and is triggered by some reason at low temperatures (the above mentioned relaxation behavior below 17K supports this assumption). It has been argued that superconductivity itself destabilizes the homogeneous metallic state and leads to the formation of (super)conducting droplets weakly linked to each other and separated by insulating (antiferromagnetic) regions. Our system is already inhomogeneous above the AF and SC transitions because the oxygen phase separation occurs at much higher temperatures but the hole concentration in OR grains is far from optimal. Therefore the occurrence of superconductivity in the OR grains might cause an additional electronic redistribution, which favors the AF transition in the OP grains. The other possibility is that both antiferromagnetism and superconductivity are triggered by a common mechanism e.g. slowing down of spin fluctuations coupled to hole motion occurring in the same temperature region.
In summary in single crystal of La<sub>2</sub>CuO<sub>4.02</sub> with limited oxygen mobility the three different La sites seen by NQR are associated with oxygen-poor, oxygen-rich and intermediate regions reminiscent the O-doped macroscopically separated system and the Sr-doped glassy or striped La<sub>2</sub>CuO<sub>4</sub> compounds. Sharper lines, well defined transition temperatures, internal fields close to bulk values and less severe wipe-out effects are seen as the main difference of the NQR data with low Sr-doped cuprates instead of O-doping. The evaluated OP grain size of the order of 8 nm shows that the crystal is phase separated on a mesoscopic scale. Although, like in the underdoped high-$`T_c`$ systems, superconductivity itself is not directly reflected in the NQR data, some features of the antiferromagnetic transition in the OP phase (a sharp transition at low temperature with $`T_NT_c`$ and <sup>139</sup>La relaxation peculiarities) are indicative for a possible coupling with the superconducting transition in this system.
We gratefully acknowledge Issa Abu-Shiekah for his help in the experiments. This work is supported in part by FOM-NWO, and NWO/INTAS-1010-CT93-0045. |
warning/0003/math-ph0003005.html | ar5iv | text | # Discrete Phase Integral Method for Five-Term Recursion Relations
## I Introduction
The purpose of this paper is to develop the formalism of the discrete phase integral (DPI), or Wentzel-Kramers-Brillouin method, for cases where the recursion relation involves five terms. Previous use of this method has, as far as we are aware, been limited to three-term recursion relations . Surprisingly, the extension to five terms is not routine, and entails novel physical and mathematical considerations. Further, once this extension is understood, no additional concepts are required in dealing with recursion relations involving still more terms.
The physical problem which led the author to consider this extension concerns the magnetic molecular cluster \[(tacn)<sub>6</sub>Fe<sub>8</sub>O<sub>2</sub>(OH)<sub>12</sub>\]<sup>8+</sup> (or just Fe<sub>8</sub> for short). This molecule has a total spin $`J=10`$ in its ground state, and crystallizes into a solid where the cluster has an approximate $`D_2`$ symmetry. The interaction between molecules is very weak, and the low temperature spin dynamics of a single molecule are well described by the Hamiltonian
$$=k_2J_z^2+(k_1k_2)J_x^2g\mu _B𝐉𝐇.$$
(1)
Here, $`𝐉`$ is a dimensionless spin operator, $`𝐇`$ is an externally applied magnetic field, and $`k_1>k_2>0`$. (Experiments reveal $`k_10.33`$ K, and $`k_20.22`$ K.)
The spectrum of the Hamiltonian (1) shows some extremely interesting features as a function of the applied field $`𝐇`$. In particular, one finds a large number of diabolical points in the $`H_x`$-$`H_z`$ plane , which have also been seen experimentally . Exactly as in the spectrum of a particle confined to a triangular box , some of the diabolical points (those arising when $`𝐇\widehat{𝐱}`$ or $`𝐇\widehat{𝐳}`$) can be related to a geometrical symmetry , but others can not. The problem was first studied by instanton methods when $`𝐇\widehat{𝐱}`$, but this method is much harder to apply for general field orientations, and the phase integral method proves to be simpler. The existence of diabolical points turns out to depend critically on having five terms in the recursion relation that we shall describe shortly, and three terms would never lead to such points. The calculations which pertain specifically to Fe<sub>8</sub> are described elsewhere , but it seems worthwhile to present the formal aspects of the work separately, as they are more generally applicable.
### A Heuristic discussion of the DPI approximation
It is useful to continue with the above example in order to introduce the DPI method. The starting point of the procedure is to write Schrödinger’s equation in the $`J_z`$ basis. Let $`|\psi =E|\psi `$, $`J_z|m=m|m`$, $`m|\psi =C_m`$, and $`m||m^{}=t_{m,m^{}}`$. Then,
$$\underset{n=m2}{\overset{m+2}{}}t_{m,n}C_n=EC_m.$$
(2)
The diagonal terms ($`t_{m,m}`$) in the above equation arise from the $`J_z^2`$ and $`J_zH_z`$ parts of $``$, those off-diagonal by one ($`t_{m,m\pm 1}`$) from the $`J_xH_x`$ and $`J_yH_y`$ parts, and those off-diagonal by two ($`t_{m,m\pm 2}`$) from the $`J_x^2`$ part.
The DPI method is applicable to a recursion relation such as (2) whenever the $`t_{m,m\pm \alpha }`$ ($`\alpha =0,1,2`$) vary sufficiently slowly with $`m`$. A physical analogy may be made with an electron hopping on a lattice with on-site energies $`t_{m,m}`$ and nearest-neighbor and next-nearest-neighbor hopping terms $`t_{m,m\pm 1}`$ and $`t_{m,m\pm 2}`$. If these quantities were independent of $`m`$, the solutions to Eq. (2) would be Bloch waves $`C_m=\mathrm{exp}(iqm)`$, with an energy
$$E=w_m+2t_{m,m+1}\mathrm{cos}q+2t_{m,m+2}\mathrm{cos}2qE(q),$$
(3)
where we have written $`w_mt_{m,m}`$ to highlight the physically different role of the on-site energy from the hopping terms. We shall use the notations $`w_m`$ and $`t_{m,m}`$ interchangably. If for fixed $`\alpha `$, the $`t_{m,m+\alpha }`$ vary slowly with $`m`$ (where the meaning of this term remains to be made precise), we expect it to be a good approximation to introduce a local Bloch wavevector, $`q(m)`$, and write $`C_m`$ as an exponential $`e^{i\mathrm{\Phi }}`$, whose phase $`\mathrm{\Phi }`$ accumulates approximately as the integral of $`q(m)`$ with increasing $`m`$, in exactly the same way that in the continuum quasiclassical method in one dimension, one writes the wavefunction as $`\mathrm{exp}(iS(x)/\mathrm{})`$, and approximates $`S(x)`$ as the integral of the local momentum $`p(x)`$.
It is obvious that the above approximation will entail the replacement of various sums by integrals, and to that end, we introduce smooth functions $`t_\alpha (m)`$ of a continuous variable $`m`$ as extensions of $`t_{m,m+\alpha }`$ such that whenever $`m`$ is an integer ,
$$t_\alpha (m)=(t_{m,m+\alpha }+t_{m,m\alpha })/2,\alpha =0,1,2.$$
(4)
We will try and choose these functions so that their derivatives are small. The precise way in which this is to be done will be discussed later, but supposing that we have been successful in finding such functions, we can seek to approximate $`C_m`$ in exact parallel with the continuum phase integral approach. The form of the solution that emerges, and which readers will readily appreciate from knowledge of the continuum case, is given by
$$C_m\frac{1}{\sqrt{v(m)}}\mathrm{exp}\left(i^mq(m^{})𝑑m^{}\right),$$
(5)
where $`q(m)`$ and $`v(m)`$ obey the equations
$`E`$ $`=`$ $`w(m)+2t_1(m)\mathrm{cos}q+2t_2(m)\mathrm{cos}(2q)_{\mathrm{sc}}(q,m),`$ (6)
$`v(m)`$ $`=`$ $`_{\mathrm{sc}}/q=2\mathrm{sin}q(m)\left(t_1(m)+4t_2(m)\mathrm{cos}q(m)\right).`$ (7)
\[Just as for the matrix elements, we define $`w(m)t_0(m)`$, and use the notation $`w(m)`$ when we want to emphasize it is as an on-site energy.\] The interpretation of these equations is exactly the same as in the continuum case. Thus, $`_{\mathrm{sc}}(q,m)`$ is a semiclassical Hamiltonian, $`q(m)`$ is a local wavevector as already mentioned, and $`v(m)`$ is the associated semiclassical electron velocity. We shall refer to Eq. (5) as the basic DPI form. Equation (6) is the eikonal or Hamilton-Jacobi equation, while Eq. (7) is the discrete counterpart of the transport equation. The presence of the lattice shows up in the $`q`$ dependence of $`_{\mathrm{sc}}(q,m)`$ through periodic functions, whereas in the continuum case, such dependence is typically of the form $`q^2`$.
As discussed by Braun , the DPI approximation has been employed in many problems in quantum mechanics where the Schrödinger equation turns into a three-term recursion relation in a suitable basis. All the types of problems as in the continuum case in can then be treated—Bohr-Sommerfeld quantization, barrier penetration, tunnel in symmetric double wells, etc. In addition, one can also use the method to give asymptotic solutions for various recursion relations of mathematical physics, such as those for the Mathieu equation, Hermite polynomials, Bessel functions, and so on. The general procedures are well known and simple to state. For any $`E`$, one solves the Hamilton-Jacobi and transport equations to obtain $`q(m)`$ and $`v(m)`$, and writes $`C_m`$ as a linear combinations of the independent solutions that result. The interesting features all arise from a single fact — that the DPI approximation breaks down at the so-called turning points. These are points where $`v(m)`$ vanishes. One must relate the DPI solutions on opposite sides of the turning point by connection formulas, and the solution of all the various types of problems mentioned above depends on judicious use of these formulas.
In this paper we will extend these ideas to five-term recursion relations, focussing especially on those features which arise over and above the three-term problem. Now, for any given $`E`$, there will be four DPI solutions (5), while in the three-term case there are only two, because the Hamilton-Jacobi equation (6) is a quartic in $`e^{iq}`$. These solutions will also break down at turning points—points where $`v(m)`$ vanishes. In contrast to the three-term case, we shall see that there are new types of turning points. It is these turning points which are responsible for the diabolical points in the spectrum of Fe<sub>8</sub>. The three-term problem turning points are analogous to those in the continuum quasiclassical method, but the new ones that we will find are not. In fact, they can only be described as lying “under the barrier” from the continuum viewpoint. These new or irregular turning points require new connection formulas, which it is our goal to provide.
The plan of our paper is as follows. We will examine the DPI approximation carefully in Sec. II, and see how it fails when $`v(m)`$ vanishes. The precise width of the failure zone is discussed in an Appendix. We will examine these failure or turning points in Sec. III, and see how the concept must be extended beyond the three-term and continuum cases. We will find that a turning point need not be a limit of the classically allowed motion, and we will categorize the different types of turning points that arise. We will conclude in Sec. IV by deriving connection formulas at the new turning points.
We will limit ourselves to problems where the matrix $`t_{m,n}`$ is real and symmetric, $`t_{m,n}=t_{n,m}`$, as it simplifies the analysis, and yet suffices to bring out all essential physical points. In our spin example, this means that we only consider fields in the $`x`$-$`z`$ plane. The extension to complex Hermitean matrices is cumbersome to carry through, but presents no difficulty of principle. We shall continue to couch our discussion in quantum mechanical language, thinking of $`E`$ as an energy eigenvalue, although from the mathematical viewpoint, this is not strictly necessary.
## II The basic DPI approximation
In this section, we will examine the DPI approximation in more detail. The argument proceeds in close analogy with the continuum case. We begin by restating the approximation in a slightly different way . Dividing Eq. (2) by $`C_m`$, and writing $`\zeta _{m+1}=C_{m+1}/C_m`$, we get
$$t_{m,m2}\zeta _{m1}^1\zeta _m^1+t_{m,m1}\zeta _m^1+t_{m,m}+t_{m,m+1}\zeta _{m+1}+t_{m,m+2}\zeta _{m+1}\zeta _{m+2}=E.$$
(8)
If $`t_{m,m+\alpha }`$ for fixed $`\alpha `$ is almost the same over some range $`K1`$ of $`m`$’s, then we will get almost the same numerical equation for the $`\zeta _m`$’s over this range, and we will clearly obtain a good approximate solution if we replace ratios like $`\zeta _{m+1}/\zeta _m`$ by unity. This leads to
$$t_{m,m2}\zeta _m^2+t_{m,m1}\zeta _m^1+t_{m,m}+t_{m,m+1}\zeta _m+t_{m,m+2}\zeta _m^2=E,$$
(9)
which is a solvable quartic equation in $`\zeta _m`$ (which is exactly like the factor $`e^{iq(m)}`$ introduced in Sec. I). The corresponding approximation for $`C_m`$ is
$$C_m\underset{k=m_a}{\overset{m}{}}\zeta _k,$$
(10)
where $`m_a`$ is a suitable starting value. One could now use this approximation to find the ratio $`\zeta _{m+1}/\zeta _m`$, substitute this value for the ratio in Eq. (8), and solve again for $`\zeta _m`$. The process could be further iterated if desired.
The product in Eq. (10) is more easily evaluated as a sum by taking logarithms. Further, if the $`\zeta _m`$’s do not vary rapidly from one $`m`$ to the next, the sum will be well approximated by an integral. This makes it necessary to introduce continuum extensions of the matrix elements $`t_{m,m+\alpha }`$. We turn therefore to this problem, and discuss the condition for slow variation more clearly. We are seeking functions $`t_\alpha (m)`$ such that
$$t_\alpha (m)=(t_{m,m+\alpha }+t_{m,m\alpha })/2,(\alpha =0,1,2),$$
(11)
whenever $`m`$ is an integer. There are infinitely many such functions, and we restrict them by imposing further conditions on a certain number of their higher derivatives. For $`t_1(m)`$, e.g., we could also demand (using dots to denote derivatives with respect to $`m`$),
$`\dot{t}_1(m)`$ $`=`$ $`t_{m,m+1}t_{m,m1},`$ (12)
$`\ddot{t}_1(m)`$ $`=`$ $`\frac{1}{2}\left(t_{m+1,m+2}t_{m,m+1}t_{m,m1}+t_{m+1,m+2}\right).`$ (13)
Similar conditions can be imposed on $`t_0(m)`$ and $`t_1(m)`$. In general, we need conditions only up to some small order for practical applications. Up to the degree of approximation in Eq. (5), e.g., we need only go up to second derivatives.
Since the matrix elements are assumed to vary slowly, we want these derivatives to be small, and this condition is best codified in terms of a small parameter $`ϵ`$, which plays the same role as $`\mathrm{}`$ in the continuum case, such that $`t_\alpha `$ is formally of order $`ϵ^0`$, $`\dot{t}_\alpha `$ of order $`ϵ`$, $`\ddot{t}_\alpha `$ of order $`ϵ^2`$, and so on. For spin Hamiltonians such as Eq. (1), this parameter is $`1/J`$. The quasiclassicality conditions then read
$$\frac{dt_\alpha }{dm}=O\left(\frac{t_\alpha (m)}{J}\right),\frac{d^2t_\alpha }{dm^2}=O\left(\frac{t_\alpha (m)}{J^2}\right),$$
(14)
etc. We will continue to use $`1/J`$ as a generic small parameter in the rest of our analysis. A problem in which one can not find functions $`t_\alpha (m)`$ obeying Eq. (14) will not be amenable to a phase integral approximation.
One small point should be kept in mind while judging orders of smallness in the spin problem, and others like it. There are natural algebraic expressions for the $`t_\alpha (m)`$’s in which $`m`$ appears only in the combination $`m/J`$ or $`m/[J(J+1)]^{1/2}`$. Thus, derivatives with respect to $`m`$ are automatically smaller by an order $`J^1`$. As $`J\mathrm{}`$, however, the classical quantity is not $`m`$, but $`m/J`$. Thus the ratio $`m/J`$ should be regarded as a quantity of order unity and not $`J^1`$.
With this lengthy preamble, we are ready to solve our recursion relation. As in the continuum case, we make the exponential substitution,
$$C_m=e^{i\mathrm{\Phi }(m)}.$$
(15)
The first approximation is obtained if we assume that $`\dot{\mathrm{\Phi }}(m)`$ varies slowly so that $`\ddot{\mathrm{\Phi }}(m)`$ may be neglected. Then $`C_{m\pm \alpha }C_me^{i\alpha \dot{\mathrm{\Phi }}}`$, and substituting this into the recurrence relation (2) with $`t_{m,m\pm \alpha }t_\alpha (m)`$, we obtain the Hamilton-Jacobi equation (6) with $`\dot{\mathrm{\Phi }}(m)=q(m)`$.
To proceed more systematically, we look for a solution for $`\mathrm{\Phi }(m)`$ as a series in inverse powers of $`J`$:
$$\mathrm{\Phi }=\mathrm{\Phi }_0+\mathrm{\Phi }_1+\mathrm{\Phi }_2+\mathrm{},$$
(16)
where,
$`\mathrm{\Phi }_n`$ $`=`$ $`O(J^{1n}),`$ (17)
$`\dot{\mathrm{\Phi }}_n`$ $`=`$ $`O(\mathrm{\Phi }_n/J),`$ (18)
$`\ddot{\mathrm{\Phi }}_n`$ $`=`$ $`O(\mathrm{\Phi }_n/J^2),`$ (19)
and so on. The successive inverse powers of $`J`$ in the derivatives are expected since we expect the $`\mathrm{\Phi }_n`$ to be simple functions of the $`t_{m,m+\alpha }`$, and we shall soon see whether or not this expectation is fulfilled. $`\mathrm{\Phi }_0`$ is our zeroth order approximation above, and so we set
$$\dot{\mathrm{\Phi }}_0(m)=q(m)$$
(20)
from the outset. As in the continuum case, we need to keep terms up to $`\mathrm{\Phi }_2`$ in order to decide if the approximation is succeeding .
Up to terms of order $`1/J^2`$ relative to the leading one, we have
$`C_{m\pm 1}`$ $`=`$ $`C_m\mathrm{exp}\left[i\left(\pm q\pm \dot{\mathrm{\Phi }}_1\pm \dot{\mathrm{\Phi }}_2+{\displaystyle \frac{1}{2}}(\dot{q}+\ddot{\mathrm{\Phi }}_1)\pm {\displaystyle \frac{1}{6}}\ddot{q}+\mathrm{}\right)\right]`$ (21)
$`=`$ $`C_me^{\pm iq}\left[1+{\displaystyle \frac{i}{2}}(\dot{q}\pm 2\dot{\mathrm{\Phi }}_1){\displaystyle \frac{1}{8}}(\dot{q}\pm 2\dot{\mathrm{\Phi }}_1)^2\pm i\dot{\mathrm{\Phi }}_2+{\displaystyle \frac{i}{2}}\ddot{\mathrm{\Phi }}_1\pm {\displaystyle \frac{i}{6}}\ddot{q}+\mathrm{}\right].`$ (22)
We now wish to substitute this form into our recursion relation. We would like to use the continuum forms $`t_\alpha (m)`$ instead of the discrete matrix elements $`t_{m,m+\alpha }`$. Just as there are infinitely many continuous functions which we could take, there are many approximants for $`t_{m,m+\alpha }`$. Note in particular, that there is no unique way to “solve” Eqs. (11)–(13) for the matrix elements in terms of the continuous functions. The simplest procedure is to take $`t_{m,m}=t_0(m)`$, $`t_{m,m\pm 1}=t_1(m\pm \frac{1}{2})`$, and $`t_{m,m\pm 2}=t_2(m\pm 1)`$. For $`t_{m,m\pm 1}`$, in particular, a Taylor expansion of this approximation gives
$$t_{m,m\pm 1}=t_1(m)\pm \frac{1}{2}\dot{t}_1(m)+\frac{1}{8}\ddot{t}_1(m)+\mathrm{},$$
(23)
where the error is of order $`J^3`$. Therefore,
$$\underset{n=m\pm 1}{}t_{m,n}C_n=t_1(m)\alpha _1(m)+\dot{t}_1(m)\alpha _2(m)+\ddot{t}_1(m)\alpha _3(m)+\mathrm{},$$
(24)
where
$`\alpha _1(m)`$ $`=`$ $`2C_m[\mathrm{cos}q+{\displaystyle \frac{i}{2}}(\dot{q}\mathrm{cos}q+2i\dot{\mathrm{\Phi }}_1\mathrm{sin}q){\displaystyle \frac{1}{8}}(\dot{q}^2+4\dot{\mathrm{\Phi }}_1^24i\ddot{\mathrm{\Phi }}_1)\mathrm{cos}q.`$ (25)
$`.{\displaystyle \frac{i}{6}}(3\dot{q}\dot{\mathrm{\Phi }}_16i\dot{\mathrm{\Phi }}_2i\ddot{q})\mathrm{sin}q],`$ (26)
$`\alpha _2(m)`$ $`=`$ $`iC_m\left[\mathrm{sin}q+{\displaystyle \frac{i}{2}}\dot{q}\mathrm{sin}q+\dot{\mathrm{\Phi }}_1\mathrm{cos}q\right],`$ (27)
$`\alpha _3(m)`$ $`=`$ $`{\displaystyle \frac{1}{4}}C_m\mathrm{cos}q.`$ (28)
Note that we have kept only terms up to order $`1/J^2`$ in $`\alpha _1`$, $`1/J`$ in $`\alpha _2`$, and $`J^0`$ in $`\alpha _3`$, since relative to $`t_1(m)`$, $`\dot{t}_1`$ and $`\ddot{t}_1`$ are of order $`1/J`$ and $`1/J^2`$, respectively.
Similarly,
$$\underset{n=m\pm 2}{}t_{m,n}C_n=t_2(m)\beta _1(m)+\dot{t}_2(m)\beta _2(m)+\ddot{t}_2(m)\beta _3(m)+\mathrm{},$$
(29)
where
$`\beta _1(m)`$ $`=`$ $`2C_m[\mathrm{cos}2q+2i(\dot{q}\mathrm{cos}2q+i\dot{\mathrm{\Phi }}_1\mathrm{sin}2q)2(\dot{q}^2+\dot{\mathrm{\Phi }}_1^2i\ddot{\mathrm{\Phi }}_1)\mathrm{cos}2q.`$ (30)
$`.{\displaystyle \frac{2i}{3}}(6\dot{q}\dot{\mathrm{\Phi }}_13i\dot{\mathrm{\Phi }}_22i\ddot{q})\mathrm{sin}2q],`$ (31)
$`\beta _2(m)`$ $`=`$ $`2iC_m\left[\mathrm{sin}2q+2i\dot{q}\mathrm{sin}2q+2\dot{\mathrm{\Phi }}_1\mathrm{cos}2q\right],`$ (32)
$`\beta _3(m)`$ $`=`$ $`C_m\mathrm{cos}2q.`$ (33)
We now substitute these relations into the recurrence relation (2), and equate equal powers of $`J`$. The terms of order $`J^0`$ obviously give Eq. (6), while those of order $`J^1`$ give, after some work,
$$2\dot{\mathrm{\Phi }}_1(t_1\mathrm{sin}q+2t_2\mathrm{sin}2q)=i\frac{d}{dm}\left(t_1\mathrm{sin}q+2t_2\mathrm{sin}2q\right).$$
(34)
The left hand side of this equation is $`\dot{\mathrm{\Phi }}_1v(m)`$ \[see Eq. (7)\], while the right hand side may be written as $`(i/2)dv(m)/dm`$. Thus, integration yields (after a suitable choice of an indefinite constant)
$$\mathrm{\Phi }_1(m)=\frac{i}{2}\mathrm{ln}v(m).$$
(35)
Note that as assumed in Eq. (18), $`\mathrm{\Phi }_1=O(J^0)`$. Pending a demonstration that $`\mathrm{\Phi }_2(m)`$ is negligible, we have arrived at the basic DPI form (5), which we now see as the first two terms of an asymptotic expansion of $`\mathrm{\Phi }`$ in inverse powers of $`J`$.
The equation for $`\mathrm{\Phi }_2(m)`$ is considerably more involved. After some analysis, we find
$$\frac{d\stackrel{~}{\mathrm{\Phi }}_2}{dm}=\frac{1}{8}\frac{d^2r}{dm^2}+\frac{i}{4}r\frac{d\dot{\mathrm{\Phi }}_1}{dm},$$
(36)
where we have defined
$`\stackrel{~}{\mathrm{\Phi }}_2=\mathrm{\Phi }_2+{\displaystyle \frac{1}{24}}{\displaystyle \frac{t_1+16t_2\mathrm{cos}q}{t_1+4t_2\mathrm{cos}q}\ddot{q}𝑑m},`$ (37)
$`r(m,q(m))={\displaystyle \frac{t_1\mathrm{cos}q+4t_2\mathrm{cos}2q}{t_1\mathrm{sin}q+2t_2\mathrm{sin}2q}}.`$ (38)
Thus,
$$\mathrm{\Phi }_2=\frac{1}{24}\frac{t_1+16t_2\mathrm{cos}q}{t_1+4t_2\mathrm{cos}q}\ddot{q}𝑑m+\frac{1}{8}\frac{dr}{dm}\frac{1}{8}r\frac{d^2}{dm^2}\mathrm{ln}vdm.$$
(39)
Based on power counting, this is indeed of order $`J^1`$, since $`\ddot{q}`$ and $`d^2\mathrm{ln}v(m)/dm^2`$ are of order $`J^2`$. Thus, $`|\mathrm{\Phi }_2|\mathrm{\Phi }_1=O(J^0)`$. However, it is plain that this condition is violated whenever $`v(m)`$ approaches zero, for then both $`\mathrm{ln}v(m)`$ and $`r(m)`$ diverge. To find the actual magnitude of $`\mathrm{\Phi }_2`$, we need to know how $`q(m)`$ and $`v(m)`$ behave near a turning point. This behavior is found in the next section \[see Eqs. (53) and (54)\]. The magnitude of $`\mathrm{\Phi }_2`$ is estimated in Appendix A, where we show that the width of the zone where DPI fails is of order $`J^{1/3}`$.
## III Turning points
We now turn to a study of the points where $`v(m)=_{\mathrm{sc}}(q,m)/q=0`$. We shall call all such points turning points in analogy with the continuum case. In contrast to that case, however, we will find that turning points are not just the limits of the classical motion for a given energy, once the notion of the classically accessible region is suitably understood.
Since we must also obey the eikonal equation (6) in addition to the condition $`v(m)=0`$, at a turning point both $`m`$ and $`q`$ are determined if $`E`$ is given. Setting $`v=0`$ in Eq. (7), we see that we must have either $`q=0`$, or $`q=\pi `$, or $`q=q_{}(m)`$, where
$$\mathrm{cos}q_{}(m)=t_1(m)/4t_2(m).$$
(40)
Substituting these value of $`q`$ in the eikonal equation, we see that a turning point arises whenever
$$E=U_0(m),U_\pi (m),\mathrm{or}U_{}(m),$$
(41)
where,
$`U_0(m)`$ $`=`$ $`_{\mathrm{sc}}(0,m)=w(m)+2t_1(m)+2t_2(m),`$ (42)
$`U_\pi (m)`$ $`=`$ $`_{\mathrm{sc}}(\pi ,m)=w(m)2t_1(m)+2t_2(m),`$ (43)
$`U_{}(m)`$ $`=`$ $`_{\mathrm{sc}}(q_{},m)=w(m)2t_2(m){\displaystyle \frac{t_1^2(m)}{4t_2(m)}}.`$ (44)
Note that $`q_{}(m)`$ may be complex for some $`m`$, but since $`\mathrm{cos}q_{}`$ is always real, $`U_{}`$ is real for all $`m`$. We shall refer to these three energy curves as critical curves.
In the one-dimensional continuum case where $`_{\mathrm{sc}}=(p^2/2m)+V(x)`$, the condition $`v(x)=0`$ is equivalent to $`E=V(x)`$. The latter condition marks the edge of the classically allowed region, $`E<V(x)`$. Let us recall why this is said to be so. A particle at a point $`x`$ where $`V(x)>E`$ must be ascribed an imaginary momentum. To understand the analogous condition in the discrete case, let us return to the analogy of an electron in a one-dimensional lattice with $`m`$-independent matrix elements $`t_{m,m+\alpha }`$. Equation (3) gives the dispersion relation for an energy band $`E(q)`$. The classicaly allowed range of energies is now defined by the limits of this band since (provided one is not too close to a band edge) a spatially localized electron with a mean energy in this range can be constructed as a wavepacket out of Bloch states with only real wavevectors. Alternatively, we could say that solutions with energy in the allowed range correspond to travelling waves, while solutions outside this range correspond to evanescent waves. These notions continue to be valid when $`t_\alpha (m)`$ are slowly varying with $`m`$. The dispersion relation may be taken as $`_{\mathrm{sc}}(q,m)`$ for any fixed value of $`m`$. The band is now $`m`$-dependent, and so in particular are the band edges, which we denote by $`U_{}(m)`$ (lower edge) and $`U_+(m)`$ (upper edge).
To find the band edges, we note that by definition, the energy at an edge is either a minimum or maximum, so $`_{\mathrm{sc}}(q,m)/q=0`$, i.e., $`v(m)=0`$. \[The converse is not true, i.e., $`v(m)=0`$ need not always define a band edge.\] Thus the wavevectors at the band edges are again to be found in the set $`q=0`$, $`\pi `$, and $`q_{}`$, with the answers depending on the signs and magnitudes of $`t_1`$ and $`t_2`$. \[We do not distinguish between $`q=\pi `$ and $`q=\pi `$, or between $`q=q_{}`$ and $`q=q_{}`$, as $`_{\mathrm{sc}}(q,m)`$ is an even function of $`q`$.\] To narrow down the number of cases to be considered, we observe that the gauge transformation $`C_m(1)^mC_m`$ changes the sign of $`t_1`$. Hence, we may assume $`t_1<0`$ without loss of generality. Consider the case $`t_2>0`$ first. Then the upper band edge is always located at $`q=\pi `$, while the lower band edge is located at $`q=0`$ if $`t_1/4t_2<1`$, and at $`q_{}=\mathrm{cos}^1(t_1/4t_2)`$ if $`1<t_1/4t_2<0`$. The case $`t_2<0`$ is completely analogous. Now the lower band edge is always at $`q=0`$, while the upper band edge is at $`q=\pi `$ for $`t_1/4t_2>1`$, and at $`q_{}=\mathrm{cos}^1(t_1/4t_2)`$ for $`0<t_1/4t_2<1`$. The different types of band energy curves that can arise are illustrated in Fig. 1. In the rest of this paper we shall carry out the analysis in detail assuming $`t_2>0`$. The case $`t_2<0`$ is easily treated in parallel, and we shall only give the final results where these are significantly different.
The conditions Eq. (41) for a turning point are analogous to the requirement that $`E=V(x)`$ in the continuum case. Correspondingly, it helps in visualization to draw all three critical curves versus $`m`$, and a horizontal line indicating the energy. Any intersection of this line with a critical curve is a turning point. See, e.g., Figs. 2 and 3.
Not every critical curve need be a band edge curve, however. Since we have chosen $`t_1(m)<0`$ and $`t_2(m)>0`$, the upper edge curve $`U_+(m)`$ is the same as $`U_\pi (m)`$ for all $`m`$, but the lower edge curve, $`U_{}(m)`$, may be $`U_0(m)`$ for some values of $`m`$, and $`U_{}(m)`$ for other values as discussed above. It turns out to be useful to introduce a dual labelling scheme for the critical curves and write
$`U_0(m)=U_i(m),U_{}(m)=U_{}(m),\mathrm{if}q_{}(0,\pi ),`$ (45)
$`U_0(m)=U_{}(m),U_{}(m)=U_f(m),\mathrm{if}q_{}(0,\pi ).`$ (46)
We have already noted that $`U_\pi (m)=U_+(m)`$. The subscripts $`i`$ and $`f`$ stand for “internal” and “forbidden”, since in the first case above, $`U_0(m)`$ lies inside the classically allowed energy range, while in the second case, $`U_{}(m)`$ lies outside this range. The turning points with $`E=U_i`$ and $`E=U_f`$ have no analogues in continuum quantum mechanical problems.
Before turning to a classification of the various turning points, however, it is useful to record some further properties of the critical curves. The first property is that $`U_0(m)U_{}(m)`$, since
$$U_0(m)U_{}(m)=\frac{1}{4t_2(m)}\left(t_1(m)+4t_2(m)\right)^2.$$
(47)
Differentiating this equation with respect to $`m`$, it follows that the case of equality, $`U_0(m)=U_{}(m)`$, happens at a point where both curves have a common tangent. These facts are illustrated in Fig. 4. Further, at the point of contact, which we denote by $`m^{}`$, $`t_1(m)/4t_2(m)=1`$, which is precisely the condition derived above for the lower band edge to change from $`q=0`$ to $`q=q_{}`$.
The second property provides an alternative way of viewing the condition $`E=U_{}(m)`$. Solving the eikonal equation (6) for $`\mathrm{cos}q`$ we obtain
$`\mathrm{cos}q(m)`$ $`=`$ $`{\displaystyle \frac{t_1(m)\pm [t_1^2(m)4t_2(m)f(m)]^{1/2}}{4t_2(m)}};`$ (48)
$`f(m)`$ $`=`$ $`w(m)2t_2(m)E.`$ (49)
Since $`\mathrm{cos}q=t_1/4t_2`$ at $`q=q_{}`$, the discriminant in Eq. (48) must vanish, and we must have
$$t_1^2(m)=4t_2(m)\left(w(m)2t_2(m)E\right)(q=q_{}).$$
(50)
It is easily verified that this equality is identical to $`E=U_{}(m)`$.
We now turn to discussing the different types of turning points:
Type $`A`$: $`E=U_{}(m)`$ when $`U_{}=U_0`$. See, e.g., Fig. 2. The region $`mm_c`$ is classically allowed. This is analogous to what happens in the conventional continuum quasiclassical method — the turning point is located at the boundary of the classically accessible region for the energy given. For $`m`$ just less than $`m_c`$, there are two solutions of the Hamilton-Jacobi equation (6) with $`q[\left(EU_{}(m)\right)/a]^{1/2}`$, where $`a=(t_1+4t_2)`$. For $`m`$ just greater than $`m_c`$, these values of $`q`$ continue on to the imaginary axis. The corresponding wavefunctions $`C_m`$ change from slowly oscillatory for $`m<m_c`$ to exponentially growing and decaying for $`m>m_c`$. The connection formulas for these solutions are exactly like those in the continuum case, and may be derived as in Refs. . Note that the other two solutions of the Hamilton-Jacobi equation evolve smoothly, and the corresponding DPI wavefunctions $`C_m`$ do not need to be “connected” across this turning point.
Type $`\overline{A}`$: $`E=U_+(m)`$. See Fig. 2 again. The region $`mm_b`$ is classically allowed. This case is physically very similar to type $`A`$ in that the turning point is at the boundary of the classically allowed and forbidden regions. Now, however, $`q\pi `$ in the transition zone, so the wavefunctions contain a rapidly oscillating factor $`(1)^m`$ in addition to all the other variation. Although the connection formulas can be derived from those for type $`A`$ turning points by means of the transformation $`C_m(1)^mC_m`$, as shown in Ref. , their detailed form has a very different superficial look.
Type $`A^{}`$: $`E=U_i(m)`$. Consider Fig. 3, and the energy $`E`$ shown there, which intersects $`U_i(m)`$ at $`m=m_c`$. In Fig. 5 we sketch energy bands for this problem for several values of $`m`$. For $`m`$ just less than $`m_c`$, $`_{\mathrm{sc}}(q,m)=E`$ in just two places, which we denote by $`\pm q(m)`$. (We do not show the solution $`q(m)`$ explicitly.) For $`m`$ just greater than $`m_c`$, two new intersections develop at $`\pm q^{}(m)0`$. Thus for $`m>m_c`$, our wavefunction consists of a sum of four basic solutions (5), all oscillatory, while for $`m<m_c`$ we have two oscillatory solutions \[associated with $`\pm q(m)`$\], and two exponentialy decaying or growing solutions \[associated with $`\pm q^{}(m)`$\]. The latter solutions must be related across the turning point by connection formulas, which are completely identical to those for type $`A`$. The fact that we may have exponentially decaying and growing solutions inside a classically allowed region is unexpected from prior experience with the continuum quantum mechanical problems, and underscores the point that this turning point has no analogue there. We are unaware if it has ever been considered in other physical situations where a continuum phase integral approach may be applied.
Type $`B`$: $`E=U_f(m)`$. This turning point is perhaps the most interesting of all. Since the energy now lies outside the classically allowed range, the point lies “under the barrier”, and at it, $`q`$ must be purely imaginary. Consider Fig. 4, and the energy $`E`$, which intersects $`U_f`$ at $`m=m_c`$. We see that for $`mm_c`$, there are two solutions to the Hamilton-Jacobi equation (6) with $`q=\pm i\kappa `$, where $`\kappa `$ is real. For $`m>m_c`$, these solutions acquire a real part as well, so that $`C_m`$ \[See Eq. (5)\] changes from a decaying (or growing) exponential to an oscillating solution with an exponentially decaying (or growing) envelope. As for type $`A^{}`$ points, this behavior has no analogue in continuum quantum mechanical problems, and we are unaware of prior analyses in other contexts. We shall derive connection formulas for this case in Sec. IV.
Type $`B^{}`$: $`E=U_{}(m)`$ when $`U_{}(m)=U_{}(m)`$. This turning point is like type $`A`$ in that the energy lies at the lower limit of the classically allowed region, but like type $`B`$ in that $`q0`$. The solutions which must be connected are purely oscillatory on the classically allowed side, with $`q\pm q_{}`$, and oscillatory exponentials (growing or decaying) on the forbidden side. The connection formulas are similar to those for case $`B`$.
Our nomenclature for the turning points may have become evident to the reader. The letter $`A`$ indicates that $`q=0`$ or $`\pi `$, while $`B`$ indicates that $`q=q^{}`$. A bar designates cases where $`q`$ has a value close to $`\pi `$, leading to a oscillatory factor in $`C_m`$ close to $`(1)^m`$, and a prime indicates cases where $`U_{}`$ is either $`U_{}`$ or $`U_+`$ at the turning point value of $`m`$. Thus in the case $`t_1<0`$, $`t_2<0`$, we would have turning points of type $`A`$, $`\overline{A}`$, $`\overline{A}^{}`$, $`\overline{B}`$, and $`\overline{B}^{}`$. Note that the mathematical aspects of the turning point, i.e., connection formulas, are governed by the value of $`q`$ at the turning point, but its physical nature is governed by whether the energy lies at the boundary (cases $`A`$, $`\overline{A}`$, $`B^{}`$, and $`\overline{B}^{}`$), in the interior (cases $`A^{}`$ and $`\overline{A}^{}`$), or in the exterior (cases $`B`$ and $`\overline{B}`$) of the classically allowed range.
It will now be apparent that in problems with three-term recursion relations, where $`_{\mathrm{sc}}(q,m)=w(m)+2t_1(m)\mathrm{cos}q`$, and $`v(m)=2t_1(m)\mathrm{sin}q(m)`$, the only turning points are at band edges, with $`q=0`$ or $`\pi `$, i.e., of type $`A`$ or $`A^{}`$. It is also apparent that no new points are involved in problems with further neighbor hopping, i.e., recursion relations with seven or more terms. Turning points are encountered whenever the energy lies on a critical curve, where this term now describes all curves in the $`E`$-$`m`$ plane on which $`_{\mathrm{sc}}(q,m)/q=0`$. As $`m`$ is varied through each turning point, two roots of the Hamilton-Jacobi equation for $`q`$ approach each other parallel to either the real or imaginary axis, coalesce, and move apart in the orthogonal direction . Consider for example the band structure in Fig. 6, which could arise from a recursion relation with seven or more terms. The accompanying critical curves are shown in Fig. 7. For the energy $`E`$ shown, if $`m<m_a`$, there is only one real solution for $`q`$, i.e. $`q_+`$. (We do not explicitly mention negative values of $`q`$.) At $`m=m_a`$, a new real value of $`q`$, $`q_2`$, enters the picture, and splits into two solutions $`q_2`$ and $`q_{2+}`$ as $`m`$ increases further.
We conclude this section by finding the general behavior of $`q(m)`$ and $`v(m)`$ near a turning point $`m=m_c`$. In its vicinity we may write
$$_{\mathrm{sc}}(q,m)=_{\mathrm{sc}}(q,m_c)+(mm_c)\frac{_{\mathrm{sc}}}{m}|_{m_c}+\mathrm{}.$$
(51)
By definition, however, $`v(m_c)=0`$, and so $`_{\mathrm{sc}}(q,m)E`$, and $`[_{\mathrm{sc}}(q,m_c)E]/q`$ both vanish. If we write $`q(m_c)=q_c`$, expand the the right hand side of Eq. (51) in powers of $`qq_c`$ as well, and retain only the leading non vanishing terms in $`qq_c`$ and $`mm_c`$, we obtain
$$_{\mathrm{sc}}(q,m)Ea(qq_c)^2+b(mm_c)+\mathrm{},$$
(52)
where $`a`$ and $`b`$ are constants. If we regard $`w(m)`$ and $`t_\alpha (m)`$ as being of order $`J^0`$, then by Eq. (14), we have $`a=O(J^0)`$, and $`b=O(1/J)`$, and therefore
$`q(m)q_c`$ $``$ $`[(mm_c)/J]^{1/2},`$ (53)
$`v(m)`$ $``$ $`[(mm_c)/J]^{1/2}.`$ (54)
These formulas prove useful when connection formulas are derived. Their usefulness is limited, however, if there is another turning point very close to $`m_c`$. In this case we should keep terms of order $`(mm_c)^2`$. The requisite analysis is very similar to that of quadratic turning points in the continuum case, but we shall not have any occasion to pursue it further. In all the calculations we have done for the Fe<sub>8</sub> or other spin problems , we have been able to sidestep the associated quadratic connection formulas by directly matching the solutions in the forbidden region to solutions of the Schrödinger equation for a harmonic oscillator.
## IV Connection Formulas for Forbidden Region Turning Points
We turn at last to the problem of finding connection formulas at the turning points. The formulas for points of type $`A`$, $`\overline{A}`$, etc. are quoted by Braun , so we will only consider points of type $`B`$ and $`\overline{B}`$. Our procedure is a small modification of that used by Schulten and Gordon . Suppose the turning point is at $`m=m_c`$, and $`q(m_c)=q_c`$. As shown in Appendix A, the DPI solution (5) fails in a window $`\mathrm{\Delta }m|mm_c|O(J^{1/3})`$, which we shall refer to as the failure zone. The first step is therefore to find another approximation that holds in the larger window (which we refer to as the central zone) $`\mathrm{\Delta }mJ^\eta `$, where $`\eta >1/3`$. To do this we write $`C_m`$ as an $`e^{iq_cm}`$ times a slowly varying factor, $`y_m`$, for which we then derive an approximate second order differential equation. The second step is to asymptotically match solutions of this differential equation to the DPI solutions in the overlap zones $`J^\eta \mathrm{\Delta }mJ^{1/3}`$ on either side of the turning point where both types of solutions are valid. The last step is to directly write down the transformation matrix between the coefficients of the linear combination of DPI solutions for $`m<m_c`$ to those for $`m>m_c`$, without having to consider the solution in the intermediate zone. We will carry out these three steps only to an order necessary to match the solutions to the accuracy represented by Eq. (5), i.e., to order $`J^0`$ in the phase $`\mathrm{\Phi }(m)`$ introduced after Eq. (2) or in Eq. (15).
Let us assume as before that $`t_1<0`$, $`t_2>0`$, and first consider turning points of type $`B`$. We will denote quantities evaluated at $`m=m_c`$ by a subscript $`c`$: $`t_1(m_c)=t_{1c}`$, $`\dot{t}_1(m_c)=\dot{t}_{1c}`$, etc. Let $`q`$ be pure imaginary for $`mm_c`$, and precisely at $`m_c`$ let us write
$$q(m_c)=i\sigma _2\kappa _c,$$
(55)
where $`\kappa _c>0`$ and $`\sigma _2=\pm 1`$. Putting this in Eqs. (40) and (6) we have
$`t_{1c}`$ $`=`$ $`4t_{2c}\mathrm{cosh}\kappa _c.`$ (56)
$`E`$ $`=`$ $`w_c+2t_{1c}\mathrm{cosh}\kappa _c+2t_{2c}\mathrm{cosh}2\kappa _c.`$ (57)
To carry out step 1, we write
$$C_me^{\sigma _2\kappa _c(mm_c)}y(m),$$
(58)
where $`\dot{y}(m)\kappa _c`$. Assuming that this is so, we write
$$C_{m\pm k}=e^{\sigma _2\kappa _c(mm_c)}e^{\sigma _2k\kappa _c}[y\pm k\dot{y}+\frac{1}{2}k^2\ddot{y}+\mathrm{}].$$
(59)
This approximation will hold provided $`\dot{y}y`$, and $`\ddot{y}\dot{y}`$ throughout the central zone $`|mm_c|J^\eta `$. \[We anticipate, in fact, that throughout this zone, $`\dot{y}(m)J^\gamma y(m)`$ where $`\gamma >0`$.\] That this is so and that higher order derivatives can be neglected in Eq. (59) will be justified post facto. Substituting Eq. (59) in Eq. (2) we obtain
$$A_0(m)\ddot{y}(m)+A_1(m)\dot{y}(m)+A_2(m)y(m)0,$$
(60)
where
$`A_0`$ $`=`$ $`{\displaystyle \underset{k=\pm 1,\pm 2}{}}{\displaystyle \frac{1}{2}}k^2e^{\sigma _2k\kappa _c}t_{m,m+k}.`$ (61)
$`A_1`$ $`=`$ $`{\displaystyle \underset{k=\pm 1,\pm 2}{}}ke^{\sigma _2k\kappa _c}t_{m,m+k},`$ (62)
$`A_2`$ $`=`$ $`{\displaystyle \underset{k=\pm 1,\pm 2}{}}e^{\sigma _2k\kappa _c}t_{m,m+k}+w_mE,`$ (63)
The differential equation that we are seeking for $`y(m)`$ need only hold in window of width $`O(J^\eta )`$ with $`\eta >1/3`$ around $`m_c`$. If we choose $`\eta <1`$, we can use the fact that $`w(m)`$ and $`t_\alpha (m)`$ are slowly varying, and use the expansions
$`t_{m,m\pm k}t_k(m)\pm {\displaystyle \frac{1}{2}}k\dot{t}_k(m),`$ (64)
$`t_k(m)t_{kc}+(mm_c)\dot{t}_{kc},`$ (65)
etc. Doing this, we obtain
$`A_0`$ $``$ $`a_1,`$ (66)
$`A_1`$ $``$ $`a_2\sigma _2(mm_c)b_2,`$ (67)
$`A_2`$ $``$ $`\sigma _2a_3+(mm_c)b_3,`$ (68)
where
$`a_1`$ $`=`$ $`t_{1c}\mathrm{cosh}\kappa _c+4t_{2c}\mathrm{cosh}2\kappa _c,`$ (69)
$`a_2`$ $`=`$ $`\dot{t}_{1c}\mathrm{cosh}\kappa _c+4\dot{t}_{2c}\mathrm{cosh}2\kappa _c,`$ (70)
$`b_2`$ $`=`$ $`2\mathrm{sinh}\kappa _c(\dot{t}_{1c}+4\dot{t}_{2c}\mathrm{cosh}\kappa _c),`$ (71)
$`a_3`$ $`=`$ $`\mathrm{sinh}\kappa _c(\dot{t}_{1c}+4\dot{t}_{2c}\mathrm{cosh}\kappa _c)={\displaystyle \frac{1}{2}}b_2,`$ (72)
$`b_3`$ $`=`$ $`\dot{w}_c+2\dot{t}_{1c}\mathrm{cosh}\kappa _c+2\dot{t}_{2c}\mathrm{cosh}2\kappa _c,`$ (73)
and where we used Eq. (57) to simplify the expression for $`A_2`$. We can also simplify the results for $`a_1`$ and $`b_3`$. Using Eq. (56) in Eq. (69), we obtain
$$a_1=4t_{2c}\mathrm{sinh}^2\kappa _c>0.$$
(74)
To simplify $`b_3`$, we recall that the discriminant in Eq. (48) vanishes at $`m=m_c`$. Expanding about $`m_c`$, we have
$$t_1^2(m)4t_2(m)f(m)=\frac{16}{J}\alpha ^2(mm_c)t_{2c}^2+O((mm_c)/J)^2,$$
(75)
where $`\alpha `$ is a positive constant. Differentiating with respect to $`m`$ and setting $`m=m_c`$, we obtain
$$t_{1c}\dot{t}_{1c}2t_{2c}\dot{f}_c2\dot{t}_{2c}f_c=\frac{8}{J}\alpha ^2t_{2c}^2.$$
(76)
Since $`f=w2t_2E`$, using Eqs. (57) and (56), we obtain $`f_c=4t_{2c}\mathrm{cosh}^2\kappa _c`$. Also, $`\dot{f}_c=\dot{w}_c2\dot{t}_{2c}`$, and the left hand side of Eq. (76) becomes
$$2t_{2c}(\dot{w}_c+2\dot{t}_{1c}\mathrm{cosh}\kappa _c+2\dot{t}_{2c}\mathrm{cosh}2\kappa _c),$$
(77)
which by Eq. (73) we recognize as $`2t_{2c}b_3`$. Therefore,
$$b_3=4\alpha ^2t_{2c}/J.$$
(78)
One can also similarly show that
$$a_3=t_{2c}\mathrm{sinh}\kappa _c\frac{d}{dm}\left(\frac{t_1}{t_2}\right)|_{m=m_c},$$
(79)
but this result is not particularly useful.
Let us now examine the order of magnitude of the various $`a`$ and $`b`$ coefficients just introduced. We first note that since $`t_{\alpha c}=O(J^0)`$ and $`\dot{t}_{\alpha c}=O(J^1)`$, Eq. (76) implies that $`\alpha =O(1)`$. It follows that $`a_1=O(J^0)`$, while $`a_2`$, $`a_3`$, $`b_2`$, and $`b_3`$ are all of order $`J^1`$.
To solve the differential equation (60) with the approximations (66)–(68) for the $`A`$’s, we eliminate the first derivative via the substitution
$$y(m)=z(m)\mathrm{exp}\frac{1}{2a_1}\left[a_2(mm_c)\frac{b_2}{2}\sigma _2(mm_c)^2\right].$$
(80)
Then $`z(m)`$ obeys
$$a_1\ddot{z}(m)+b_3^{}(mm_c^{})z_m=0,$$
(81)
where we have dropped a term of order $`(mm_c)^2/J^2`$ in the coefficient of $`z_m`$, and where
$`b_3^{}`$ $`=`$ $`b_3+\sigma _2{\displaystyle \frac{a_2b_2}{2a_1}},`$ (82)
$`m_c^{}`$ $`=`$ $`m_c+{\displaystyle \frac{a_2^2}{4a_1b_3}}.`$ (83)
This is Airy’s differential equation, and the general solution can be written as a linear combination of $`\mathrm{Ai}(\zeta ^{})`$ and $`\mathrm{Bi}(\zeta ^{})`$, where
$$\zeta ^{}=\left(\frac{b_3^{}}{a_1}\right)^{1/3}(mm_c^{}).$$
(84)
We can now assess the validity of our approximation for $`y(m)`$. First, let us ask for the order of $`\dot{y}`$ relative to $`y`$. From the known behavior of the Airy functions, this is $`J^{1/3}`$ in the failure zone $`\mathrm{\Delta }mJ^{1/3}`$, and of order $`(\mathrm{\Delta }m/J)^{1/2}`$ in the overlap zone $`J^\eta \mathrm{\Delta }mJ^{1/3}`$. We thus see that the higher order terms in Eqs. (66)–(68) are indeed smaller than those retained. In the same way the terms dropped in the differential equation for $`z(m)`$ can be seen to be small. Second, the $`d^3y/dm^3`$ term which was neglected in the differential equation (60) is of order $`(\mathrm{\Delta }m)^{3/2}/J^{1/2}`$ in the overlap zone. Since $`\eta <1`$, this term is smaller than the $`\ddot{y}`$ term, and higher order derivatives are smaller still.
The next step is to match the solution (58) and (80) using the known asymptotic forms of the Airy functions $`\mathrm{Ai}`$ and $`\mathrm{Bi}`$, onto the DPI forms for $`m<m_c`$ and $`m>m_c`$. The matching zones can be taken to be any regions in which $`J^\eta \mathrm{\Delta }mJ^{1/3}`$, where $`1>\eta >1/3`$. The leading behavior of the Airy functions is either an exponential or a sine or cosine of $`(\mathrm{\Delta }m)^{3/2}/J^{1/2}`$. We can ensure that the term $`b_2(mm_c)^2/4a_1`$ in the exponential in Eq. (80) is inconsequential on both sides if we choose $`\mathrm{\Delta }mJ^{1/2}`$. Accordingly we take the matching zone as
$$J^{1/2}|mm_c|J^{1/3},$$
(85)
In this zone we can approximate $`y(m)z(m)`$, and also neglect the small corrections in Eqs. (82) and (83). This amounts to saying that
$$C_me^{\sigma _2\kappa _c(mm_c)}[c_1\mathrm{Ai}(\zeta )+c_2\mathrm{Bi}(\zeta )],$$
(86)
where $`c_1`$ and $`c_2`$ are two arbitrary constants, and
$$\zeta =\left(\frac{b_3}{a_1}\right)^{1/3}(mm_c).$$
(87)
Let us first match Eq. (86) to the DPI solutions for $`m<m_c`$. In order to treat all four solutions at the same time, we rewrite Eq. (48) as
$$\mathrm{cos}q(m)=\frac{t_1(m)+\sigma _1[t_1^2(m)4t_2(m)f(m)]^{1/2}}{4t_2(m)},(m<m_c)$$
(88)
where $`f(m)=w(m)2t_2(m)E`$, and $`\sigma _1=\pm 1`$. The wavevector $`q(m)`$ thus depends on both $`\sigma _1`$ and $`\sigma _2`$, and hence, so does the velocity $`v(m)`$. We do not bother to write down these forms explicitly, except to note that $`i\sigma _1\sigma _2v(m)`$ is positive. The DPI solutions may therefore be written as
$$C_m=\frac{A_{\sigma _1\sigma _2}}{2\sqrt{i\sigma _1\sigma _2v(m)}}\mathrm{exp}i_{m_c}^mq(m^{})𝑑m^{},$$
(89)
where $`A_{\sigma _1\sigma _2}`$ is a real constant which is introduced as a notational aid in distinguishing the different cases. In the matching zone, it follows from Eqs. (88, (55), (75), and (7), that
$`\mathrm{cos}q(m)`$ $``$ $`\mathrm{cosh}\kappa _c+\alpha \sigma _1((m_cm)/J)^{1/2},`$ (90)
$`q(m)`$ $``$ $`i\sigma _2\kappa _c+i\sigma _1\sigma _2{\displaystyle \frac{\alpha }{\mathrm{sinh}\kappa _c}}((m_cm)/J)^{1/2},`$ (91)
$`v(m)`$ $``$ $`8i\alpha \sigma _1\sigma _2\mathrm{sinh}\kappa _c|t_{2c}|((m_cm)/J)^{1/2}.`$ (92)
(We have written $`|t_{2c}|`$ instead of $`t_{2c}`$ with a view to including the case $`t_1<0`$, $`t_2<0`$, which we shall consider later.) Therefore,
$$C_m\frac{A_{\sigma _1\sigma _2}}{4\sqrt{2\alpha t_{2c}\mathrm{sinh}\kappa _c}}\left(\frac{m_cm}{J}\right)^{1/4}\mathrm{exp}\left[\sigma _2\kappa _c(mm_c)+\frac{2}{3}\frac{\alpha \sigma _1\sigma _2}{\mathrm{sinh}\kappa _c}\frac{(m_cm)^{3/2}}{J^{1/2}}\right].$$
(93)
Since $`\zeta 1`$ for $`m<m_c`$ in the zone (85), we may use the asymptotic forms
$`\mathrm{Ai}(\zeta )`$ $``$ $`{\displaystyle \frac{1}{2\sqrt{\pi }}}\zeta ^{1/4}\mathrm{exp}\left({\displaystyle \frac{2}{3}}\zeta ^{3/2}\right),`$ (94)
$`\mathrm{Bi}(\zeta )`$ $``$ $`{\displaystyle \frac{1}{\sqrt{\pi }}}\zeta ^{1/4}\mathrm{exp}\left({\displaystyle \frac{2}{3}}\zeta ^{3/2}\right).`$ (95)
Using Eq. (87) and comparing with Eq. (93), we see that we must use the function $`\mathrm{Ai}`$ when $`\sigma _2=\sigma _1`$, and $`\mathrm{Bi}`$ when $`\sigma _2=\sigma _1`$. We can write
$$C_m=KA_{\sigma _1\sigma _2}e^{\sigma _2\kappa _c(mm_c)}\times \{\begin{array}{cc}\mathrm{Bi}(\zeta ),\hfill & \sigma _2=\sigma _1\text{;}\hfill \\ 2\mathrm{A}\mathrm{i}(\zeta ),\hfill & \sigma _2=\sigma _1\text{,}\hfill \end{array}$$
(96)
with
$$K=\frac{\sqrt{\pi }}{4\sqrt{2\alpha t_{2c}\mathrm{sinh}\kappa _c}}\left(\frac{a_1}{b_3}\right)^{1/12}J^{1/4}$$
(97)
To connect Eq. (96) with the DPI solutions for $`m>m_c`$, we make use of the reality principle, namely that the solution to the basic recursion relation (2) must be real for $`m>m_c`$ if it is real for $`m<m_c`$. Since the solutions to Eq. (48) now lead to complex $`q`$, it is clear that we must add two DPI solutions in order to get a real result. To this end we define
$$q_a(m)=i\sigma _2\kappa (m)+\chi (m),m>m_c$$
(98)
where $`\kappa (m)`$ and $`\chi (m)`$ are both real and positive, $`\kappa (m_c)=\kappa _c`$, and
$`\mathrm{cosh}\kappa \mathrm{cos}\chi `$ $`=`$ $`t_1/4|t_2|,`$ (99)
$`\mathrm{sinh}\kappa \mathrm{sin}\chi `$ $`=`$ $`(4t_2ft_1^2)^{1/2}/4|t_2|.`$ (100)
In terms of these quantities, we have
$`\mathrm{cos}q_a(m)`$ $`=`$ $`\mathrm{cosh}\kappa (m)\mathrm{cos}\chi (m)i\sigma _2\mathrm{sinh}\kappa (m)\mathrm{sin}\chi (m),`$ (101)
$`s_a(m)`$ $``$ $`i\sigma _2v_a(m)=8|t_2(m)|\mathrm{sinh}\kappa (m)\mathrm{sin}\chi (m)\mathrm{sin}q_a(m).`$ (102)
\[As before, we have written $`|t_2(m)|`$ instead of $`t_2(m)`$ with a view to subsequently including the case $`t_1<0`$, $`t_2<0`$.\] The corresponding quantities for the complex conjugate solution are given by $`q_b(m)=q_a^{}(m)`$, $`s_b(m)=s_a^{}(m)`$. The DPI solution into which Eq. (89) continues for $`m>m_c`$ may therefore be written as
$$C_m=\frac{1}{2}B_{\sigma _1\sigma _2}[\frac{1}{\sqrt{s_a(m)}}\mathrm{exp}(i_{m_c}^mq_a(m^{})dm^{}+i\mathrm{\Delta }_{\sigma _1\sigma _2})+\mathrm{c}.\mathrm{c}.]$$
(103)
where $`B_{\sigma _1\sigma _2}`$ is real, $`\mathrm{\Delta }_{\sigma _1\sigma _2}`$ is a phase to be found, and c.c. denotes “complex conjugate”.
It remains to compare the asymptotic forms of Eqs. (96) and (103) in the overlap zone (85), and thus find $`\mathrm{\Delta }_{\sigma _1\sigma _2}`$ and the relation between $`A_{\sigma _1\sigma _2}`$ and $`B_{\sigma _1\sigma _2}`$. Using the asymptotic forms for $`\mathrm{Ai}`$ and $`\mathrm{Bi}`$, we have
$$C_mKA_{\sigma _1\sigma _2}\frac{e^{\sigma _2\kappa _c(mm_c)}}{\sqrt{\pi }}(\zeta )^{1/4}\times \{\begin{array}{cc}\mathrm{cos}\left[\frac{2}{3}(\zeta )^{3/2}+\frac{\pi }{4}\right],\hfill & \sigma _2=\sigma _1\text{;}\hfill \\ 2\mathrm{sin}\left[\frac{2}{3}(\zeta )^{3/2}+\frac{\pi }{4}\right],\hfill & \sigma _2=\sigma _1\text{.}\hfill \end{array}$$
(104)
To write Eq. (103) in a similar form, we first note that in the matching zone,
$`q_a(m)`$ $``$ $`i\sigma _2\kappa _c+\sigma _2{\displaystyle \frac{\alpha }{\mathrm{sinh}\kappa _c}}((mm_c)/J)^{1/2},`$ (105)
$`s_a(m)`$ $``$ $`8t_{2c}\alpha \mathrm{sinh}\kappa _c\left({\displaystyle \frac{mm_c}{J}}\right)^{1/2}\mathrm{exp}\left[i\sigma _2{\displaystyle \frac{\pi }{2}}i\sigma _2\alpha {\displaystyle \frac{\mathrm{cosh}\kappa _c}{\mathrm{sinh}^2\kappa _c}}\left({\displaystyle \frac{mm_c}{J}}\right)^{1/2}\right].`$ (106)
Therefore,
$$i_{m_c}^mq_a(m^{})𝑑m^{}=\sigma _2\kappa _c\mathrm{\Delta }m+i\frac{2}{3}(\zeta )^{3/2},$$
(107)
where we have used Eqs. (74), (78), and (87) to simplify the term in $`(\zeta )^{3/2}`$. Compared to this term, the correction proportional to $`(\mathrm{\Delta }m)^{1/2}`$ in the exponent for $`s_a(m)`$ is of higher order in $`1/J`$ in the matching zone, and can be ignored. Doing this, and making use of Eqs. (87) and (97), we get
$$C_m\frac{K}{\sqrt{\pi }}B_{\sigma _1\sigma _2}(\zeta )^{1/4}e^{\sigma _2\kappa _c\mathrm{\Delta }m}[\mathrm{exp}(i\frac{2}{3}(\zeta )^{3/2}i\frac{\pi \sigma _2}{4}+i\mathrm{\Delta }_{\sigma _1\sigma _2})+\mathrm{c}.\mathrm{c}.].$$
(108)
Comparing with Eq. (104), we see that $`B_{\sigma _1\sigma _1}=A_{\sigma _1\sigma _1}/2`$, while $`B_{\sigma _1,\sigma _1}=A_{\sigma _1,\sigma _1}`$. Further, $`\mathrm{\Delta }_{\sigma _1\sigma _2}`$ vanishes if $`\sigma _1=1`$, and equals $`\pm \pi /2`$ if $`\pm \sigma _2=\sigma _1=1`$. These results can be summarized as
$`B_{\sigma _1\sigma _2}`$ $`=`$ $`{\displaystyle \frac{2\delta _{\sigma _1\sigma _2}}{2}}A_{\sigma _1\sigma _2},`$ (109)
$`\mathrm{\Delta }_{\sigma _1\sigma _2}`$ $`=`$ $`{\displaystyle \frac{\pi }{4}}(1+\sigma _1)\sigma _2.`$ (110)
Putting together Eqs. (89), (103), (109), and (110), the connection formula may be written in the final form
$`{\displaystyle \frac{A_{\sigma _1\sigma _2}}{2\sqrt{i\sigma _1\sigma _2v(m)}}}\mathrm{exp}i{\displaystyle _{m_c}^m}q(m^{})𝑑m^{}C_m`$ (111)
$`(1{\displaystyle \frac{1}{2}}\delta _{\sigma _1\sigma _2})A_{\sigma _1\sigma _2}[{\displaystyle \frac{1}{\sqrt{s_a(m)}}}\mathrm{exp}(i{\displaystyle _{m_c}^m}q_a(m^{})dm^{}+i{\displaystyle \frac{\pi }{4}}(1+\sigma _1)\sigma _2)+\mathrm{c}.\mathrm{c}.].`$ (112)
The wavevectors $`q`$ and $`q_a`$, and the velocity and speed, $`v(m)`$ and $`s_a`$, to be used for a given set of signs $`(\sigma _1,\sigma _2)`$ are given by Eqs. (91) and (105).
We use the double arrow notation $`C_m`$ advocated by Heading and Dingle to emphasize the bidirectionality of the connection formula. We refer readers to these authors for lucid discussions of this point, but since even as masterly and authoratative a text as Landau and Lifshitz states that connection formulas may only be used in one direction, it may be worth paraphrasing their remarks briefly. Thus, as stated by Heading, the notation means that there is a solution $`C_m`$ to the recursion relation (2) with the stated asymptotic behaviors for $`m<m_c`$ and $`m>m_c`$. And, as stressed by Dingle, the formula merely states that a certain exponentially growing on one side of the turning point, which is free from the growing component, goes over into a certain oscillatory solution (with an exponentially growing envelope) on the other side, and vice versa. Likewise for the growing component. It says nothing about whether or not we can use the formula in both directions in physical problems where we do not have complete information. If, e.g., a solution with $`\sigma _1=\sigma _2=1`$ for $`m<m_c`$ contains an admixture of the $`\sigma _1=\sigma _2=1`$ solution of small but indeterminable magnitude, then we can say nothing about the amplitude or the phase of the oscillatory factor for $`m>m_c`$. It is in these situations, where information is incomplete, that the reservations about the unidirectionality of the connection formulas are relevant.
The connection formula for the case where $`t_1<0`$ and $`t_2<0`$ can be derived in exact parallel by making minor modifications to the intermediate steps. The final form is sufficiently different to be worth giving separately:
$`{\displaystyle \frac{A_{\sigma _1\sigma _2}e^{i\pi m_c}}{2\sqrt{i\sigma _1\sigma _2v(m)}}}\mathrm{exp}i{\displaystyle _{m_c}^m}q(m^{})𝑑m^{}C_m`$ (113)
$`(1{\displaystyle \frac{1}{2}}\delta _{\sigma _1\sigma _2})A_{\sigma _1\sigma _2}[{\displaystyle \frac{e^{i\pi m_c}}{\sqrt{s_a(m)}}}\mathrm{exp}(i{\displaystyle _{m_c}^m}q_a(m^{})dm^{}i{\displaystyle \frac{\pi }{4}}(1\sigma _1)\sigma _2)+\mathrm{c}.\mathrm{c}.].`$ (114)
The quantities $`v(m)`$ and $`s_a(m)`$ are given by the same formal expressions as before, but now
$`q(m)`$ $``$ $`\pi +i\sigma _2\kappa _c+i\sigma _1\sigma _2{\displaystyle \frac{\alpha }{\mathrm{sinh}\kappa _c}}((m_cm)/J)^{1/2},`$ (115)
$`q_a(m)`$ $`=`$ $`\pi +i\sigma _2\kappa (m)+\chi (m).`$ (116)
In Eq. (116), $`\kappa (m)`$ and $`\chi (m)`$ are also given by the same formal expressions as before, i.e., Eqs. (99) and (100).
###### Acknowledgements.
This work is supported by the NSF via grant number DMR-9616749. I am indebted to Wolfgang Wernsdorfer and Jacques Villain for useful discussions and correspondence about Fe<sub>8</sub>.
## A Width of failure zone of DPI approximation
We have seen in Sec. II that the DPI approximation is valid everywhere except near turning points. Let us now estimate the size of the region where it fails. Suppose that $`q=0`$ or $`\pi `$ at the turning point, $`m_c`$. As shown in Eqs. (53) and (54), $`q(m)`$ and $`v(m)`$ both vary as $`[(mm_c)/J]^{1/2}`$ for $`m`$ near $`m_c`$. Thus $`\ddot{q}(mm_c)^{3/2}J^{1/2}`$, and the first term in $`\mathrm{\Phi }_2`$ is of order $`[(mm_c)J]^{1/2}`$. Next, note that $`r[J/(mm_c)]^{1/2}`$, so that the second and third terms are both of order $`J^{1/2}(mm_c)^{3/2}`$. The first term is subdominant to these two, and therefore, the condition that $`\mathrm{\Phi }_21`$ reduces to
$$|mm_c|J^{1/3}.$$
(A1)
In the case where the second factor in $`v(m)`$ vanishes, we have $`t_1+4t_2\mathrm{cos}q[(mm_c)/J]^{1/2}`$, and $`qq_c[(mm_c)/J]^{1/2}`$. Thus the integrand of the first term in $`\mathrm{\Phi }_2`$ varies as $`1/(mm_c)^2`$, and the term itself is of order $`1/(mm_c)`$. However, $`r[J/(mm_c)]^{1/2}`$ as before, so that the second and third terms are again of order $`J^{1/2}(mm_c)^{3/2}`$, and much greater than the first term. The condition for validity of the DPI form is again given by Eq. (A1).
The condition can be written in alternative form by noting that near the region of failure, $`r(m)1/v(m)`$, and that $`\dot{v}\dot{q}`$. Thus $`dr/dm\dot{q}/v^2`$, and the last term in $`\mathrm{\Phi }_2`$ can also be shown to be of order $`\dot{q}/v^2`$. Thus the condition for validity can be written as
$$\dot{q}(m)v^2(m).$$
(A2) |
warning/0003/astro-ph0003222.html | ar5iv | text | # New Evidence for the Complex Structure of the Red Giant Branch in 𝜔 CentauriBased on observations collected at ESO, La Silla, Chile
## 1. Introduction
The globular cluster $`\omega `$ Centauri (NGC 5139) is the most luminous and massive object among the Galactic Globular Clusters (GGC), and surely the most peculiar one in terms of structure, kinematics and stellar content. It is the most flattened GGC, displaying also a decrease of ellipticity in the most internal region (Geyer, Nelles & Hopp 1983 - GNH83), and it has a significant rotation (Merrit, Meylan & Mayor 1997).
The most interesting anomaly is its chemical inhomogeneity (first revealed by Dickens and Woolley 1967 and spectroscopically confirmed by Freeman & Rodgers 1975); since then a number of extensive spectroscopic surveys (Norris, Freeman & Mighell 1996 - NFM96, Suntzeff & Kraft 1996 - SK96) have shown that $`\omega `$ Cen is the only GGC for which a multi-component heavy element distribution has been identified. Although SK96 found a single peaked distribution with an extended tail towards high metallicities, NFM96 showed that the distribution is at least bimodal with a main metal poor component at $`[Fe/H]1.6`$, a second smaller peak at $`[Fe/H]1.2`$, and a long tail extending up to $`[Fe/H]0.5`$. Furthermore, the most metal rich stars ($`[Ca/H]>1.0`$) have been found to be more centrally concentrated than the bulk of the cluster population (Norris et al. 1997 - N97, and SK96). It has also been suggested that the kinematical properties (N97) and the spatial distribution of the two metallicity groups differ significantly (N97, Jurcsik 1998 - J98). This puzzling scenario has usually been explained either in terms of self-enrichment processes (Freeman 1993) and/or merging events (N97, J98).
As part of a long term project specifically devoted to the study of the global stellar population in a sample of GGCs, we obtained wide field $`B`$, $`I`$ photometry in $`\omega `$ Cen. The complete data set will be presented in a forthcoming paper, while in this letter we concentrate on the complex structure of the RGB.
## 2. The RGB of $`\omega `$ Cen
The data have been obtained on January 1999 and July 1999 at the 2.2m ESO-MPI telescope at La Silla (Chile), using the Wide Field Imager (WFI) which has a total field of view of $`34^{}\times 33^{}`$. The images were taken through the $`B`$ filter and $`I_{853}`$, a Medium Band Filter (cf. WFI@2.2 Manual) which avoids the most pronounced atmospheric emission lines. After applying the standard bias and flat field correction, we used DAOPHOT II and the psf-fitting algorithm ALLSTAR (Stetson 1994) for the stellar photometry. The photometric calibration was performed using 50 photoelectric standard stars in the selected areas SA98 and SA95 (Landolt 1992).
Figure 1 shows the $`(B,BI)`$ CMDs for each of the 8 WFI chips separately. The cluster center is roughly centered on chip $`\mathrm{\#}2`$. More than 220,000 stars have been plotted in Figure 1: to our knowledge, this is the largest stellar sample ever observed in $`\omega `$ Cen. The most striking feature of the CMDs presented in Figure 1 is the existence of a complex structure in the brighter part of the Red Giant Branch (RGB): at least two main components are visible.
Particularly notable is the presence of a narrow sequence, significantly redder and more bent than the bulk of the “main” RGB stars, which we call the anomalous RGB (hereafter RGB-a). Note that the RGB-a is visible only in the CMD from chip #2 (where the cluster center is located) and to a much lesser extent in the nearby chip #1 and possibly #3 (see the arrows in Figure 1(a),(b)). We shall come back to the spatial distribution of the RGB stars in the next Section. While we are writing this Letter, a $`(V,BV)`$ CMD showing the same RGB structure has been published by Lee et al. (1999): the reality of this feature is thus confirmed by two independent surveys based on different photometric systems.
The morphology of the RGB-a and its position in the CMD strongly suggests that it must be populated by stars much more metal rich than the $`\omega `$ Cen bulk population. We can directly check this hypothesis using the $`[Ca/H]`$ catalog by NFM96 (kindly provided by J. Norris). We have cross-identified the stars in our catalog and in the NFM96 one, and marked the common stars in Figure 2 (panel (a)) using different symbols for different metallicities (see figure caption).
Six stars belonging to the RGB-a are in common with the NFM96 sample (namely stars $`ROA300`$, $`ROA447`$, $`ROA500`$, $`ROA512`$, $`ROA517`$, $`ROA523`$, adopting the Woolley 1966 nomenclature): these stars have $`[Ca/H]=0.1\pm 0.1`$. They are the most metal rich stars in the NFM96 catalog, though they do not correspond to any of the peaks shown in the NFM96 $`[Ca/H]`$ distribution (e.g. their Figure 12). Figure 2 suggests that the RGB-a represents an additional, very metal rich component, located at the extreme tail of the abundance distribution in $`\omega `$ Cen. Moreover, the large color baseline of the present survey allows us to clearly separate the most metal rich component from the remaining $`\omega `$ Cen RGB stars and, thus, to estimate its contribution to the global population of the cluster. By comparing the star counts along the RGBs, in the same magnitude range ($`B<16`$), we find that the RGB-a represents $`5\%`$ of the whole stellar content. It is important to note that the six RGB-a stars for which we know the radial velocities from the Mayor et al. (1997) catalog (kindly provided by G. Meylan) have $`v_{rad}`$ = $`233\pm 9kms^1`$. This value is fully compatible with the mean value $`<v_{rad}>=233\pm 17kms^1`$ for the 471 stars in the Mayor et al. (1997) catalog, showing that the RGB-a stars are indeed members of the $`\omega `$ Cen system (see also Lloyd Evans, 1983 - LE83).
To further investigate the distribution of the RGB stars in the CMD, we computed the distance of each of them from the mean ridge line of the main metal poor component (MP-MRL) in several magnitude bins. The MP-MRL has been obtained following the procedure described in Ferraro et al. (1999a). A preliminary selection of the stars belonging to the bluest ridge of the RGB (excluding HB, AGB, and the reddest RGB components) has initially been performed by eye. Then we fitted the selected stars with polynomials, and the procedure was iterated (each time, rejecting all stars at $`>2\sigma `$ from the best fit line), until the result was considered acceptable. In doing this, we used the $`(I,BI)`$ CMD, since in this plane the upper part of the RGB is less bent than in $`(B,BI)`$, and a low order polynomial excellently reproduces the RGB shape.
Then we defined the observable $`\delta x`$ as the geometrical distance of each star from the adopted MP-MRL. The distribution of $`\delta x`$ for the RGB stars in the magnitude range $`11.9<I<12.5`$ is plotted in Figure 2 (panel (b)). Positive and negative values of $`\delta x`$ are assumed for stars redder and bluer than the MP-MRL, respectively. The distribution is clearly asymmetric; besides the main peak, secondary peaks are well visible. The estimated typical $`[Ca/H]`$ abundances for the different components \[as in panel (a)\] are marked in panel (b). Four different components can be identified in Figure 2. They can be grouped in three main samples as follows: (i) RGB-MP for the most metal poor component, at $`[Ca/H]1.4`$; (ii) RGB-MInt for the metal intermediate component: this sample includes the secondary peak ($`\delta x0.08`$) in Figure 2(b), which corresponds to $`[Ca/H]1.0`$, and the metal rich tail ($`\delta x0.20.3`$), corresponding to $`[Ca/H]0.5`$; (iii) RGB-a for the extreme metal rich component, at $`\delta x0.4`$ in Figure 2(b), i.e. $`[Ca/H]0.05`$, clearly separated from the other stellar populations. The first two components correspond to the peaks visible in Figure 12 in NFM96, while the RGB-a is a third, new component of the stellar population of $`\omega `$ Cen (see also Lee et al. 1999).
To better describe the characteristics of the “multiple” $`\omega `$ Cen RGB in terms of the “classical” abundance parameter $`[Fe/H]`$, we used a set of 60 giants in our sample which have both $`[Ca/H]`$ (from NFM96) and $`[Fe/H]`$ abundances (from SK96, kindly provided by N. Suntzeff). From these, we derived an empirical relation linking the calcium abundance to iron, in terms of $`[Fe/H]`$ on the Zinn & West (1984) scale. The procedure yields the following results for the four peaks labeled in Figure 2 (panel(b)): $`[Fe/H]_{ZW}1.7,1.3,0.8`$ and $`0.4`$, respectively.
It is interesting to note that the spectra of all of the six metal rich stars quoted above exhibit strong $`BaII`$ lines, and the brightest three (namely $`ROA300`$, $`ROA447`$ and $`ROA513`$) show also strong $`ZrO`$ bands (LE83), a very unusual occurrence for globular cluster giants. The fact that high metallicity giants have strong $`Ba`$ lines fully agrees with the increase of the $`[Ba/Fe]`$ ratio with metallicity found by Norris & Da Costa (1995), and suggests that intermediate mass ($`M<10M_{}`$) stars may have contributed significantly to the enrichment of the material from which RGB-a stars formed <sup>1</sup><sup>1</sup>1Both $`Ba`$ and $`Zr`$ are elements produced by neutron s-capture processes. s-process elements are brought up to the surface during the third dredge up occurring at late stages of the asymptotic giant branch (AGB) evolutionary phase of intermediate mass stars (see the discussion in Smith, Cunha & Lambert 1995). However LE83 noted that s-element rich giants in $`\omega `$ Cen show different characteristics with respect to normal AGB stars and first suggested that the anomalous s-elements enrichment could be primordial.. This, in turn, implies a significant delay ($`0.5Gyr`$) between the formation of the stellar population that enriched the medium and the one associated with RGB-a (Norris & Da Costa 1995, Lee et al. 1999).
## 3. Spatial distribution anomalies
The correlation between metal content and kinematics found by N97, and the spatial asymmetry between metal rich and metal poor bright red giants found by J98, are a possible evidence of a past merging event, in which a small metal rich object has been captured by a larger system, mainly metal poor. Note that none of the quoted studies have identified the RGB-a as a distinct population, since only a few stars belonging to this branch have been measured by NFM96 and listed by J98. The discovery of a population so clearly separated from the main branches makes the hypothesis that $`\omega `$ Cen hosts an external population captured in the past more and more appealing, even if it is worth noticing, according to note 3 by N97, that neither the large metal spread nor the s-process abundances can be easily explained in terms of a simple merging event between ordinary globular clusters.
These considerations prompted us to investigate the spatial distribution of the three RGB components, taking advantage of our photometric sample, which is much larger ($`N_{RGB}>3500`$) and more complete than the N97 and J98 ones.
As a first step, we evaluated the centroid of the stars of the entire sample; by adopting the procedure described in Montegriffo et al. (1995) and Ferraro et al. (1999b), we found it to be at pixel $`(700\pm 20,1900\pm 20)`$. The centroids of all three of the RGB components appear to coincide within the observational errors.
Then, we investigated the radial distributions; special care has been devoted to avoid contamination by the field population which can affect the radial trend, expecially in the lower density external region. To this purpose, we limited our analysis to the most internal region ($`r<10^{}`$), and to the brighter portion ($`B<15.5`$) of the RGB, which is expected to be little contaminated. The comparison of the cumulative radial distributions for the selected RGB samples shows that the RGB-MInt and RGB-a stars have a very similar distribution, and both seem to be more concentrated than the RGB-MP. A KS test gives an $`8\%`$ probability that the radial distribution of the coadded (RGB-MInt $`+`$ RGB-a) sample and the RGB-MP sample are drawn from the same parent distribution. Although this evidence is marginal, it is in good agreement with that found by N97. Still, this simple radial distribution analysis implicitly assumes that the three populations are distributed with the same (spherical) geometry. In view of the strong ellipticity of $`\omega `$ Cen, we decided to follow an alternative approach.
In order to avoid the sparsely populated outer regions, we divided the central region of the cluster ($`r<13^{}`$ from the center) into a grid of $`13\times 13`$ square boxes, each $`500\times 500`$ pixels wide. Inside each $`(i,j)th`$ box, we counted the number of stars belonging to the different samples ($`N_{RGBMP}^{ij}`$, $`N_{RGBMInt}^{ij}`$ and $`N_{RGBa}^{ij}`$, respectively), thus obtaining a density map for each component. The resulting isodensity contour maps, down to a fixed density limit (4 stars per box), are shown in Fig. 3. The distributions of the RGB-a sample (heavy solid lines in panel (a)) and of the RGB-MInt (heavy solid lines in panel (b)) are compared with the RGB-MP distribution (light dashed lines in both panels). The elongation of the RGB-MP isopleth along the X-axis is clearly visible, reflecting the well known elliptical shape of the whole system: our X-axis is approximately oriented along the E-W direction, i.e., the major axis of the cluster. It is worth noticing that the ellipticity for the RGB-MP component turns out to be $`ϵ0.2`$, even in the most internal region of the cluster ($`2^{}4^{}`$). On the other hand, both the RGB-a and RGB-MInt isopleth show a very different direction of maximum elongation, nearly perpendicular to the RGB-MP one (i.e., along the N-S direction). Furthermore, a very peculiar substructure is visible in the RGB-a contours (see panel(a)), resembling the tidal tails which should be expected to form around a disrupting stellar system (e.g., Meylan, Leon & Combes 1999).
We took advantage of the fact that the major axis of the RGB-MP ellipsoid is nearly oriented along our X-axis, while the RGB-MInt and RGB-a ones are nearly oriented along the Y-axis. We can thus study the distribution along the two main symmetry axes by simply comparing the distributions along our X and Y coordinates with respect to the the common centroid. The main results of this analysis are: (1) the RGB-MP sample is significantly more concentrated in the Y direction than in X (confidence level $`99.5\%`$, according to a KS test), as expected for an ellipsoid elongated in the X-axis direction; (2) the RGB-MInt and RGB-a do not share the elongation in the X direction of the RGB-MP stars, and look more elongated in Y than the RGB-MP ones (c.l. $`99.6\%`$); (3) Fig. 3 (panel (b)) suggests that the RGB-MInt stars distribution is strongly asymmetric along the Y-axis, being more elongated towards the North direction.
These results could help to shed some light on previous puzzling evidence. In particular: point (1) and (2) could explain the decrease of ellipticity in the inner $`5^{}`$ of the cluster reported by GNH83, and point (3) could help to understand the bizarre spatial asymmetry found by J98.
Furthermore, the similarity in the spatial distributions may indicate a physical association between the RGB-a and RGB-MInt, while the differences shown in Figure 3 suggest a different origin of these two components with respect to the main metal poor one (RGB-MP). If this result finds additional support, and the primordial origin of the s-elements found in the RGB-a stars is definitely established, then the RGB-a and RGB-MInt populations can be interpreted as two successive phases of the self-enrichment history of an independent stellar system (a giant molecular cloud or a gas rich protocluster), now sunk into the center of $`\omega `$ Cen.
In conclusion, we can say that the observational evidence discussed in this Letter seem to suggest that both self-enrichment processes and a merging event should be invoked to explain the complex structure of the RGB in $`\omega `$ Cen.
We thank N. Suntzeff, R. Kraft, J. Norris, G. Meylan and J. Jurcsik for providing their data in computer readable form, and Luca Pasquini and Alvio Renzini for many stimulating discussions. We are also indebted to an anonymous referee for the helpful suggestions on the Ba and ZrO anomalies and to Tad Pryor that referred the paper for what concerns dynamics for his detailed report. The financial support of the Ministero della Università e della Ricerca Scientifica e Tecnologica (MURST) to the project Stellar Dynamics and Stellar Evolution in Globular Clusters and to the project Treatment of large-format astronomical images is kindly acknowledged. EP thanks the ESO Imaging Survey Visitor Program and the EIS team for providing technical support. FRF gratefully acknowledges the hospitality of the Visitor Program during his stay at ESO, when most of this work has been carried out. |
warning/0003/astro-ph0003383.html | ar5iv | text | # The late afterglow and host galaxy of GRB 990712 1footnote 11footnote 1Based on observations collected at the European Southern Observatory, La Silla, Chile (ESO Programme 64.O–0019) and on observations with the NASA/ESA Hubble Space Telescope, obtained from the data archive at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc. under NASA contract NAS5-26555.
## 1 Introduction
The spatial association of GRB 980425 with the unusual Type Ib/c supernova SN1998bw at $`z=0.0085`$ provided the first tantalizing evidence that some gamma-ray bursts (GRBs) are related to the end-stages of the lives of massive stars (Galama et al., 1998). Recent evidence for similar supernova (SN) signatures in the late ($`15(1+z)`$ days) light curves of the genuine cosmological GRBs GRB 970228 (Dar, 1999; Reichart, 1999; Galama et al., 1999a) and GRB 980326 (Castro-Tirado & Gorosabel, 1999; Bloom et al., 1999a) indicates that at least some long-duration GRBs are related to SN explosions. A GRB–SN association suggests that the progenitors of GRBs are short-lived and that GRBs die where they were born—in the star-forming regions of their host galaxies (Paczyński, 1998).
Holland & Hjorth (1999) found evidence from Hubble Space Telescope (HST) Space Telescope Imaging Spectrograph (STIS) imaging for a spatial coincidence between GRB 990123 and a star-forming region in its host galaxy. The association of GRBs with star-forming regions is important for models of their progenitors and can be used to probe the physics of star formation and the global star-formation history of the Universe (Mao & Mo, 1998; Totani, 1999; Blain & Natarajan, 2000).
GRB 990712 was first localized by BeppoSAX and detected as having the strongest $`X`$-ray afterglow observed to date (Heise et al., 1999). Bakos et al. (1999) discovered a bright, decaying optical afterglow (OA) ($`R=19.4\pm 0.1`$) four hours after the burst. Galama et al. (1999b) measured a preliminary redshift of $`z=0.430\pm 0.005`$ from a set of absorption and emission lines, which makes it the nearest GRB with a secure redshift that has been observed to date (apart from SN1998bw). ESO New Technology Telescope images obtained 3.7 days after the burst led Hjorth et al. (1999a) to hypothesize the existence of a bright host galaxy with $`R=22`$ on the grounds of an apparent leveling off of the light curve relative to the suspected power-law decline ($`\alpha =1.05`$) (Kemp & Halpern, 1999) of the OA. Subsequent ESO Very Large Telescope imaging (Hjorth et al., 1999b) confirmed the leveling off of the light curve, and yielded evidence for the existence of an extended object contributing to the flux at the position of the GRB. Hjorth et al. (1999b) predicted that the existence of a SN would lead to a bump in the light curve around 1 August 1999, and that a SN model could be distinguished from a no-SN model in late HST and ground-based imaging, as the OA would be brighter and the host fainter in the SN scenario than in the no-SN scenario. These predictions are presented in Sahu et al. (2000), which reports the discovery and early light curve of the OA of GRB 990712.
In this Letter we present late HST imaging, as well as ground-based imaging and spectroscopy, aimed at testing these predictions. At the time of the HST observations the $`R`$-band magnitudes of the host galaxy and the OA are predicted to be $`22.25\pm 0.05`$ and $`<23.91\pm 0.05`$ in the SN scenario, and $`21.75\pm 0.05`$ and $`25.39\pm 0.1`$ in the no-SN scenario. We assume a standard Friedman cosmology with $`H_0=65\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$, $`\mathrm{\Omega }_0=0.2`$, and $`\mathrm{\Lambda }=0`$. At $`z=0.4337`$ this corresponds to a scale of 5.6 proper kpc per arcsecond, a luminosity distance of 2.37 Gpc, a distance modulus of 41.88, and a look-back time of 4.9 Gyr. Including a cosmological constant of $`\mathrm{\Lambda }=0.8`$ increases these values by $`10`$%.
## 2 Late Ground-Based Imaging
Direct images were obtained with the DFOSC on the Danish 1.54-m telescope at La Silla on 7 ($`R`$ band) and 9 ($`V`$ band) October 1999 UT, i.e., 87 and 89 days after the burst. Exposure times were $`3\times 15`$ minutes in each band and the seeing full-width at half-maximum (FWHM) was $`1\stackrel{}{\mathrm{.}}8`$. The photometry was carried out using SExtractor v2.0.13 (Bertin & Arnouts, 1996). The calibration was tied to the internal reference stars of Sahu et al. (2000) and gave $`R=21.92\pm 0.08`$ and $`V=22.40\pm 0.08`$ for the galaxy at the location of the GRB. Two more $`R`$ images (120 sec and 180 sec) were obtained with the ESO 3.6-m telescope on 12 November 1999 UT (123 days after the burst) in $`0\stackrel{}{\mathrm{.}}7`$ seeing (see §4). These images yielded $`R=21.91\pm 0.05`$ for the host galaxy. The ground-based images thus showed no signs of a transient source. The photometry is consistent with host magnitudes of $`R=21.91\pm 0.04`$ and $`V=22.40\pm 0.08`$.
## 3 HST Imaging
### 3.1 The STIS Data
HST observations of the OA were made on 29 August 1999 UT, 47.7 days after the burst, as part of the Cycle 8 program GO-8189. Six 620 second exposures were taken with the STIS in each of its 50CCD (clear, hereafter referred to as CL) and F28X50LP (long pass, hereafter referred to as LP) modes. The CCD gain was set to 1 e<sup>-</sup>/ADU, giving a read-out noise of 4 e<sup>-</sup>/pixel, and the data was processed through the standard STIS pipeline. We retrieved this data from the HST Data Archive and combined the images using the Dither (v1.2) software (Fruchter & Hook, 2000) as implemented in IRAF<sup>2</sup><sup>2</sup>2Image Reduction and Analysis Facility (IRAF), a software system distributed by the National Optical Astronomy Observatories (NOAO). (v2.11.1)/STSDAS (v2.0.2). We used “pixfrac” $`=0.5`$, and a final output scale of $`0\stackrel{}{\mathrm{.}}0254`$/pixel. Figure 1 shows the drizzled CL image of the probable host galaxy.
Conversion from counts to $`AB`$ magnitudes was achieved using the zero points given in the STIS Instrument Handbook. We assumed that the galaxy had a power-law spectrum of the form $`F_\nu (\nu )=k\nu ^\beta `$, where $`k`$ is constant, and converted the $`AB`$ magnitudes to standard Johnson $`V`$\- and Kron-Cousins $`R`$-band magnitudes using
$$M=m_{\mathrm{CL}}+K_M+48.6+2.5\beta \mathrm{log}_{10}\left(\nu _{\mathrm{CL}}/\nu _M\right),$$
(1)
where $`M`$ represents the $`V`$\- or $`R`$-band magnitude, as appropriate, $`m_{\mathrm{CL}}`$ and $`m_{\mathrm{LP}}`$ are the instrumental $`AB`$ magnitudes measured in the CL and LP filters, $`K_M`$ is the appropriate zero point, $`\nu _{\mathrm{CL}}`$ and $`\nu _{\mathrm{LP}}`$ are the central frequencies of the CL and LP filters, and $`\beta =0.4(m_{\mathrm{CL}}m_{\mathrm{LP}})/\mathrm{log}_{10}\left(\nu _{\mathrm{CL}}/\nu _{\mathrm{LP}}\right)`$.
### 3.2 Photometry of the Optical Afterglow
We estimated the total $`AB`$ magnitude of the OA on both the CL and the LP images by using the DaoPhot/AllStar II (Stetson, 1987) photometry package. A point-spread function (PSF) was constructed in the manner described in §4 of Holland & Hjorth (1999). The OA was identified by matching the coordinates of the OA given by Sahu et al. (2000) to the CL and LP STIS images, and looking for point sources inside the Sahu et al. (2000) error circle (see Fig. 1). The AllStar “sharp” statistic is related to the angular size of an object relative to a PSF. Isolated point sources will have “sharp” values of $`0`$ while extended sources will have “sharp” values $`0.1`$. The only point source (“sharp” $`=0.000`$ in the CL filter and $`0.053`$ in the LP filter) on the STIS images that lies within the error circle is located $`0\stackrel{}{\mathrm{.}}033`$ ($`=1.3`$ drizzled STIS pixels) from the Sahu et al. (2000) position for the OA. Therefore, we conclude that this point source is the OA for GRB 990712. The brighter, point-like feature 0$`\stackrel{}{\mathrm{.}}`$24 to the southeast of the OA has “sharp” $`=0.990`$ in the CL filter which suggests that it is the nucleus of the galaxy and not a point source (see § 5).
Aperture corrections were performed by using the DaoPhot II addstar routine to generate an artificial star with the same instrumental magnitude as the OA and measuring the total flux in apertures with radii of $`1\stackrel{}{\mathrm{.}}108`$ (CL) and $`0\stackrel{}{\mathrm{.}}963`$ (LP). Tables 14.3 and 14.5 of the STIS Instrument Handbook suggest that these radii correspond to $`100`$% of the encircled energy in the PSF. The OA magnitudes are $`V=25.69\pm 0.02`$ and $`R=25.23\pm 0.09`$, which yields a color of $`VR=0.46\pm 0.09`$. The Galactic reddening towards the OA ($`b^{\mathrm{II}}=40\stackrel{}{\mathrm{.}}19`$, $`l^{\mathrm{II}}=315\stackrel{}{\mathrm{.}}28`$) is $`E_{BV}=0.033`$ (Schlegel et al., 1998). We used $`A_V=0.11`$ and $`A_R=0.08`$ to obtain extinction-corrected magnitudes of $`V_0=25.58\pm 0.02`$ and $`R_0=25.15\pm 0.09`$.
A visual examination of each STIS image, after the PSF for the OA has been subtracted, shows that we have slightly oversubtracted the OA. We estimate that this oversubtraction has resulting in our magnitudes being overestimated by at most $`0.23`$ mag in the $`V`$ band and $`0.40`$ mag in the $`R`$ band. Therefore, the true magnitudes of the OA are $`V_0=25.58`$$`25.81`$ and $`R_0=25.15`$$`25.55`$. The corresponding range in color is $`(VR)_0=0.26`$$`0.43`$, and in spectral index is $`1.24\beta _{\mathrm{OA}}0.37`$.
## 4 Spectroscopy of the Host
We used the ESO Faint Object Spectrograph and Camera 2 (EFOSC2), mounted at the cassegrain focus of the ESO 3.6-m telescope, to obtain a 30-minute optical spectrum of the host galaxy of GRB 990712 on 12 November 1999 UT. The weather conditions were good, with $`0\stackrel{}{\mathrm{.}}7`$ seeing and photometric sky. We observed in binned mode, with an effective pixel size of $`0\stackrel{}{\mathrm{.}}32`$ and a resolution of $`4.28`$ Å/pixel (or $`R550`$) between 4000 Å and 8000 Å. The slit was $`1\stackrel{}{\mathrm{.}}2`$ wide and oriented along the parallactic angle. The data were reduced with IRAF and the CTIO long-slit package. A spectrum of the standard star LTT9491 (Landolt, 1992) was used to perform the relative flux calibration. As the standard was observed with a relatively narrow slit ($`1\stackrel{}{\mathrm{.}}2`$), we tied the absolute flux calibration to our measurement of the $`V`$-band flux for the galaxy.
The spectrum (Fig. 2) exhibits four prominent narrow (unresolved) emission lines which we identify as \[O II\]$`\lambda `$3727, H$`\beta \lambda `$4861, and \[O III\]$`\lambda \lambda `$4959,5007. Table 1 lists the observed- and rest-frame wavelengths, $`\lambda `$ and $`\lambda _0`$, the redshifts, $`z`$, the observed- and rest-frame line widths, $`W`$ and $`W_0`$, and the line fluxes, $`f`$, for each line. The mean redshift derived from these lines is $`z=0.4337\pm 0.0004`$ where the uncertainty is the $`3\sigma `$ uncertainty in the mean. We detect no trace of stellar absorption features such as the Ca H and K ($`\lambda \lambda `$3968,3933) lines or the 4000 Å break, and we do not detect any obvious intervening absorption systems along the line of sight.
We note that the \[O III\]$`\lambda `$5007 line lies near at the red edge of the bandpass of the $`R`$ filter. This may complicate the interpretation of late-time $`R`$-band light curves of the OA when the data is collected using different instruments, and when the total flux is dominated by the light of the host galaxy.
## 5 The Host Galaxy
The host galaxy is an extended source with elliptical isophotes, a bright, concentrated nucleus, and a bright, extended feature to the northwest of the nucleus. The nucleus has a FWHM of $`0\stackrel{}{\mathrm{.}}11`$ ($`=0.62`$ kpc) in the CL filter, which is only slightly wider than the PSF. There is some indication that the isophotes twist to the north at the northwest end of the galaxy and to the south at the southeast end.
The integrated $`V`$\- and $`R`$-band magnitudes of the galaxy were obtained by subtracting the light from the OA and performing aperture photometry with an aperture of radius $`2\stackrel{}{\mathrm{.}}5`$ in each of the CL and LP images. The measured spectral index is $`\beta _{\mathrm{gal}}=2.69`$, and Eq. 1 yields $`V=22.51\pm 0.04`$ and $`R=21.80\pm 0.06`$. Correcting for Galactic extinction, and assuming no internal extinction in the host galaxy, gives $`(VR)_0=0.68\pm 0.07`$. For $`z=0.4337`$ the observed $`R`$-band is approximately equivalent to the rest-frame $`B`$-band, so $`M_B19.5`$. Lilly et al. (1995) finds $`M_B^{}=21.23`$ in the rest frame for blue galaxies with redshifts of $`0.2z0.5`$. This corresponds to the host galaxy for GRB 990712 having $`L_B0.2L_B^{}`$ where $`L_B^{}`$ is the $`B`$-band luminosity of a typical blue galaxy at $`z=0.4337`$.
We estimated the star-formation rate (SFR) in the host galaxy using Eq. 2 of Madau et al. (1998) as described in Holland & Hjorth (1999). The total SFR is $`0.29`$$`0.45_{\mathrm{}}`$ yr<sup>-1</sup>, depending on the assumed initial mass function (IMF). The SFR is corrected for extinction in our Galaxy, but it assumes that there is no dust, or obscured star formation, in the host galaxy. This is probably a poor assumption so our derived SFR should be considered to be a lower limit on the true SFR in the host. The implied \[O II\] luminosity (Table 1), corrected for Galactic extinction, is $`L_{3727}=6.3\times 10^{40}`$ erg s<sup>-1</sup>. If we assume that the strength of the \[O II\] line is related to star formation then this corresponds to a SFR of $`0.88\pm 0.25_{\mathrm{}}`$ yr<sup>-1</sup> (Kennicutt, 1998). This is 2–3 times as large as the SFR derived from the continuum flux, possibly indicating internal extinction in the host or a contribution from non-thermal emission. The derived SFR (from the \[O II\] flux) is $`20`$% of the SFR found by Bloom et al. (1999b) for the host of GRB 990123 and comparable to that of the host of GRB 970508 (Bloom et al., 1998). The specific SFR per unit luminosity of the GRB 990712 host galaxy is $`0.4`$ times that of the host galaxies of GRB 990123 and GRB 970508.
The OA is located in the bright, extended source $`0\stackrel{}{\mathrm{.}}242`$ ($`=1.4`$ kpc) to the northwest ($`50\mathrm{°}`$ north of east) of the nucleus (see Fig. 1). This region has a FWHM of $`0\stackrel{}{\mathrm{.}}28`$ ($`=1.6`$ kpc), which is significantly more extended than the STIS PSF (FWHM $`=0\stackrel{}{\mathrm{.}}09`$). The surface brightness, after subtracting the OA and correcting for Galactic extinction, is $`\mu _{V,0}=15.83\pm 0.01`$ and $`\mu _{R,0}=15.04\pm 0.01`$ for an integrated color of $`(VR)_0=0.79\pm 0.01`$. The OA is located $`0\stackrel{}{\mathrm{.}}025`$ ($`140`$ pc) south of the center of this extended structure. The total $`V`$-band flux in the feature, after subtracting the flux from the \[O II\] emission line, is $`0.323\pm 0.003`$ $`\mu `$Jy. Assuming a power-law spectrum with $`\beta =2.93`$ the SFR is $`0.03`$$`0.05_{\mathrm{}}`$ yr<sup>-1</sup> depending on the IMF. Again, we wish to stress that this is a lower limit on the true SFR in the feature. The estimated SFR is approximately one third of the SFR that Holland & Hjorth (1999) found for the star-forming region that coincides with the position of GRB 990123 whereas the diameters and luminosities are comparable.
The spectroscopic data demonstrate that all emission lines, both forbidden and permitted, have narrow widths that are consistent with a Seyfert 2 galaxy (the instrumental resolution only gives an upper limit on the rest-frame velocity widths of a few hundred $`\mathrm{km}\mathrm{s}^1`$). The ratio \[O III\]/H$`\beta `$ is greater than three, which is indicative of an active galaxy (Shuder & Osterbrock, 1981). The object lies on the borderline between H II regions and narrow-line AGNs in the $`\mathrm{log}(`$\[O III\]$`/\mathrm{H}\beta )`$ vs. $`\mathrm{log}(`$\[O II\]/\[O III\]) diagram—uncorrected for extinction, see Baldwin et al. (1981). This suggests that the host galaxy of GRB 990712 may be a Seyfert 2 galaxy, but further spectroscopic observations will be needed to confirm this. For example, a measurement of $`\mathrm{log}(`$\[O I\]$`/\mathrm{H}\alpha )`$ from near-IR spectroscopy would discriminate between a star-forming galaxy or a Seyfert 2 (Veilleux & Osterbrock, 1987).
## 6 The Late Afterglow and the Possible GRB 990712–Supernova Association
Table 2 summarizes the predicted and observed $`V`$\- and $`R`$-band magnitudes for the OA and its host galaxy in both the pure power-law (PL), and power law + supernova (PL+SN) scenarios. The magnitudes for the host galaxy are the weighted means of the magnitudes that we obtained from the Danish 1.54-m and ESO 3.6-m telescopes, and the HST. Allowing for small systematic uncertainties, the observed magnitudes and colors for the host and OA are consistent with the predictions of the pure power-law model, which indicates that the OA followed a slow power-law decay with a constant index of $`\alpha 1.0`$, with no significant late-time break towards steeper decay. Our results are inconsistent with the presence of a SN like SN1998bw. The discrepancy between the observations and the predictions of the PL+SN scenario suggests that a SN in the late-time light curve of GRB 990712 would have had to be $`>1.5`$ mag fainter than SN1998bw to be consistent with the data presented in this Letter. Therefore, we conclude that either GRB 990712 did not produce a SN, or that the flux received from the SN was much smaller than expected from scaling SN1998bw to $`z=0.4337`$.
If there was no SN then this supports the view that long-duration GRBs have more than one type of progenitor (Livio & Waxman, 1999). GRB 970508 is the only other GRB for which there is some evidence that a SN was not present. However, the paucity of observations around the time of the predicted peak luminosity of the SN makes it difficult to unambiguously rule out a SN contribution in GRB 970508’s light curve. Therefore, GRB 990712 provides the first solid evidence that not all long-duration GRBs are associated with standard-candle SNe. There are some similarities between GRB 990712 and GRB 970508 that suggest that they may be members of the same class of bursts. Both had shallow late-time decay slopes ($`\alpha 1.0`$), both appear to lack a 1998bw-type SN, both bursts were strong $`X`$-ray sources, and both bursts have host galaxies that are significantly fainter than $`L^{}`$.
If there was a SN associated with GRB 990712, then its peak intensity was $`>1.5`$ mag fainter than that of SN1998bw. Type Ib/c SNe are poor standard candles since their predicted peak intensities can vary by 1–2 mag, with a mean peak $`B`$-band magnitude that is 1.16 mag fainter than that of SN1998bw (van den Bergh & Tammann, 1991). Moreover, the predicted peak magnitude for a SN associated with GRB 990712 depends on the cosmological model, the exact spectral shape of SN1998bw at very short wavelengths, the width of the lightcurve peak, the possible extinction in the host galaxy, and the possible evolution of SNe with redshift. A large sample of GRB afterglows with measured redshifts and detectable SN signatures is needed to establish the intrinsic luminosity distribution of SNe accompanying GRBs.
FC is supported by Chilean grant FONDECYT/3990024. Additional funding from the European Southern Observatory is gratefully acknowledged. This work was supported by the Danish Natural Science Research Council (SNF). |
warning/0003/nucl-th0003009.html | ar5iv | text | # Compton scattering in a unitary approach with causality constraints
## I Introduction
In Ref. a relativistic covariant model was presented for pion-nucleon scattering in which constraints due to unitarity were taken into account through the use of the K-matrix formalism with a non-perturbative dressing of the $`\pi NN`$-vertex. An approach based on the use of dispersion relations was employed in this dressing. It was shown that as a result of the dressing the effective form factors were softened. Here we extend this work to include processes involving photons. The fact that our procedure is based on the use of dispersion relations, thus incorporating analyticity, gives a considerable advantage. It is known that in photon induced processes important constraints are imposed by the condition that the amplitude for the process has to be analytic, especially at energies near the pion-production threshold. The usage of dispersion relations allows for an implementation of these analyticity constraints in a unitary K-matrix approach. In quantum field theory analyticity is based on the condition of causality and is a fundamental property of the S-matrix.
As with any process involving photons one should take care to obey gauge invariance of the amplitude since otherwise low-energy theorems may be violated. In the present approach we used the minimal substitution procedure. We discuss this method in some detail and present general formulas.
In the procedure the effects of dressing are expressed in terms of form factors and self-energies. The present work is focused on the $`\gamma NN`$-vertex. Electromagnetic vertices of the nucleon with one or both nucleons off-shell have been studied in the past. The method of dispersion relations was applied in Refs.. Dynamical models based on a perturbative dressing of the vertex with meson loops, within effective Lagrangian approaches, were developed in Refs.. The role of off-shell nucleon-photon form factors has been investigated, for example, in models for proton-proton bremsstrahlung and virtual Compton scattering . Analyticity considerations have been used in Refs. to construct amplitudes for Compton scattering from those of pion photoproduction through the application of dispersion relations. The present approach incorporates dispersion relations (or analyticity considerations) in a more microscopic approach to pion and photon induced reactions on the proton.
The present model consists of two stages implemented in an iteration procedure to reach self-consistency. In the first stage effective two-, three- and four-point Green’s functions are built which incorporate non-perturbative dressing due to non-pole parts of loop diagrams. At each iteration step, the imaginary parts of the loop integrals are found by applying Cutkosky rules . In doing so, only the intermediate states with one nucleon and one pion are taken into account. The real parts are constructed using dispersion relations . The dispersion integrals converge due to the sufficiently fast falloff of the $`\pi NN`$ form factors in the loop diagrams. The resulting $`\gamma NN`$-vertex is normalized in such a way that, at the point where both nucleons are on-shell, it reproduces the physical anomalous magnetic moment of the nucleon.
In the second stage a K-matrix formalism is employed to calculate the T-matrix, where the kernel, the K-matrix, is built 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 of loop integrals are taken into account. The T-matrix obtained from thus constructed K-matrix will contain the principal-value parts of the same loop integrals which were included in the dispersion calculation for the form factors and self-energies, implementing analyticity in the K-matrix framework. 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.
The $`\gamma NN`$-vertex must satisfy the Ward-Takahashi identity. This is achieved in our model by including a loop diagram with a four-point $`\gamma \pi NN`$-vertex (the contact term). The latter is constructed based on the dressed $`\pi NN`$-vertex using the minimal substitution prescription (various constructions of contact terms can be found in Refs.). Such a procedure leads to a unique result only for the longitudinal (with respect to the photon momentum) part of the four-point vertex. To investigate the role of the transverse terms, we calculated the electromagnetic form factors utilizing two different $`\gamma \pi NN`$ vertices. To provide current conservation in the description of Compton scattering also a contact $`\gamma \gamma NN`$ term is built using the minimal substitution prescription.
The model is geared to the calculation of pion-photoproduction and Compton scattering on the nucleon. To study effects of the dressing we compare the $`f_{EE}^1`$ partial wave amplitude for Compton scattering obtained using the dressed $`\gamma NN`$ and $`\pi NN`$ vertices and nucleon propagator with that of a calculation using bare vertices and the free propagator. We also calculated the electric polarizability of the proton.
In Section II we outline the construction of the K-matrix in a coupled-channel unitary description of Compton scattering, pion photoproduction and pion-nucleon scattering. Details of the dressing procedure are given in Sections II.A and II.B where special attention is payed to vertices with photons. Numerical results on $`\gamma NN`$-vertices, expressed in terms of half-off-shell form factors, are given in Section III. In Section IV the formalism is applied to Compton scattering where we focus on the effects of dressing on observables. Conclusions are given in Section V.
## II Structure of the K-matrix
Our model is based on the K-matrix formalism and to explain our procedure we work in a simple model space with only the nucleon, pion and photon degrees of freedom. Only the one-pion threshold discontinuities are taken into account explicitly.
To describe simultaneously pion-nucleon scattering, pion photoproduction and Compton scattering the scattering matrix has two indices corresponding to the channel in the initial and final state, $`𝒯_{c^{}c}`$, where the indices can be $`\pi `$ or $`\gamma `$ for the channels $`\pi N`$ or $`\gamma N`$, respectively. The Bethe-Salpeter equation for the scattering matrix can be written as
$$𝒯_{c^{}c}=V_{c^{}c}+\underset{c^{\prime \prime }}{}V_{c^{}c^{\prime \prime }}𝒢_{c^{\prime \prime }}𝒯_{c^{\prime \prime }c},$$
(1)
where $`V_{c^{}c}`$ is the sum of irreducible diagrams describing the process $`cc^{}`$ and $`𝒢_{c^{\prime \prime }}`$ is the two-body propagator pertinent to the channel $`c^{\prime \prime }`$. $`𝒢_{c^{\prime \prime }}`$ contains the pole contribution $`i\delta _{c^{\prime \prime }}`$ which is imaginary, according to Cutkosky rules, and the regular (principal-value) part $`𝒢_{c^{\prime \prime }}^R`$ which is real,
$$𝒢_{c^{\prime \prime }}=𝒢_{c^{\prime \prime }}^R+i\delta _{c^{\prime \prime }}.$$
(2)
The K-matrix can be introduced as the solution of the equation
$$K_{c^{}c}=V_{c^{}c}+\underset{c^{\prime \prime }}{}V_{c^{}c^{\prime \prime }}𝒢_{c^{\prime \prime }}^RK_{c^{\prime \prime }c}.$$
(3)
According to this formula, the loop diagrams contributing to the K-matrix contain only the principal-value part of the two-particle propagator. We assume throughout that $`V`$ is a sum of tree-level diagrams. The remaining pole contribution enters explicitly in the equation for the T-matrix expressed in terms of the K-matrix,
$$𝒯_{c^{}c}=K_{c^{}c}+\underset{c^{\prime \prime }}{}K_{c^{}c^{\prime \prime }}i\delta _{c^{\prime \prime }}𝒯_{c^{\prime \prime }c},$$
(4)
which can be obtained from Eqs.(1-3). A formal solution of Eq. (4) can be written as (suppressing the channel indices)
$$𝒯=K(1Ki\delta )^1,$$
(5)
from which it follows that the S-matrix, $`S=1+2i𝒯`$, will be unitary provided $`K`$ is hermitian.
Eq. (3) suggests an interpretation of the K-matrix in terms of a dressing of a potential $`V_{c^{}c}`$ with principal-value parts of loop integrals. To illustrate this for the case of Compton scattering we choose $`V_{\gamma \gamma }`$ and $`V_{\gamma \pi }`$ as the sum of s- and u- and t-channel tree diagrams plus a possible four-point vertex, where the free nucleon propagator and bare nucleon-photon vertices are used. Up to second order in $`V_{c^{}c}`$ and leading order in the electromagnetic coupling constant, $`K_{\gamma \gamma }`$ can be written as
$$K_{\gamma \gamma }^{(2)}=V_{\gamma \gamma }+V_{\gamma \pi }𝒢_\pi ^RV_{\pi \gamma }.$$
(6)
The set of diagrams corresponding to the right-hand side of this equation is depicted in Fig. (2). The notation $`ss`$, $`su`$ etc. for the loop diagrams refer to their structure in terms of the s-, u- , t-channel and contact tree diagrams contributing to $`V_{\gamma \pi }`$. The index $`\mathrm{𝑅𝑒}`$ at the loops indicates that only the principal-value integrals are taken into account, in accordance with Eq. (6). Consequently, the self-energy functions and form factors parametrizing these loops are real functions. One can see that the one-particle reducible diagrams in Fig. (2) (diagrams ss, su, st, sc, us, uc, ts, cs and cu) are part of a dressing of the nucleon propagator and half-off-shell nucleon-photon vertices. The other diagrams in Fig. (2), which are one-particle irreducible, are necessary to ensure the gauge invariance of $`K_{\gamma \gamma }^{(2)}`$.
The above description of $`K_{\gamma \gamma }^{(2)}`$ serves as an introduction to the dressing procedure described below. In the full model dressing up to infinite order is taken into account, expressed in terms of an integral equation. Gauge invariance is maintained through the introduction of an appropriate contact term. It should be pointed out that Eq. (3) dictates that only principal-value parts of the loop integrals (or, equivalently, only the real parts of the form factors and self-energy functions) are taken into account in the iteration procedure for the vertices and the nucleon propagator.
To summarize, in our procedure we construct the $`K`$ matrix which enters in Eq. (5) as the sum of tree-level diagrams (those for $`K_{\gamma \gamma }`$ and $`K_{\gamma \pi }`$ are depicted in Fig. (1)) where dressed nucleon propagators, dressed nucleon-pion, dressed nucleon-photon vertices, and contact terms (for gauge invariance) are used. The use of dressed quantities is implied by Eq. (3), where, due to taking the principal value integrals, the effect of dressing can be expressed in terms of purely real form factors or self-energy functions. These real functions are obtained by applying dispersion relations to the one-particle reducible pole contributions from Eq. (4), thereby implementing analyticity (causality) constraints in the calculation of the T-matrix. The procedure followed is discussed in detail in the following sections.
### A The dressing procedure
The most general $`\gamma NN`$-vertex for a real photon with momentum $`q=p^{}p`$, in which the outgoing nucleon is on the mass shell, $`p^2=m^2`$, can be written<sup>*</sup><sup>*</sup>*The notation of Ref. is used throughout this paper. as
$$e\mathrm{\Gamma }_\mu (p)=e\underset{l=\pm }{}\left\{\gamma _\mu F_1^l(p^2)+i\frac{\sigma _{\mu \nu }q^\nu }{2m}F_2^l(p^2)\right\}\mathrm{\Lambda }_l(p),$$
(7)
where $`e`$ and $`m`$ are the elementary electric charge and the mass of the nucleon and
$$\mathrm{\Lambda }_\pm (p)\frac{\pm p/+m}{2m}.$$
(8)
The isospin structure of the form factors is taken as $`F=F^s+\tau _3F^v`$. The dressing of this vertex is expressed in terms of a system of integral equations, shown diagrammatically in Fig. (3),
$$\mathrm{\Gamma }_{\mu ,R}(p)=\mathrm{\Gamma }_\mu ^0(p)+\text{D.I.}\left\{\mathrm{\Gamma }_{\mu ,I}[1]+\mathrm{\Gamma }_{\mu ,I}[2]+\mathrm{\Gamma }_{\mu ,I}[3]\right\},$$
(9)
where “D.I.” implies taking a dispersion integral, $`\mathrm{\Gamma }_{\mu ,R}`$ contains only the real parts of the form factors and each of the terms will be discussed in detail in the following. This equation expresses the dressing of a bare vertex $`\mathrm{\Gamma }_\mu ^0(p)`$ with an infinite series of pion loops. The bare vertex is taken as
$$\mathrm{\Gamma }_\mu ^0(p)=(\gamma _\mu \widehat{e}_N+i\widehat{\kappa }_B\frac{\sigma _{\mu \nu }q^\nu }{2m}),$$
(10)
where $`\widehat{e}_N=(1+\tau _3)/2`$ and $`\widehat{\kappa }_B=\kappa _B^s+\tau _3\kappa _B^v`$ is the bare anomalous magnetic moment of nucleon, adjusted to provide the normalization Eq. (37) of the dressed vertex. The solution of Eq. (9) is obtained by requiring self-consistency in an iteration procedure. We consider irreducible vertices, which implies that the external propagators are not included in the dressing of the vertices.
Every iteration step proceeds as follows. The imaginary or pole contributions of the loop integrals for both the propagators and the vertices are obtained by applying cutting rules . Since the outgoing nucleon is on-shell, the only kinematically allowed cuts are those shown in Fig. (3). In calculating these pole contributions, we retain only real parts of the form factors and nucleon self-energies from the previous iteration step as required by Eq. (3). We note that all integrals on the right-hand of Eq. (9), except the one over $`\mathrm{\Gamma }_{\mu ,I}[2]`$, are inhomogeneities of the equation because they do not depend on the $`\gamma NN`$-vertex. Therefore, they need to be calculated only once. The dressed $`\pi NN`$-vertices and the nucleon propagator are taken from Ref. where they have been constructed using a compatible procedure.
The real parts of the form factors are calculated at every iteration step by applying dispersion relations to the imaginary parts just calculated.
This procedure is repeated to obtain a converged solution. The convergence criterion is imposed for a normalized root-mean-square difference $`d_n`$ for the form factors between two subsequent iteration steps $`n`$ and $`n+1`$. The convergence criterion is that $`d_n<10^8`$ for a large number of iterations.
One of the advantages of the use of cutting rules is that throughout the solution procedure we need vertices with only one virtual nucleon (half-off-shell vertices), as can be seen from Fig. (3). In other words, the knowledge of full-off-shell form factors will not be required for the calculation of the pole contributions to the loop integrals. Also for the construction of the K-matrix only half-off-shell vertices are required.
Since in the dressing of the $`\pi NN`$-vertex a bare form factor is required for regularization, the described procedure obeys analyticity only approximately (the bare $`\gamma NN`$-vertex does not contain form factors see Eq. (10)). The influence of the singularities of the bare form factor can be diminished in the kinematical region of interest by a rather large width. This is consistent with the fact that the bare form factor represents physics left out from the model and thus should vary at an energy scale larger than the heaviest meson included explicitly.
### B The loop integrals
The pole contribution $`\mathrm{\Gamma }_{\mu ,I}[1]`$, of the first loop integral on the right hand side of Eq. (9) comes from cutting the nucleon propagator $`S(pk)`$ and the pion propagator $`D(k^2)`$, i.e. from putting the corresponding particles on their mass-shell (see Fig. (3)). According to Cutkosky rules , we replace $`S(pk)`$ with $`2i\pi (p/k/+m)\delta ((pk)^2m^2)\mathrm{\Theta }(p_0k_0)`$ and $`D(k^2)`$ with $`2i\pi \delta (k^2m_\pi ^2)\mathrm{\Theta }(k_0)`$, where $`m_\pi `$ is the pion mass.
The half-off-shell pion-nucleon vertex for an incoming nucleon with momentum $`p`$ and an on-shell outgoing nucleon ($`p^{}`$) entering in the expressions is written as
$$\mathrm{\Gamma }_{5,\alpha }(p)=\tau _\alpha \mathrm{\Gamma }_5(p)=\tau _\alpha \gamma _5\left[G_1(p^2)+\frac{p/m}{m}G_2(p^2)\right],$$
(11)
where $`k=pp^{}`$ is the momentum of the pion. The functions $`G_{1,2}(p^2)`$ are the (half-off-shell) form factors in the nucleon-pion vertex. In the approach adopted in we find that the dependence of the form factor on the pion momentum is small which is therefore ignored. The pion-nucleon coupling constant is taken from Ref. , $`g=13.02`$.
Denoting $`g_iReG_i`$, the pole contribution reads
$`\mathrm{\Gamma }_{\mu ,I}[1]`$ $`=`$ $`{\displaystyle \frac{2\tau _3}{8\pi ^2}}{\displaystyle }d^4k\gamma _5g_1(m^2)(p/k/+m)\gamma _5[g_1(p^2)+{\displaystyle \frac{p/m}{m}}g_2(p^2)]{\displaystyle \frac{2k_\mu +q_\mu }{(k+q)^2m_\pi ^2}}`$ (13)
$`\times \delta ((pk)^2m^2)\mathrm{\Theta }(p_0k_0)\delta (k^2m_\pi ^2)\mathrm{\Theta }(k_0),`$
where $`m_\pi `$ is the pion mass. The pion-photon vertex $`\mathrm{\Gamma }_{\mu ,\alpha \beta }(k^{},k)`$ is chosen such that it satisfies the Ward-Takahashi identity with the free pion propagator,
$$\mathrm{\Gamma }_{\mu ,\alpha \beta }(k^{},k)=(\widehat{e}_\pi )_{\alpha \beta }(k_\mu +k_\mu ^{}),$$
(14)
where the pion charge operator $`(\widehat{e}_\pi )_{\alpha \beta }=iϵ_{\alpha \beta 3}`$.
Using the notation introduced in Eq. (A1), one can write
$$\mathrm{\Gamma }_{\mu ,I}[1]=\underset{i=1}{\overset{4}{}}c^i[1](e_i)_\mu $$
(15)
where it has implicitly been assumed that the final nucleon is on the mass-shell and where
$`c^i[1]`$ $`=`$ $`{\displaystyle \frac{r(p^2)}{16\pi p^2}}\mathrm{\Theta }(p^2(m+m_\pi )^2){\displaystyle _1^1}𝑑xV^i(x){\displaystyle \frac{g_1(m^2)}{(k+q)^2m_\pi ^2}},`$ (16)
$`V^i(x)`$ $`=`$ $`\tau _3{\displaystyle \underset{j=1}{\overset{6}{}}}(E^1)_j^i(\theta ^j)^\mu \text{,}\mathrm{\Lambda }_+(p^{})(p/+k/+m)[g_1(p^2)`$ (18)
$`+{\displaystyle \frac{p/m}{m}}g_2(p^2)](2k_\mu +q_\mu )`$
where the brackets $`,`$ are defined in Eq. (A2), the $`(\theta ^j)^\mu `$ is the basis in the dual space, $`q=p^{}p`$ and $`r(p^2)=\sqrt{\lambda (p^2,m^2,m_\pi ^2)}`$, with the Källén function $`\lambda (x,y,z)(xyz)^24yz`$. The integral in Eq. (16) is a Lorentz-scalar and therefore can be evaluated in any frame of reference. We choose the rest frame of the incoming nucleon, i.e. we put $`p_\mu =W\delta _{\mu 0}`$, where $`W=\sqrt{p^2}`$ is the invariant mass of the off-shell nucleon. Furthermore we introduced $`x`$, the cosine of the polar angle between the three-vectors $`(\stackrel{}{q})`$ and $`\stackrel{}{k}`$. The integral in Eq. (16) is done numerically.
The term $`\mathrm{\Gamma }_{\mu ,I}[2]`$ depends on the unknown half-off-shell $`\gamma NN`$-vertex and therefore has to be considered in the context of the iteration procedure. As explained above, when calculating $`\mathrm{\Gamma }_{\mu ,I}^{n+1}[2]`$, the pole contribution to the n+1<sup>st</sup> iteration for $`\mathrm{\Gamma }_\mu [2]`$, we retain only the real parts of $`F_i^{\pm ,n}(p^2)`$ from the previous iteration as well as of the nucleon-pion form factors and the functions $`\alpha (p^2)`$ and $`\xi (p^2)`$ parametrizing the renormalized dressed nucleon propagator
$$S(p)=\left[\alpha (p^2)(p/\xi (p^2))\right]^1.$$
(19)
The functions $`\alpha (p^2)`$, $`\xi (p^2)`$, as well as $`G_{1,2}(p^2)`$, were calculated in Ref.. Using the same approach as for $`\mathrm{\Gamma }_{\mu ,I}[1]`$ we write
$$\mathrm{\Gamma }_{\mu ,I}^{n+1}[2]=\underset{i=1}{\overset{4}{}}c^{i,n+1}[2](e_i)_\mu $$
(20)
where
$`c^{i,n+1}[2]`$ $`=`$ $`{\displaystyle \frac{r(p^2)}{32\pi p^2}}\mathrm{\Theta }(p^2(m+m_\pi )^2){\displaystyle _1^1}𝑑x{\displaystyle \frac{U^i(x)}{\alpha ((p^{}k)^2)[(p^{}k)^2\xi ^2((p^{}k)^2)]}},`$ (21)
$`U^i(x)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{6}{}}}(E^1)_j^i(\theta ^j)^\mu \text{,}\mathrm{\Lambda }_+(p^{})\{\tau _\alpha \gamma _5[g_1((p^{}k)^2)+{\displaystyle \frac{p/^{}k/m}{m}}g_2((p^{}k)^2)]`$ (25)
$`\times (p/^{}k/+\xi ((p^{}k)^2))\left\}\right\{\mathrm{\Lambda }_+(p^{}k)[\gamma _\mu f_1^+((p^{}k)^2)`$
$`+i{\displaystyle \frac{\sigma _{\mu \nu }q^\nu }{2m}}f_2^+((p^{}k)^2)]+\mathrm{\Lambda }_{}(p^{}k)[\gamma _\mu f_1^{}((p^{}k)^2)+i{\displaystyle \frac{\sigma _{\mu \nu }q^\nu }{2m}}f_2^{}((p^{}k)^2)]\}`$
$`\times (p/k/+m)\tau _\alpha \gamma _5[g_1(p^2)+{\displaystyle \frac{p/m}{m}}g_2(p^2)]\},`$
where $`f_i^\pm (p^2)Re(F^\pm )_i^n(p^2)`$. At the required kinematics the functions $`\alpha (p^2)`$ and $`\xi (p^2)`$ are real. Note that the $`f`$’s contain isospin operators.
The term $`\mathrm{\Gamma }_{\mu ,I}[3]`$ contains a “contact” $`\gamma \pi NN`$-vertex. We build such a vertex applying the procedure of minimal substitution (see Appendices B and C for details) to the dressed half-off-shell $`\pi NN`$-vertex,
$`\left(\mathrm{\Gamma }_{\gamma \pi NN}\right)_\alpha ^\mu (p^{},p,q)=`$ (26)
$`\tau _\alpha \widehat{e}\left\{{\displaystyle \frac{2p^\mu +q^\mu }{(p+q)^2p^2}}\left[\mathrm{\Gamma }^5(p+q)\mathrm{\Gamma }^5(p)\right]+\gamma ^5{\displaystyle \frac{g_2((p+q)^2)}{m}}\left[\gamma ^\mu q/{\displaystyle \frac{2p^\mu +q^\mu }{(p+q)^2p^2}}\right]\right\}`$ (27)
$`\widehat{e}\tau _\alpha \{{\displaystyle \frac{2p^\mu q^\mu }{(p^{}q)^2p^2}}[\overline{\mathrm{\Gamma }^5}(p^{}q)+\overline{\mathrm{\Gamma }^5}(p^{})]+[\gamma ^\mu +{\displaystyle \frac{2p^\mu q^\mu }{(p^{}q)^2p^2}}q/]{\displaystyle \frac{g_2((p^{}q)^2)}{m}}\gamma ^5\},`$ (28)
where Eqs. (C3,C5), $`p^{}=p+q`$ and Eq. (11) with $`\overline{\mathrm{\Gamma }^5}(p)=\gamma _0\left(\mathrm{\Gamma }^5(p)\right)^{}\gamma _0`$ have been used. Using the same notation as before we obtain
$$\mathrm{\Gamma }_{\mu ,I}[3]=\underset{i=1}{\overset{4}{}}c^i[3](e_i)_\mu ,$$
(29)
with
$`c^i[3]`$ $`=`$ $`{\displaystyle \frac{r(p^2)}{64\pi p^2}}\mathrm{\Theta }(p^2(m+m_\pi )^2)\left[(3\tau _3){\displaystyle _1^1}𝑑xW_1^i(x)+3(1+\tau _3){\displaystyle _1^1}𝑑xW_2^i(x)\right],`$ (30)
$`W_1^i(x)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{6}{}}}(E^1)_j^i(\theta ^j)^\mu \text{,}\mathrm{\Lambda }_+(p^{})\{(2p_\mu 2k_\mu +q_\mu )[{\displaystyle \frac{g_{12}((p^{}k)^2)g_{12}(m^2)}{(p^{}k)^2m^2)}}`$ (33)
$`{\displaystyle \frac{p/k/}{m}}{\displaystyle \frac{g_2((p^{}k)^2)g_2(m^2)}{(p^{}k)^2m^2}}]{\displaystyle \frac{\gamma _\mu }{m}}g_2((p^{}k)^2)\}(mp/+k/)`$
$`\times \{g_{12}(p^2)+{\displaystyle \frac{p/}{m}}g_2(p^2)\},`$
$`W_2^i(x)`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{6}{}}}(E^1)_j^i(\theta ^j)^\mu \text{,}\mathrm{\Lambda }_+(p^{})\{(2p_\mu ^{}q_\mu )[{\displaystyle \frac{g_{12}(p^2)g_{12}(m^2)}{p^2m^2}}`$ (35)
$`+{\displaystyle \frac{g_2(p^2)g_2(m^2)}{p^2m^2}}{\displaystyle \frac{p/^{}}{m}}]+{\displaystyle \frac{\gamma _\mu }{m}}g_2(p^2)\}(mp/+k/)\{g_{12}(p^2)+{\displaystyle \frac{p/}{m}}g_2(p^2)\},`$
where $`g_{12}g_1g_2`$, and $`r(p^2)`$ is defined as in Eq. (18). An alternative expression for $`\mathrm{\Gamma }_{\mu ,I}[3]`$ is obtained if, instead of Eq. (C5), one uses the contact term Eq. (C8). The choice of the contact term has an influence on the nucleon-photon form factors, as will be described below.
Since the half-off-shell form factors are analytic in the complex $`p^2`$-plane with the cut from the pion threshold $`(m+m_\pi )^2`$ to infinity, dispersion relations can be used to construct the real parts from the imaginary parts. In our model the imaginary parts of the form factors $`F_2^\pm `$ vanish at infinity. At every iteration step we thus write
$$\text{Re}F_2(p^2)=\widehat{\kappa }_B+\frac{𝒫}{\pi }_{(m+m_\pi )^2}^{\mathrm{}}𝑑p^2\frac{\text{Im}F_2(p^2)}{p^2p^2},$$
(36)
where we have dropped the superscripts <sup>±</sup> of the form factors to keep the expression transparent. The constant $`\widehat{\kappa }_B`$ originates from the first term on the right-hand side of Eq. (9). Note that, according to Eq. (10), $`\kappa _B^{s,v}`$ are chosen the same for the form factors $`F_2^+(p^2)`$ and $`F_2^{}(p^2)`$. They are fixed by the requirement that the vertex reproduces the physical anomalous isoscalar and isovector magnetic moment when both nucleons are on-shell,
$$F_2^{+,s}(m^2)=0.06\text{ and }F_2^{+,v}(m^2)=1.85.$$
(37)
In terms of the parametrization of the propagator, Eq. (19), the Ward-Takahashi identity Eq. (D1) gives
$$(F_1^{})^{s,v}(p^2)=\frac{\alpha (p^2)}{2}$$
(38)
and
$$(F_1^+)^{s,v}(p^2)=\frac{\alpha (p^2)m}{p^2m^2}\left[\frac{p^2+m^2}{2m}\xi (p^2)\right].$$
(39)
where $`lim_{p^2m^2}(F_1^+)^{s,v}(p^2)`$ is finite because $`lim_{p^2m^2}\xi (p^2)=m`$ due to the correct location of the pole of the renormalized propagator. In Ref. the loop contribution to the self-energy vanishes in the limit $`p^2\mathrm{}`$ and therefore $`lim_{p^2\mathrm{}}(F_1^\pm )^{s,v}(p^2)Z_2/2=0`$. $`Z_2`$ is the nucleon-field renormalization constant. The form factors $`(F_1^\pm )^{s,v}(p^2)`$ thus obey the dispersion relations (omitting the superscripts)
$$\text{Re}F_1(p^2)=\frac{Z_2}{2}+\frac{𝒫}{\pi }_{(m+m_\pi )^2}^{\mathrm{}}𝑑p^2\frac{\text{Im}F_1(p^2)}{p^2p^2},$$
(40)
applied at every iteration step.
## III The form factors
The results presented in this section are obtained using the $`\pi NN`$-vertex Eq. (11) and the nucleon propagator calculated in Ref.. The solution for the nucleon-pion form factors depends on the choice of the bare vertex. However, the half-width of this form factor should not exceed a rather well defined maximum for the procedure to converge. For the present calculation of the form factors in the $`\gamma NN`$-vertex, we chose the solution in which the bare pseudovector form factor is given by Eq. (23) of Ref., a di-pole with half-width $`\mathrm{\Lambda }^2=1.28\text{GeV}^2`$. We also did the calculation using the other choice of the bare $`\pi NN`$-vertex, given by Eq. (24) of Ref., with the half-width $`\mathrm{\Lambda }^2=1.33\text{GeV}^2`$ (not shown). We found that the results for the nucleon-photon form factors do not depend significantly on the choice of the $`\pi NN`$-vertex.
Since the $`\gamma NN`$-vertex obeys the Ward-Takahashi identity, the form factors $`F_1^\pm (p^2)`$ are uniquely determined, see Eqs. (38,39), by the functions parametrizing the nucleon propagator calculated in Ref.. We checked numerically that these are satisfied by the converged vertex. One of the consequences of the Ward-Takahashi identity is that $`(F_1^\pm )^s=(F_1^\pm )^v`$ and therefore $`F_1^\pm =0`$ for the neutron-photon vertex. The form factors $`F_1^\pm (p^2)`$ in the proton-photon vertex are depicted in Fig. (4). They do not depend on the choice of the $`\gamma \pi NN`$-vertex.
The dominant contribution to the form factors $`F_2`$ is due to $`\mathrm{\Gamma }_\mu [1]`$. Since this term is an inhomogeneity of the equation, the bulk of the magnitude of the form factors is already generated in the first iteration. This, however, does not mean that the other integrals on the right-hand side of Eq. (9) are of minor importance. In particular, they are crucial for satisfying the Ward-Takahashi identity.
In Fig. (5) the imaginary and real parts of the form factors $`F_2^+(p^2)`$ (the solid line) and $`F_2^{}(p^2)`$ (the dotted line) are shown for the proton. The slope of $`ImF_2^{}(p^2)`$ at the pion threshold, $`p^2=(m+m_\pi )^2`$, is much steeper as compared to that of $`ImF_2^+(p^2)`$. As a consequence of this, we obtain a pronounced cusp-like behavior of $`ReF_2^{}(p^2)`$ at the threshold. The reason is that in pion photoproduction the $`E_{0^+}`$ multipole has a finite value at threshold while other multipoles tend to zero. Since this multipole corresponds to spin and parity $`1/2^{}`$ in the coupled nucleon-photon channel it contributes to the imaginary part of $`F_2^{}`$ which now obtains a term proportional to the 3-momentum of the cut intermediate pion. The real part of $`F_2^{}`$ calculated from a dispersion integral thus exhibits a pronounced cusp structure, contrary to the case of $`F_2^+`$ ($`F_2^+`$ is associated with the positive energy component of the off-shell nucleon carrying $`J^\pi =1/2^+`$). The magnitude of the cusp in $`F_2^{}`$ depends thus on the magnitude of the $`E_{0^+}`$ multipole multiplied by a weighted difference of the pseudoscalar and pseudovector coupling strengths in the pion-nucleon vertex.
The form factors in Fig. (5) are calculated using the $`\gamma \pi NN`$ contact term of Eq. (28) when evaluating $`\mathrm{\Gamma }_{\mu ,I}[3]`$ in Eq. (9). An alternative form of the $`\gamma \pi NN`$-vertex is obtained by using Eq. (C8) instead of Eq. (C5). The difference between these two contact terms, Eq. (C10), is transverse to the photon momentum. To illustrate the influence of the different choices of the contact terms on the $`\gamma NN`$-vertex, in Fig. (5), right panel, we show the form factors $`F_2^\pm (p^2)`$ calculated using the alternative contact term. From a comparison of left and right panels in Fig. (5), it follows that the different choices of the contact term affect mainly $`F_2^{}(p^2)`$. The form factor $`F_2^+(p^2)`$ is normalized at $`p^2=m^2`$ to the physical anomalous magnetic moment of the nucleon and is only slightly sensitive to the choice of the contact term. The difference between the two choices for the contact terms shows most strongly in the $`E_{0^+}`$ multipole in pion-photoproduction which explains why mainly $`F_2^{}(p^2)`$ is affected. In a full calculation one would fix the ambiguity in the contact term from a calculation of pion-photoproduction. This goes, however, beyond the scope of the present work but will be discussed in a forthcoming publication.
The results for the form factors $`F_2^\pm (p^2)`$ in the neutron-photon vertex are shown in Fig. (6) for the two choices of the $`\gamma \pi NN`$-vertex. The conclusions drawn above for the proton apply qualitatively to this case as well.
The renormalization conditions, Eq. (37), are fulfilled by adjusting the bare renormalization constants $`\kappa _B^{s,v}`$ defined in Eq. (10). We obtain $`\kappa _B^s=0.03`$ and $`\kappa _B^v=1.51`$ if the contact term Eq. (28) is used and $`\kappa _B^s=0`$ and $`\kappa _B^v=1.6`$ for the alternative contact term.
The off-shell form factors by themselves cannot unambiguously be extracted from experiment. In particular, they can be changed by a redefinition of the nucleon field. At the same time, the field redefinition will in general also change the four- and higher-point vertices. It is known that the S-matrix is independent of the representation of the fields . Therefore, in a consistent calculation of the observables, two- three- and higher-point Green’s functions should be treated using the same model assumptions and representation of the fields. The link between off-shell effects and contact interactions was emphasized in , where pion-nucleon scattering, Compton scattering by a pion and bremsstrahlung processes were considered. The form factors in the present approach are constructed consistently with the nucleon self-energy and the K-matrix, using the same representation to treat all these quantities. A certain care should however be excersized in applying these form factors in other calculations.
## IV Application in Compton scattering
As an example of the application of the formalism, Compton scattering off the nucleon will be calculated. Since only a very restricted model space is included in the present calculation, we do not make a direct comparison with experimental data. For a definitive comparison with experiment, other important degrees of freedom, such as the $`\mathrm{\Delta }`$-resonance, would have to be included. This extension of the model is in progress.
The amplitude for the Compton scattering process is obtained through solving Eq. (5) in a partial wave basis . The K-matrix matrix elements are constructed as a sum of tree-level Feynman diagrams where, however, dressed vertices and propagators are used. The tree-level diagrams for Compton scattering can be written as the sum of three contributions depicted in Fig. (1),
$$K_{\mu \nu }(q,k)=K_{\mu \nu }^s(q,k)+K_{\mu \nu }^u(q,k)+K_{\mu \nu }^c(q,k).$$
(41)
The incoming and outgoing nucleon momenta are $`p`$ and $`p^{}`$, respectively. The momenta of the incoming and outgoing photons are $`k^\nu `$ and $`q^\mu `$ so that energy and momentum conservation reads $`p^{}=p+k+q`$. The pole contributions to Compton scattering are given by the s- and u-channel diagrams. $`K_{\mu \nu }^c(q,k)`$ denotes the matrix element of the contact term given by Eq. (C24). This term is added to obtain a gauge-invariant matrix elements and is constructed using the minimal substitution procedure. Obeying gauge invariance is important for satisfying low-energy theorems for the matrix elements.
The contact term Eq. (C24) is, however, not unique and a purely transverse contribution may be added,
$$K_{\mu \nu }^c^{}(q,k)=\overline{u}(p^{})\mathrm{\hspace{0.17em}4}i\widehat{e}^2\left\{F_c((p+k)^2)(e_4)_\mu (q,p^{}q)(\overline{e_4})_\nu (k,p+k)+\left[\begin{array}{c}\mu \nu \\ qk\end{array}\right]\right\}u(p),$$
(42)
where the operator $`(e_4)_\mu (q,p^{}q)`$ is given by Eq. (A1) (with photon momentum $`q`$ and off-shell nucleon momentum $`p^{}q`$). The form factor is given by
$$F_c(p^2)=\frac{𝒫}{\pi }_{(m+m_\pi )^2}^{\mathrm{}}\frac{dp^2}{p^2p^2}\frac{\left[\text{Im}\stackrel{~}{F}_2^{}(p^2)\right]^2}{\text{Tr}\left[\mathrm{\Lambda }_{}\stackrel{~}{\mathrm{\Sigma }}_I(p^2)\right]},$$
(43)
where $`\stackrel{~}{F}_2^{}`$ and $`\stackrel{~}{\mathrm{\Sigma }}_I`$ are calculated from expressions similar to those for the negative energy form factor $`F_2`$ and the imaginary part of the nucleon self-energy, except that in the pion vertex, Eq. (11), we have put $`G_2=0`$. The reason for adding this term is to take into account a contribution arising from cutting the pion line in the “handbag” loop diagram. The contribution from this diagram, generated in the K-matrix procedure, gives rise to a sharply increasing imaginary contribution to the Compton amplitude and thus to a pronounced cusp structure in the corresponding real part.
We checked numerically that the calculated amplitudes obey current conservation which was also proven analytically. In addition the calculations obey constraints imposed by low-energy theorems.
Since we work in a restricted model space we postpone a complete comparison with experiment to a future publication where the model will be extended to include the $`\mathrm{\Delta }`$-isobar degree of freedom and t-channel meson exchanges. Here we limit ourselves to the $`f_{EE}^1`$ amplitude which shows a highly non-trivial behavior at the pion production threshold. Our results for the $`f_{EE}^1`$ amplitude (the solid line in Fig. (7)) shows a distinct cusp structure at the pion-production threshold. This cusp is generated by the analyticity condition we imposed. For comparison, the dashed line shows the partial wave amplitude calculated using the K-matrix built with the bare vertices and the free nucleon propagator, $`K_{c^{}c}=V_{c^{}c}`$, thereby neglecting the principal-value parts of the loop integrals contributing to the T-matrix (see Eqs. (3,4)) which does not show the cusp. The effect of unitarization on the real part of this amplitude are small, indistinguishable in the figure. We also show the results extracted from pion-photoproduction data through the application of analyticity consideration. A similar cusp structure is also seen in the analysis of Ref.. Extending our model with other degrees of freedom will add a smooth function to $`f_{EE}^1`$, changing the value at the higher energies but should not affect the cusp structure.
Assuming that Compton scattering is dominated by dipole process, the electric polarizability $`\overline{\alpha }`$ can be extracted from $`f_{EE}^1`$,
$$f_{EE}^1=f_B+\frac{\overline{\alpha }}{3}E_\gamma ^2,$$
(44)
where $`f_B`$ is the Born amplitude. Using this relation we extract $`\overline{\alpha }=11.610^4fm^3`$, which is close to the value obtained from chiral-perturbation theory.
## V Conclusions
We have presented a model for Compton scattering on the nucleon. In this model special attention is payed to observing analyticity in addition to unitarity, crossing symmetry and gauge invariance. The model is formulated in terms of half-off-shell form factors in the vertices and a nucleon self-energy which carry the non-perturbative dressing due to the non-pole contributions of pion-loop diagrams. The pole contributions are taken into account through the use of a K-matrix formalism.
The key element of the model is an integral equation which describes dressing of the $`\gamma NN`$-vertex with an infinite number of pion loops. In the solution procedure we take advantage of unitarity and analyticity considerations by using dispersion relations to obtain the real parts of the form factors from their imaginary parts. The latter, in turn, are obtained by applying cutting rules , with only the one-pion-nucleon discontinuities of the loop integrals taken into account. The dependence of the form factors on the four-momentum squared of the off-shell nucleon deviates from a monopole- (or dipole-) like shape adopted often in phenomenological applications. In particular, a characteristic feature of our results is a cusp-like structure of the form factors in the vicinity of the one-pion threshold, showing most clearly in the magnetic form factors corresponding to negative-energy states of the off-shell nucleon.
One of the important requirements for the electromagnetic vertex is obeying the Ward-Takahashi identity, which relates the vertex with the nucleon propagator. We have included a four-point $`\gamma \pi NN`$ term in our model to obey this condition. In a theory with nucleon-pion form factors such a term is always necessary. We construct a $`\gamma \pi NN`$-vertex using the prescription of minimal substitution. Terms in the contact vertex which are transverse to the photon momentum are not uniquely determined. As an example of this ambiguity, we have constructed two contact terms with different transverse components. We used these two contact terms in the calculation of the half-off-shell nucleon-photon vertex and found that the negative-energy magnetic form factors are influenced noticeably by the choice of the contact term, while the effect on the positive-energy form factors is rather small.
It should be emphasized that off-shell vertices (as any general Green’s functions, for that matter) depend not only on the model used to calculate them, but also on the representation of fields in the Lagrangian. In contrast, the measurable physical observables are obtained from the scattering matrix and are therefore oblivious to the representation of the Lagrangian (see, e.g., ). Even though information on the half-off-shell vertices cannot be unambiguously extracted from experiment, they are important for the calculation of observables.
We have argued that the vertices and propagator generated in the present dressing procedure are consistent with a coupled-channel K-matrix approach to Compton scattering, pion photoproduction and pion scattering. We have shown effects of the dressing on the cross section for real Compton scattering. An extension of the model to include additional degrees of freedom is in progress.
###### Acknowledgements.
This work is part of the research program of the “Stichting voor Fundamenteel Onderzoek der Materie” (FOM) with financial support from the “Nederlandse Organisatie voor Wetenschappelijk Onderzoek” (NWO). We would like to thank Alex Korchin, Rob Timmermans and John Tjon for discussions.
## A Projection method
The calculation of the imaginary parts of the form factors (the pole contributions in Eq. (9)) is formulated in terms of the following projection procedure. The half-off-shell vertex Eq. (7) can be regarded as a vector in a four-dimensional linear space. For the sake of generality we present here the procedure for a virtual photon where the vertex is a vector in a six-dimensional space $`V_6`$. with the basis
$$\begin{array}{cc}(e_1)_\mu =\mathrm{\Lambda }_+(p^{})\gamma _\mu \mathrm{\Lambda }_+(p),\hfill & (e_2)_\mu =\mathrm{\Lambda }_+(p^{})\gamma _\mu \mathrm{\Lambda }_{}(p),\hfill \\ (e_3)_\mu =\mathrm{\Lambda }_+(p^{})i\frac{\sigma _{\mu \nu }q^\nu }{2m}\mathrm{\Lambda }_+(p),\hfill & (e_4)_\mu =\mathrm{\Lambda }_+(p^{})i\frac{\sigma _{\mu \nu }q^\nu }{2m}\mathrm{\Lambda }_{}(p),\hfill \\ (e_5)_\mu =\mathrm{\Lambda }_+(p^{})\frac{q_\mu }{m}\mathrm{\Lambda }_+(p),\hfill & (e_6)_\mu =\mathrm{\Lambda }_+(p^{})\frac{q_\mu }{m}\mathrm{\Lambda }_{}(p),\hfill \end{array}$$
(A1)
defined over a ring of complex-valued functions (form factors). For the case of a real photon, as discussed in the present paper, basis vectors $`e_5`$ and $`e_6`$ can be truncated from the space. Thus, to find contributions to the imaginary parts of the form factors $`F_i^\pm `$ from the integral $`\mathrm{\Gamma }_{\mu ,I}[k],k=1\mathrm{}4`$, amounts to finding the coefficients of the expansion in the basis Eq. (A1).
The dual space $`V_6^{}`$ can be defined as spanned over the basis $`(\theta ^i)^\mu =g^{\mu \lambda }\overline{(e_i})_\lambda `$, where the over-lining denotes the Dirac conjugate of an operator, $`\overline{A}\gamma _0A^{}\gamma _0`$. We define the scalar product of $`\omega ^\mu V_6^{}`$ and $`v_\mu V_6`$ as
$$\omega ^\mu ,v_\mu =\text{Tr}(\omega ^\mu v_\mu ),$$
(A2)
with a tacit summation over $`\mu `$ and the trace taken in spinor space. For the evaluation of the traces we used the algebraic programming system REDUCE . Now if
$$v_\mu =\underset{i=1}{\overset{6}{}}c^i(e_i)_\mu ,$$
(A3)
then the coefficients are obtained from the formula
$$c^k=\underset{l=1}{\overset{6}{}}(E^1)_l^k(\theta ^l)^\mu ,v_\mu ,$$
(A4)
where the matrix $`E_j^i=(\theta ^i)^\mu ,(e_j)_\mu `$. The coefficients $`c^k`$ are the form factors (or, more precisely, contributions to the imaginary parts of the form factors). Thus, we identify
$$\begin{array}{c}c^1=ImF_1^+,c^2=ImF_1^{},c^3=ImF_2^+,\hfill \\ c^4=ImF_2^{},c^5=ImF_3^+,c^6=ImF_3^{}.\hfill \end{array}$$
(A5)
## B Minimal substitution
The minimal substitution in momentum space amounts to the following replacement of the nucleon momentum, $`P_\mu \stackrel{~}{P}_\mu =P_\mu \widehat{e}A_\mu `$, where $`P_\mu `$ has to be considered as an operator acting on the right and $`\widehat{e}=e\widehat{e}_Ne(1+\tau _3)/2`$. If in a given term $`P_\mu `$ is the rightmost operator and thus acts on the field of the incoming nucleon, it gives $`p_\mu `$ which has c-number components. Our procedure closely follows that of Ref.. Throughout this appendix we assume that the electromagnetic field $`A_\mu `$ carries the four-momentum $`q_\mu `$ directed inwards the vertex, $`[P_\nu ,A_\mu ]=q_\nu A_\mu `$. We thus obtain
$$P^2A_\mu =A_\mu (P+q)^2=A_\mu (p+q)^2=(p+q)^2A_\mu ,$$
(B1)
where for ease of writing the nucleon spinor fields have been omitted. More generally, for any smooth function $`f(p^2)`$ one obtains
$$f(P^2)A_\mu =A_\mu f((p+q)^2).$$
(B2)
Under the minimal substitution, the nucleon momentum squared changes as
$$P^2\stackrel{~}{P}^2=p^22\widehat{e}Ap\widehat{e}Aq+O(A^2)=p^2\widehat{e}A^\mu (2p_\mu +q_\mu )+O(A^2).$$
(B3)
Collecting the coefficients of the terms linear in $`A^\mu `$ results in the photon vertex. This procedure in indicated by the symbol $``$, i.e.
$$p^2\widehat{e}(2p_\mu +q_\mu ),$$
(B4)
which reads that upon minimal substitution a $`p^2`$ term in an n-point Green’s function generates an (n-point+photon) Green’s function corresponding to the vertex $`\mathrm{\Gamma }_\mu =\widehat{e}(2p_\mu +q_\mu )`$.
To generalize this for an arbitrary function $`f(p^2)`$, we first consider the following combination:
$`P^2\stackrel{~}{P}^2`$ $`=`$ $`p^42\widehat{e}q^2Ap2\widehat{e}App^24\widehat{e}qpAp\widehat{e}q^2Aq\widehat{e}Aqp^2`$ (B6)
$`2\widehat{e}Aqqp+O(A^2)=p^4\widehat{e}A^\mu (2p_\mu +q_\mu )(p+q)^2+O(A^2),`$
where Eq. (B3) has been used. The next step is to find the result of the minimal substitution in the monomials $`p^{2n},n=1,2,3\mathrm{}`$. Using Eqs. (B1,B3,B6), we have
$`P^{2n}\stackrel{~}{P}^{2n}`$ $`=`$ $`p^{2n}[2\widehat{e}Ap+\widehat{e}Aq]\stackrel{n1}{\stackrel{}{P^2\mathrm{}P^2}}P^2[2\widehat{e}Ap+\widehat{e}Aq]\stackrel{n2}{\stackrel{}{P^2\mathrm{}P^2}}`$ (B8)
$`\mathrm{}\stackrel{n1}{\stackrel{}{P^2\mathrm{}P^2}}[2\widehat{e}Ap+\widehat{e}Aq]+O(A^2)`$
$`=`$ $`p^{2n}\widehat{e}A^\mu (2p_\mu +q_\mu ){\displaystyle \underset{m=0}{\overset{n1}{}}}(p+q)^{2m}p^{2(n1m)}+O(A^2).`$ (B9)
The corresponding vertex is thus given by
$$p^{2n}\widehat{e}(2p_\mu +q_\mu )\frac{(p+q)^{2n}p^{2n}}{(p+q)^2p^2},$$
(B10)
where the identity
$$\underset{l=0}{\overset{k1}{}}x^{2l}y^{2(kl1)}=\frac{y^{2k}x^{2k}}{y^2x^2},$$
(B11)
has been used. Since a generic function $`f(p^2)`$ can be formally expanded in powers of $`p^2`$,
$$f(p^2)=\underset{k}{}a_kp^{2k},$$
(B12)
we obtain
$$f(p^2)\widehat{e}(2p_\mu +q_\mu )\frac{f((p+q)^2)f(p^2)}{(p+q)^2p^2}.$$
(B13)
Minimal substitution in $`p/`$ results in
$$p/\widehat{e}\gamma _\mu .$$
(B14)
Under minimal substitution the product $`f(p^2)p/`$ changes as
$`f(P^2)/P`$ $``$ $`f(\stackrel{~}{P}^2)\stackrel{~}{/P}=f(\stackrel{~}{P}^2)p/f((p+q)^2)\widehat{e}A^\mu \gamma _\mu +O(A^2)`$ (B15)
$`=`$ $`f(p^2)p/\widehat{e}A^\mu (2p_\mu +q_\mu ){\displaystyle \frac{f(p+q)^2)f(p^2)}{(p+q)^2p^2}}p/\widehat{e}A^\mu \gamma _\mu f((p+q)^2)+O(A^2),`$ (B16)
and hence
$$f(p^2)p/\widehat{e}(2p_\mu +q_\mu )\frac{f(p+q)^2)f(p^2)}{(p+q)^2p^2}p/\widehat{e}\gamma _\mu f((p+q)^2).$$
(B17)
Some other useful formulas are stated without proof,
$`g(pk)`$ $``$ $`\widehat{e}k_\mu {\displaystyle \frac{g((p+q)k)g(pk)}{qk}},`$ (B18)
$`{\displaystyle \frac{1}{(p+k)^2p^2}}`$ $``$ $`\widehat{e}{\displaystyle \frac{2k_\mu }{[(p+k+q)^2(p+q)^2][(p+k)^2p^2]}},`$ (B19)
$`f(p^2)g(pk)`$ $``$ $`\widehat{e}\{(2p_\mu +q_\mu ){\displaystyle \frac{f((p+q)^2)f(p^2)}{(p+q)^2p^2}}g(pk)`$ (B21)
$`+k_\mu f((p+q)^2){\displaystyle \frac{g((p+q)k)g(pk)}{qk}}\},`$
$`p_\mu `$ $``$ $`\widehat{e}g_{\mu \nu }.`$ (B22)
where $`k`$ is the momentum of an uncharged third particle and $`g(pk)`$ is a generic function.
The formulas for minimal substitution in $`P^{}`$, the momentum associated with the outgoing nucleon, are analogous to the above, except that everywhere $`q`$ should be replaced by $`q`$.
Please note that the terms generated by this minimal substitution procedure are free from poles in the limit of $`q0`$.
## C Vertices from minimal substitution
### 1 The $`\gamma \pi NN`$-vertex
Minimal substitution in the $`\pi NN`$-vertex is discussed separately for the pseudoscalar vertex,
$$\left(\mathrm{\Gamma }_{\pi NN}^{ps}\right)_\alpha =\tau _\alpha \gamma ^5f(p^2)+f(p^2)\tau _\alpha \gamma ^5,$$
(C1)
and the pseudovector vertex,
$$\left(\mathrm{\Gamma }_{\pi NN}^{pv1}\right)_\alpha =\tau _\alpha \gamma ^5g(p^2)p/+p/^{}g(p^2)\gamma ^5\tau _\alpha .$$
(C2)
The sum of these reduces for the half-off-shell vertex Eq. (11) for $`f(p^2)=G_1(p^2)G_2(p^2)G_1(m^2)/2`$ and $`g(p^2)=G_2(p^2)/m`$. Minimal substitution in Eqs. (C1,C2) gives
$$\left(\mathrm{\Gamma }_{\gamma \pi NN}^{ps}\right)_\alpha ^\mu =\gamma ^5\left\{\tau _\alpha \widehat{e}(2p^\mu +q^\mu )\frac{f((p+q)^2)f(p^2)}{(p+q)^2p^2}+\widehat{e}\tau _\alpha (2p^\mu q^\mu )\frac{f((p^{}q)^2)f(p^2)}{(p^{}q)^2p^2}\right\},$$
(C3)
and
$`\left(\mathrm{\Gamma }_{\gamma \pi NN}^{pv1}\right)_\alpha ^\mu =`$ $`\tau _\alpha \widehat{e}\gamma ^5\left\{(2p^\mu +q^\mu )p/{\displaystyle \frac{g((p+q)^2)g(p^2)}{(p+q)^2p^2}}+\gamma ^\mu g((p+q)^2)\right\}`$ (C5)
$`\widehat{e}\tau _\alpha \{(2p^\mu q^\mu ){\displaystyle \frac{g((p^{}q)^2)g(p^2)}{(p^{}q)^2p^2}}p/^{}+\gamma ^\mu g((p^{}q)^2)\}\gamma ^5,`$
respectively where $`p^{}=p+q`$.
Minimal substitution in
$$\left(\mathrm{\Gamma }_{\pi NN}^{pv2}\right)_\alpha =\tau _\alpha \gamma ^5p/g(p^2)+g(p^2)p/^{}\gamma ^5\tau _\alpha ,$$
(C6)
gives a different contact term (because $`\stackrel{~}{P}_\mu `$ and $`\stackrel{~}{P}^2`$ do not commute),
$`\left(\mathrm{\Gamma }_{\gamma \pi NN}^{pv2}\right)_\alpha ^\mu =`$ $`\tau _\alpha \widehat{e}\gamma ^5\{(2p^\mu +q^\mu )(p/+q/){\displaystyle \frac{g((p+q)^2)g(p^2)}{(p+q)^2p^2}}+\gamma ^\mu g(p^2)\}`$ (C8)
$`\widehat{e}\tau _\alpha \{(2p^\mu q^\mu ){\displaystyle \frac{g((p^{}q)^2)g(p^2)}{(p^{}q)^2p^2}}(p/^{}q/)+\gamma ^\mu g(p^2)\}\gamma ^5.`$
The difference between the vertices in Eq. (C5) and Eq. (C8) equals
$`\mathrm{\Delta }^\mu `$ $`=`$ $`\tau _\alpha \widehat{e}\gamma ^5\left\{(2p^\mu +q^\mu )q/{\displaystyle \frac{g((p+q)^2)g(p^2)}{(p+q)^2p^2}}+\gamma ^\mu [g(p^2)g((p+q)^2)]\right\}`$ (C10)
$`\widehat{e}\tau _\alpha \{(2p^\mu q^\mu ){\displaystyle \frac{g((p^{}q)^2)g(p^2)}{(p^{}q)^2p^2}}(q/)+\gamma ^\mu [g(p^2)g((p^{}q)^2)]\}\gamma ^5,`$
which is orthogonal to the photon momentum, $`q\mathrm{\Delta }=0`$. This presents one example of the known ambiguity in constructing such contact vertices: terms orthogonal to the photon momentum are not uniquely determined by the minimal substitution prescription.
### 2 The $`\gamma \gamma NN`$-vertex
As a first step, the $`\gamma NN`$-vertex needs to be constructed which reduces to the appropriate half-off-shell vertex and in addition obeys the Ward identity. It is constructed through minimal substitution in the inverse dressed nucleon propagator,
$$S^1(p)=\frac{1}{2}[\alpha (p^2)p/+p/\alpha (p^2)]+\beta (p^2)$$
(C11)
where $`\beta (p^2)=\alpha (p^2)\xi (p^2)`$. We can use Eqs. (B13,B17) to write the nucleon-photon vertex obtained by the minimal substitution as
$$\mathrm{\Gamma }_\mu ^{min}(p^{},p)=\widehat{e}_N\left\{\frac{p_\mu ^{}+p_\mu }{p^2p^2}[S^1(p^{})S^1(p)]+\frac{\alpha (p^2)+\alpha (p^2)}{2}[\gamma _\mu q/\frac{2p_\mu +q_\mu }{p^2p^2}]\right\}.$$
(C12)
where $`p^{}=p+q`$. This vertex clearly satisfies the Ward-Takahashi identity Eq. (D1). In principle, both nucleons can be off-shell in this vertex.
A general form for the vertex can now be written as
$`\mathrm{\Gamma }_\mu (p^{},p)`$ $`=`$ $`\mathrm{\Gamma }_\mu ^{min}(p^{},p)+[q/,\gamma _\mu ]\{F(p^2)(p/m)+H(p^2)\}`$ (C13)
$`+`$ $`\{(p/^{}m)F(p^2)+H(p^2)\}[q/,\gamma _\mu ].`$ (C14)
To obtain the half-off-shell vertex with the outgoing on-shell nucleon, we apply Eq. (C14) to a positive-energy spinor $`\overline{u}(p^{})`$ on the left, $`\overline{u}(p^{})p/^{}=\overline{u}(p^{})m`$. Equating the resulting half-off-shell vertex to Eq. (7) the functions $`F(p^2)`$ and $`H(p^2)`$ can be determined,
$`(F)^{s,v}(p^2)={\displaystyle \frac{(F_2^+)^{s,v}(p^2)(F_2^{})^{s,v}(p^2)}{8m^2}}+{\displaystyle \frac{\alpha (p^2)\alpha (m^2)}{8(p^2m^2)}}`$ (C15)
$`(H)^{s,v}(p^2)={\displaystyle \frac{2(F_2^+)^{s,v}(p^2)(F_2^+)^{s,v}(m^2)}{8m}}+{\displaystyle \frac{\alpha (p^2)m+\beta (p^2)}{4(p^2m^2)}}+{\displaystyle \frac{\alpha (m^2)1}{16m}},`$ (C16)
where an analogue of the Gordon identity has been used in the form
$$\overline{u}(p^{})(p_\mu +p_\mu ^{})=\overline{u}(p^{})(\gamma _\mu p/+\gamma _\mu mi\sigma _{\mu \lambda }q^\lambda ).$$
(C17)
To obtain the contact $`\gamma \gamma NN`$-vertex we perform a minimal substitution in Eq. (C14), with a second photon field carrying a momentum $`k`$ and polarization index $`\nu `$. Since both incoming $`p`$ and outgoing $`p^{}`$ nucleons are on the mass shell in Compton scattering, we need only the matrix element of the contact $`\gamma \gamma NN`$-vertex between the positive-energy spinors of the incoming and outgoing nucleons,
$`K_{\mu ,\nu }^{ct}(q,k)`$ $`=`$ $`\overline{u}(p^{})\mathrm{\Gamma }_{\mu \nu }^{ct}(q,k)u(p)=\overline{u}(p^{})i\widehat{e}^2\{{\displaystyle \frac{\alpha ((p+q)^2)m+\beta ((p+q)^2)}{2[(p+q)^2m^2]}}`$ (C24)
$`\times \left[{\displaystyle \frac{(p_\mu +p_\mu ^{}k_\mu )(p_\nu +p_\nu ^{}+q_\nu )}{(p+q)^2m^2}}{\displaystyle \frac{(p_\mu +p_\mu ^{}+k_\mu )(p_\nu +p_\nu ^{}q_\nu )}{(p+k)^2m^2}}+2g_{\mu \nu }\right]`$
$`+{\displaystyle \frac{\alpha ((p+q)^2)\alpha (m^2)}{2[(p+q)^2m^2]}}\left[(p_\nu +p_\nu ^{}+q_\nu )\gamma _\mu +(p_\mu +p_\mu ^{}k_\mu )\gamma _\nu \right]`$
$`+{\displaystyle \frac{H((p+q)^2)H(m^2)}{(p+q)^2m^2}}[[k/,\gamma _\nu ](p_\mu +p_\mu ^{}k_\mu )+(p_\nu +p_\nu ^{}+q_\nu )[q/,\gamma _\mu ]]`$
$`+F((p+q)^2)[[k/,\gamma _\nu ]\gamma _\mu +\gamma _\nu [q/,\gamma _\mu ]]+\left[\begin{array}{c}\mu \nu \\ qk\end{array}\right]\}u(p),`$
where $`p^{}=p+k+q`$ and the notation introduced in Eqs. (C11,C15,C16) has been used. In Eq. (C24) $`H=H^s+H^v`$ (and analogously for F) since the contact term vanishes for the neutron. The contact term is explicitly crossing symmetric due to the last term in Eq. (C24).
## D Gauge invariance of the method
The Ward-Takahashi identity, a consequence of gauge invariance, imposes an important constraint on the nucleon-photon vertex,
$$q\mathrm{\Gamma }(p^{},p)=\widehat{e}_N\left[S^1(p^{})S^1(p)\right],$$
(D1)
with the photon momentum $`q^\mu =p^\mu p^\mu `$. In the following we prove that the photon vertex obtained in our procedure obeys the Ward-Takahashi identity Eq. (D1).
Initially we assume that the $`\gamma NN`$-vertex on the right-hand side of Eq. (9) obeys the Ward identity. As a first step we first construct a tree level pion-photoproduction amplitude and prove its gauge invariance. The amplitude is written as a sum of s- ,u-, and t-channel contributions and a contact term (see Fig. (1)),
$$K_\alpha ^\mu =\underset{i=s,u,t,c}{}K_{i,\alpha }^\mu .$$
(D2)
The incoming pion caries momentum $`k`$, the outgoing photon $`q`$ while $`p`$ and $`p^{}`$ are the on-mass-shell momenta of the incoming and outgoing nucleons, respectively ($`p^{}=p+k+q`$). Contracting each term in Eq. (D2) with the photon momentum yields
$`q_\mu K_{s,\alpha }^\mu `$ $`=`$ $`\tau _\alpha \widehat{e}_N\overline{u}(p^{})\mathrm{\Gamma }^5(p^{}k)u(p),`$ (D3)
$`q_\mu K_{u,\alpha }^\mu `$ $`=`$ $`\widehat{e}_N\tau _\alpha \overline{u}(p^{})\overline{\mathrm{\Gamma }^5}(p^{}q)u(p),`$ (D4)
$`q_\mu K_{t,\alpha }^\mu `$ $`=`$ $`\tau _\beta (\widehat{e}_\pi )_{\beta \alpha }\overline{u}(p^{})\gamma ^5gu(p),`$ (D5)
$`q_\mu K_{c,\alpha }^\mu `$ $`=`$ $`\tau _\alpha \widehat{e}_N\overline{u}(p^{})\left[\mathrm{\Gamma }^5(p^{}k)\gamma ^5g\right]u(p)`$ (D7)
$`\widehat{e}_N\tau _\alpha \overline{u}(p^{})\left[\overline{\mathrm{\Gamma }^5}(p^{}q)+\gamma ^5g\right]u(p),`$
where we have used Eq. (D1) and the fact that for an on-shell nucleon with momentum $`p`$, $`\overline{u}(p)S^1(p)=0=S^1(p)u(p)`$. We also used the normalization condition for the nucleon-pion vertex with both nucleons on-shell, $`\overline{u}(p^{})\mathrm{\Gamma }^5(m)u(p)=\overline{u}(p^{})\gamma ^5gu(p)`$. Adding Eqs. (D3-D7) and using $`[\widehat{e}_N,\tau _\alpha ]=\tau _\beta (\widehat{e}_\pi )_{\beta \alpha }`$, we obtain the desired result, $`q_\mu K_\alpha ^\mu =0`$ .
The gauge invariance of this pion-photoproduction amplitude is used to show that the solution of Eq. (9) obeys the Ward-Takahashi identity for the nucleon-photon vertex. This can be done in a transparent way with the help of diagrammatic expressions. The pole contribution to the vertex is given by the sum of cut loop diagrams entering in the dispersion integral in Eq. (9). (We assume here that the convergence of the procedure has been reached.) This sum can be rewritten by adding and subtracting an additional diagram containing the pole contribution to the self-energy in the incoming nucleon leg, as shown in Fig. (8) (top). Index $`I`$ on the left-hand side (l.h.s.) of this equation indicates that only the pole contribution to the vertex is considered. To evaluate the scalar product of the r.h.s. with the photon momentum $`q_\mu `$, it is convenient to rewrite the equation as shown in Fig. (8) (bottom). Here, a common sub-diagram, which is a nucleon-pion vertex, has been extracted from the r.h.s., and the asterisk indicates that an integration is tacitly understood over the phase space of the cut (on-shell) nucleon and pion lines. Such separation of a sub-diagram is consistent with the interpretation of Cutkosky rules as a unitarity condition . Note also that the Dirac spinor $`\overline{u}(p^{})`$ is explicitly identified with the outgoing nucleon line. The sum of diagrams in the brackets is the scattering amplitude $`K_\alpha ^\mu `$ for pion photoproduction considered above, which is gauge invariant, i.e. $`q_\mu K_\alpha ^\mu =0`$. Therefore, only the last diagram on the r.h.s. contributes to
$$q_\mu \overline{u}(p^{})\mathrm{\Gamma }_I^\mu (p)=q_\mu \overline{u}(p^{})\mathrm{\Gamma }^\mu (p)S(p)\mathrm{\Sigma }_I(p)=\widehat{e}_N\overline{u}(p^{})\mathrm{\Sigma }_I(p),$$
(D8)
where $`\mathrm{\Sigma }_I(p)`$ stands for the pole contribution to the nucleon self-energy and we have used Eq. (D1) for the vertex on the r.h.s.. Eq. (D8) corresponds precisely to the Ward identity for the pole contribution of the vertex since the pole contribution to $`S_0^1(p)=(p/m)`$ is zero and $`S^1(p)=S_0^1(p)\mathrm{\Sigma }(p)`$. |
warning/0003/cs0003017.html | ar5iv | text | # The lexicographic closure as a revision process
## Introduction and Preliminaries
The methodological connections between the areas of nonmonotonic reasoning, i.e., the process by which an agent may, possibly, withdraw previously derived conclusions upon enlarging her set of hypotheses (?), and belief revision, i.e., the process by which an agent changes her beliefs upon discovering some new information (??), are well-known (see, for example, (????)). As a consequence, it is possible to translate particular problems in one area into problems in the other. One particular problem in nonmonotonic reasoning is the question of default entailment, i.e., when should we regard one item of so-called “default knowledge” (hereafter just “default”), i.e., an expression of the form $`\theta \varphi `$ standing for “if $`\theta `$ then normally (or usually, or typically) $`\varphi `$”, as “following from” a given set of defaults. Several answers to this question have been proposed in the literature (such as in (???????), to name but a few) but none of them (with the exception of the last named) seem to attempt any explicit connection with belief revision. The aim of this paper is to make a start on such a connection by showing how one particular method of default entailment, namely the lexicographic closure construction (??) can be given a formulation in terms of a certain method of belief revision which was first given in (?) and studied further in (?). In the process, we uncover one or two interesting avenues for further research on both sides.
The plan of this paper is as follows. Firstly, in the next section we formally pose the basic question of default entailment outlined above and describe the lexicographic closure. The set of defaults defined by the lexicographic closure, considered as a binary relation, forms a rational consequence relation (in the sense of (?)). The section following this introduces the theory of belief revision and the important notion of epistemic entrenchment relation (E-relation for short) which it utilises. Also in this section we describe the correspondence between E-relations and rational consequence relations. Next, with the aid of this correspondence, we describe Nayak’s operation of revision. Nayak proposes to model revision of an epistemic state (represented as an E-relation) by an arbitrary set of sentences by first converting this set into an E-relation and then revising by this relation. We present one particular method for generating an E-relation from a set of sentences and show our main result: that, given this method, the E-relation corresponding to the lexicographic closure can be obtained by revising the initial epistemic state (which we take to be the E-relation in which the only sentences believed are the tautologies) firstly by the set of (the material counterparts of) those defaults which are the least specific, then those defaults which are the next-least specific and so on up to the set of the most specific defaults. After this we give our ideas for possible further study before offering some short concluding remarks.
Before we get started, let us fix our notation. Throughout this paper, $`L`$ is an arbitrary but fixed propositional language built up from a finite set of propositional variables using the usual connectives $`\neg ,,,,`$ and $``$. Semantics is provided by the (finite) set $`W`$ of propositional worlds. For $`\theta L`$ we set $`S_\theta =\{wWw\theta \}`$, i.e., $`S_\theta `$ is the set of worlds which satisfy $`\theta `$. Given $`E\{\varphi \}L`$ we write $`E\varphi `$ whenever $`_{\theta E}S_\theta S_\varphi `$ and let $`Cn(E)`$ denote the set $`\{\varphi E\varphi \}`$. As usual we write $`\theta \varphi `$ rather than $`\{\theta \}\varphi `$ etc. while, for any $`wW`$ and $`EL`$ we set $`\text{sent}_E(w)=\{\theta Ew\theta \}`$. Finally, for an arbitrary set $`X`$ we use $`|X|`$ to denote the cardinality of $`X`$.
## The Lexicographic Closure of a Set of Defaults
Suppose we have somehow learnt that an intelligent agent believes some finite set of defaults $`\mathrm{\Delta }=\{\lambda _i\chi _i\lambda _i,\chi _iL,i=1,\mathrm{},l\}`$. In this case what other assertions of this form should we conclude our agent believes? Or, put another way, what is the binary relation $`^\mathrm{\Delta }`$ on $`L`$ where $`\theta ^\mathrm{\Delta }\varphi `$ holds iff we can conclude, on the basis of $`\mathrm{\Delta }`$, that if $`\theta `$ is true then, normally, $`\varphi `$ is also true? In this paper, one answer to this question which we are particularly interested in is the lexicographic closure construction which was proposed independently in both (?) and (?). We describe this construction now.
Throughout this paper we assume that $`\mathrm{\Delta }`$ is an arbitrary but fixed, finite set of defaults. For this paper we also make the simplifying assumption that $`\mathrm{\Delta }`$ is “consistent”, in the sense that its set of material counterparts $`\mathrm{\Delta }^{}=\{\lambda \chi \lambda \chi \mathrm{\Delta }\}`$ is consistent. Using a procedure given in (?) (or, equivalently, in (?)) we may partition $`\mathrm{\Delta }`$ into $`\mathrm{\Delta }=(\mathrm{\Delta }_0,\mathrm{},\mathrm{\Delta }_n)`$, where the $`\mathrm{\Delta }_i`$ correspond, in a precise sense, to “levels of specificity” – given a default $`\delta \mathrm{\Delta }`$, the larger the $`i`$ for which $`\delta \mathrm{\Delta }_i`$, the more specific are the situations to which $`\delta `$ is applicable. Following (?), we call this partition the Z-partition of $`\mathrm{\Delta }`$. Like many methods of default entailment (see (?) for several examples), the lexicographic closure can be based on a method of choosing maximal consistent subsets of $`\mathrm{\Delta }^{}`$. More precisely the lexicographic closure is a member of a family of consequence relations $`_{}^\mathrm{\Delta }`$, where $``$ is an ordering on $`2^\mathrm{\Delta }`$, and, for all $`\theta ,\varphi L`$, we have
$$\begin{array}{ccc}\hfill \theta _{}^\mathrm{\Delta }\varphi & \text{iff}& \text{for all}\mathrm{\Gamma }\mathrm{\Delta }\text{such that}\mathrm{\Gamma }^{}\{\theta \}\text{is}\hfill \\ & & \text{consistent and}\mathrm{\Gamma }\text{is }\text{-maximal amongst}\hfill \\ & & \text{such subsets, we have}\mathrm{\Gamma }^{}\{\theta \}\varphi .\hfill \end{array}$$
To specify the lexicographic closure we instantiate the order $``$ above, with the help of the Z-partition, as follows: Given subsets $`A,B\mathrm{\Delta }`$ let $`A_i=A\mathrm{\Delta }_i`$ and $`B_i=B\mathrm{\Delta }_i`$ for each $`i=0,\mathrm{},n`$. We define an ordering $`_{lex}`$ on $`2^\mathrm{\Delta }`$ by:
$$\begin{array}{ccc}\hfill A_{lex}B& \text{iff}& \text{there exists }i\text{ such that}|A_i|<|B_i|\text{and,}\hfill \\ & & \text{for all}j>i,|A_j|=|B_j|.\hfill \end{array}$$
(The reason for the name “lexicographic closure” should now be clear.) The lexicographic closure $`_{lex}^\mathrm{\Delta }`$ is then just defined to be $`_{_{lex}}^\mathrm{\Delta }`$.
How successful is $`_{lex}^\mathrm{\Delta }`$ in achieving the goals of default reasoning? We refer the reader to (?) for the details. However, the internal, closure properties of $`_{lex}^\mathrm{\Delta }`$ can be summed up by the following proposition, which can be found jointly in (?) and (?).
###### Proposition 1
The binary relation $`_{lex}^\mathrm{\Delta }`$ is a rational consequence relation (see (??)). Furthermore, $`_{lex}^\mathrm{\Delta }`$ is consistency preserving, i.e., for all $`\theta `$, $`\theta _{lex}^\mathrm{\Delta }`$ implies $`\theta `$.
Now we already know (see, for example, (??)) that rational consequence relations may be represented by finite sequences $`\stackrel{}{𝒰}=(𝒰_0,\mathrm{},𝒰_k)`$ of mutually disjoint subsets of $`W`$ in the following sense: Given such a sequence $`\stackrel{}{𝒰}`$ and $`\theta L`$ we set $`\text{rank}^\stackrel{}{𝒰}(\theta )=`$ the least $`i`$ such that $`𝒰_iS_\theta \mathrm{}`$. If no such $`i`$ exists then we set $`\text{rank}^\stackrel{}{𝒰}(\theta )=\mathrm{}`$. If we then define a binary relation $`_\stackrel{}{𝒰}`$ on $`L`$ by setting<sup>1</sup><sup>1</sup>1Note the first clause includes the case $`\text{rank}^\stackrel{}{𝒰}(\theta \neg \varphi )=\mathrm{}`$ and $`\text{rank}^\stackrel{}{𝒰}(\theta )\mathrm{}`$.
$$\begin{array}{cccc}\hfill \theta _\stackrel{}{𝒰}\varphi & \text{iff}& \text{either}& \text{rank}^\stackrel{}{𝒰}(\theta )<\text{rank}^\stackrel{}{𝒰}(\theta \neg \varphi )\hfill \\ & & \text{or}& \text{rank}^\stackrel{}{𝒰}(\theta )=\mathrm{}\hfill \end{array}$$
then $`_\stackrel{}{𝒰}`$ forms a rational consequence relation, while moreover every rational consequence relation arises in this way from some sequence $`\stackrel{}{𝒰}`$.<sup>2</sup><sup>2</sup>2Such sequences are clearly equivalent to the ranked models used to characterise rational consequence relations in (?). The intuition behind the sequences $`\stackrel{}{𝒰}`$ is that they represent a “ranking” of the worlds in $`W`$ according to their plausibility – the lower the $`i`$ for which $`w𝒰_i`$, the more plausible, in relation to the other worlds, it is considered to be. If $`w𝒰_i`$ for all $`i`$ then we may take $`w`$ to be considered “impossible”.
One thing to note about the definition of $`_\stackrel{}{𝒰}`$ given above is that we allow $`\mathrm{}`$ to appear, possibly more than once, in $`\stackrel{}{𝒰}`$.<sup>3</sup><sup>3</sup>3This approach carries us very close to the “semi-quantitative” approaches of (???), which use an explicit ranking function as a starting point rather than deriving one from a sequence of world-sets. Our approach, though, remains squarely qualitative in character. This freedom comes in useful when proving some of our results. It also has the effect that the mapping $`\stackrel{}{𝒰}_\stackrel{}{𝒰}`$ detailed above is not injective – given a rational consequence relation $``$ there will be many (in fact infinitely many) sequences $`\stackrel{}{𝒰}`$ such that $`=_\stackrel{}{𝒰}`$.<sup>4</sup><sup>4</sup>4Since clearly we can insert as many copies of $`\mathrm{}`$ into the sequence $`(𝒰_0,\mathrm{},𝒰_k)`$ as we wish without changing the relation $`_\stackrel{}{𝒰}`$. Another thing to note about $`_\stackrel{}{𝒰}`$ is that $`_\stackrel{}{𝒰}`$ will be consistency preserving iff $`_{i=0}^k𝒰_i=W`$, while it will be trivial, i.e., will satisfy $`\theta _\stackrel{}{𝒰}\varphi `$ for all $`\theta `$ and $`\varphi `$, iff $`_{i=0}^k𝒰_i=\mathrm{}`$. We make the following definitions:
###### Definition 1
Let $`\stackrel{}{𝒰}=(𝒰_0,\mathrm{},𝒰_k)`$ be a finite sequence of mutually disjoint subsets of $`W`$. We shall say that $`\stackrel{}{𝒰}`$ is full iff $`_{i=0}^k𝒰_i=W`$ and that $`\stackrel{}{𝒰}`$ is empty iff $`_{i=0}^k𝒰_i=\mathrm{}`$. We let $`\mathrm{{\rm Y}}`$ denote the set of all such $`\stackrel{}{𝒰}`$ which are either full or empty.
Hence Proposition 1 tells us that there must exist a full sequence $`\stackrel{}{𝒰}\mathrm{{\rm Y}}`$ such that $`\theta _{lex}^\mathrm{\Delta }\varphi `$ iff $`\theta _\stackrel{}{𝒰}\varphi `$. What form does $`\stackrel{}{𝒰}`$ take here? The answer is given in (?) and (?) (and is, in fact, used to define $`_{lex}^\mathrm{\Delta }`$ in the latter). In this paper we show that we can arrive at this answer via a different route.
## Belief Revision and Epistemic Entrenchment
Belief revision is concerned with the following problem: How should an agent revise her beliefs upon receiving some new information which may, possibly, contradict some of her current beliefs? The most popular basic framework within which this question is studied is the one laid down by Alchourŕon, Gärdenfors and Makinson (AGM) in (?). In that framework an agent’s epistemic state is represented as a logically closed set of sentences called a belief set, and the new information, or epistemic input, is represented as a single sentence. AGM propose a number of postulates which a reasonable operation of revision should satisfy. In particular, the revised belief set should contain the epistemic input and should be consistent.<sup>5</sup><sup>5</sup>5Unless the epistemic input itself is inconsistent. See (?) for the full list of postulates with detailed discussion. In order to meet these requirements, in the general case when the input is inconsistent with the prior belief set, the agent is forced to give up some of her prior beliefs. One way of determining precisely which sentences the agent should give up in this situation is to assign to the agent an E-relation $``$ on $`L`$ (see, for example, (?????)).
The intuitive meaning behind E-relations is that $`\varphi \psi `$ should hold iff the agent finds it at least as easy to give up $`\varphi `$ as she does $`\psi `$, i.e., her belief in $`\psi `$ is at least as entrenched as her belief in $`\varphi `$. In cases of conflict the agent should then give up those sentences which are less entrenched. In what follows we use $``$ to denote the strict part of $``$, i.e, $`\theta \varphi `$ iff $`\theta \varphi `$ and not($`\varphi \theta `$). We follow (?) in formally defining E-relations as follows:
###### Definition 2
An epistemic entrenchment relation (E-relation) (on $`L`$) is a relation $`L\times L`$ which satisfies the following conditions for all $`\theta ,\varphi ,\psi L`$,
(E1) If $`\theta \varphi `$ and $`\varphi \psi `$ then $`\theta \psi `$ (transitivity) (E2) If $`\theta \varphi `$ then $`\theta \varphi `$ (dominance) (E3) $`\theta \theta \varphi `$ or $`\varphi \theta \varphi `$ (conjunctiveness) (E4) Given there exists $`\psi L`$ such that $`\psi `$, if $`\theta \varphi `$ for all $`\theta L`$, then $`\varphi `$ (maximality)
If there is no $`\psi L`$ such that $`\psi `$, equivalently, if $`\theta \varphi `$ holds for all $`\theta ,\varphi `$, then we call $``$ the absurd E-relation. The original definition of E-relation, such as is found in (?), is given relative to a belief set. However, as is noted in (?), E-relations contain enough information by themselves for the belief set to be extracted from it. The belief set $`Bel()`$ associated with the E-relation $``$ is defined as:
$$Bel()=\{\begin{array}{cc}\{\theta \theta \}\hfill & \text{if}\theta ,\text{for some}\theta ,\hfill \\ L\hfill & \text{otherwise.}\hfill \end{array}$$
The belief set associated with an E-relation was called its epistemic content in (?).
### E-relations and Rational Consequence
We now bring in the connection between E-relations, as they have been defined here, and rational consequence relations. The following result is virtually the same as one given in (?).
###### Proposition 2
Let $``$ be a rational consequence relation which is either consistency preserving or trivial. If we define, from $``$, a binary relation $`_{}`$ on $`L`$ by setting, for all $`\theta ,\varphi L`$,
$$\theta _{}\varphi \text{iff}\neg \theta \neg \varphi \mid ̸\theta \text{or}\neg \varphi ,$$
(1)
then $`_{}`$ forms an E-relation. Conversely if, given an E-relation $``$ we define a binary relation $`_{}`$ on $`L`$ by setting, for all $`\theta ,\varphi L`$,
$$\theta _{}\varphi \text{iff}\neg \theta \neg \theta \varphi \text{or}\neg \theta $$
then $`_{}`$ forms a rational consequence relation which is either consistency preserving or trivial. Furthermore the identity $`=_{_{}}`$ holds.
So there is a bijection between rational consequence relations which are either consistency preserving or trivial, and E-relations. Essentially they are different ways of describing the same thing, and so an operation for changing one automatically gives us an operation for changing the other. This observation is at the heart of the present paper. Given $`\stackrel{}{𝒰}\mathrm{{\rm Y}}`$ we shall denote by $`_\stackrel{}{𝒰}`$ the E-relation defined from $`_\stackrel{}{𝒰}`$ via (1) above. Since we have already seen that rational consequence relations which are either consistency preserving or trivial are characterised by the sequences in $`\mathrm{{\rm Y}}`$, Proposition 2 leads us to the following result.
###### Proposition 3
Let $``$ be a binary relation on $`L`$. Then $``$ is an E-relation iff $`=_\stackrel{}{𝒰}`$ for some $`\stackrel{}{𝒰}\mathrm{{\rm Y}}`$.
Note again that $`_\stackrel{}{𝒰}=_\stackrel{}{𝒱}`$ does not imply $`\stackrel{}{𝒰}=\stackrel{}{𝒱}`$. Also note that $`_\stackrel{}{𝒰}`$ will be absurd iff $`\stackrel{}{𝒰}`$ is empty. It is straightforward to prove the following.
###### Proposition 4
Let $`\stackrel{}{𝒰}\mathrm{{\rm Y}}`$ and $`\theta L`$. Then $`\theta Bel(_\stackrel{}{𝒰})`$ iff $`_\stackrel{}{𝒰}\theta `$.
## Revision of E-relations
Nayak (?) deviates from the basic AGM framework in two ways. Firstly, in order to help us deal with iterated revision (see (???)), he argues that we need not only a description of the new belief set which results from a revision, but also a new E-relation which can then guide any further revision. Thus we should enlarge our epistemic state to consist of a belief set together with an E-relation and then perform revision on this larger state. In fact, since, as we have seen, the belief set may be determined from the E-relation, we may take our epistemic states to be just E-relations.<sup>6</sup><sup>6</sup>6In this context of iterated revision, the consideration of more comprehensive epistemic states of which a belief set is but one component has also been suggested in (?) and (?). Secondly, he suggests that the epistemic input should consist not of a single sentence, but rather another E-relation. (See (?) for motivation.) He claims it is then possible, in his framework, to capture the revision of E-relations by arbitrary sets of sentences $`E`$ by first converting the set $`E`$ into a suitable E-relation $`_E`$ and then revising by $`_E`$. We shall discuss this point further in the next section. In this section we shall use the characterisation of E-relations given in Proposition 3 to describe Nayak’s proposal of how one E-relation should be revised by another to obtain a new E-relation. The ideas behind this formulation can also be seen in (?).
Let $`_K`$ be the prior E-relation and let $`_E`$ be the input E-relation. By Proposition 3, we know that there exist $`\stackrel{}{𝒰},\stackrel{}{𝒱}\mathrm{{\rm Y}}`$ such that $`_K=_\stackrel{}{𝒰}`$ and $`_E=_\stackrel{}{𝒱}`$. Hence we may reduce the question of entrenchment revision to a question of how to revise one sequence of world-sets by another. More precisely, we can define a sequence revision function $`:\mathrm{{\rm Y}}\times \mathrm{{\rm Y}}\mathrm{{\rm Y}}`$, where $`\stackrel{}{𝒰}\stackrel{}{𝒱}`$ is the result of revising $`\stackrel{}{𝒰}`$ by $`\stackrel{}{𝒱}`$, and then simply lift this to an entrenchment revision function by setting
$$_K_E=_{\stackrel{}{𝒰}\stackrel{}{𝒱}}.$$
(2)
(The context will always make it clear whether we are considering $``$ as an operation on sequences or an operation on E-relations.) All this must be independent of precisely which $`\stackrel{}{𝒰}`$ and $`\stackrel{}{𝒱}`$ are chosen to represent $`_K`$ and $`_E`$ respectively. The definition for the sequence revision function $``$ which we choose, motivated purely in order to arrive at Nayak’s entrenchment revision function, is the following:
###### Definition 3
We define the function $`:\mathrm{{\rm Y}}\times \mathrm{{\rm Y}}\mathrm{{\rm Y}}`$ by setting , for all $`\stackrel{}{𝒰}=(𝒰_0,\mathrm{},𝒰_k)`$ and $`\stackrel{}{𝒱}=(𝒱_0,\mathrm{},𝒱_m)`$,
$$\stackrel{}{𝒰}\stackrel{}{𝒱}=\{\begin{array}{c}\begin{array}{c}(𝒰_0𝒱_0,𝒰_1𝒱_0,\mathrm{},𝒰_k𝒱_0,\hfill \\ 𝒰_0𝒱_1,𝒰_1𝒱_1,\mathrm{},𝒰_k𝒱_1,\hfill \\ \mathrm{},\hfill \\ 𝒰_0𝒱_m,𝒰_1𝒱_m,\mathrm{},𝒰_k𝒱_m).\hfill \end{array}\text{if}\stackrel{}{𝒰}\text{is full}\hfill \\ \stackrel{}{𝒱}\text{otherwise.}\hfill \end{array}$$
Clearly it is the case that $`\stackrel{}{𝒰}\stackrel{}{𝒱}`$ is always full, unless $`\stackrel{}{𝒱}`$ is empty, in which case so is $`\stackrel{}{𝒰}\stackrel{}{𝒱}`$. Hence we certainly have $`\stackrel{}{𝒰}\stackrel{}{𝒱}\mathrm{{\rm Y}}`$. The following proposition assures us that $``$, when lifted to an operation on E-relations, is well-defined.
###### Proposition 5
Let $`\stackrel{}{𝒰}_i,\stackrel{}{𝒱}_i\mathrm{{\rm Y}}`$ for $`i=1,2`$. Then $`_{\stackrel{}{𝒰}_1}=_{\stackrel{}{𝒰}_2}`$ and $`_{\stackrel{}{𝒱}_1}=_{\stackrel{}{𝒱}_2}`$ implies $`_{\stackrel{}{𝒰}_1\stackrel{}{𝒱}_1}=_{\stackrel{}{𝒰}_2\stackrel{}{𝒱}_2}`$.
From now on we will follow Nayak and use $`_{KE}`$ as an abbreviation for $`_K_E`$. The authors of (?) propose the following postulates for the revision of E-relations:
(E$`1^{}`$) $`_{KE}`$ is an E-relation. (E$`2^{}`$) If $`\theta _E\varphi `$ then $`\theta _{KE}\varphi `$. (E$`3^{}`$) If both $`\theta _E\varphi `$ and $`\varphi _E\theta `$ and if, for all $`\lambda `$, $`\chi `$ such that $`\theta \varphi \chi `$ and $`\theta \chi `$, we have $`\lambda _K\chi `$ iff $`\lambda _E\chi `$, then $`\theta _{KE}\varphi `$ iff $`\theta _K\varphi `$.
We refer the reader to (?) for the justification of these postulates. Any operation of revision of E-relations which satisfies the above three conditions is called a well-behaved entrenchment revision operation in (?), where it is shown that there is, in fact, precisely one well-behaved entrenchment revision operation, namely the one given in (?). Thus the above three postulates serve to characterise Nayak’s revision method. Our revision operation, defined by Definition 3 via (2) above, also satisfies (E$`1^{}`$)–(E$`3^{}`$) and hence is semantically equivalent to the operation constructed in (?).
###### Theorem 1
If we set $`_{KE}=_{\stackrel{}{𝒰}\stackrel{}{𝒱}}`$ where $`\stackrel{}{𝒰}`$ ($`\stackrel{}{𝒱}`$) is chosen so that $`_K=_\stackrel{}{𝒰}`$ ($`_E=_\stackrel{}{𝒱}`$) then the operator $``$ satisfies (E$`1^{}`$), (E$`2^{}`$) and (E$`3^{}`$).
One advantage of this particular formulation is that it is relatively easy to show properties of the well-behaved entrenchment revision operation $``$. For example, the following proposition regarding sequence revision is straightforward to prove.
###### Proposition 6
Let $`\stackrel{}{𝒰},\stackrel{}{𝒱},\stackrel{}{𝒲}\mathrm{{\rm Y}}`$ and suppose $`\stackrel{}{𝒱}`$ is not empty. Then $`(\stackrel{}{𝒰}\stackrel{}{𝒱})\stackrel{}{𝒲}=\stackrel{}{𝒰}(\stackrel{}{𝒱}\stackrel{}{𝒲})`$.
This proposition, in turn, gives us the following interesting associativity property of the induced entrenchment revision operation.
###### Proposition 7
Let $`_i`$ be an E-relation for $`i=1,2,3`$. Then, if $`_2`$ is not absurd, we have $`(_1_2)_3=`$ $`_1(_2_3)`$.
## Generating E-relations from Sets of Sentences
As we said in the last section, Nayak proposes that his way of revising one E-relation by another allows a way of modelling the revision of an E-relation by a set of sentences $`E`$ by first converting, according to some suitable method, the set $`E`$ into an E-relation $`_E`$ and then revising by $`_E`$. The question of which “suitable method” we should use for generating $`_E`$ is clearly an interesting question in itself. A strong feeling is that the relation $`_E`$ should adequately convey the informational content of $`E`$, but what does this mean? An obvious first requirement of $`_E`$ would seem to be $`Bel(_E)=Cn(E)`$, but there are different ways in which this can be achieved. The definition which Nayak seems to advocate is the following, based on an idea in (?), and expressed via its strict part.
$$\begin{array}{ccc}\hfill \theta _E\varphi & \text{iff}& E\vDash ̸,\vDash ̸\theta \text{and}\text{for all}E^{}E\text{such that}\hfill \\ & & E^{}\{\neg \varphi \}\text{is consistent, there exists}\hfill \\ & & E^{\prime \prime }E\text{such that}E^{}E^{\prime \prime }\text{and}E^{\prime \prime }\{\neg \theta \}\hfill \\ & & \text{is consistent.}\hfill \end{array}$$
The clause “$`E\vDash ̸`$” in the above merely ensures that if $`E`$ is inconsistent then $`_E`$ is absurd, while the clause “$`\vDash ̸\theta `$” ensures that tautologies are maximally entrenched. The main body of the definition essentially says that $`\varphi `$ should be strictly more entrenched than $`\theta `$ iff each $``$-maximal subset of $`E`$ which fails to imply $`\varphi `$ may be strictly enlarged to a subset of $`E`$ which fails to imply $`\theta `$. The problem with defining $`_E`$ in this way is that it will fail, in general, to be an E-relation. In particular it will not necessarily satisfy (E1).<sup>7</sup><sup>7</sup>7It should be noted, however, that $`_E`$ so defined does still enjoy several interesting properties. In fact it belongs to Rott’s family of generalized E-relations (?). How can we modify/extend it so as to obtain an E-relation? The possibility we choose is to compare the sets which fail to imply $`\theta `$ and $`\varphi `$ by cardinality rather than inclusion:<sup>8</sup><sup>8</sup>8Possibilities in this spirit are also discussed in (?) (Section 2) and (?) (Section 8). See also the closely related Section 5 of (?).
###### Definition 4
Given a set $`EL`$, define a relation $`_EL\times L`$ by, for all $`\theta ,\varphi L`$,
$$\begin{array}{ccc}\hfill \theta _E\varphi & \text{iff}& E\vDash ̸,\vDash ̸\theta \text{and}\text{for all}E^{}E\text{such that}\hfill \\ & & E^{}\{\neg \varphi \}\text{is consistent, there exists}\hfill \\ & & E^{\prime \prime }E\text{such that}|E^{}|<|E^{\prime \prime }|\text{and}\hfill \\ & & E^{\prime \prime }\{\neg \theta \}\text{is consistent.}\hfill \end{array}$$
Note that this definition does indeed extend the “old” definition given above. That $`_E`$ defined by Definition 4 is a genuine E-relation will follow once we have found a sequence $`\stackrel{}{𝒰}\mathrm{{\rm Y}}`$ such that $`_E=_\stackrel{}{𝒰}`$. We do this as follows. Let us assume for simplicity that $`E`$ is finite with $`|E|=k`$. Then, for each $`i=0,\mathrm{},k`$, we set
$$𝒰_i^E=\{\begin{array}{cc}\{wW|\text{sent}_E(w)|=ki\}\hfill & \text{if}E\vDash ̸\hfill \\ \mathrm{}\hfill & \text{otherwise.}\hfill \end{array}$$
So, in the principal case when $`E`$ is consistent, $`𝒰_i^E`$ contains those worlds which satisfy precisely $`ki`$ elements of $`E`$. Let $`\stackrel{}{𝒰}^E=(𝒰_0^E,\mathrm{},𝒰_k^E)`$.
###### Proposition 8
If $`E`$ then $`\stackrel{}{𝒰}^E`$ is empty, while if $`E\vDash ̸`$ then $`\stackrel{}{𝒰}^E`$ is full (and so, either way, $`\stackrel{}{𝒰}^E\mathrm{{\rm Y}}`$). In both cases we have $`_E=_{\stackrel{}{𝒰}^E}`$. Hence $`_E`$ is an E-relation.
Note that, with this notation, we have $`\stackrel{}{𝒰}^{\mathrm{}}=(W)`$. Hence we can think of $`_{\mathrm{}}`$ as being the initial epistemic state in which each world is equally plausible.
How does $`_E`$ portray the informational content of $`E`$? The sequence $`\stackrel{}{𝒰}^E`$ shows us clearly. First of all it is easy to see that $`_E`$ satisfies the basic requirement of $`Bel(_E)=Cn(E)`$ (in particular the only sentences believed in $`_{\mathrm{}}`$ are the tautologies) since the most plausible worlds in $`\stackrel{}{𝒰}^E`$, i.e., the worlds in $`𝒰_0^E`$, are precisely those worlds which satisfy every sentence in $`E`$. The big question is how does $`\stackrel{}{𝒰}^E`$ classify the worlds which do not satisfy every sentence in $`E`$? The answer is that it considers one such world more plausible than another iff it satisfies strictly more sentences in $`E`$. This makes the relation $`_E`$ dependent on the syntactic form, not just the semantic form, of $`E`$, i.e., we can have $`Cn(E_1)=Cn(E_2)`$ without necessarily having $`_{E_1}=_{E_2}`$. One situation where this method might be deemed suitable is if we want to regard the elements of $`E`$ as items of information coming from different, independent sources.
From now on, for the special case when $`E`$ is a singleton, we shall write $`_\theta `$ rather than $`_{\{\theta \}}`$ etc. We have the following partial generalisation of Proposition 4.
###### Proposition 9
Let $`\stackrel{}{𝒰}\mathrm{{\rm Y}}`$ be full and let $`\theta ,\varphi L`$. Then $`\varphi Bel(_\stackrel{}{𝒰}_\theta )`$ iff $`\theta _\stackrel{}{𝒰}\varphi `$.
We are now ready to give the sequence $`\stackrel{}{𝒰}`$ such that $`\theta _\stackrel{}{𝒰}\varphi `$ iff $`\theta _{lex}^\mathrm{\Delta }\varphi `$. Let $`(\mathrm{\Delta }_0,\mathrm{},\mathrm{\Delta }_n)`$ be the Z-partition of $`\mathrm{\Delta }`$. Then, to obtain our special $`\stackrel{}{𝒰}`$ we start at the sequence $`(W)`$ and then successively revise, using our sequence revision function $``$, by $`\stackrel{}{𝒰}^{\mathrm{\Delta }_i^{}}`$ for $`i=0,1,\mathrm{},n`$. Recalling that $`(W)=\stackrel{}{𝒰}^{\mathrm{}}`$ we may give our main result. Recall that we are assuming $`\mathrm{\Delta }`$ is finite and that $`\mathrm{\Delta }^{}`$ is consistent.
###### Theorem 2
Let $`\mathrm{\Delta }`$ be a set of defaults with associated Z-partition $`(\mathrm{\Delta }_0,\mathrm{},\mathrm{\Delta }_n)`$. Then, for all $`\theta ,\varphi L`$, we have $`\theta _{lex}^\mathrm{\Delta }\varphi `$ iff $`\theta _{\stackrel{}{𝒰}^{\mathrm{}}\stackrel{}{𝒰}^{\mathrm{\Delta }_0^{}}\mathrm{}\stackrel{}{𝒰}^{\mathrm{\Delta }_n^{}}}\varphi `$.
Note that, by Proposition 6 and the assumption that $`\mathrm{\Delta }^{}`$ is consistent, the term $`\stackrel{}{𝒰}^{\mathrm{}}\stackrel{}{𝒰}^{\mathrm{\Delta }_0^{}}\mathrm{}\stackrel{}{𝒰}^{\mathrm{\Delta }_n^{}}`$ is independent of the bracketing. Similar remarks apply (using Proposition 7) to the next result. Using Propositions 8 and 9 we may re-express Theorem 2 as:
###### Corollary 1
Let $`\mathrm{\Delta }`$ be a set of defaults with associated Z-partition $`(\mathrm{\Delta }_0,\mathrm{},\mathrm{\Delta }_n)`$. Then, for all $`\theta ,\varphi L`$, we have $`\theta _{lex}^\mathrm{\Delta }\varphi `$ iff $`\varphi Bel(_{\mathrm{}}_{\mathrm{\Delta }_0^{}}\mathrm{}_{\mathrm{\Delta }_n^{}}_\theta )`$.
If we go further and actually identify a revision of the form $`_E`$ with $`E`$ then we have the following characterisation of the lexicographic closure.
###### Corollary 2
Let $`\mathrm{\Delta }`$ be a set of defaults with associated Z-partition $`(\mathrm{\Delta }_0,\mathrm{},\mathrm{\Delta }_n)`$. Then, for all $`\theta ,\varphi L`$, we have $`\theta _{lex}^\mathrm{\Delta }\varphi `$ iff $`\varphi Bel(_{\mathrm{}}\mathrm{\Delta }_0^{}\mathrm{}\mathrm{\Delta }_n^{}\theta )`$.
Hence, using this particular method of revision and this particular way of interpreting revision by a set of sentences, we have shown that $`\theta _{lex}^\mathrm{\Delta }\varphi `$ iff $`\varphi `$ is believed after first successively revising the initial epistemic state by the set of sentences $`\mathrm{\Delta }_i^{}`$ for $`i=0,1,\mathrm{},n`$, and then revising by $`\theta `$.
## Further Work
The developments in the previous sections have raised a couple of questions regarding both belief revision and default entailment. Firstly, while there have been several papers published concerned with iterated revision by single sentences, and also some concerned with revision by sets of sentences,<sup>9</sup><sup>9</sup>9Either directly (e.g. (?)) or indirectly, via the study of contraction by a set of sentences (e.g. (?)). See (?) for a description of contractions and their close relationship with revision. there seems to be little in the way of any systematic study of iterated revision by sets of sentences.<sup>10</sup><sup>10</sup>10An exception, in a slightly more complex framework, is (?). Darwiche and Pearl (?) provide a postulational approach to the question of iterated revision of epistemic states by single sentences. In this approach they take the concept of epistemic state to be primitive, assuming only that from each such state $`\mathrm{\Psi }`$ we may extract a belief set (in the usual AGM sense of the term) $`B(\mathrm{\Psi })`$ representing the set of sentences accepted in that state. For example Darwiche and Pearl’s second postulate may be stated as
$$\text{If}\varphi \neg \theta \text{then}B((\mathrm{\Psi }\theta )\varphi )=B(\mathrm{\Psi }\varphi ).$$
(For the other postulates and their justifications see (?).) It is not difficult to see that, if we identify epistemic state here with E-relation and take $`B()=Bel()`$, then the method proposed by Nayak, on its restriction to single sentences<sup>11</sup><sup>11</sup>11We obviously interpret single sentences here as singleton sets. satisfies all of Darwiche and Pearl’s postulates. However, it also satisfies some interesting properties in the general case. For example, given an E-relation $``$ and $`E_1E_2L`$ such that $`E_2`$ is consistent, we have $`(E_2)E_1=(E_2E_1)E_1`$. In particular, if $`\{\theta ,\varphi \}`$ is consistent, we have $`(\{\theta ,\varphi \})\varphi =(\theta )\varphi `$. (Note this is a stronger statement than just $`Bel((\{\theta ,\varphi \})\varphi )=Bel((\theta )\varphi )`$.) The question of whether this, or any other, property of iterated revision by sets is desirable seems to be a question worth investigating. Another question is: Can we, by modifying the various parameters involved in this revision process, model any of the other existing methods of default entailment, apart from the lexicographic closure, or even construct new ones? For example, given our set of defaults $`\mathrm{\Delta }`$ and its Z-partition $`(\mathrm{\Delta }_0,\mathrm{},\mathrm{\Delta }_n)`$, let $`\mathrm{\Theta }_i=_{ij}\mathrm{\Delta }_j`$ for each $`i=1,\mathrm{},n`$. Then, by the above comments, we may rewrite Corollary 2 as
$$\theta _{lex}^\mathrm{\Delta }\varphi \text{iff}\varphi Bel(_{\mathrm{}}\mathrm{\Theta }_0^{}\mathrm{}\mathrm{\Theta }_n^{}\theta ).$$
We conjecture that if we now replace each $`\mathrm{\Theta }_i^{}`$ in the above by $`\mathrm{\Theta }_i^{}`$ (i.e., the conjunction, in some order, of the sentences in $`\mathrm{\Theta }_i^{}`$), then we obtain the rational closure (?) (which is semantically equivalent to System Z (?)) of $`\mathrm{\Delta }`$, instead of the lexicographic closure. This and other variations are the subject of ongoing study. Finally, note that, since we assumed at the outset that our language $`L`$ is based on only finitely many propositional variables, and also that $`\mathrm{\Delta }`$ is a finite set of defaults, we have not needed in this paper to confront the question of revision by infinite sets of sentences. It remains to be seen to what extent the ideas in this paper can be extended to cover this more general situation.<sup>12</sup><sup>12</sup>12For one treatment of this topic, and its relation with nonmonotonic inference from infinite sets of premises, see (?).
## Conclusion
In this paper we have taken a particular model of default reasoning – the lexicographic closure – and re-cast it in terms of iterated belief revision by sets of sentences, using the particular, independently motivated, revision model of Nayak. In the process of doing this, a couple of interesting avenues for further exploration have suggested themselves. In particular, the questions of which properties of iterated multiple revision should be deemed desirable, and of how we may apply the principles underlying the AGM theory of belief revision in the context of default reasoning.
## Acknowledgements
This work is supported by the DFG project “Computationale Dialektik” within the DFG research group “Kommunikatives Verstehen”. Much of this paper was written while the author was a researcher at the Max-Planck-Institute for Computer Science in Saarbrücken, Germany. The author would like to thank Emil Weydert, Michael Freund, Hans Rott, Leon van der Torre and the anonymous referees for helpful comments and suggestions. |
warning/0003/hep-th0003250.html | ar5iv | text | # The Image of Self Intersecting QCD Strings in Four Dimensions
## I Introduction
Quantum Chromodynamics (QCD) is widely regarded as the correct theory of strong interactions. However, inherently non-perturbative effects such as confinement remain hidden in the theory due to the nature of the strong coupling of QCD. In 1973, t’Hooft proposed the parameter $`N_c`$ of the color gauge group be treated as a free parameter, and considered the limit $`N_c\mathrm{}`$ applied to the expansion of a gauge theory. The resulting expansion gives Feynman diagrams which possess the same topology as the quantized dual string model with quarks at the string ends. Also, Wilson showed that lattice gauge theory in the strong-coupling approximation is a confining theory due to the formation of color charged strings formed by Faraday flux lines. However this strong coupling approximation cannot be made in the continuous theory. Later Makeenko and Migdal as well as Gervais and Nevue were able to establish to first order in the lattice spacing an equivalence between a Wilson loop
$`\mathrm{\Phi }(x^\mu (\tau ))=trT\mathrm{exp}({\displaystyle A_\nu \frac{dX^\nu }{d\sigma }𝑑\sigma })`$
and solutions to the single string Schroedinger equation, provided that the gauge field satisfied the Yang-Mills equations of motion. These indirect probes suggests that the effective action for QCD in the strong coupling regime is a string-like theory. Although the precise string theory that governs strongly coupled QCD is unknown we expect this string theory to possess certain qualitative features that are characteristic of QCD. Polyakov and Balachandran et al. examined the effect of adding a term proportional to the extrinsic curvature to the Nambu-Goto action so that $`\mathrm{\Theta }`$ vacua effects could be incorporated into the string ansatz. The extrinsic curvature term admits self-intersecting immersions as solutions and Nair and Mazur relate the self-intersections number to the instanton number.
In this note we will examine these self-intersecting string immersions while in a fixed frame in four-dimensions to see if these configurations possess any realistic topological properties that we expect from an effective action of QCD. We will focus our attention on the torus knot solutions of Robertson in order to be explicit. We will show that there exists topologies that supports monopole/anti-monopole fields whose flux is constrained to a finite string (tube) and in some cases to strings that are infinite in length. Furthermore we will show that there exists topologies that can support pair production and annihilation. By this we mean that the immersed world sheets have remnants that appear as finite segments of strings that emerge and dissolve in a fixed frame. These segments are related to an integer $`q`$ that appears in the self-dual solutions that we consider. We are able to show that $`q=111`$ puts an upper limit on the production of fermions. We then examine the pullback of $`\gamma _5`$ onto the world sheet and show that there can be chiral symmetry breaking “bubbles” around the points where the immersion self-intersects with itself. This helps strengthen the relationship between intersecting strings on the one hand and QCD instantons on the other. Next we verify the conjecture of Robertson by showing that for torus knot solutions that the intersection number is given by $`\nu =4(pq)`$. We end by providing a program in MAPLE that annimates the torus knot and torus link solutions.
## II Topological Support For QCD Processes From Self-Dual Strings
One of the questions we ask is what topologies are provided by the string to support the gauge fields and quarks in four dimensions. We are interested in the classical features of the self-intersecting instantons since these feature should appear as first order quantum corrections to the effective action of QCD in the strong coupling regime. In order to appreciate the behavior and support of the strings we need to sit in a frame and examine the topologies that the immersed string unfolds. Since the self-dual solutions have non-trivial winding number any features we find related to the self-intersection will be protected from quantum fluctuations since winding number is a topological propoerty of the string configuration. To proceed we will look at the Nambu-Goto action and include modifications introduced by Polyakov. In order to minimize the contributions of crumpled string world sheets to the effective action of QCD, Polyakov proposed the action following;
$$S=\mu \sqrt{g}𝑑\sigma 𝑑\tau +\frac{1}{\alpha _0}\sqrt{g}K_b^{Aa}K_{Aa}^b𝑑\sigma 𝑑\tau $$
(1)
where $`K_b^{Aa}`$ is the extrinsic curvature tensor. However the presence of the extrinsic curvature term not only acts as a regulator for bosonic strings but also provides an avenue for the introductions of the topological instanton solutions. Furthermore it has been shown that the presence of the extrinsic curvature is necessary for the effective action of free fermions projected onto the world sheet.
The last summand to the above action can be expressed in an alternate form as;
$`S_1`$ $`=`$ $`{\displaystyle \frac{1}{\alpha _0}}{\displaystyle \sqrt{g}g^{ab}_at_{\mu \nu }_bt^{\mu \nu }d\sigma d\tau }`$ (2)
$`S_1`$ $`=`$ $`{\displaystyle \frac{1}{2\alpha _0}}{\displaystyle \sqrt{g}g^{ab}((_at_{\mu \nu }_a\stackrel{~}{t}^{\mu \nu })(_bt_{\mu \nu }_b\stackrel{~}{t}^{\mu \nu })\pm (_a\stackrel{~}{t}_{\mu \nu }_bt^{\mu \nu }))𝑑\sigma 𝑑\tau }`$ (4)
$`\text{where }t^{\mu \nu }={\displaystyle \frac{ϵ^{ab}}{\sqrt{g}}}_aX^\mu _bX^\nu `$
As with QCD, the action is minimized by searching for self-dual solutions, viz. $`_at^{\mu \nu }=_at_{}^{}{}_{}{}^{\mu \nu }`$. Wheater showed that any surface embedded in four-dimensional space-time which is a complex curve is a solution to the equations of motion for the extrinsic curvature term. Robertson showed one explicit example of a self-dual string instanton is a (p,q) torus knot. These are examples of superminimal immersions into $`R^4`$. Pawelczyk has extended the classification of instantons and includes the Euler character of the surface as well . A torus is parametrized by two variables, $`u`$ and $`v`$, representing local coordinates on the surface. The torus knot is the curve on the torus defined by the constraint $`u^p+v^q=0`$. Figure 1 shows an example, the (3,2) torus knot.
For a (p,q) torus knot instanton solution the string vector is
$$X^\mu =[\mathrm{}(z^p),\mathrm{}(z^p),\mathrm{}(z^q),\mathrm{}(z^q)],$$
(5)
where $`z=\tau +i\sigma `$ and $`\mathrm{}`$ and $`\mathrm{}`$ are the real and imaginary parts respectively. Note that Robertson has shown somewhat more generally that a string vector of the form
$`X^\mu =[\mathrm{}(F(z)),\mathrm{}(F(z)),\mathrm{}(G(z)),\mathrm{}(G(z))]`$
with $`F(z)`$ and $`G(z)`$ as any functions analytic in $`z`$ are also solutions to the equations of motion. These solutions automatically satisfy the Euclidean string equations
$$_\tau _\tau X^\mu +_\sigma _\sigma X^\mu =0,$$
(6)
as well as the Euclidean constraint equations
$$_\sigma X^\mu _\tau X^\mu =0,$$
(7)
and
$$(_\sigma X^\mu )^2=(_\tau X^\nu )^2.$$
(8)
To study these solutions, we create computer animated plots of the solutions over a finite time interval. We start with a particular (p,q) torus knot solution such as the (3,2) knot. This is a topologically non-trivial solution with self-intersection number $`\nu =4`$. The solution takes the form
$`X^\mu (\sigma ,\tau )=[3\tau ^2\sigma \sigma ^3,\tau ^33\tau \sigma ^2,\sigma ^2\tau ^2,2\tau \sigma ]=[x,y,z,t]`$
$`X^\mu `$ is a mapping from ($`\sigma ,\tau `$) space to ($`x,y,z,t`$) space. We choose the last coordinate to be our time parameter and invert the the last term and replace $`\tau `$ in $`X^\mu (\sigma ,\tau )`$ with the solution in terms of $`X^4=t`$ and $`\sigma `$. (One may chose other coordinates for the time such as $`X^2`$ where the $`p=1`$ cases correspond to the usual choice where $`\tau `$ is the time parameter.) In the case of the (3,2) torus knot, the inversion of the last term gives $`\tau =\frac{t}{2\sigma }`$; substitution into $`X^\mu `$ gives
$`X^\mu (\sigma ,t)=[{\displaystyle \frac{3t^2}{4\sigma }}\sigma ^3,{\displaystyle \frac{t^3}{8\sigma ^3}}+{\displaystyle \frac{3t\sigma }{2}},\sigma ^2{\displaystyle \frac{t^2}{4\sigma ^2}},t].`$
For each value of the time $`X_4=t`$ we generate a plot of the string. These plots are combined to form the animation. The ranges chosen were $`\sigma =[1,1]`$, $`t=[1,1]`$. The open string is symmetric in our case but we are careful not to include characteristics that are due to this symmetry of the solution. We have marked the image of $`\sigma =1`$ and $`\sigma =1`$ in order to keep track of the string endpoints on the Reimann sheet..
There are several characterisitics that we observed:
* As opposed to finite string segments, the majority of sigments are semi-inifinte in length. This is due mainly to the inversion of the time parameter. We could have chosen $`X^2`$ to be the time component. Then for the small sector corresponding to $`p=1`$, $`t`$ would be $`\tau `$. However all other sectors would exhibit the presence of semi-infinite segments. These segments may be thought of as flux tubes that carry the monopole and anti-monopole gauge fields from a point out to infinity. One can construct gauge invariant objects that can live on these topologies (see Eq.).
* The strings interact by exchanging flux lines. Whenever two semi-infinite segments, say A and B, touch the result is that A will pass its segment above the intersection point onto B while sewing the segment that B had above the intersection point onto its lower portion. This exchange of flux lines can also take place with segments that are finite.
* For the case when $`p`$ is even and $`q=2`$ then the image is that of a single semi-infinite segment that is actually the image of the string projected back onto itself. In this case if one had assigned a quark at $`\sigma =1`$ and an anti-quarks at $`\sigma =1`$ then the endpoint of this type of image would correspond to a $`\overline{q}q`$ bound state with gauge field lines moving up and moving down the flux line that goes out to spatial infinity effectively canceling the contribution of the gauge field in the Wilson line (see Eq.). Such configurations are expected to contribute to the $`<\overline{q}q>`$ condensate. Examples of this are the (2,2), (4,2), (6,2) and (8,2).
* One of the most interesting observations is the appearance and disappearance of finite segments of lines. These segments can appear at one point, interact with some segment by exchanging flux lines and then disappear at another point. The emerging segment can have a $`\overline{q}q`$ pair attached to the ends providing topology for pair productions and pair annihilation. We will discuss this process in some detail shortly. Nice examples are (4,3), (5,3) and (1,3) torus knots.
* Surfaces that do not intersect $`p=q`$, appear as infinite or semi-infinte segments. The multiple images of the endpoints can be seen embedded in the string segment itself undergoing pair production and annihilation along the flux tube itself. This activity can also be seen in torus knots were $`p`$ is not relatively primed to $`q`$. Examples are (6,3), (4,2), and (8,4).
* Finally we note that the image of $`\sigma =\pm 1`$ does not always map into any endpoint of the image.
Figure 3 shows the (3,2) torus knot instanton at time $`t=0.60`$. The ends of the string are labeled with points, and the string is colored blue to represent $`\sigma >0`$ and red to represent $`\sigma <0`$ in order to visually distinguish these parts. The bounding box containing the plot has dimensions $`x=\mathrm{3..3},y=\mathrm{3..3},z=\mathrm{3..3}`$. In this frame we see the string broken in two halves; where part of the string around $`\sigma =0`$ leaves the bounding box and is mapped out to infinity.
Here is an example of a knot with negative intersection number. The vertical strands are semi-infinite in length. Here the broken strands are an imaging artifact. Just right of the center one can see the emergence of a finite segment. This segment will go on to exchange flux lines with the semi-infinite flux lines and then collapse back into the vacuum. These finite strands provide support for pair production and annihilation.
Figure 5 shows a sequence of frames of the (4,3) torus knot instanton. Initially, we see two string segments; at time $`t1.0`$, however, a pair of points appear. As time continues to evolve these points develop a string segment connecting the two. This finite segment interacts with the pair of strings initially seen, then the two original segments combine and disappear. This behavior is reminiscent of the early meson string model, in which a meson was seen as a $`q\overline{q}`$ pair connected by a string of gluon flux. In this interpretation, it appears as though a $`q\overline{q}`$ pair is created out of the vacuum, interacts strongly with a pair of gluon flux lines, then is annihilated back to the vacuum in the presence of a monopole/anti-monopole pair. Here it is the original string segment that we interpret as monopole-anti-monopole pairs. We emphasize that these are open string solutions that provide this topology as opposed to closed stings. The new requirement is that the opens strings be self intersecting.
The structure of the images suggests that the quarks interact with each other through non-local interactions. We would like for the strings to correspond to flux tubes of gauge fields and to have quarks on the endpoints. Lets fix the gauge, $`A_0=0`$ and consider the Wilson line integral given by the following path-ordered exponential,
$$U(\overline{X}(\sigma _+,t),\overline{X}(\sigma _{},t))=𝒫\mathrm{exp}(_\sigma _{}^{\sigma _+}A_i\frac{dX^i}{d\sigma }𝑑\sigma ).$$
(9)
Here $`\overline{X}(\sigma _+)`$ and $`\overline{X}(\sigma _{})`$ are the end points of the string as seen from a fixed frame and $`\sigma `$ is used to parameterize the string for each time, $`t`$. From the image of the string immersions it appears that non-local point particle-like interaction Lagrangians such
$$L=\overline{q}(\overline{X}(\sigma _+,t))U(\overline{X}(\sigma _+,t),\overline{X}(\sigma _{},t))q(\overline{X}(\sigma _{},t)).$$
(10)
can be supported. In order to account for the semi-infinite segments we can imagine taking one of the endpoints of the Wilson line out to spatial infinity. At spatial infinity the gauge fields approach zero and the Wilson line goes to the identity. The semi-infinite segment then can support a quark attached to the end of the segment with a monopole or anti-monopole gauge field confined to the segment. One can build an operator that respect the residual gauge invariance with
$$\overline{q}(\overline{X}(\sigma _+,t))U(\overline{X}(\sigma _+,t),\mathrm{}),$$
(11)
or either
$$U(\mathrm{},\overline{X}(\sigma _{},t))q(\overline{X}(\sigma _{},t)).$$
(12)
The lines that are infinite in extent correspond to both Wilson line endpoints being taken to spatial infinity. These configurations seem to be the catalyst for pair production and annihilation. There might be suppression to these infinite and semi-infinite segments if one uses the same “area law” arguments one uses to suppress largely separated quark paths in pair production if the flux tubes are too far from each other. However very close pairs $`\overline{q}U`$ and $`Uq`$ may be significant and could contribute to $`<\overline{q}q>`$. To sum up the overall picture that these string suggests in that the general vacuum structure will consist of infinite and semi-infinite flux tubes as well as emerging and disappearing finite tubes and stable finite strings. The gauge fields will be confined to the surfaces of these segments suggesting that monopole and anti-monopole are pervasive and where pair production seems to be catalyzed by the presence of these monopoles/anti-monopoles. The presence of the flux tubes pouring their gauge flux out at spatial infinity is already suggested by the electric Meissner effect . The exchange of flux tubes between pair-produced segments and the other segments suggest the charge-exchange scattering processes between fermions and charged Dirac magnetic monopoles .
## III $`\overline{q}q`$ Production
The topology that supports the pair production and annihilation is not due directly to string self-intersection but is related to how the time parameter $`X^4`$ takes its values from the $`\sigma \tau `$ parameter space. This is related to the fact that the solution is self-dual, however. Let us consider the multi-valued relationship of $`\tau `$ with $`\sigma `$ for a given time, $`X_4`$. Since the $`X_4`$ coordinate only depends on $`q`$, we can examine a time sequence of the relationship of $`\sigma `$ vs. $`\tau `$. To be explicit let us consider any solution where $`q=3`$.
Here one sees that at very early times ($`t<<1`$), $`\tau `$ is a single valued function for values of $`\sigma `$ between $`1`$ and $`1`$. As time moves on, this function becomes multi-valued (here at $`t=2`$) for values where $`\sigma `$ takes values from -1 to 1. Now the physical $`\sigma \tau `$ parameter space is split into three distinct regions corresponding to three distinct images in the frame. As time moves on these region move toward the origin until at $`t=0`$ all three regions merge for an “instant” and the regions begin to recede. At some later time (here $`t=2`$) the $`\tau `$ parameter is again single valued with respect to the physically relevant values for $`\sigma `$. This picture highlights two issues about the self-dual solution that may be realized in the strong coupling regime of QCD. The first is how the multivalued nature of $`\tau `$ vs. $`\sigma `$ can cause pair production and annihilation and the second being the singularity of the map at $`t=0`$, the point of self-intersection, leads to regions where chiral symmetry can be broken. These singular points are where the string segments are allowed to exhange flux lines and also where the determinant of the induced metric vanishes. The details of this last remark will be explored in the next section. For now we will take a closer look at the issues related to production and annihilation.
From figure , one sees that from the time of the production ($`t=2`$) until the time of annihilation ($`t=2)`$, four units of time have transpired. Here we can assume that the constant $`\mu `$ from the action in Eq., sets the space and time scale since $`\mu `$ has dimensions of $`[L^2]`$. From the uncertainty principle we can estimate a typical value of $`\frac{1}{\sqrt{\mu }}`$ since four quarks are produced in a time $`4\frac{1}{\sqrt{\mu }}`$. This would imply that $`\mu (16M_q)^2`$ where $`M_q`$ is the quark mass. The $`q=3`$ set of torus knots correspond to the minimal pair-producing configuration. However higher $`q`$ instantons could easily surpass the bounds from the uncertainty principle by producing more quarks in nearly equal times to that of the $`q=3`$ solutions. For this reason we need to check the production capabilities of the $`q>3`$ solutions.
In order to determine the time it takes for higher $`q`$ configurations to produce and annihilate quarks we need only to ask at what time, $`t_{real}`$ are all the roots of $`\tau (\sigma =1,t_{real})`$ are real. For every value of $`q`$ there are $`q`$ regions that must merge into the $`(1<\sigma <1)`$ region and then move out again. The total production/annihilation time would then be $`2\times t_{real}`$. As an example lets examine the $`q=4`$ case.
In this case the pair production begins at $`t=\frac{8}{3\sqrt{3}}`$ where $`\tau `$ takes values of $`\frac{2}{\sqrt{3}}`$, $`\frac{1}{\sqrt{3}}`$ and $`\frac{1}{\sqrt{3}}`$. The time for total pair production is then $`\frac{16}{3\sqrt{3}}`$ time units or about three units of time. In this amount of time again four quarks or anti-quarks are produced and annihilated. As $`q`$ increases it empirically appears that the amount of time to undergo full pair production monotonically decreases but reaches a plateau. One can show that the total production times for different values of $`q`$ are:
$`q=3,`$ $`t_{real}=2`$ (13)
$`q=4,`$ $`t_{real}={\displaystyle \frac{8}{3\sqrt{3}}}`$ (14)
$`q=5,`$ $`t_{real}=1.37`$ (15)
$`q=6,`$ $`t_{real}=1.285`$ (16)
$`q=10,`$ $`t_{real}=1.147`$ (17)
$`q=100,`$ $`t_{real}=1.012`$ (18)
$`q=111,`$ $`t_{real}=.608`$ (19)
We observe that $`q=111`$ puts an upper limit on the value of $`q`$ that will produce any new pairs. For all values greater that $`q=111`$ some of the roots for $`\tau `$ remain complex even when $`t=0`$. By analogy with the $`q=3,4`$ and $`5`$ cases we see that there can be at most 112 quarks/anti-quarks produced in $`1.2`$ seconds by any instanton. Since the total number of quarks is limited by all torus knot instantons we can again use the uncertainty principle to get an even better estimate of $`\mu `$. We find that $`\mu (112M_q)^2`$.
Animations of several $`(p,q)`$ solutions with $`X^4`$ chosen as the time parameter can be found at http://www-hep.physics.uiowa.edu/~bacus/research.html.
## IV Chiral Symmetry Breaking Processes
As we discussed earlier, the self intersection of the world sheet can cause singular points to exist and lead to zeros in the determinant of the induced metric $`g_{ab}=_aX^\nu _bX_\nu `$. Self duality for the string instantons imply that $`t^{\mu \nu }`$ from Eq.(4) satisfy the constraint
$$t^{\mu \nu }=\frac{1}{2}ϵ^{\mu \nu \lambda \rho }t_{\lambda \rho }+C^{\mu \nu },$$
(20)
where $`C^{1,2}=C^{2,1}=C^{3,4}=C^{4,3}=1`$ and all other entries are zero.
As we are interested in chiral symmetry, we would like to look at the pullback of $`\gamma _5`$ onto the string world sheet. In terms of functions on the world sheet we may write
$$\stackrel{~}{\gamma }_5=\frac{ϵ^{cd}}{4\sqrt{g}}\widehat{ϵ}^{EF}_cX^\mu _dX^\nu N_E^\lambda N_F^\rho \frac{1}{4}[\gamma _\mu ,\gamma _\nu ][\gamma _\lambda ,\gamma _\rho ].$$
(21)
In the above $`N_A^\mu `$ are elements of the normal bundle of the string world sheet and $`\widehat{ϵ}^{AB}`$ is the volume form associated with the normal bundle. By using self duality we can eliminate the tangent vectors from the above expression and write,
$$\stackrel{~}{\gamma }_5=\frac{1}{4}\widehat{ϵ}^{EF}C^{\mu \nu }N_E^\lambda N_F^\rho ϵ_{\mu \nu \lambda \rho }\gamma _5.$$
(22)
The normal vectors satisfy the conditions that $`_aX^\mu N_E^\mu =0,`$ and $`N_E^\mu N^{\mu F}=\delta _E^F,`$. These conditions as well as the self-duality conditions imply that a solution for the tangent and normal vectors is
$`_\sigma X^\mu `$ $`=(_\sigma X^1,_\sigma X^2,_\sigma X^3,_\sigma X^4)`$ (23)
$`_\tau X^\mu `$ $`=(_\sigma X^2,_\sigma X^1,_\sigma X^4,_\sigma X^3)`$ (24)
$`N_1^\mu `$ $`={\displaystyle \frac{1}{\sqrt{g}}}(_\sigma X^3,_\sigma X^4,_\sigma X^1,_\sigma X^2)`$ (25)
$`N_2^\mu `$ $`={\displaystyle \frac{1}{\sqrt{g}}}(_\sigma X^4,_\sigma X^3,_\sigma X^2,_\sigma X^1),`$ (26)
where $`g`$ is the determinant of the induced metric. For these solutions $`\sqrt{g}=g_{\sigma \sigma }`$. Without loss of generality we can take the (3,2) torus knot as an example. Here the torus knot determinant is
$`\sqrt{g}=9(\tau ^4+\sigma ^4)12\tau ^2\sigma ^2+4\sigma ^2+4\tau ^2.`$
Clearly this vanishes at $`\sigma =0,\tau =0`$. This corresponds to the point where the string self-intersects. At this point the normal bundle is ill-defined and two copies of the tangent vectors are imaged at this point. The pullback of $`\gamma _5`$ is also ill-defined when the determinant vanishes. Elsewhere on the manifold $`\stackrel{~}{\gamma }_5`$ can be used to construct a projection operator for chiral symmetry. However these points where the determinant vanish act as tiny “bubbles” where chiral symmetry is broken. The transition of the normal bundle going through this singularity has the effect of imposing reflections on some but not necessarily all of the components. In the (3,2) case the relationship of the normal vectors just before and just after is given by:
$$N_{1}^{\mu }{}_{before}{}^{}=\frac{1}{\sqrt{g}}(_\sigma X^3,_\sigma X^4,_\sigma X^1,_\sigma X^2)$$
(27)
while
$$N_{1}^{\mu }{}_{after}{}^{}=\frac{1}{\sqrt{g}}(_\sigma X^3,_\sigma X^4,_\sigma X^1,_\sigma X^2)$$
(28)
flipping the second and fourth components. Similar changes happen in both the tangent vectors and the other normal. To show that the orientation has changed consider the vector $`\overline{M}=\overline{N}_1\times \overline{N}_2`$ for the (3,2) knot. Just after the singularity, $`M_yM_y`$ while all other components remain the same. This corresponds to a change in the orientation. This pictures corresponds to the field theoretic case where the core of monopoles can be thought of as bubbles where chiral symmetry is broken. The S-wave of the fermions interact with this core which leads to processes such as those seen in .
## V The Self-Intersection Number
It is conjectured in that the self-intersection number of the $`(p,q)`$ torus is $`\nu =4(pq)`$. Here we explicitly calculate the self-intersection number $`\nu `$ for a general (p,q) torus-knot solution proving the conjecture. Consider an open string where $`(L\sigma L)`$ and $`(\mathrm{}<\tau <\mathrm{})`$. Starting from the general definition of the self-intersection number
$`\nu ={\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\tau {\displaystyle _L^L}𝑑\sigma _at^{\mu \nu }_at^{\mu \nu }`$
which we can express in terms of the string solution $`X(\sigma ,\tau )`$ as
$`\nu ={\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\tau {\displaystyle _L^L}𝑑\sigma _a({\displaystyle \frac{ϵ^{cd}}{\sqrt{g}}}_cX^\mu _dX^\nu )_a({\displaystyle \frac{ϵ^{cd}}{\sqrt{g}}}_cX^\mu _dX^\nu )`$
Using the general (p,q) torus-knot solution
$`X(\sigma ,\tau )=[(\sigma ^2+\tau ^2)^{p/2}\mathrm{sin}(p\mathrm{tan}^1({\displaystyle \frac{\sigma }{\tau }})),(\sigma ^2+\tau ^2)^{p/2}\mathrm{cos}(p\mathrm{tan}^1({\displaystyle \frac{\sigma }{\tau }})),`$
$`(\sigma ^2+\tau ^2)^{q/2}\mathrm{cos}(q\mathrm{tan}^1({\displaystyle \frac{\sigma }{\tau }})),(\sigma ^2+\tau ^2)^{q/2}\mathrm{sin}(q\mathrm{tan}^1({\displaystyle \frac{\sigma }{\tau }}))]`$
we find
$`\nu ={\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\tau {\displaystyle _L^L}𝑑\sigma {\displaystyle \frac{8p^2q^2(pq)^2(\sigma ^2+\tau ^2)^{p+q1}}{(p^2(\sigma ^2+\tau ^2)^p+q^2(\sigma ^2+\tau ^2)^q)^2}}`$
If we let $`r^2\sigma ^2+\tau ^2`$,
$`\nu `$ $`=`$ $`{\displaystyle \frac{1}{\pi }}{\displaystyle _{\pi /2}^{\pi /2}}𝑑\theta {\displaystyle _0^{\frac{L}{\mathrm{cos}(\theta )}}}𝑑r{\displaystyle \frac{8p^2q^2(pq)^2r^{2(p+q1)+1}}{(p^2r^{2p}+q^2r^{2q})^2}}`$
$`=`$ $`{\displaystyle \frac{4((pq)qp)^2}{\pi }}{\displaystyle _{\pi /2}^{\pi /2}}d\theta {\displaystyle \frac{1}{(pq)q^2(q^2r^{2(pq)}+p^2)}}|_0^{\frac{L}{\mathrm{cos}(\theta )}}`$
Let $`\varphi 2\theta +\pi `$,
$`\nu ={\displaystyle \frac{2(pq)}{\pi }}{\displaystyle _0^{2\pi }}{\displaystyle \frac{1}{(\frac{p}{q}L^{(pq)})^2(\frac{1\mathrm{cos}(\varphi )}{2})^{(pq)}+1}}𝑑\varphi 2(pq)`$
Note the following relation;
$`{\displaystyle _0^{2\pi }}f(\mathrm{sin}\theta ,\mathrm{cos}\theta )𝑑\theta =i{\displaystyle _{\frac{unit}{circle}}}f({\displaystyle \frac{zz^1}{2i}},{\displaystyle \frac{z+z^1}{2}}){\displaystyle \frac{dz}{z}}`$
Then
$`\nu `$ $`=`$ $`i{\displaystyle \frac{2(pq)}{\pi }}{\displaystyle _{\frac{unit}{circle}}}{\displaystyle \frac{dz}{z((\frac{p}{q}L^{(pq)})^2(\frac{1}{2}(1\frac{z+z^1}{2}))^{(pq)}+1)}}2(pq)`$
$`=`$ $`4(qp){\displaystyle \frac{4(qp)}{\pi }}\pi {\displaystyle \text{Residues in unit circle}}`$
$`=`$ $`4(qp)(1{\displaystyle \text{Residues in unit circle}})`$
Now we must find the singular points of $`f(z)=\frac{1}{z((\frac{q}{p}L^{pq})^2(\frac{1}{2}(1\frac{z+z^1}{2})^{pq}+1)}`$; they are
$`z`$ $`=`$ $`0`$
$`z`$ $`=`$ $`12(({\displaystyle \frac{q}{p}})^2(L)^{2(pq)})^{\frac{1}{qp}}\pm 2\sqrt{(({\displaystyle \frac{q}{p}})^2(L)^{2(pq)})^{\frac{1}{qp}}+({\displaystyle \frac{q}{p}}(L)^{(pq)})^{\frac{4}{qp}}}`$
Evaluating the residue $`(zz_0)f(z)|_{z=z_0}`$ gives zero; then we have
$`\nu =4(qp)`$
for (p,q) torus-knot solutions. This confirms the conjecture of Robertson .
## VI Computer Program
The software described in this section was used to implement and generate the animated solutions and to perform some lengthy calculations. Two symbolic computation systems were used; Maple and Mathematica.
Listing 1 shows the Maple source code which generates raw ($`x,y,z`$) coordinates for a (p,q) torus knot solution and exports them to a file. The user selects the values for $`p`$ and $`q`$, the frame numbers to start and end at, the number of data points to use for each frame, and the ranges for the parameters $`t`$ and $`\sigma `$.
### Listing 1
#----------------------------------------------------------------------#
\# Animation of the string instanton solutions. This program generates a#
\# .dat numeric data file for each frame of the animation sequence. #
\# Usage (on UNIX): maple 4D\_TorusKnot\_Data.txt & #
\# #
\# Bob Bacus c.1997 #
#------------------------- User input parameters ----------------------#
p:=3: # F=zp̂ #
q:=2: # G=-zq̂ #
startframe:=0: # Frame to start from #
endframe:=400: # Frame to end on #
points:=400: # Number of data points per frame #
lmin:=-3:lmax:=3: # Range for the time parameter #
smin:=-1:smax:=1: # Range for the length parameter #
\# ---------------------------- Main Program ---------------------------#
gc(300000):interface(screenwidth=500):Digits:=4:
readlib(unassign):readlib(write):
unassign(’t’):unassign(’l’):unassign(’n’):unassign(’u’):unassign(’s’):
assume(t,real):assume(s,real):
z:=t+I\*s:
F:=expand(zp̂):G:=expand(-zq̂):
imf:=Im(F):ref:=Re(F):reg:=Re(G):img:=Im(G): # X=\[Im(F),Re(F),Re(G),Im(G)\]
l:=(lmax-lmin)/(endframe-startframe)\*n+lmin: # n represents frame number; l is the time
v:=(smax-smin)/(points-1):
for n from startframe by 1 to endframe do
open(cat(convert(p,string),‘x‘,convert(q,string),‘-‘,convert(n,string),‘.dat‘)):
writeln(evalf(l)):
writeln(points):
for i from 1 to q-1 do
writeln(cat(‘Solution #‘,convert(i,string))):
for u from smin by v to smax do
r:=\[fsolve(subs(s=u,img)=l,t,complex)\]:r:=r\[i\]:
X:=expand(subs(t=r,imf)):Y:=expand(subs(t=r,ref)):Z:=expand(subs(t=r,reg)):
writeln(evalf(subs(s=u,\[X,Y,Z\]))):
od;
od;
close():
od;
unassign(’t’):unassign(’l’):unassign(’n’):unassign(’u’):
close():
### Listing 2
#----------------------------------------------------------------------
\# Animation of the string instanton solutions. This program generates a
\# .gif graphics file for each frame of the animation sequence.
\#
\# Bob Bacus c.1997
#------------------------- User input parameters ----------------------
p:=1: # F= zp̂
q:=2: # G=-zq̂
startframe:=0: # frame to start with
endframe:=400: # frame to end with
#-----------------------------------------------------------------------
space:=‘ ‘:gc(300000):interface(plotdevice=gif):
readlib(write):readlib(unassign):
for n from startframe by 1 to endframe do
d:=\[\]:
file:=cat(convert(p,string),‘x‘,convert(q,string),‘-‘,
convert(n,string),‘.dat‘):
indx:=parse(readline(file)):
if indx\>=0 then tp:=cat(‘p=‘,convert(p,string), ‘ q=‘,convert(q,string),‘ Time Index : ‘,substring(convert(indx,string),1..7),substring(space,1..7-length(substring(convert(
indx,string),1..7)))) else tp:=cat(‘p=‘,convert(p,string),‘ q=‘,convert(q,string),‘
Time Index:‘,substring(convert(indx,string),1..8),substring(space,1..8-length(substring(
convert(indx,string),1..8)))) fi:
points:=parse(readline(file)):
for i from 1 to q-1 do
readline(file):
j:=1:
for s from 1 to points/2 do
pt:=parse(readline(file)):
if s=1 then if linalg\[iszero\](map(Im,pt)) then
smin:=pt :b(j):=\[pt\] else b(j):=\[\] : smin:=\[2.999,2.999,2.999\] fi fi:
if s\>1 then if linalg\[iszero\](map(Im,pt)) then b(j):=\[op(b(j)),pt\] else j:=j+1 :b (j):=\[\]: fi fi:
od;
k:=1:
c(1):=\[\]:
for s from points/2+1 to points do
pt:=parse(readline(file)):
if s\<points then if linalg\[iszero\](map(Im,pt)) then c(k):=\[op(c(k)),pt\]
else k:=k+1: c(k):=\[\]: fi fi:
if s=points then if linalg\[iszero\](map(Im,pt)) then smax:=pt: c(k):=\[op(c(k)),pt\]
else smax:=\[2.999,2.999,2.999\] fi fi:
od;
L1:=\[\]:
L2:=\[\]:
for h from 1 to j do
if nops(b(h))\>0 then L1:=\[op(L1),b(h)\] fi:
od:
for h from 1 to k do
if nops(c(h))\>0 then L2:=\[op(L2),c(h)\] fi:
od:
if nops(L1)\>0 then L1:=op(1,L1): else L1:=\[\]: fi:
if nops(L2)\>0 then L2:=op(1,L2): else L1:=\[\]: fi:
d:=\[op(d),PLOT3D(POINTS(\[2.999,2.999,2.999\]),COLOR(RGB,0,0,0)),
PLOT3D(POINTS(\[smin,smax\]),COLOR(RGB,0,0,0)),PLOT3D(CURVES(L1),COLOR(RGB,1,0,0)),PLOT3D(CURVES(L2),
COLOR(RG B,0,0,1))\]:
od:
interface(plotoutput=cat(convert(p,string),‘x‘,convert(q,string),
‘-‘,convert(n,string ),‘.gif‘)):
plots\[display\](d,view=\[-3..3,-3..3,-3..3\],projection=.1,orientation=,axes=boxe d,labels=\[’x’,’y’,’z’\],scaling=constrained,titlefont=\[COURIER,10\],title=tp);
test:=readline(file):
unassign(’l’):unassign(’b’):unassign(’points’):close():
od:
## VII Acknowledgements
V.G.J. Rodgers thanks V.P. Nair for discussion.
## REFERENCES
1. G. t’Hooft, Nucl. Phys. B72 (1973) 461
2. K. G. Wilson, Phys. Rev. D 10 (1974) 2445
3. J.-L. Gervais and A. Neveu, Phys. Lett. B80 (1979) 255
4. Yu M. Makeenko and A.A. Migdal, Phys. Lett. B88 (1979) 135
5. A. A. Belavin, A. M. Polyakov, A. Schwartz, Y. Tyupkin, Phys. Lett. 59B (1975) 85
6. E. Witten, Phys. Rev. Lett. 38 (1977) 121
7. A. M. Polyakov, Nucl. Phys. B268 (1986) 406
8. A.P. Balachandran, F. Lizzi, G. Sparano, Nucl. Phys. B263 (1986) 608
9. P.O. Mazur and V.P. Nair, Nucl. Phys. B284 (1986) 146
10. J. Ambjorn and B. Durhuus, Phys. Lett. B188 (1987) 253
11. R.Parthasarathy and K.S.Viswanathan, Lett.Math.Phys. 48 (1999) 243
12. J. F. Wheater, Phys. Lett. B208 (1988) 388
13. G. Robertson, Phys. Lett. B 226 (1989) 244
14. B.G. Konopelchenko and G. Landolfi, Phys.Lett. B459 (1999) 522
15. J.Pawelcyzk, Phy. Rev. Lett 74 (1995) 3924; Phys. Lett. B387 (1996) 287; Nucl. Phys. B491, (1997) 515
16. V.P. Nair and C. Rosenzweig, Phys. Lett. B135 (1984) 450; Phy. Rev. D31 (1985) 401
17. Allan S. Blaer, Norman H. Christ, and Ju-Fei Tang, Phy. Rev. D25 (1982) 25
18. C.G. Callan, Jr. Phys. Rev. D25 (1982) 2141 |
warning/0003/astro-ph0003055.html | ar5iv | text | # The High-Ionization Nuclear Emission-Line Region of Seyfert Galaxies
## 1. INTRODUCTION
Seyfert galaxies have been broadly classified into two classes based on the presence or absence of broad emission lines in their optical spectra (Khachikian & Weedman 1974): Seyfert galaxies with broad lines are type 1 (hereafter S1) while those without broad lines are type 2 (S2). According to the current unified model of Seyfert nuclei (Antonucci & Miller 1985; see for a review Antonucci 1993), this difference between S1 and S2 can be explained as follows. The broad-line region (BLR) is located in the very inner region (e.g., a typical radial distance from the central black hole is $`r`$ 0.01 pc; e.g., Peterson 1993) and is surrounded by a geometrically and optically thick dusty torus. Therefore, the visibility of the central engine as well as the BLR is strongly affected by the viewing angle toward the dusty torus and then the difference between S1s and S2s is naturally understood. Indeed, this unified scheme has been supported by various observational results, for example, obscured X-ray emission in S2s (Awaki et al. 1991; Rush et al. 1996), colors of mid-infrared (MIR) emission (Pier & Krolik 1992, 1993; Murayama, Mouri, & Taniguchi 2000), MIR luminosity distributions (Heckman, Chambers, & Postman 1992; Maiolino et al. 1995), polarized broad emission lines mentioned below, and the results of multi wavelength observational tests (Mulchaey et al. 1994). In order to understand Seyfert nuclei and active galactic nuclei (AGNs) more comprehensively, any new observational tests toward the unified model are very important.
In addition to the traditional two types of Seyfert nuclei, it is known that some Seyfert nuclei show intermediate properties between S1 and S2; type 1.2 (S1.2), type 1.5 (S1.5), type 1.8 (S1.8), and type 1.9 (S1.9) (Osterbrock & Koski 1976; Osterbrock 1977, 1981b; Cohen 1983; Winkler 1992; Whittle 1992), which show both the narrow and broad components in the Balmer emission lines. It is also noted that the objects without BLR in their optical spectra (i.e., S2s) do not comprise a simple population. First, some S2s show a broad Pa$`\beta `$ line (Goodrich, Veilleux, & Hill 1994; Hill, Goodrich, & Depoy 1996; Veilleux, Goodrich, & Hill 1997), providing evidence for highly reddened BLRs in these objects. Second, the hidden BLR is detected only in a part ($``$ 20%) of S2s in the polarized optical spectra (Antonucci & Miller 1985; Miller & Goodrich 1990; Tran, Miller, & Kay 1992; Kay 1994; Tran 1995a, 1995b, 1995c); the survey promoted by Lick Observatory found 10 S2s with the hidden broad line among 50 S2s.
Another important type of Seyfert nuclei is narrow-line Seyfert 1 galaxies (NLS1s; Davidson & Kinman 1978). Optical emission-line properties of the NLS1s are summarized as follows (e.g., Osterbrock & Pogge 1985). (1) The Balmer lines are only slightly broader than the forbidden lines such as \[O iii\]$`\lambda `$5007 (typically less than 2000 km s<sup>-1</sup>). This property makes NLS1s a distinct type of S1s. (2) The \[O iii\]$`\lambda `$5007/H$`\beta `$ intensity ratio is smaller than 3. This criterion has introduced to discriminate S1s from S2s by Shuder & Osterbrock (1981). (3) They present strong Fe ii emission lines which are often seen in S1s but generally not in S2s. And, (4) the soft X-ray spectra of NLS1s are very steep (Puchnarewicz et al. 1992; Boller, Brandt, & Fink 1996; Wang, Brinkmann, & Bergeron 1996) and highly variable (Boller et al. 1996; Turner et al. 1999a). Because of these complex properties, it has not yet been fully understood what NLS1s are in the context of the current unified model of Seyfert nuclei while various models for NLS1s have been proposed (see for reviews Boller et al. 1996; Taniguchi, Murayama, & Nagao 1999).
Recently, Murayama & Taniguchi (1998a; hereafter MT98a) have found that S1s have excess \[Fe vii\]$`\lambda `$6087 emission with respect to S2s. This means that a significantly large fraction of the high-ionization nuclear emission-line region (HINER; Binette 1985; Murayama, Taniguchi, & Iwasawa 1998) traced by \[Fe vii\]$`\lambda `$6087 resides in a viewing-angle dependent region; i.e., the inner wall of dusty tori (Murayama & Taniguchi 1998b; see also Pier & Voit 1995). Accordingly, it turns out that the HINER provides the indicator of the viewing angle for dusty tori of Seyfert nuclei. In this paper, we report on our statistical analysis of the HINER in the various types of Seyfert nuclei.
## 2. DATA
### 2.1. Classification of Seyfert Nuclei
As mentioned in Section 1, there are a number of sub-types of Seyfert nuclei. Summarizing their definitions and properties, we broadly re-classify all the objects in the following way (see Table 1). 1) The type of S1.2 is included in the type of S1. These Seyferts together with typical S1s are abbreviated as BLS1s (broad-line type 1 Seyferts) because there is another type of S1s; i.e., NLS1s. 2) The type of S1.5 is kept as a distinct type because Seyfert galaxies belonging to this type are more numerous than those of the other intermediate-type Seyferts. 3) The types of S1.8 and S1.9 are basically included into the type of S2. The BLRs in these galaxies are reddened more seriously than those in both BLS1s and S1.5s. In this respect, S2s with the BLR detected only in their infrared spectra (e.g., broad Pa$`\beta `$ emission; hereafter S2<sub>NIR-BLR</sub>) share the same BLR properties. Therefore, we refer these types of Seyferts as type 2 Seyferts with the reddened BLR (S2<sub>RBLR</sub>). 4) Another important type of S2s is S2s with the hidden BLR which is detected in optical polarized spectra<sup>1</sup><sup>1</sup>1 This type is referred either as S3 (Tran 1995a), as S1h (Véron-Cetty & Véron 1998), or as S2<sup>+</sup> (Taniguchi & Anabuki 1999). . In this paper, we refer this type as S2<sub>HBLR</sub>. 5) S2s either with the reddened BLR or with the hidden BLR are also referred as S2<sup>+</sup>; i.e., S2<sup>+</sup> = S2<sub>RBLR</sub> \+ S2<sub>HBLR</sub>. 6) In contrast, S2s without any evidence for the BLR are referred as S2<sup>-</sup> following Taniguchi & Anabuki (1999). And, 7) both types of S2<sup>+</sup> and S2<sup>-</sup> are referred as S2<sub>total</sub> when necessary; i.e., S2<sub>total</sub> = S2<sup>+</sup> \+ S2<sup>-</sup>. Our classification scheme is summarized in Table 1.
In some cases, the classification is assigned differently to a certain Seyfert nucleus among the literature. Therefore, in Table 2, we compare the types of our sample objects with those given in some previous papers (Dahari & De Robertis 1988a; Stephens 1989; Whittle 1992; Cruz-González et al. 1994; Véron-Cetty & Véron 1998). The type adopted in this paper for each galaxy is given in the last column of Table 2. The objects classified as NLS1s in the previous literature (Osterbrock & Pogge 1985; Stephens 1989; Boller et al. 1996; Véron-Cetty & Véron 1998; Vaughan et al. 1999) are categorized as NLS1.
### 2.2. Data
In order to investigate HINER properties of various types of Seyfert galaxies, we are interested in the following high-ionization emission lines; \[Fe vii\]$`\lambda `$6087, \[Fe x\]$`\lambda `$6374<sup>2</sup><sup>2</sup>2It is noted that Osterbrock (1977) misidentified \[Fe x\]$`\lambda `$6374 as Fe ii $`\lambda `$6369 (see Osterbrock 1981a)., and \[Fe xi\]$`\lambda `$7892. In addition to these lines, we are also interested in the following low-ionization emission lines; \[O iii\]$`\lambda `$5007, \[S ii\]$`\lambda \lambda `$6717,6731, and \[O i\]$`\lambda `$6300, because of the comparison with the high-ionization emission lines. These lines are simply referred as \[Fe vii\], \[Fe x\], \[Fe xi\], \[O iii\], \[O i\], and \[S ii\], respectively. Here we should mention that some fraction of the \[O iii\] emission arises from the inner wall of dusty tori (Pier & Voit 1995; Murayama & Taniguchi 1998b). Therefore, it seems better to use more low-ionization emission lines such as \[O i\] or \[S ii\] as a normalization emission line. This is the reason why we have compiled the data of not only \[O iii\] but also \[O i\] and \[S ii\]. Though \[N ii\]$`\lambda `$6583 is also one of important low-ionization emission lines, we do not use this line because the deblending \[N ii\] from H$`\alpha `$ may not be well done if the spectral resolution is not so high. We have compiled the emission line data from the literature (Table 3) which are spectroscopic studies at wavelengths covering the emission lines of our interests. The number of compiled objects is 227; i.e., 31 NLS1s, 58 S1s, 67 S1.5s, 31 S2<sup>+</sup>s, and 40 S2<sup>-</sup>s.
The detection rates of \[Fe vii\], \[Fe x\], and \[Fe xi\] in the sample are given in Table 4. The fraction of objects with at least one of these high-ionization lines detected are also given. This table shows that the detection rate for the S2<sup>+</sup> (83.9 %) is higher than those for the other types (38.7 % $``$ 64.2 %). Although the reason for this fact is not clear, it may be partly because the average redshift of the S2<sup>+</sup>s is smaller than those of the other types (see section 2.3.1) and accordingly those objects might be observed with higher S/N. Here we mention that we do not use any upper limit data in our study.
We choose the objects which show \[Fe vii\] and/or \[Fe x\] from the sample, and consequently, the object number of our sample is 124 including 9 radio-loud galaxies<sup>3</sup><sup>3</sup>3In this paper, we define the radio-loud galaxy as the one which satisfies the criterion of R $`>`$ 500, where R is the ratio of radio ($`\lambda `$ = 6 cm) to optical (B band) flux density. Here the R is defined as follows; the optical flux density $`S_{\mathrm{opt}}`$ at B-band are calculated from the relation B = –2.5 $`\mathrm{log}S_{\mathrm{opt}}`$ – 48.36 (Schmidt & Green 1983), and R is derived from dividing the radio flux density at the wavelength of 6 cm by this $`S_{\mathrm{opt}}`$. ; i.e., 12 NLS1s, 23 BLS1s, 43 S1.5s (including 3 radio-loud galaxies), 27 S2<sup>+</sup>s (including 2 radio-loud galaxies), and 19 S2<sup>-</sup>s (including 4 radio-loud galaxies). Although the sample is not a statistically complete one in any sense, the data set is the largest one for the study of HINER ever compiled.
In Table 5, the redshift, the apparent B magnitude, the absolute B magnitude<sup>4</sup><sup>4</sup>4In this paper, we adopt a Hubble constant H<sub>0</sub> = 50 km s<sup>-1</sup> Mpc<sup>-1</sup> and a deceleration parameter q<sub>0</sub> = 0., the radio flux density at the wavelength of 6 cm, the ratio of radio to optical flux density, the 60 $`\mu `$m luminosity, the \[O iii\] luminosity, and the references for the \[O iii\] luminosity and the emission-line flux ratios are given for each galaxy. Those magnitudes are taken from Véron-Cetty & Véron (1998), who mentioned that they had chosen the magnitudes in the smallest possible diaphragm as they were interested in the nuclei rather than in the galaxy itself. Table 3 describes the references for Table 5.
The emission-line flux ratios for each object are given in Table 6. Each ratio is the averaged value among the references. Since it is often difficult to measure the narrow Balmer component for S1s accurately, there might be the systematic error if we make reddening corrections using the Balmer decrement method (e.g., Osterbrock 1989) for all the types of Seyferts. Therefore we do not make the reddening correction for the objects in our sample. The effect of dust extinction on our result is discussed in section 3.4.
### 2.3. Selection Bias
Because we do not impose any selection criteria upon our sample, it is necessary to check whether or not the various samples are appropriate for our comparative study. Systematic difference of the redshift distribution, the intrinsic AGN power distribution, and the excitation degree of the narrow-line region (NLR) gas among the various Seyfert types may cause possible biases, thus we investigate these distributions.
#### 2.3.1 Redshift
The average redshifts and 1$`\sigma `$ deviations for each type are 0.0351$`\pm `$0.0315 for the NLS1s, 0.0550$`\pm `$0.0450 for the BLS1s, 0.0378$`\pm `$0.0401 for the S1.5s, 0.0243$`\pm `$0.0315 for the S2<sup>+</sup>s, and 0.0353$`\pm `$0.0309 for the S2<sup>-</sup>s. We show the histograms of the redshift in Figure 1. It is noted that the average redshifts of the S1 and the S1.5 sample are a little higher than those of the other samples. In order to investigate whether or not the frequency distributions of the redshift are statistically different among the types of Seyferts, we apply the Kolmogorov-Smirnov (KS) statistical test (Press et al. 1988). The null hypothesis is that the redshift distributions among the NLS1s, the BLS1s, the S1.5s, the S2<sup>+</sup>s and the S2<sup>-</sup>s come from the same underlying population. The results are summarized in Table 7. We give the KS probabilities for the class of S2<sub>total</sub>, which means S2<sup>+</sup> and S2<sup>-</sup> since the numbers of the S2<sup>+</sup>s and the S2<sup>-</sup>s in our sample are not so large. We give two KS probabilities for each combination; the first line gives the KS probabilities for the case including the radio-loud objects while the second line gives those for the case without the radio-loud objects. The results are nearly the same for these two cases in each combination. The results of the KS test suggest that the redshifts of the S1s are systematically higher than those of the other samples.
In this paper, our main attention is addressed to the visibility of the torus HINER emission among the different Seyfert types. Since the torus HINER is located in the inner 1 pc region around the central engine, the larger average redshift of the S1s may not affect the visibility of the torus HINER. If the S1s could have intense circumnuclear star-forming regions, such emission would contribute to the line emission. However, since it is known that S1s tend to have few such circumnuclear star-forming regions (Pogge 1989; Oliva et al. 1995; Heckman et al. 1995; González Delgado et al. 1997; Hunt et al. 1997), such contamination is expected to be negligibly small. Therefore, we conclude that our later analyses are free from the redshift difference among the samples.
#### 2.3.2 Luminosity
Current unified model of Seyfert galaxies require anisotropical nuclear radiation. This may cause systematical differences of intrinsic AGN power highly depending on selection criteria. Comparison of emission lines among different Seyfert types might suffer from this bias of intrinsic luminosity. Therefore, we investigate whether or not the intrinsic AGN power is systematically different among the different Seyfert types using the luminosities which are regarded as isotropic emission reprocessed from the nuclear radiation. We use IRAS 60$`\mu `$m and low-ionization emission lines as such isotropic emission.
We firstly check the distributions of the 60$`\mu `$m luminosity among the samples. The 60$`\mu `$m luminosity is thought to scale the nuclear continuum radiation which is absorbed and re-radiated by the dusty torus. Therefore the distribution of the 60$`\mu `$m luminosity reflects that of the intrinsic luminosity. The histograms of the 60$`\mu `$m luminosity are shown in Figure 2. There appears to be no systematic difference among the types of Seyferts. We apply the KS test where the null hypothesis is that the distribution of the 60$`\mu `$m luminosity among the various types of Seyferts come from the same underlying population. The results suggest that there is no systematic difference of the 60$`\mu `$m luminosity among the samples (see Table 8). This means that there is no bias concerning to the intrinsic luminosity in our sample.
However, the 60$`\mu `$m luminosity might be contaminated with the influence of circumnuclear star formation. Hence we secondly investigate the luminosity of the low-ionization emission lines. Because most of the flux of the low-ionization emission lines is radiated from the NLRs, it is thought to be almost independent with the viewing angle. Therefore the luminosity of a low-ionization emission line is a good tool to investigate the intrinsic power of the AGN. As shown in Figure 3, the intensity distributions of the low-ionization emission lines appear to be indistinguishable among the samples. We apply the KS test where the null hypothesis is that the luminosity of the low-ionization emission lines among various types of Seyferts come from the same underlying population. The results suggest that there is no systematic difference of the luminosity of the low-ionization emission lines among the samples (see Table 9). Therefore we conclude that there is little difference of the distribution of the intrinsic luminosity among the samples.
#### 2.3.3 Excitation of the NLR gas
There is another problem concerning our comparisons in this paper. In our study, we assume that the excitation degree of the NLRs is similar among the samples when we compare various line ratios. In order to confirm the validity of this assumption, we investigate whether or not the physical property of the NLRs is different among the various types of Seyferts. In Figure 4, we show the diagram of the intensity ratios of \[S ii\]/\[O iii\] versus \[O i\]/\[O iii\]. The diagram shows that there is little difference of the excitation degree of the NLRs among the types of Seyferts (see also Cohen 1983). This guarantees the validity of the statistical comparisons in our study.
## 3. RESULTS
Emission-line ratios of AGNs have been often discussed in the form normalized by the narrow component of Balmer lines; e.g., \[O iii\]/H$`\beta `$, \[N ii\]/H$`\alpha `$, and so on (e.g., Veilleux & Osterbrock 1987). However, since we investigate emission-line properties of S1s together with S2s, we cannot use the usual emission-line ratios in our analysis. Therefore, following the manner of MT98a, we investigate intensity ratios between a HINER line and a low-ionization forbidden emission line which is thought to be independent of the viewing-angle.
### 3.1. The Relative Strength of the \[Fe vii\] Emission
We show the histograms of the line ratios of \[Fe vii\] to \[O iii\], \[S ii\] and \[O i\], for the NLS1s, the BLS1s, the S1.5s, the S2<sup>+</sup>s, and the S2<sup>-</sup>s in Figure 5. Both the NLS1s and the BLS1s tend to have stronger \[Fe vii\] emission than the S2<sup>+</sup>s and the S2<sup>-</sup>s, being consistent with the result of MT98a. It is interesting to note that the S1.5s show a marginal nature between the S1s and the S2s.
In order to investigate whether or not the differences of the emission-line ratios among the samples are statistically real, we apply the KS test. The null hypothesis is that the observed distributions of the intensity ratios of \[Fe vii\] to the low-ionization emission lines among the NLS1s, the BLS1s, the S1.5s, the S2<sup>+</sup>s and the S2<sup>-</sup>s come from the same underlying population. The results are summarized in Table 10. It is noted that there is almost no difference between the KS probabilities in the case of including the radio-loud galaxies and excluding those objects.
The KS test leads to the following results. 1) Both the NLS1s and the BLS1s have higher \[Fe vii\] strengths than the S2<sup>+</sup>s and the S2<sup>-</sup>s although the statistical significance is marginally low for the NLS1s when we use the \[Fe vii\]/\[O i\] ratio. 2) There is no statistical difference in the relative \[Fe vii\] strength between the NLS1s and the BLS1s. 3) There is no statistical difference in the relative \[Fe vii\] strength between the S2<sup>+</sup>s and the S2<sup>-</sup>s. 4) There is no statistical difference in the relative \[Fe vii\] strength between the S1.5s and the S1s (i.e., the NLS1s and the BLS1s) although the statistical significance is marginally low for the BLS1s when we use the \[Fe vii\]/\[O iii\] ratio. And, 5) the S1.5s have higher \[Fe vii\] strengths than the S2<sup>+</sup>s and the S2<sup>-</sup>s. Because the frequency distributions of the luminosity of the low-ionization emission lines are indistinguishable among the samples (see Figure 3), the excess of the line ratios of \[Fe vii\] to the low-ionization emission lines in both the NLS1s and the BLS1s with respect to the S2<sup>+</sup>s and the S2<sup>-</sup>s is thought to be due not to the depression of the emission of the low-ionization emission lines but to the excess of the \[Fe vii\] emission.
It is considered that the inclination effect is responsible for the above results. In order to confirm this, we investigate whether or not the relative strength of the high-ionization emission lines correlate with the redshift or the intrinsic power of each AGN. As shown in Figure 6, there appears to be no correlation between the relative \[Fe vii\] strength and the redshift, the absolute B magnitude, the 60$`\mu `$m luminosity, and the \[O iii\] luminosity. Therefore the various comparisons of the emission-line flux ratios described in this paper are thought to be valid although there is slight difference in the average redshifts of the samples (see Section 2.3.1).
We also investigate the HINER properties of the S1.8s, S1.9s, S2<sub>NIR-BLR</sub> and S2<sub>HBLR</sub>, respectively. The results are shown in Figure 7. There can be seen no systematic trend. Therefore, these four sub-types are indistinguishable in the HINER properties.
### 3.2. The Relative Strength of the \[Fe x\] Emission
We present the histograms of the intensity ratios of \[Fe x\] to low-ionization emission lines in Figure 8. We apply the KS test where the null hypothesis is that the observed distributions of the intensity ratios of \[Fe x\] to the low-ionization emission lines among various types of Seyferts come from the same underlying population. The results are given in Table 11. There is also no difference between the KS probabilities in the case of including the radio-loud galaxies and excluding those objects.
The KS test leads to the following results. 1) Both the NLS1s and the BLS1s have higher \[Fe x\] strengths than the S2<sup>+</sup>s and the S2<sup>-</sup>s. However, the statistical significance is much worse than that using the \[Fe vii\] emission. 2) There is no statistical difference in the relative \[Fe x\] strength between the NLS1s and the BLS1s. 3) There is no statistical difference in the relative \[Fe x\] strength between the S2<sup>+</sup>s and the S2<sup>-</sup>s 4) There is no statistical difference in the relative \[Fe x\] strength between the S1.5s and the S1s (i.e., the NLS1s and the BLS1s). And, 5) it is not clear whether or not there is statistical difference in the relative \[Fe x\] strength between the S1.5s and the S2s (i.e., S2<sup>+</sup>s + S2<sup>-</sup>s). These results are not consistent with those using the \[Fe vii\] emission. This point will be discussed in next section.
In Figure 9, we show the diagrams of the redshift, the absolute B magnitude, the 60$`\mu `$m luminosity, and the \[O iii\] luminosity versus the relative \[Fe x\] strength. Similar to the case mentioned in the previous section, there is no correlation between the relative \[Fe x\] strength and the redshift, the absolute B magnitude, the 60$`\mu `$m luminosity, and the \[O iii\] luminosity. This result also assures the validity of our comparative study.
We also investigate the HINER properties of the S1.8s, S1.9s, S2<sub>NIR-BLR</sub> and S2<sub>HBLR</sub>, respectively. The result is shown in Figure 10. Again, there can be seen no systematic trend in this figure. Therefore, these four subclasses are indistinguishable in the HINER properties.
### 3.3. \[Fe vii\] versus \[Fe x\]
We investigate whether or not the \[Fe x\]/\[Fe vii\] ratio is different among the samples. The frequency distributions of this ratio are shown in Figure 11. We apply the KS test where the null hypothesis is that the observed distributions of the \[Fe x\]/\[Fe vii\] ratio among the various types of Seyferts come from the same underlying population. The results are given in Table 12. Although there seem to be a marginal tendency that the NLS1s have higher \[Fe x\]/\[Fe vii\] ratios than the other types of Seyferts, the KS test shows that this is not statistically real.
### 3.4. Effects of the Dust Extinction
As mentioned in section 2.2, no reddening correction has been made for all the observed emission line ratios analyzed here. However, it is known that the dust extinction is larger on average in S2s than in S1s (Dahari & De Robertis 1988a, 1988b). In order to see how the extinction affects the line ratios, we summarize the shifts of the line ratios for the following three cases; $`A_V`$ = 1.0, 5.2, and 10.0 (see Table 13). The case of $`A_V`$ = 5.2 corresponds to that of the Circinus galaxy (Oliva et al. 1994). In these estimates, we use the Cardelli’s extinction curve (Cardelli, Clayton, & Mathis 1989).
Since the wavelength of \[O i\] is relatively close to those of \[Fe vii\] and \[Fe x\], the effect of dust extinction is negligibly small even for the case of $`A_V`$ = 10.0 when the \[O i\] intensity is used as a normalizer. When normalized by \[O iii\], the effect of dust extinction leads to higher \[Fe vii\]/\[O iii\] and \[Fe x\]/\[O iii\] ratios. Because the dust extinction is larger on average in S2s than in S1s, the observational results show that these ratios are higher in the S1s than in the S2s even if the effect of the dust extinction is taken into account. On the other hand, the excess of \[Fe vii\]/\[S ii\] in both the NLS1s and the BLS1s with respect to the S2s would be extinguished if the extinction of the S2s is systematically larger (e.g., $`A_V`$ = 10) than that of the S1s. However, the average difference of the extinction between S1s and S2s is about 1 mag (Dahari & De Robertis 1988a; see also De Zotti & Gaskell 1985). Hence it is unlikely that the S2s analyzed here suffer from such larger extinction systematically. Therefore, we conclude that the results obtained in our analysis are not so seriously affected by the dust extinction.
## 4. DISCUSSION
### 4.1. The HINER in the NLS1s
Our analysis has shown that; 1) the NLS1s have higher \[Fe vii\] and \[Fe x\] strengths than the S2<sup>+</sup>s and the S2<sup>-</sup>s although the statistical significance is worse when using the \[Fe x\] emission, and 2) there are no statistical differences in the relative strength of \[Fe vii\] and \[Fe x\] between the NLS1s and the BLS1s. Several previous works suggested that strong high-ionization emission lines are often seen in NLS1s (Davidson & Kinman 1978; Osterbrock & Pogge 1985; Nagao et al. 2000). Our analysis has statistically confirmed for the first time that the HINER emission lines of the NLS1s are significantly stronger than those of the S2s. Accordingly this suggests that the NLS1s are viewed from a more face-on view toward dusty tori than the S2s. On the other hand, the second result means that there is no systematic difference in the viewing angle toward the dusty torus between the NLS1s and the BLS1s from a statistical point of view.
Many theoretical models have been proposed to explain the properties of NLS1s (e.g., Boller et al. 1996; Taniguchi et al. 1999 and references therein). Any model is required to satisfy the statistical properties of the HINER presented in this paper; i.e., the narrow line width of NLS1s cannot be explained assuming the obscuration of broad component with dusty torus. For example, Giannuzzo & Stripe (1996) mentioned a possibility that the NLS1s may be objects seen from relatively large inclination angles and thus only outer parts of the BLR can be seen, being responsible for the narrow line width. However such models appear difficult to explain the property of HINER in the NLS1s consistently.
Since our second result means that the viewing angle toward the dusty torus is nearly the same on average between the NLS1s and the BLS1s, it is possible to propose the following model if the BLR observed in optical spectra has a disk-like configuration. Suppose that the rotational axis of the BLR is different from that of the dusty torus. In this case, the BLR line width of S1s depends on the viewing angle toward the BLR disk. However, the line width does not depend on the the viewing angle toward the dusty torus unless the BLR is not hidden by the dusty torus. This model is schematically shown in Figure 12. Our results appear consistent with this model.
It is noted that the BLR emission may arise from outer parts of a warped accretion disk (Shields 1977; Nishiura, Murayama, & Taniguchi 1998; see also for a review Osterbrock 1989). Such warping of accreting gas disks may be driven by the effect of radiation pressure force (Pringle 1996, 1997). Indeed, it has been recently shown that accreting gas clouds probed water vapor maser emission at 22 GHz show evidence for significant warping (Miyoshi et al. 1995; Begelman & Bland-Hawthorn 1997). Therefore, it is likely that the rotation axis of the BLR is not necessarily to align to that of the dusty torus.
Recently, Turner, George, & Netzer (1999b) reported on the observation of the NLS1 Akn 564. They estimated the viewing angle toward the accretion disk $``$60° using a model for asymmetric Fe K$`\alpha `$ line profile and mentioned that this result is contrary to the hypothesis that NLS1s are viewed from pole-on view. However, if the rotation axis of the accretion disk is different from those of the BLR and the dusty torus (e.g., Nishiura, Murayama, & Taniguchi 1998), their observation is not inconsistent with the HINER properties presented in this paper.
### 4.2. The HINER in the S1.5s
The S1.5 is widely recognized as a distinct class of Seyferts observationally (Osterbrock & Koski 1976; Cohen 1983). However, the nature of this type of Seyferts have not yet been fully understood. Comparing the HINER properties of the S1.5s with those of the other types of Seyferts, we discuss the nature of S1.5s.
Our analysis shows that; 1) there is no statistical difference in the relative \[Fe vii\] and \[Fe x\] strengths between the S1.5s and the S1s (i.e., NLS1s + BLS1s), but 2) the S1.5s have higher \[Fe vii\] strengths than the S2<sup>+</sup>s and the S2<sup>-</sup>s although this tendency is not confirmed in the relative \[Fe x\] strength. In summary, as shown in Figures 5 and 8, although the S1.5s have an intermediate property in the HINER line strengths between the S1s and S2s, the relative HINER line strengths cover the whole observed ranges of both the S1s and the S2s. Therefore, there are three alternative ideas to explain these observational properties. The first idea is that the S1.5s are seen from an intermediate viewing angle between S1s and S2s; i.e., a significant part of the BLR is obscured by a dusty torus, resulting in a composite profile consisting of both the narrow-line region (NLR) and BLR emission. The second idea is that some S1.5s are basically S1s but a significant part of the BLR emission is accidentally obscured by dense, clumpy gas clouds. The third idea is that some S1.5s are basically S2s but a part of the BLR emission can be seen from some optically-thin regions of the dusty torus.
As shown in Figure 5, the majority of S1.5s has nearly the same relative \[Fe vii\] strengths as those of the S1s, being consistent with the second idea. The remaining minority can be explained either by the first idea or by the third one. Yet, it seems important to mention that the origin of S1.5s may be heterogeneous. Finally, it is also important to mention that the latter two ideas may explain why some Seyfert nuclei show the so-called type switching between S1 and S2; e.g., NGC 4151 (Penston & Perez 1984; Ayani & Maehara 1991). The reason for this is as follows. It seems likely that the dusty torus consists of rather small blobs which are orbiting around the central engine. When a blob is passing the line of sight to the central engine, the BLR can be obscured if the blob is optically thick enough to hide it. If we assume that the blob is located at a radial distance of 0.1 pc from the central engine and the mass of the supermassive black hole is 10<sup>7</sup> M, the Keplerian velocity is estimated to be V$`{}_{\mathrm{rot}}{}^{}`$ 660 km s<sup>-1</sup>. Since the typical time scale of the observed type switchings is $``$ 10 years, this blob could move 2 $`\times `$ 10<sup>16</sup> cm. This is almost consistent with the typical size of the BLR, $``$ 0.01 pc (e.g., Peterson 1993). This idea also suggests that a typical size of such blobs is $``$ 0.01 pc.
### 4.3. The HINER in the “S2<sup>+</sup>s”
As presented in section 3, there is no statistical difference in the relative strengths of \[Fe vii\] and \[Fe x\] between S2<sup>+</sup>s and S2<sup>-</sup>s . Among the subtypes of S2<sup>+</sup>s (i.e., S1.8, S1.9, S2<sub>NIR-BLR</sub> and S2<sub>HBLR</sub>), there is no systematic difference in the HINER properties (see Figures 7 and 10). On the other hand, the S2<sup>+</sup>s have weaker \[Fe vii\] strength than the S1.5s (Table 10). These facts suggest that there is a systematic difference in the viewing angle toward dusty tori between the S1.5s and the S2<sup>+</sup>s; i.e., the S2<sup>+</sup>s might be those which are seen with large inclination angle and the emission radiated from the BLR reach us through the occasionally thin dusty tori. All these arguments imply that S2<sup>+</sup>s are viewed from large inclination angles, leading to more significant extinction of the BLR emission with respect to BLS1s and S1.5s. This appears consistent with earlier implications (e.g., Miller & Goodrich 1990; Heisler, Lumsden, & Bailey 1997).
### 4.4. The Nature of HINER Traced by \[Fe x\]
As presented in sections 3.1 and 3.2, both the NLS1s and the BLS1s have higher HINER emission-line strengths than the S2<sup>+</sup>s and the S2<sup>-</sup>s. However, comparing the KS probabilities given in Table 10 and Table 11, in particular those concerning to the relative intensities normalized by \[O iii\], this tendency is much more prominent in the analysis using \[Fe vii\] rather than \[Fe x\]. As proposed by MT98a, the excess \[Fe vii\] emission in the S1s appears attributed to the significant contribution from the torus HINER. Therefore, the weaker excess emission in \[Fe x\] implies that the major \[Fe x\] emitting region may be not the torus HINER but either the clumpy NLR HINER or the extended HINER or both because the latter two HINERs show less viewing angle dependence (see MT98a).
Another interesting property related to the \[Fe x\] emission is the observed \[Fe x\]/\[Fe vii\] ratios (see Figure 11). The average ratios are compared among the sample in Table 14. It is remarkable that some S1s and S2s have very higher ratios; e.g., \[Fe x\]/\[Fe vii\] $``$ 1. Therefore, in order to investigate the origin of the \[Fe x\] emission, it is interesting to compare the observed line ratios of \[Fe x\]/\[Fe vii\] with several theoretical predictions. Because simple one-zone models are known to predict too weak high-ionization emission lines (e.g., Pelat, Alloin, & Bica 1987; Dopita et al. 1997), we investigate multi-component photoionization models; 1) optically thin multi-cloud model (Ferland & Osterbrock 1986), and 2) the locally optimally emitting cloud model (LOC model; Ferguson et al. 1997). As shown in Table 14, these models predict smaller line ratios. On the other hand, the low-density interstellar matter (ISM) model by Korista & Ferland (1989) predicts higher line ratios of \[Fe x\]/\[Fe vii\]. They mentioned that such HINER will be observed out to 1 – 2 kpc. Indeed, extended HINERs have been found in some Seyfert galaxies; NGC 3516 (Golev et al. 1995), Tololo 0109$``$383 (Murayama et al. 1998), and NGC 4051 (Nagao et al. 2000). Alternatively, shock models may also be responsible for the observed higher \[Fe x\]/\[Fe vii\] ratios. Viegas-Aldrovandi & Contini (1989) calculated those ratios introducing the shock component. As shown in Table 14, such models can also explain those higher line ratios.
Which is the appropriate model for the higher line ratios of \[Fe x\]/\[Fe vii\], highly ionized low-density ISM or shock-driven ionization? In order to distinguish these two models, we compare the observed \[Fe xi\]/\[Fe x\] ratios with model results in Table 14. The \[Fe xi\] emission is observed in only 19 Seyfert nuclei (e.g., Grandi 1978; Cohen 1983; Penston et al. 1984; Erkens et al. 1997).As shown in Table 14, the low-density ISM models of Korista & Ferland (1989) appear consistent with the observed \[Fe xi\]/\[Fe x\]ratios. Though Viegas-Aldrovandi & Contini (1989) did not calculate this line ratio, Evans et al. (1999) mentioned that the ionization state in the emission-line region ionized by shocks is somewhat lower than that ionized by a typical nonthermal continuum. Therefore, it is likely that some Seyfert galaxies have the extended HINER described by Korista & Ferland (1989), being responsible for the unusually strong \[Fe x\] emission. This idea also explains why the excess \[Fe x\] emission is less significant in the S1s than the excess \[Fe vii\] emission.
## 5. CONCLUDING REMARKS
The anisotropic property of the radiation from the HINER traced \[Fe vii\] reported by MT98a has been statistically confirmed using the larger sample. The line ratios of \[Fe x\] to the low-ionization emission lines show a rather isotropic property with respect to those of \[Fe vii\] to the low-ionization emission lines. This may be interpreted by an idea that a significant fraction of the \[Fe x\] emission arises from low-density ISM as suggested by Korista & Ferland (1989). We note that the \[Fe x\] emission is not suitable to investigate the viewing angle toward the dusty tori of Seyfert nuclei.
We have also investigated the HINER properties of the intermediate-type of Seyfert nuclei. Using the frequency distributions of the line ratios of \[Fe vii\] to the low-ionization emission lines, we find the following suggestions. (1) The NLS1s are viewed from a more face-on orientation toward dusty tori than the S2s. (2) The line ratios of S1.5s are distributed in a wide range from the smallest value of the S2s to the largest value of the S1s. This suggests that the S1.5s are heterogeneous populations. (3) The HINER properties of the S1.8s, the S1.9s and the objects showing a broad Pa$`\beta `$ line or polarized broad Balmer lines are considerably different from those of the S1s. These facts mean that the “S2<sup>+</sup>” objects are those which are seen from a large inclination angle and their BLR emission comes through optically-thin line of sights toward the dusty tori.
We would like to thank the anonymous referee for useful comments and suggestions and Yuji Ikeda, Shingo Nishiura and Yasuhiro Shioya for useful advice. This research has made use of the NED (NASA extragalactic database) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under construct with the National Aeronautics and Space Administration. TM is supported by a Research Fellowship from the Japan Society for the Promotion of Science for Young Scientists. A part of this work was made when YT visited the Astronomical Data Analysis Center (ADAC) of the National Astronomical Observatory of Japan. YT thanks the staff of ADAC, in particular Shin-ichi Ichikawa, for their kind hospitality. This work was financially supported in part by Grant-in-Aids for the Scientific Research (Nos. 10044052, and 10304013) of the Japanese Ministry of Education, Culture, Sports, and Science. |
warning/0003/hep-th0003006.html | ar5iv | text | # A SHORT SURVEY OF NONCOMMUTATIVE GEOMETRY
## Introduction
The origin of noncommutative geometry is twofold.
On the one hand there is a wealth of examples of spaces whose coordinate algebra is no longer commutative but which have obvious relevance in physics or mathematics. The first examples came from phase space in quantum mechanics but there are many others, such as the leaf spaces of foliations, the duals of nonabelian discrete groups, the space of Penrose tilings, the Brillouin zone in solid state physics, the noncommutative tori which appear naturally in M-theory compactification, and the Adele class space which is a natural geometric space carrying an action of the analogue of the Frobenius for global fields of zero characteristic. Finally various models of space-time itself are interesting examples of noncommutative spaces.
On the other hand the stretching of geometric thinking imposed by passing to noncommutative spaces forces one to rethink about most of our familiar notions. The difficulty is not to add arbitrarily the adjective quantum to our geometric words but to develop far reaching extensions of classical concepts, ranging from the simplest which is measure theory, to the most sophisticated which is geometry itself.
## Measure theory
The extension of the classical concepts has been achieved a long time ago by operator algebraists as far as measure theory is concerned. The theory of nonabelian von Neumann algebras is indeed a far reaching extension of measure theory, whose main surprise is that such an algebra $`M`$ inherits from its noncommutativity a god-given time evolution.
It is given by the group homomorphism, ()
$$\delta :\mathrm{Out}(M)=\mathrm{Aut}(M)/\mathrm{Int}(M)$$
(1)
from the additive group $``$ to the group of automorphism classes of $`M`$ modulo inner automorphisms.
This uniqueness of the, a priori state dependent, modular automorphism group of a state, together with the earlier work of Powers, Araki-Woods and Krieger were the first steps which eventually led to the complete classification of approximately finite dimensional factors (also called hyperfinite).
They are classified by their module,
$$\mathrm{Mod}(M)\underset{}{}_+^{},$$
(2)
which is a virtual closed subgroup of $`_+^{}`$ in the sense of G. Mackey, i.e. an ergodic action of $`_+^{}`$.
The classification involves three independent parts,
* The definition of the invariant $`\mathrm{Mod}(M)`$ for arbitrary factors.
* The equivalence of all possible notions of approximate finite dimensionality.
* The proof that Mod is a complete invariant and that all virtual subgroups are obtained.
The module of a factor $`M`$ was first defined () as a closed subgroup of $`_+^{}`$ by the equality
$$S(M)=\underset{\phi }{}\mathrm{Spec}(\mathrm{\Delta }_\phi )_+,$$
(3)
where $`\phi `$ varies among (faithful, normal) states on $`M`$ and the operator $`\mathrm{\Delta }_\phi `$ is the modular operator of the Tomita-Takesaki theory ().
The virtual subgroup $`\mathrm{Mod}(M)`$ is the flow of weights ( ) of $`M`$. It is obtained from the module $`\delta `$ as the dual action of $`_+^{}`$ on the abelian algebra,
$$C=\text{Center of}(M>_\delta ),$$
(4)
where $`M>_\delta `$ is the crossed product of $`M`$ by the modular automorphism group $`\delta `$.
This takes care of (A), to describe (B) let us simply state the equivalence () of the following conditions
$$\begin{array}{cc}& M\text{is the closure of the union of an increasing sequence of}\\ & \text{finite dimensional algebras.}\end{array}$$
(5)
$$\begin{array}{cc}& M\text{is complemented as a subspace of the normed space of}\\ & \text{all operators in a Hilbert space.}\end{array}$$
(6)
The condition (5) is obviously what one would expect for an approximately finite dimensional algebra. Condition (6) is similar to amenability for discrete groups and the implication (6) $``$ (5) is a very powerful tool.
Besides the reduction from type III to type II ( ), the proof of (C) involves the uniqueness of the approximately finite dimensional factor of type $`\mathrm{II}_{\mathrm{}}`$, , the classification of its automorphisms for the $`\mathrm{III}_\lambda `$ case, and the results of Krieger for the $`\mathrm{III}_0`$ case. The only case which was left open in 1976 was the $`\mathrm{III}_1`$ case, which was reduced to a problem on the bicentralizer of states , this problem was finally settled by U. Haagerup in . Since then, the subject of von-Neumann algebras has undergone two major revolutions, thanks first to the famous work of Vaughan Jones on subfactors and then to the pioneering work of Dan Voiculescu who created and developped the completely new field of free probability theory.
Von Neumann algebras arise very naturally in geometry from foliated manifolds $`(V,F)`$. The von Neumann algebra $`L^{\mathrm{}}(V,F)`$ of a foliated manifold is easy to describe, its elements are random operators $`T=(T_f)`$, i.e. bounded measurable families of operators $`T_f`$ parametrized by the leaves $`f`$ of the foliation. For each leaf $`f`$ the operator $`T_f`$ acts in the Hilbert space $`L^2(f)`$ of square integrable densities on the manifold $`f`$. Two random operators are identified if they are equal for almost all leaves $`f`$ (i.e. a set of leaves whose union in $`V`$ is negligible). The algebraic operations of sum and product are given by,
$$(T_1+T_2)_f=(T_1)_f+(T_2)_f,(T_1T_2)_f=(T_1)_f(T_2)_f,$$
(7)
i.e. are effected pointwise.
All types of factors occur from this geometric construction and the continuous dimensions of Murray and von-Neumann play an essential role in the longitudinal index theorem.
Finally we refer to for the role of approximately finite dimensional factors in number theory as the missing Brauer theory at Archimedean places.
## Topology
The development of the topological ideas was prompted by the work of Israel Gel’fand, whose C\* algebras give the required framework for noncommutative topology. The two main driving forces were the Novikov conjecture on homotopy invariance of higher signatures of ordinary manifolds as well as the Atiyah-Singer Index theorem. It has led, through the work of Atiyah, Singer, Brown, Douglas, Fillmore, Miscenko and Kasparov ( ) to the recognition that not only the Atiyah-Hirzebruch K-theory but more importantly the dual K-homology admit Hilbert space techniques and functional analysis as their natural framework. The cycles in the K-homology group $`K_{}(X)`$ of a compact space X are indeed given by Fredholm representations of the C\* algebra A of continuous functions on X. The central tool is the Kasparov bivariant K-theory. A basic example of C\* algebra to which the theory applies is the group ring of a discrete group and restricting oneself to commutative algebras is an obviously undesirable assumption.
For a $`C^{}`$ algebra $`A`$, let $`K_0(A)`$, $`K_1(A)`$ be its $`K`$ theory groups. Thus $`K_0(A)`$ is the algebraic $`K_0`$ theory of the ring $`A`$ and $`K_1(A)`$ is the algebraic $`K_0`$ theory of the ring $`AC_0()=C_0(,A)`$. If $`AB`$ is a morphism of $`C^{}`$ algebras, then there are induced homomorphisms of abelian groups $`K_i(A)K_i(B)`$. Bott periodicity provides a six term $`K`$ theory exact sequence for each exact sequence $`0JAB0`$ of $`C^{}`$ algebras and excision shows that the $`K`$ groups involved in the exact sequence only depend on the respective $`C^{}`$ algebras. As an exercice to appreciate the power of this abstract tool one should for instance use the six term $`K`$ theory exact sequence to give a short proof of the Jordan curve theorem.
Discrete groups, Lie groups, group actions and foliations give rise through their convolution algebra to a canonical $`C^{}`$ algebra, and hence to $`K`$ theory groups. The analytical meaning of these $`K`$ theory groups is clear as a receptacle for indices of elliptic operators. However, these groups are difficult to compute. For instance, in the case of semi-simple Lie groups the free abelian group with one generator for each irreducible discrete series representation is contained in $`K_0C_r^{}G`$ where $`C_r^{}G`$ is the reduced $`C^{}`$ algebra of $`G`$. Thus an explicit determination of the $`K`$ theory in this case in particular involves an enumeration of the discrete series.
We introduced with P. Baum () a geometrically defined $`K`$ theory which specializes to discrete groups, Lie groups, group actions, and foliations. Its main features are its computability and the simplicity of its definition. In the case of semi-simple Lie groups it elucidates the role of the homogeneous space $`G/K`$ ($`K`$ the maximal compact subgroup of $`G`$) in the Atiyah-Schmid geometric construction of the discrete series . Using elliptic operators we constructed a natural map from our geometrically defined $`K`$ theory groups to the above analytic (i.e. $`C^{}`$ algebra) $`K`$ theory groups. Much progress has been made in the past years to determine the range of validity of the isomorphism between the geometrically defined $`K`$ theory groups and the above analytic (i.e. $`C^{}`$ algebra) $`K`$ theory groups. We refer to the three Bourbaki seminars (, , ) for an update on this topic.
## Differential Topology
The development of differential geometric ideas, including de Rham homology, connections and curvature of vector bundles, etc… took place during the eighties thanks to cyclic cohomology which came from two different horizons ( ). This led for instance to the proof of the Novikov conjecture for hyperbolic groups , but got many other applications. Basically, by extending the Chern-Weil characteristic classes to the general framework it allows for many concrete computations of differential geometric nature on noncommutative spaces. It also showed the depth of the relation between the above classification of factors and the geometry of foliations. For instance, using cyclic cohomology together with the following simple fact,
$$\begin{array}{cc}& \text{“A connected group can only act trivially on a homotopy}\\ & \text{invariant cohomology theory”,}\end{array}$$
(8)
one proves (cf. ) that for any codimension one foliation $`F`$ of a compact manifold $`V`$ with non vanishing Godbillon-Vey class one has,
$$\mathrm{Mod}(M)\text{has finite covolume in}_+^{},$$
(9)
where $`M=L^{\mathrm{}}(V,F)`$ and a virtual subgroup of finite covolume is a flow with a finite invariant measure.
In its simplest form, cyclic cohomology is the cohomology theory obtained from the cochain complex of ($`n+1`$)-linear form on $`𝒜`$, $`n`$ arbitrary, such that
$$\phi (a^0,a^1,\mathrm{},a^n)=(1)^n\phi (a^1,a^2,\mathrm{},a^0)a_j𝒜,$$
(10)
with coboundary operator given by
$`(b\phi )(a^0,\mathrm{},a^{n+1})=`$ (11)
$`{\displaystyle \underset{0}{\overset{n}{}}}(1)^j\phi (a^0,\mathrm{},a^ja^{j+1},\mathrm{},a^{n+1})+(1)^{n+1}\phi (a^{n+1}a^0,a^1,\mathrm{},a^n)`$
Its first important role is to provide invariants of $`K`$-theory classes as follows. Given an n-dimensional cyclic cocycle on $`𝒜`$, n even, the following scalar is invariant under homotopy for projectors (idempotents) $`EM_n(𝒜)`$,
$$\phi _n(E,E,\mathrm{},E)$$
(12)
where $`\phi `$ has been uniquely extended to $`M_n(𝒜)`$ using the trace on $`M_n()`$, as in (9) below. This defines a $`\text{pairing}K(𝒜),HC(𝒜)`$ between cyclic cohomology and K-theory.
When we take $`𝒜=C^{\mathrm{}}(M)`$ for a manifold $`M`$ and let
$$\phi (f^0,f^1,\mathrm{},f^n)=C,f^0df^1df^2\mathrm{}df^nf^j𝒜$$
(13)
where $`C`$ is an n-dimensional closed de Rham current, the above invariant is equal to (up to normalization)
$$C,Ch(E)$$
(14)
where $`Ch(E)`$ is the Chern character of the vector bundle $`E`$ on $`M`$ whose fiber at $`xM`$ is the range of $`E(x)M_n()`$. In this example we see that for any permutation of $`\{0,1,\mathrm{},n\}`$ one has:
$$\phi (f^{\sigma (0)},f^{\sigma (1)},\mathrm{},f^{\sigma (n)})=\epsilon (\sigma )\phi (f^0,f^1,\mathrm{},f^n)$$
(15)
where $`\epsilon (\sigma )`$ is the signature of the permutation. However when we extend $`\phi `$ to $`M_n(𝒜)`$ as $`\phi _n=\phi \mathrm{Tr}`$,
$$\phi _n(f^0\mu ^0,f^1\mu ^1,\mathrm{},f^n\mu ^n)=\phi (f^0,f^1,\mathrm{},f^n)\mathrm{Tr}(\mu ^0\mu ^1\mathrm{}\mu ^n)$$
(16)
the property (8) only survives for cyclic permutations. This is at the origin of the name, cyclic cohomology, given to the corresponding cohomology theory.
Both the Hochschild and Cyclic cohomologies of the algebra $`𝒜=C^{\mathrm{}}(M)`$ of smooth functions on a manifold $`M`$ were computed in , , thus showing how to extend the familiar differential geometric notions to the general noncommutative case according to the following dictionnary:
$$\begin{array}{cc}\text{Space}& \text{Algebra}\\ \\ \text{Vector bundle}& \text{Finite projective module}\\ \\ \text{Differential form}& \text{(Class of) Hochschild cycle}\\ \\ \text{DeRham current}& \text{(Class of) Hochschild cocycle}\\ \\ \text{DeRham homology}& \text{Cyclic cohomology}\\ \\ \text{Chern Weil theory}& \text{Pairing}K(𝒜),HC(𝒜)\end{array}$$
$`(A)`$
A simple example of cyclic cocycle on a nonabelian group ring is provided by the following formula. Any group cocycle $`cH^{}(B\mathrm{\Gamma })=H^{}(\mathrm{\Gamma })`$ gives rise to a cyclic cocycle $`\phi _c`$ on the algebra $`𝒜=\mathrm{\Gamma }`$
$$\phi _c(g_0,g_1,\mathrm{},g_n)=\{\begin{array}{c}0\text{if}g_0\mathrm{}g_n1\\ c(g_1,\mathrm{},g_n)\text{if}g_0\mathrm{}g_n=1\end{array}$$
(17)
where $`cZ^n(\mathrm{\Gamma },)`$ is suitably normalized, and 17 is extended by linearity to $`\mathrm{\Gamma }`$.
Cyclic cohomology has an equivalent description by means of the bicomplex $`(b,B)`$ which is given by the following operators acting on multi-linear forms on $`𝒜`$,
$`(b\phi )(a^0,\mathrm{},a^{n+1})=`$ (18)
$`{\displaystyle \underset{0}{\overset{n}{}}}(1)^j\phi (a^0,\mathrm{},a^ja^{j+1},\mathrm{},a^{n+1})+(1)^{n+1}\phi (a^{n+1}a^0,a^1,\mathrm{},a^n)`$
$`B=AB_0,B_0\phi (a^0,\mathrm{},a^{n1})=\phi (1,a^0,\mathrm{},a^{n1})(1)^n\phi (a^0,\mathrm{},a^{n1},1)`$ (19)
$`(A\psi )(a^0,\mathrm{},a^{n1})={\displaystyle \underset{0}{\overset{n1}{}}}(1)^{(n1)j}\psi (a^j,a^{j+1},\mathrm{},a^{j1}).`$
The pairing between cyclic cohomology and K-theory is given in this presentation by the following formula for the Chern character of the class of an idempotent $`e`$, up to normalization one has
$$Ch_n(e)=(e1/2)ee\mathrm{}e,$$
(20)
where $`e`$ appears 2n times in the right hand side of the equation.
At the conceptual level, cyclic cohomology is a way to embed the nonadditive category of algebras and algebra homomorphisms in an additive category of modules. The latter is the additive category of $`\mathrm{\Lambda }`$-modules where $`\mathrm{\Lambda }`$ is the cyclic category. Cyclic cohomology is then obtained as an $`Ext`$ functor ().
The cyclic category is a small category which can be defined by generators and relations. It has the same objects as the small category $`\mathrm{\Delta }`$ of totally ordered finite sets and increasing maps which plays a key role in simplicial topology. Let us recall (we shall use it later) that $`\mathrm{\Delta }`$ has one object $`[n]`$ for each integer $`n`$, and is generated by faces $`\delta _i,[n1][n]`$ (the injection that misses $`i`$), and degeneracies $`\sigma _j,[n+1][n]`$ (the surjection which identifies $`j`$ with $`j+1`$), with the relations,
$$\delta _j\delta _i=\delta _i\delta _{j1}\text{for}i<j,\sigma _j\sigma _i=\sigma _i\sigma _{j+1}ij$$
(21)
$$\sigma _j\delta _i=\{\begin{array}{cc}\delta _i\sigma _{j1}& i<j\\ 1_n& \text{if}i=j\text{or}i=j+1\\ \delta _{i1}\sigma _j& i>j+1.\end{array}$$
To obtain $`\mathrm{\Lambda }`$ one adds for each $`n`$ a new morphism $`\tau _n,[n][n]`$ such that,
$$\begin{array}{ccc}\tau _n\delta _i=\delta _{i1}\tau _{n1}& 1in,& \tau _n\delta _0=\delta _n\\ \\ \tau _n\sigma _i=\sigma _{i1}\tau _{n+1}& 1in,& \tau _n\sigma _0=\sigma _n\tau _{n+1}^2\\ \\ \tau _n^{n+1}=1_n.\end{array}$$
(22)
The original definition of $`\mathrm{\Lambda }`$ (cf. ) used homotopy classes of non decreasing maps from $`S^1`$ to $`S^1`$ of degree 1, mapping $`/n`$ to $`/m`$ and is trivially equivalent to the above.
Given an algebra $`A`$ one obtains a module over the small category $`\mathrm{\Lambda }`$ by assigning to each integer $`n0`$ the vector space $`C^n`$ of $`n+1`$-linear forms $`\phi (x^0,\mathrm{},x^n)`$ on $`A`$, while the basic operations are given by
$$\begin{array}{ccc}(\delta _i\phi )(x^0,\mathrm{},x^n)& =& \phi (x^0,\mathrm{},x^ix^{i+1},\mathrm{},x^n),i=0,1,\mathrm{},n1\\ \\ (\delta _n\phi )(x^0,\mathrm{},x^n)& =& \phi (x^nx^0,x^1,\mathrm{},x^{n1})\\ \\ (\sigma _j\phi )(x^0,\mathrm{},x^n)& =& \phi (x^0,\mathrm{},x^j,1,x^{j+1},\mathrm{},x^n),j=0,1,\mathrm{},n\\ \\ (\tau _n\phi )(x^0,\mathrm{},x^n)& =& \phi (x^n,x^0,\mathrm{},x^{n1}).\end{array}$$
(23)
These operations satisfy the relations (21) and (22). This shows that any algebra $`A`$ gives rise canonically to a $`\mathrm{\Lambda }`$-module and allows to interpret the cyclic cohomology groups $`HC^n(A)`$ as $`Ext^n`$ functors. All of the general properties of cyclic cohomology such as the long exact sequence relating it to Hochschild cohomology are shared by Ext of general $`\mathrm{\Lambda }`$-modules and can be attributed to the equality of the classifying space $`B\mathrm{\Lambda }`$ of the small category $`\mathrm{\Lambda }`$ with the classifying space $`BS^1`$ of the compact one-dimensional Lie group $`S^1`$. One has
$$B\mathrm{\Lambda }=BS^1=P_{\mathrm{}}()$$
(24)
For group rings $`𝒜=\mathrm{\Gamma }`$ as above the cyclic cohomology bicomplex corresponds exactly () to the bicomplex computing the $`S^1`$-equivariant cohomology of the free loop space of the classifying space $`B\mathrm{\Gamma }`$, which is in essence dual to the space of irreducible representations of $`\mathrm{\Gamma }`$.
In the recent years J. Cuntz and D. Quillen ( ) have developed a powerful new approach to cyclic cohomology which allowed them to prove excision in full generality. A great deal of activity has also been generated around the work of Maxim Kontsevich on deformation theory and the Deligne conjecture on the fine structure of the algebra of Hochschild cochains (see ).
## Geometry
The basic data of Riemannian geometry consists of a manifold $`M`$ whose points are locally labeled by a finite number of real coordinates $`\{x^\mu \}`$ and a metric, which is given by the infinitesimal line element:
$$ds^2=g_{\mu \nu }dx^\mu dx^\nu .$$
(25)
The distance between two points $`x,yM`$ is given by
$$d(x,y)=\text{Inf}\{\text{Length}\gamma |\gamma \text{is a path between }x\text{ and }y\}$$
(26)
where
$$\text{Length}\gamma =_\gamma 𝑑s.$$
(27)
One of the main virtues of Riemannian geometry is to be flexible enough to give a good model of space-time in general relativity (up to a sign change) while simple notions of Euclidean geometry continue to make sense. Homogeneous spaces which are geometries in the sense of the Klein program are too restrictive to achieve that goal. For instance the idea of a straight line gives rise to the notion of geodesic and the geodesic equation
$$\frac{d^2x^\mu }{dt^2}=\mathrm{\Gamma }_{\nu \rho }^\mu \frac{dx^\nu }{dt}\frac{dx^\rho }{dt}$$
(28)
where $`\mathrm{\Gamma }_{\nu \rho }^\mu =\frac{1}{2}g^{\mu \alpha }(g_{\alpha \nu ,\rho }+g_{\alpha \rho ,\nu }g_{\nu \rho ,\alpha })`$, gives the Newton equation of motion of a particle in the Newtonian potential $`V`$ provided one uses the metric $`dx^2+dy^2+dz^2(1+2V(x,y,z))dt^2`$ instead of the Minkowski metric (cf. for the more precise formulation). The next essential point is that the differential and integral calculus is available and allows to go from the local to the global.
The central notion of noncommutative geometry, comes from the identification of the noncommutative analogue of the two basic concepts in Riemann’s formulation of Geometry, namely those of manifold and of infinitesimal line element. Both of these noncommutative analogues are of spectral nature and combine to give rise to the notion of spectral triple and spectral manifold, which will be described in detail below. We shall first describe an operator theoretic framework for the calculus of infinitesimals which will provide a natural home for the line element $`ds`$.
## Calculus and Infinitesimals
It was recognized at an early stage of the development of noncommutative geometry that the formalism of quantum mechanics gives a natural home both to infinitesimals (the compact operators in Hilbert space) and to the integral (the logarithmic divergence in an operator trace) thus allowing for the generalization of the differential and integral calculus which is vital for the development of the general theory.
The following is the beginning of a long dictionary which translates classical notions into the language of operators in the Hilbert space $``$:
$$\begin{array}{cc}\text{Complex variable}& \text{Operator in}\\ & \\ \text{Real variable}& \text{Selfadjoint operator}\\ & \\ \text{Infinitesimal}& \text{Compact operator}\\ & \\ \text{Infinitesimal of order}\alpha & \text{Compact operator with characteristic values}\\ & \mu _n\text{satisfying }\mu _n=O(n^\alpha ),n\mathrm{}\\ \text{Integral of an infinitesimal }& T=\text{ Coefficient of logarithmic}\\ \text{of order 1}& \text{divergence in the trace of }T.\end{array}$$
The first two lines of the dictionary are familiar from quantum mechanics. The range of a complex variable corresponds to the spectrum of an operator. The holomorphic functional calculus gives a meaning to $`f(T)`$ for all holomorphic functions $`f`$ on the spectrum of $`T`$. It is only holomorphic functions which operate in this generality which reflects the difference between complex and real analysis. When $`T=T^{}`$ is selfadjoint then $`f(T)`$ has a meaning for all Borel functions $`f`$.
The size of the infinitesimal $`T𝒦`$ is governed by the order of decay of the sequence of characteristic values $`\mu _n=\mu _n(T)`$ as $`n\mathrm{}`$. In particular, for all real positive $`\alpha `$ the following condition defines infinitesimals of order $`\alpha `$:
$$\mu _n(T)=O(n^\alpha )\text{when}n\mathrm{}$$
(29)
(i.e. there exists $`C>0`$ such that $`\mu _n(T)Cn^\alpha n1`$). Infinitesimals of order $`\alpha `$ also form a two–sided ideal and moreover,
$$T_j\text{of order}\alpha _jT_1T_2\text{of order}\alpha _1+\alpha _2.$$
(30)
Hence, apart from commutativity, intuitive properties of the infinitesimal calculus are fulfilled.
Since the size of an infinitesimal is measured by the sequence $`\mu _n0`$ it might seem that one does not need the operator formalism at all, and that it would be enough to replace the ideal $`𝒦`$ in $`()`$ by the ideal $`c_0()`$ of sequences converging to zero in the algebra $`\mathrm{}^{\mathrm{}}()`$ of bounded sequences. A variable would just be a bounded sequence, and an infinitesimal a sequence $`\mu _n,\mu _n0`$. However, this commutative version does not allow for the existence of variables with range a continuum since all elements of $`\mathrm{}^{\mathrm{}}()`$ have a point spectrum and a discrete spectral measure. Only noncommutativity of $`()`$ allows for the coexistence of variables with Lebesgue spectrum together with infinitesimal variables. As we shall see shortly, it is precisely this lack of commutativity between the line element and the coordinates on a space that will provide the measurement of distances.
The integral is obtained by the following analysis, mainly due to Dixmier (), of the logarithmic divergence of the partial traces
$$Trace_N(T)=\underset{0}{\overset{N1}{}}\mu _n(T),T0.$$
(31)
In fact, it is useful to define $`Trace_\mathrm{\Lambda }(T)`$ for any positive real $`\mathrm{\Lambda }>0`$ by piecewise affine interpolation for noninteger $`\mathrm{\Lambda }`$.
Define for all order 1 operators $`T0`$
$$\tau _\mathrm{\Lambda }(T)=\frac{1}{\mathrm{log}\mathrm{\Lambda }}_e^\mathrm{\Lambda }\frac{Trace_\mu (T)}{\mathrm{log}\mu }\frac{d\mu }{\mu }$$
(32)
which is the Cesaro mean of the function $`\frac{Trace_\mu (T)}{\mathrm{log}\mu }`$ over the scaling group $`_+^{}`$.
For $`T0`$, an infinitesimal of order 1, one has
$$Trace_\mathrm{\Lambda }(T)C\mathrm{log}\mathrm{\Lambda }$$
(33)
so that $`\tau _\mathrm{\Lambda }(T)`$ is bounded. The essential property is the following asymptotic additivity of the coefficient $`\tau _\mathrm{\Lambda }(T)`$ of the logarithmic divergence (33):
$$|\tau _\mathrm{\Lambda }(T_1+T_2)\tau _\mathrm{\Lambda }(T_1)\tau _\mathrm{\Lambda }(T_2)|3C\frac{\mathrm{log}(\mathrm{log}\mathrm{\Lambda })}{\mathrm{log}\mathrm{\Lambda }}$$
(34)
for $`T_j0`$.
An easy consequence of (34) is that any limit point $`\tau `$ of the nonlinear functionals $`\tau _\mathrm{\Lambda }`$ for $`\mathrm{\Lambda }\mathrm{}`$ defines a positive and linear trace on the two–sided ideal of infinitesimals of order $`1`$,
In practice the choice of the limit point $`\tau `$ is irrelevant because in all important examples $`T`$ is a measurable operator, i.e.:
$$\tau _\mathrm{\Lambda }(T)\text{converges when }\mathrm{\Lambda }\mathrm{}.$$
(35)
Thus the value $`\tau (T)`$ is independent of the choice of the limit point $`\tau `$ and is denoted
$$T.$$
(36)
The first interesting example is provided by pseudodifferential operators $`T`$ on a differentiable manifold $`M`$. When $`T`$ is of order 1 in the above sense, it is measurable and $`T`$ is the non-commutative residue of $`T`$ (). It has a local expression in terms of the distribution kernel $`k(x,y)`$, $`x,yM`$. For $`T`$ of order $`1`$ the kernel $`k(x,y)`$ diverges logarithmically near the diagonal,
$$k(x,y)=a(x)\mathrm{log}|xy|+0(1)(\text{for}yx)$$
(37)
where $`a(x)`$ is a 1–density independent of the choice of Riemannian distance $`|xy|`$. Then one has (up to normalization),
$$T=_Ma(x).$$
(38)
The right hand side of this formula makes sense for all pseudodifferential operators (cf. ) since one can see that the kernel of such an operator is asymptotically of the form
$$k(x,y)=a_k(x,xy)a(x)\mathrm{log}|xy|+0(1)$$
(39)
where $`a_k(x,\xi )`$ is homogeneous of degree $`k`$ in $`\xi `$, and the 1–density $`a(x)`$ is defined intrinsically.
The same principle of extension of $``$ to infinitesimals of order $`<1`$ works for hypoelliptic operators and more generally as we shall see below, for spectral triples whose dimension spectrum is simple.
## Manifolds
As we shall see shortly this framework gives a natural home for the analogue of the infinitesimal line element $`ds`$ of Riemannian geometry, but we need first to exhibit its compatibility with the notion of manifold.
It was recognized long ago by geometors that the main quality of the homotopy type of a manifold, (besides being defined by a cooking recipee) is to satisfy Poincaré duality not only in ordinary homology but also in K-homology. Poincaré duality in ordinary homology is not sufficient to describe homotopy type of manifolds () but D. Sullivan () showed (in the simply connected PL case of dimension $`5`$ ignoring 2–torsion) that it is sufficient to replace ordinary homology by $`KO`$–homology.
The characteristic property of differentiable manifolds which is carried over to the noncommutative case is Poincaré duality in $`KO`$–homology.
Moreover, $`K`$-homology admits, as we saw above, a fairly simple definition in terms of Hilbert space Fredholm representations.
In the general framework of Noncommutative Geometry the confluence of the Hilbert space incarnation of the two notions of metric and fundamental class for a manifold led very naturally to define a geometric space as given by a spectral triple:
$$(𝒜,,D)$$
(40)
where $`𝒜`$ is an involutive algebra of operators in a Hilbert space $``$ and $`D`$ is a selfadjoint operator on $``$. The involutive algebra $`𝒜`$ corresponds to a given space $`M`$ like in the classical duality “Space $``$ Algebra” in algebraic geometry. The infinitesimal line element in Riemannian geometry is given by the equality
$$ds=1/D,$$
(41)
which expresses the infinitesimal line element $`ds`$ as the inverse of the Dirac operator $`D`$, hence under suitable boundary conditions as a propagator.
The significance of $`D`$ is two-fold. On the one hand it defines the metric by the above equation, on the other hand its homotopy class represents the K-homology fundamental class of the space under consideration. The exact measurement of distances is performed as follows, instead of measuring distances between points using the formula (5.2) we measure distances between states $`\phi ,\psi `$ on $`\overline{𝒜}`$ by a dual formula. This dual formula involves sup instead of inf and does not use paths in the space
$$d(\phi ,\psi )=Sup\{|\phi (a)\psi (a)|;a𝒜,[D,a]1\}.$$
(42)
A state, is a normalized positive linear form on $`𝒜`$ such that $`\phi (1)=1`$,
$$\phi :\overline{𝒜},\phi (a^{}a)0,a\overline{𝒜},\phi (1)=1.$$
(43)
In the commutative case the points of the space coincide with the characters of the algebra or equivalently with its pure states (i.e. the extreme points of the convex compact set of states). As it should, this formula gives the geodesic distance in the Riemannian case. The spectral triple $`(𝒜,,D)`$ associated to a compact Riemannian manifold $`M`$, $`K`$-oriented by a spin structure, is given by the representation
$$(f\xi )(x)=f(x)\xi (x)xM,f𝒜,\xi $$
(44)
of the algebra $`𝒜`$ of functions on $`M`$ in the Hilbert space
$$=L^2(M,S)$$
(45)
of square integrable sections of the spinor bundle. The operator $`D`$ is the Dirac operator (cf. ). The commutator $`[D,f]`$, for $`f𝒜=C^{\mathrm{}}(M)`$ is the Clifford multiplication by the gradient $`f`$ and its operator norm is:
$$[D,f]=Sup_{xM}f(x)=\text{Lipschitz norm}f.$$
(46)
Let $`x,yM`$ and $`\phi ,\psi `$ be the corresponding characters: $`\phi (f)=f(x)`$, $`\psi (f)=f(y)\text{for all}f𝒜.`$ Then formula (42) gives the same result as formula (5.2), i.e. it gives the geodesic distance between $`x`$ and $`y`$.
Unlike the formula (5.2) the dual formula (42) makes sense in general, namely, for example for discrete spaces and even for totally disconnected spaces.
The second role of the operator $`D`$ is to define the fundamental class of the space $`X`$ in K-homology, according to the following table,
$$\begin{array}{cc}\text{Space }X& \text{Algebra}𝒜\\ & \\ K_1(X)& \text{Stable homotopy class of the spectral}\\ & \text{triple}(𝒜,,D)\\ K_0(X)& \text{Stable homotopy class of }/2\text{graded}\\ & \text{ spectral triple}\end{array}$$
(i.e. for $`K_0`$ we suppose that $``$ is $`/2`$–graded by $`\gamma `$, where $`\gamma =\gamma ^{}`$, $`\gamma ^2=1`$ and $`\gamma a=a\gamma a𝒜`$, $`\gamma D=D\gamma `$).
This description works for the complex $`K`$–homology which is 2-periodic. We shall come back later to its refinement to $`KO`$-homology.
## Operator theoretic Index Formula
Before entering in the detailed discussion of the spectral notion of manifold let us mention the local index formula. This result allows, using the infinitesimal calculus, to go from local to global in the general framework of spectral triples $`(𝒜,,D)`$.
The Fredholm index of the operator $`D`$ determines (in the odd case) an additive map $`K_1(𝒜)\stackrel{\phi }{}`$ given by the equality
$$\phi ([u])=Index(PuP),uGL_1(𝒜)$$
(47)
where $`P`$ is the projector $`P=\frac{1+F}{2}`$, $`F=Sign(D)`$.
This map is computed by the pairing of $`K_1(𝒜)`$ with the following cyclic cocycle
$$\tau (a^0,\mathrm{},a^n)=Trace(a^0[F,a^1]\mathrm{}[F,a^n])a^j𝒜$$
(48)
where $`F=\text{Sign}D`$ and we assume that the dimension $`p`$ of our space is finite, which means that $`(D+i)^1`$ is of order $`1/p`$, also $`np`$ is an odd integer. There are similar formulas involving the grading $`\gamma `$ in the even case, and it is quite satisfactory ( ) that both cyclic cohomology and the chern Character formula adapt to the infinite dimensional case in which the only hypothesis is that $`exp(D^2)`$ is a trace class operator.
It is difficult to compute the cocycle $`\tau `$ in general because the formula (48) involves the ordinary trace instead of the local trace $``$ and it is crucial to obtain a local form of the above cocycle.
This problem is solved by a general formula which we now describe
Let us make the following regularity hypothesis on $`(𝒜,,D)`$
$$a\text{and }[D,a]Dom\delta ^k,a𝒜$$
(49)
where $`\delta `$ is the derivation $`\delta (T)=[|D|,T]`$ for any operator $`T`$.
We let $``$ denote the algebra generated by $`\delta ^k(a)`$, $`\delta ^k([D,a])`$. The usual notion of dimension of a space is replaced by the dimension spectrum which is a subset of $``$. The precise definition of the dimension spectrum is the subset $`\mathrm{\Sigma }`$ of singularities of the analytic functions
$$\zeta _b(z)=Trace(b|D|^z)Rez>p,b.$$
(50)
The dimension spectrum of a manifold $`M`$ is the set $`\{0,1,\mathrm{},n\}`$, $`n=dimM`$; it is simple. Multiplicities appear for singular manifolds. Cantor sets provide examples of complex points $`z`$ in the dimension spectrum.
We assume that $`\mathrm{\Sigma }`$ is discrete and simple, i.e. that $`\zeta _b`$ can be extended to $`/\mathrm{\Sigma }`$ with simple poles in $`\mathrm{\Sigma }`$.
We refer to for the case of a spectrum with multiplicities. Let $`(𝒜,,D)`$ be a spectral triple satisfying the hypothesis (49) and (50). The local index theorem is the following, :
1. The equality
$$P=Res_{z=0}Trace(P|D|^z)$$
defines a trace on the algebra generated by $`𝒜`$, $`[D,𝒜]`$ and $`|D|^z`$, where $`z`$.
2. There is only a finite number of non–zero terms in the following formula which defines the odd components $`(\phi _n)_{n=1,3,\mathrm{}}`$ of a cocycle in the bicomplex $`(b,B)`$ of $`𝒜`$,
$$\phi _n(a^0,\mathrm{},a^n)=\underset{k}{}c_{n,k}a^0[D,a^1]^{(k_1)}\mathrm{}[D,a^n]^{(k_n)}|D|^{n2|k|}a^j𝒜$$
where the following notations are used: $`T^{(k)}=^k(T)`$ and $`(T)=D^2TTD^2`$, $`k`$ is a multi-index, $`|k|=k_1+\mathrm{}+k_n`$,
$$c_{n,k}=(1)^{|k|}\sqrt{2i}(k_1!\mathrm{}k_n!)^1((k_1+1)\mathrm{}(k_1+k_2+\mathrm{}+k_n+n))^1\mathrm{\Gamma }\left(|k|+\frac{n}{2}\right).$$
3. The pairing of the cyclic cohomology class $`(\phi _n)HC^{}(𝒜)`$ with $`K_1(𝒜)`$ gives the Fredholm index of $`D`$ with coefficients in $`K_1(𝒜)`$.
For the normalization of the pairing between $`HC^{}`$ and $`K(𝒜)`$ see . In the even case, i.e. when $``$ is $`/2`$ graded by $`\gamma `$,
$$\gamma =\gamma ^{},\gamma ^2=1,\gamma a=a\gamma a𝒜,\gamma D=D\gamma ,$$
there is an analogous formula for a cocycle $`(\phi _n)`$, $`n`$ even, which gives the Fredholm index of $`D`$ with coefficients in $`K_0`$. However, $`\phi _0`$ is not expressed in terms of the residue $``$ because it is not local for a finite dimensional $``$.
## Diffeomorphism invariant Geometry
The power of the above operator theoretic local trace formula lies in its generality. We showed in how to use it to compute the index of transversally hypoelliptic operators for foliations (). This allows to give a precise meaning to diffeomorphism invariant geometry on a manifold M, by the construction of a spectral triple $`(𝒜,,D)`$ where the algebra $`𝒜`$ is the crossed product of the algebra of smooth functions on the finite dimensional bundle $`P`$ of metrics on M by the natural action of the diffeomorphism group of M. The operator $`D`$ is an hypoelliptic operator which is directly associated to the reduction of the structure group of the manifold $`P`$ to a group of triangular matrices whose diagonal blocks are orthogonal. By construction the fiber of $`P\stackrel{\pi }{}M`$ is the quotient $`F/O(n)`$ of the $`GL(n)`$–principal bundle $`F`$ of frames on $`M`$ by the action of the orthogonal group $`O(n)GL(n)`$. The space $`P`$ admits a canonical foliation: the vertical foliation $`VTP`$, $`V=Ker\pi _{}`$ and on the fibers $`V`$ and on $`N=(TP)/V`$ the following Euclidean structures. A choice of $`GL(n)`$–invariant Riemannian metric on $`GL(n)/O(n)`$ determines a metric on $`V`$. The metric on $`N`$ is defined tautologically: for every $`pP`$ one has a metric on $`T_{\pi (p)}(M)`$ which is isomorphic to $`N_p`$ by $`\pi _{}`$.
The computation of the local index formula for diffeomorphism invariant geometry was quite complicated even in the case of codimension 1 foliations: there were innumerable terms to be computed; this could be done by hand, by 3 weeks of eight hours per day tedious computations, but it was of course hopeless to proceed by direct computations in the general case. Henri and I finally found how to get the answer for the general case after discovering that the computation generated a Hopf algebra $`(n)`$ which only depends on n= codimension of the foliation, and which allows to organize the computation provided cyclic cohomology is suitably adapted to Hopf algebras.
Hopf algebras arise very naturally from their actions on noncommutative algebras . Given an algebra $`A`$, an action of the Hopf algebra $``$ on $`A`$ is given by a linear map,
$$AA,hah(a)$$
satisfying $`h_1(h_2a)=(h_1h_2)(a)`$, $`h_i`$, $`aA`$ and
$$h(ab)=h_{(1)}(a)h_{(2)}(b)a,bA,h.$$
(51)
where the coproduct of $`h`$ is,
$$\mathrm{\Delta }(h)=h_{(1)}h_{(2)}$$
(52)
In concrete examples, the algebra $`A`$ appears first, together with linear maps $`AA`$ satisfying a relation of the form (51) which dictates the Hopf algebra structure. This is exactly what occured in the above example (see for the description of $`(n)`$ and its relation with Diff($`^n`$)).
The theory of characteristic classes for actions of $``$ extends the construction of cyclic cocycles from a Lie algebra of derivations of a $`C^{}`$ algebra $`A`$, together with an invariant trace $`\tau `$ on $`A`$.
This theory was developped in in order to solve the above computational problem for diffeomorphism invariant geometry but it was shown in that the correct framework for the cyclic cohomology of Hopf algebras is that of modular pairs in involution. It is quite satisfactory that exactly the same structure emerged from the analysis of locally compact quantum groups. The resulting cyclic cohomology appears to be the natural candidate for the analogue of Lie algebra cohomology in the context of Hopf algebras. We fix a group-like element $`\sigma `$ and a character $`\delta `$ of $``$ with $`\delta (\sigma )=1`$. They will play the role of the module of locally compact groups.
We then introduce the twisted antipode,
$$\stackrel{~}{S}(y)=\delta (y_{(1)})S(y_{(2)}),y,\mathrm{\Delta }y=y_{(1)}y_{(2)}.$$
(53)
We associate a cyclic complex (in fact a $`\mathrm{\Lambda }`$-module, where $`\mathrm{\Lambda }`$ is the cyclic category), to any Hopf algebra together with a modular pair in involution. By this we mean a pair ($`\sigma `$, $`\delta `$) as above, such that the ($`\sigma `$, $`\delta `$)-twisted antipode is an involution,
$$(\sigma ^1\stackrel{~}{S})^2=I.$$
(54)
Then $`_{(\delta ,\sigma )}^{\mathrm{}}=\{^n\}_{n1}`$ equipped with the operators given by the following formulas (Diffeomorphism invariant Geometry)–(57) defines a module over the cyclic category $`\mathrm{\Lambda }`$. By transposing the standard simplicial operators underlying the Hochschild homology complex of an algebra, one associates to $``$, viewed only as a coalgebra, the natural cosimplicial module $`\{^n\}_{n1}`$, with face operators $`\delta _i:^{n1}^n`$,
$`\delta _0(h^1\mathrm{}h^{n1})=1h^1\mathrm{}h^{n1}`$
$`\delta _j(h^1\mathrm{}h^{n1})=h^1\mathrm{}\mathrm{\Delta }h^j\mathrm{}h^n,1jn1,`$
(55)
$`\delta _n(h^1\mathrm{}h^{n1})=h^1\mathrm{}h^{n1}\sigma `$
and degeneracy operators $`\sigma _i:^{n+1}^n`$,
$$\sigma _i(h^1\mathrm{}h^{n+1})=h^1\mathrm{}\epsilon (h^{i+1})\mathrm{}h^{n+1},0in.$$
(56)
The remaining two essential features of a Hopf algebra –product and antipode – are brought into play, to define the cyclic operators $`\tau _n:^n^n`$,
$$\tau _n(h^1\mathrm{}h^n)=(\mathrm{\Delta }^{n1}\stackrel{~}{S}(h^1))h^2\mathrm{}h^n\sigma .$$
(57)
The theory of characteristic classes applies to actions of the Hopf algebra on an algebra endowed with a $`\delta `$-invariant $`\sigma `$-trace. A linear form $`\tau `$ on $`A`$ is a $`\sigma `$-trace under the action of $``$ iff one has,
$$\tau (ab)=\tau (b\sigma (a))a,bA.$$
A $`\sigma `$-trace $`\tau `$ on $`A`$ is $`\delta `$-invariant under the action of $``$ iff
$$\tau (h(a)b)=\tau (a\stackrel{~}{S}(h)(b))a,bA,h.$$
The definition of the cyclic complex $`HC_{(\delta ,\sigma )}^{}()`$ is uniquely dictated in such a way that the following defines a canonical map from $`HC_{(\delta ,\sigma )}^{}()`$ to $`HC^{}(A)`$,
$$\begin{array}{c}\gamma (h^1\mathrm{}h^n)C^n(A),\gamma (h^1\mathrm{}h^n)(x^0,\mathrm{},x^n)=\\ \\ \tau (x^0h^1(x^1)\mathrm{}h^n(x^n)).\end{array}$$
## Hopf algebras, Renormalization and the Riemann-Hilbert problem
I have been for many years fascinated by those topics in theoretical physics which combine mathematical sophistication together with validation by experiments. A prominent example is Quantum Field Theory, not in its abstract formulation but in its computational power, as a mysterious new calculus, known as perturbative renormalization. It is heartening that some hard workers ( ) continue to dig in the bottom of that mine and actually find gold. I had the luck to meet one of them, Dirk Kreimer, and to join him in trying to unveil the secret beauty of these computations.
Dirk Kreimer showed ( ) that for any quantum field theory, the combinatorics of Feynman graphs is governed by a Hopf algebra $``$ whose antipode involves the same algebraic operations as in the Bogoliubov-Parasiuk-Hepp recursion and the Zimmermann forest formula.
His Hopf algebra is commutative as an algebra and we showed in that it is the dual Hopf algebra of the envelopping algebra of a Lie algebra $`\underset{¯}{G}`$ whose basis is labelled by the one particle irreducible Feynman graphs. The Lie bracket of two such graphs is computed from insertions of one graph in the other and vice versa. The corresponding Lie group $`G`$ is the group of characters of $``$.
We also showed that, using dimensional regularization, the bare (unrenormalized) theory gives rise to a loop
$$\gamma (z)G,zC$$
(58)
where $`C`$ is a small circle of complex dimensions around the integer dimension $`D`$ of space-time. Our main result ( ) which relies on all the previous work of Dirk is that the renormalized theory is just the evaluation at $`z=D`$ of the holomorphic part $`\gamma _+`$ of the Birkhoff decomposition of $`\gamma `$.
The Birkhoff decomposition is the factorization
$$\gamma (z)=\gamma _{}(z)^1\gamma _+(z)zC$$
(59)
where we let $`CP_1()`$ be a smooth simple curve, $`C_{}`$ the component of the complement of $`C`$ containing $`\mathrm{}C`$ and $`C_+`$ the other component. Both $`\gamma `$ and $`\gamma _\pm `$ are loops with values in $`G`$,
$$\gamma (z)Gz$$
and $`\gamma _\pm `$ are boundary values of holomorphic maps (still denoted by the same symbol)
$$\gamma _\pm :C_\pm G.$$
(60)
The normalization condition $`\gamma _{}(\mathrm{})=1`$ ensures that, if it exists, the decomposition (2) is unique (under suitable regularity conditions). It is intimately tied up to the classification of holomorphic $`G`$-bundles on the Riemann sphere $`P_1()`$ and for $`G=\mathrm{GL}_n()`$ to the Riemann-Hilbert problem. The Riemann-Hilbert problem comes from Hilbert’s $`21^{\mathrm{st}}`$ problem which he formulated as follows:
* “Prove that there always exists a Fuchsian linear differential equation with given singularities and given monodromy.”
In this form it admits a positive answer due to Plemelj and Birkhoff. When formulated in terms of linear systems of the form,
$$y^{}(z)=A(z)y(z),A(z)=\underset{\alpha S}{}\frac{A_\alpha }{z\alpha },$$
(61)
where $`S`$ is the given finite set of singularities, $`\mathrm{}S`$, the $`A_\alpha `$ are complex matrices such that
$$A_\alpha =0$$
(62)
to avoid singularities at $`\mathrm{}`$, the answer is not always positive (, ), but the solution exists when the monodromy matrices $`M_\alpha `$ are sufficiently close to 1. It can then be explicitly written as a series of polylogarithms.
For $`G=\mathrm{GL}_n()`$ the existence of the Birkhoff decomposition (2) is equivalent to the vanishing,
$$c_1(L_j)=0$$
(63)
of the Chern numbers $`n_j=c_1(L_j)`$ of the holomorphic line bundles of the Birkhoff-Grothendieck decomposition,
$$E=L_j$$
(64)
where $`E`$ is the holomorphic vector bundle on $`P_1()`$ associated to $`\gamma `$, i.e. with total space:
$$(C_+\times ^n)_\gamma (C_{}\times ^n).$$
(65)
When $`G`$ is a simply connected nilpotent complex Lie group the existence (and uniqueness) of the Birkhoff decomposition (2) is valid for any $`\gamma `$. When the loop $`\gamma :CG`$ extends to a holomorphic loop: $`C_+G`$, the Birkhoff decomposition is given by $`\gamma _+=\gamma `$, $`\gamma _{}=1`$. In general, for $`zC_+`$ the evaluation,
$$\gamma \gamma _+(z)G$$
(66)
is a natural principle to extract a finite value from the singular expression $`\gamma (z)`$. This extraction of finite values coincides with the removal of the pole part when $`G`$ is the additive group $``$ of complex numbers and the loop $`\gamma `$ is meromorphic inside $`C_+`$ with $`z`$ as its only singularity.
As I mentionned earlier our main result is that the renormalized theory is just the evaluation at $`z=D`$ of the holomorphic part $`\gamma _+`$ of the Birkhoff decomposition of the loop given by the unrenormalized theory $`\gamma `$.
We showed that the group $`G`$ is a semi-direct product of an easily understood abelian group by a highly non-trivial group closely tied up with groups of diffeomorphisms, thanks to the relation that we had uncovered in between the Hopf algebra of rooted trees and the Hopf algebra $``$ of section 9 involved in the computation of the index formula. The analysis of the relation between these two groups is intimately connected with the renormalization group and anomalous dimensions, this will be the content of our coming paper .
## Spectral Manifolds
Let us now turn to manifolds and explain by giving concrete examples the content of our characterization () of spectral triples associated to ordinary Riemannian manifolds. It will be crucial that it applies to any Riemannian metric with fixed volume form. What we shall show in particular is that even in that classical case there is a definite advantage in dealing with the slightly noncommutative algebra of matrices of functions. The pair given by the algebra and the Dirac operator is then the solution of a remarkably simple polynomial equation. We shall also give a very natural ”quantization” of the volume form of the manifolds which will appear most naturally in our examples, namely the spheres $`S^n`$ for n=1, 2 and 4.
Let us start with the simplest example, namely, let us show that the geometry of the circle $`S^1`$ of length $`2\pi `$ is completely specified by the presentation:
$$U^1[D,U]=1,\text{where}UU^{}=U^{}U=1.$$
(67)
Of course $`D`$ is as above an unbounded selfadjoint operator. We let $`𝒜`$ be the algebra of smooth functions of the single element $`U`$. One has $`S^1=\text{Spectrum}(𝒜)`$ as one easily checks using the invariance of the spectrum of $`U`$ by rotations implied by the above equation. Any element $`a`$ of $`𝒜`$ is of the form $`a=f(U)`$ and one has
$$[D,a]=U\left(\frac{}{U}f\right)(U)=g(U)$$
(68)
and thus,
$$[D,a]=\underset{X}{Sup}|g(x)|,g=U\frac{}{U}f.$$
(69)
This shows that the metric on $`S^1=\text{Spectrum}(𝒜)`$ given by (6.3) is the standard Riemannian metric of length $`2\pi `$. Let us now assume that $`ds=D^1`$ is an infinitesimal of order 1. It is easy to see that this holds iff the commutant of the algebra generated by $`U`$ and $`D`$ is finite dimensional. We then claim that
$$f|ds|=n\pi ^1f(x)\sqrt{g}𝑑xf𝒜$$
(70)
where the metric $`g`$ on $`S^1`$ is the above Riemannian metric and where the integer n is the index
$$n=Index(PUP),$$
(71)
where $`P`$ is the projector $`P=\frac{1+F}{2}`$, $`F=Sign(D)`$. This formula is simple to prove directly since it is enough to check it for irreducible pairs $`U,D`$ in which case the spectrum of $`D`$ is of the form,
$$Spec(D)=+\lambda $$
(72)
for some $`\lambda `$, while $`U`$ is the shift.
It is important for our later purpose to understand that it is a special case of the general index formula. Indeed both sides of 70 are translation invariant and the equality for $`f=1`$ follows from
$$Index(PUP)=1/2U^1[D,U]|ds|$$
(73)
which follows from the following expression for the n-dimensional Hochschild class of the Chern character of a spectral triple of dimension n,
$$\tau _n(a^0,\mathrm{},a^n)=a^0[D,a^1]\mathrm{}[D,a^n]|D|^na^j𝒜$$
where we insert a $`\gamma `$ in the even case. This formula is weaker than the local index formula of section 8 since it only gives the n-dimensional Hochschild class of the character, but it has the superiority to hold in full generality, with no assumption on the dimension spectrum. It is easy to use it to compute the index pairing with K-theory classes which come from the algebraic K-theory group $`K_n`$ since the Chern character of such classes is an n-dimensional Hochschild cycle. In the above toy example, $`U`$ defines an element in $`K_1(𝒜)`$ and its Chern character is the 1-dimensional Hochschild cycle $`U^1U`$ so that 73 follows.
Of course this toy example is a bit too simple, but the above K-theory discussion tells us how to proceed to higher dimension by relying on the formula (13) of section 4 for the Chern character and requiring the vanishing of the lower components.
It is crucial that we do not restrict ourselves to the homogeneous case.
We shall now show that all geometries with fixed total area on the 2-sphere $`S^2`$ are indeed described by the following even analogues of equation (1),
$$e\frac{1}{2}=0,\left(e\frac{1}{2}\right)[D,e][D,e]=\gamma $$
(74)
where, as above, $`D=D^{}`$ is an unbounded selfadjoint operator and $`e`$, $`e^{}=e`$, $`e^2=e`$ is a selfadjoint idempotent.
The right hand side of (74) namely $`\gamma `$, is the $`/2`$ grading of the Hilbert space $``$ which is a characteristic feature of even dimensions, as we saw above. One has,
$$\gamma ^2=1,\gamma =\gamma ^{},\gamma e=e\gamma ,D\gamma =\gamma D.$$
(75)
We still need to explain the symbol $`T`$ for any operator $`T`$ in $``$. We fix a subalgebra $`M()`$ isomorphic to $`M_2()`$ and let,
$$T=E_M(T)$$
(76)
where $`E_M`$ is the conditional expectation onto its commutant $`M^{}`$, given for instance as the integral over its unitary group of the conjugates $`uTu^{}`$ of $`T`$. We assume that $`D`$ and $`\gamma `$ commute with $`M`$,
$$DM^{},\gamma M^{}.$$
(77)
One has the factorization $`()=M_2()M^{}`$, and any $`T()`$ can be uniquely written as,
$$T=\epsilon _{ij}T^{ij}T^{ij}M^{}$$
(78)
where $`\epsilon _{ij}`$ are the usual matrix units in $`M_2()`$. We can apply (78) to $`T=e`$ and we let $`𝒜`$ be the algebra of operators generated by the components $`e^{ij}`$ of $`e`$. Let us show that $`𝒜`$ is abelian and is the algebra of functions on the 2-sphere $`S^2`$.
We let $`t=e^{11}`$, $`z=e^{12}`$ so that
$$e^{22}=1t,e^{21}=z^{}$$
(79)
using $`e\frac{1}{2}=0`$ and $`e=e^{}`$. Also $`t=t^{}`$ and $`0\underset{=}{<}t\underset{=}{<}\mathrm{\hspace{0.17em}1}`$ follow from $`e=e^{}`$ and $`e^2=e`$. Thus $`e=\left[\begin{array}{cc}t& z\\ z^{}& (1t)\end{array}\right]`$ and the equation $`e^2=e`$ means that $`t^2+zz^{}=t`$, $`tz+z(1t)=z`$, $`z^{}t+(1t)z^{}=z^{}`$, $`z^{}z+(1t)^2=(1t)`$. This shows that $`zz^{}=z^{}z`$ and that $`tz=zt`$ so that $`𝒜`$ is abelian.
It also shows that the joint spectrum $`X`$ of $`t`$ and $`z`$ in $`\times `$ is a compact subset of
$$\{(t,z)[0,1]\times ;(t^2t)+|z|^2=0\}=P_1().$$
(80)
Let us now compute the left hand side of (74) using $`e=\left[\begin{array}{cc}t& z\\ z^{}& (1t)\end{array}\right]`$ and the notation
$$dx=[D,x].$$
(81)
We just expand the product of matrices,
$$\left[\begin{array}{cc}\left(t\frac{1}{2}\right)& z\\ z^{}& \left(\frac{1}{2}t\right)\end{array}\right]\left[\begin{array}{cc}dt& dz\\ dz^{}& dt\end{array}\right]\left[\begin{array}{cc}dt& dz\\ dz^{}& dt\end{array}\right]$$
(82)
and take the sum of the diagonal elements. We get the terms,
$`\left(t{\displaystyle \frac{1}{2}}\right)(dtdt+dzdz^{})+z(dz^{}dtdtdz^{})`$
$`+`$ $`z^{}(dtdzdzdt)+\left({\displaystyle \frac{1}{2}}t\right)(dz^{}dz+dtdt)`$
$`=`$ $`\left(t{\displaystyle \frac{1}{2}}\right)(dzdz^{}dz^{}dz)+z(dz^{}dtdtdz^{})`$
$`+`$ $`z^{}(dtdzdzdt).`$
Thus the second equation (74) is equivalent to,
$`\left(t{\displaystyle \frac{1}{2}}\right)([D,z][D,z^{}][D,z^{}][D,z])+`$ (83)
$`z([D,z^{}][D,t][D,t][D,z^{}])+`$
$`z^{}([D,t][D,z][D,z][D,t])=\gamma .`$
Equivalently we can write it as,
$$\pi (c)=\gamma $$
(84)
where $`c`$ is the Hochschild 2-cycle,
$$cZ_2(𝒜,𝒜)$$
(85)
given by the formula,
$$c=\left(t\frac{1}{2}\right)(zz^{}z^{}z)+z(z^{}ttz^{})+z^{}(tzzt)$$
(86)
and where $`\pi `$ is the canonical map from Hochschild chains to operators (),
$$\pi (a^0a^1\mathrm{}a^n)=a^0[D,a^1]\mathrm{}[D,a^n]a^j𝒜.$$
(87)
We let $`vC^{\mathrm{}}(S^2,^2T^{})`$ be the 2-form on $`S^2=P_1()`$ associated to the Hochschild class of $`c`$ (). It is given up to normalization by,
$$v=\frac{1}{12t}dzd\overline{z}$$
(88)
and vanishes nowhere on $`S^2`$.
We shall now show that any Riemannian metric $`g`$ on $`S^2`$ whose associated volume form is equal to $`v`$,
$$\sqrt{g}d^2x=v$$
(89)
gives canonically a solution to our equations (74)–(77).
It is very important for our later considerations on gravity that not only the round metric but all possible metrics fulfilling (89) actually appear as solutions.
The solution associated to a given metric $`g`$ fulfilling (89) is constructed as follows, one lets
$$=L^2(S^2,S)^2$$
(90)
be the direct sum of two copies of the space of $`L^2`$ spinors on $`S^2`$. The algebra $`M`$ is just,
$$M=M_2().$$
(91)
The operator $`D`$ is given by,
$$D=/1$$
(92)
where $`/`$ is the Dirac operator (of the metric $`g`$). Finally the $`/2`$ grading is
$$\gamma =\gamma _51$$
(93)
where $`\gamma _5`$ is the chirality operator on spinors. We identify $`S^2`$ with $`P_1()`$ which is the space
$$P_1()=\{xM_2(),x^2=x=x^{},\text{trace}x=1\}$$
(94)
and we let
$$eC^{\mathrm{}}(S^2)M_2()$$
(95)
be the corresponding selfadjoint idempotent in $``$ where $`C^{\mathrm{}}(S^2)`$ is acting by multiplication operators in $`L^2(S^2,S)`$.
One has (75)–(77) by construction as well as $`e\frac{1}{2}=0`$ using (94).
Let us check the second equality of (74), or rather the equivalent form (17). For any $`f𝒜=C^{\mathrm{}}(S^2)`$ one has,
$$[D,f]=df1$$
(96)
where $`df=[/,f]`$ is the Clifford multiplication by the differential of the function $`f`$.
For any $`f^0,f^1,f^2𝒜`$ one has,
$$f^0([D,f^1][D,f^2][D,f^2][D,f^1])=\rho \gamma $$
(97)
where $`\rho `$ is the smooth function such that
$$f^0df^1df^2=\rho \sqrt{g}d^2x$$
(98)
where $`\sqrt{g}d^2x`$ is the volume form of the metric $`g`$. By (89) we have $`\sqrt{g}d^2x=v`$ and by construction of $`v`$ as the 2-form associated to the class of $`C`$ we get from (97), (98) that
$$\pi (c)=\rho \gamma ,v=\rho v$$
(99)
i.e. $`\rho =1`$.
This is enough to check that any Riemannian metric $`g`$ on $`S^2`$ with volume form equal to $`v`$ does give a solution of equations (74)–(77).
To establish the converse one still needs technical assumptions in order to use the theorem of (), the main additional hypothesis being the order one condition which requires,
$$[[D,e^{ij}],e^k\mathrm{}]=0i,j,k,\mathrm{}.$$
(100)
Let us show now that the index formula (4) admits a perfect analogue in the general framework of solutions of (74)–(77), assuming the following control of the dimension,
$$ds=D^1\text{is of order}\frac{1}{2},$$
(101)
i.e. the $`n^{\mathrm{th}}`$ characteristic value $`\mu _n(D^1)`$ is of order of $`n^{1/2}`$ as $`n\mathrm{}`$.
One has $`eM_2(𝒜)`$ and the chern character of $`e`$ in the cyclic homology bicomplex $`(b,B)`$ is given by its components,
$$\left(e\frac{1}{2}\right)\underset{2n}{\underset{}{e\mathrm{}e}}=\mathrm{ch}_{2n}(e)$$
(102)
where the $``$ means that we take the $`M_2()`$ trace of the corresponding elements.
Let us recall the index formula,
$$\text{Index}D_e^+=\mathrm{ch}(e),\mathrm{ch}(D)$$
(103)
which computes the index of the compression $`eD^+e`$ of $`D^+:\frac{1+\gamma }{2}\frac{1\gamma }{2}`$, in terms of the pairing between cyclic homology and cyclic cohomology. In general this requires the full knowledge of the chern character $`\mathrm{ch}(D)`$ in cyclic cohomology.
However in our case (74) shows that $`\mathrm{ch}_0(e)=0`$, so that $`\mathrm{ch}_2(e)`$ is a Hochschild cycle. Moreover by (101) all the higher components of $`\mathrm{ch}(D)`$ vanish and () its component of degree 2, $`\mathrm{ch}_2(D)`$ has a Hochschild class given by,
$$\tau _2(a^0,a^1,a^2)=\gamma a^0[D,a^1][D,a^2]D^2.$$
(104)
The integral $``$ is a trace and when specializing (104) to $`a^j=e`$ we can replace the integrand by its average $`\left(e\frac{1}{2}\right)[D,e][D,e]D^2=\gamma D^2`$.
Since $`\gamma ^2=1`$ we thus obtain,
$$ds^2=\text{Index}D_e^+.$$
(105)
In particular the area, taken in suitable units, is “quantized” by this equation since the index is always an integer.
This simple fact will take more meaning in the 4-dimensional case where the Einstein-Hilbert action will appear.
To close the discussion of this 2-dimensional example we note that the natural algebra here is not $`𝒜`$ but rather $`M_2(𝒜)`$ which admits an amazingly simple presentation. It is generated by $`M_2()`$ and $`e`$ with the only relations
$$e=e^{},e^2=e,e\frac{1}{2}=0$$
(106)
where $``$ is the conditional expectation on the commutant of the subalgebra $`M_2()`$.
Indeed the above computations show that the $`C^{}`$ algebra generated by $`M_2()`$ and $`e`$ with the relations (106) is,
$$C(S^2,M_2())=C(S^2)M_2().$$
(107)
Let us now move on to the 4-dimensional case.
We first determine the $`C^{}`$ algebra generated by $`M_4()`$ and a projection $`e=e^{}`$ such that $`e\frac{1}{2}=0`$ as above and whose matrix expression (78) is of the form,
$$[e^{ij}]=\left[\begin{array}{cc}q_{11}& q_{12}\\ q_{21}& q_{22}\end{array}\right]$$
(108)
where each $`q_{ij}`$ is a $`2\times 2`$ matrix of the form,
$$q=\left[\begin{array}{cc}\alpha & \beta \\ \beta ^{}& \alpha ^{}\end{array}\right].$$
(109)
Since $`e=e^{}`$, both $`q_{11}`$ and $`q_{22}`$ are selfadjoint, moreover since $`e\frac{1}{2}=0`$, we can find $`t=t^{}`$ such that,
$$q_{11}=\left[\begin{array}{cc}t& 0\\ 0& t\end{array}\right],q_{22}=\left[\begin{array}{cc}(1t)& 0\\ 0& (1t)\end{array}\right].$$
(110)
We let $`q_{12}=\left[\begin{array}{cc}\alpha & \beta \\ \beta ^{}& \alpha ^{}\end{array}\right]`$, we then get from $`e=e^{}`$,
$$q_{21}=\left[\begin{array}{cc}\alpha ^{}& \beta \\ \beta ^{}& \alpha \end{array}\right].$$
(111)
We thus see that the commutant $`𝒜`$ of $`M_4()`$ is generated by $`t,\alpha ,\beta `$ and we need to find the relations imposed by the equality $`e^2=e`$.
In terms of $`e=\left[\begin{array}{cc}t& q\\ q^{}& 1t\end{array}\right]`$, the equation $`e^2=e`$ means that $`t^2t+qq^{}=0`$, $`t^2t+q^{}q=0`$ and $`[t,q]=0`$. This shows that $`t`$ commutes with $`\alpha `$, $`\beta `$, $`\alpha ^{}`$ and $`\beta ^{}`$ and since $`qq^{}=q^{}q`$ is a diagonal matrix
$$\alpha \alpha ^{}=\alpha ^{}\alpha ,\alpha \beta =\beta \alpha ,\alpha ^{}\beta =\beta \alpha ^{},\beta \beta ^{}=\beta ^{}\beta $$
(112)
so that the $`C^{}`$ algebra $`𝒜`$ is abelian, with the only further relation, (besides $`t=t^{}`$),
$$\alpha \alpha ^{}+\beta \beta ^{}+t^2t=0.$$
(113)
This is enough to check that,
$$𝒜=C(S^4)$$
(114)
where $`S^4`$ appears naturally as quaternionic projective space,
$$S^4=P_1().$$
(115)
The original $`C^{}`$ algebra is thus,
$$B=C(S^4)M_4().$$
(116)
The analogue of (74) is,
$$\left(e\frac{1}{2}\right)[D,e]^{2n}=0,n=0,1\text{and}=\gamma \text{for}n=2.$$
(117)
As above we assume,
$$DM^{},\gamma M^{}$$
(118)
where $`M=M_4()`$ is the algebra of $`4\times 4`$ matrices.
We shall first check by a direct computation that the equality $`\left(e\frac{1}{2}\right)[D,e]^2=0`$ is automatic with our choice of $`e`$ (108). We use (81) for notational convenience and first compute exactly as in (82), with $`z`$ replaced by $`q=\left[\begin{array}{cc}\alpha & \beta \\ \beta ^{}& \alpha ^{}\end{array}\right]`$. We thus get,
$`(e1/2)[D,e]^2`$ $`=(t{\displaystyle \frac{1}{2}})(dqdq^{}dq^{}dq)`$
$`+q(dq^{}dtdtdq^{})+q^{}(dtdqdqdt)`$
where the expectation in the right hand side is relative to $`M_2()`$.
The diagonal elements of $`\omega =dqdq^{}`$ are
$$\omega _{11}=d\alpha d\alpha ^{}+d\beta d\beta ^{},\omega _{22}=d\beta ^{}d\beta +d\alpha ^{}d\alpha $$
while for $`\omega ^{}=dq^{}dq`$ we get,
$$\omega _{11}^{}=d\alpha ^{}d\alpha +d\beta d\beta ^{},\omega _{22}^{}=d\beta ^{}d\beta +d\alpha d\alpha ^{}.$$
It follows that, since $`t`$ is diagonal,
$$\left(t\frac{1}{2}\right)(dqdq^{}dq^{}dq)=0.$$
(120)
The diagonal elements of $`qdq^{}dt=\rho `$ are
$$\rho _{11}=\alpha d\alpha ^{}dt+\beta d\beta ^{}dt,\rho _{22}=\beta ^{}d\beta dt+\alpha ^{}d\alpha dt$$
while for $`\rho ^{}=q^{}dqdt`$ they are
$$\rho _{11}^{}=\alpha ^{}d\alpha dt+\beta d\beta ^{}dt,\rho _{22}^{}=\beta ^{}d\beta dt+\alpha d\alpha ^{}dt.$$
Similarly for $`\sigma =qdtdq^{}`$ and $`\sigma ^{}=q^{}dtdq`$ one gets the required cancellations so that,
$$\left(e\frac{1}{2}\right)[D,e]^2=0,$$
(121)
holds irrespective of the operator $`D`$ fulfilling (118).
As in (88) we let $`v`$ be the natural volume form on $`S^4`$ given by,
$$v=\frac{1}{12t}d\alpha d\overline{\alpha }d\beta d\overline{\beta }.$$
(122)
We shall now show that any Riemannian metric $`g`$ on $`S^4`$ whose associated volume form is $`v`$ gives a solution to (117), (118), thus,
$$\sqrt{g}d^4x=v.$$
(123)
For this we proceed exactly as in (90)–(99) above and we need to check that the Hochschild cycle $`c`$ obtained in the computation of
$$\left(e\frac{1}{2}\right)[D,e]^4=\pi (c)$$
(124)
is totally antisymmetric, i.e. of the form,
$$c=\underset{i,\sigma }{}\epsilon (\sigma )a_0^ia_{\sigma (1)}^i\mathrm{}a_{\sigma (4)}^i$$
(125)
where $`\sigma `$ ranges through all 24 permutations of $`\{1,\mathrm{},4\}`$. With the above notations one has,
$$\left(e\frac{1}{2}\right)[D,e]^4=\left[\begin{array}{cc}t\frac{1}{2}& q\\ q^{}& \frac{1}{2}t\end{array}\right]\left[\begin{array}{cc}dt& dq\\ dq^{}& dt\end{array}\right]^4$$
(126)
and the sum of the diagonal elements is,
$`\left(t{\displaystyle \frac{1}{2}}\right)((dt^2+dqdq^{})^2+(dtdqdqdt)(dq^{}dtdtdq^{}))`$
$``$ $`\left(t{\displaystyle \frac{1}{2}}\right)((dt^2+dq^{}dq)^2+(dq^{}dtdtdq^{})(dtdqdqdt))`$
$`+`$ $`q((dq^{}dtdtdq^{})(dt^2+dqdq^{})+(dq^{}dq+dt^2)(dq^{}dtdtdq^{}))`$
$`+`$ $`q^{}((dt^2+dqdq^{})(dtdqdqdt)+(dtdqdqdt)(dq^{}dq+dt^2)).`$
Since $`t`$ and $`dt`$ are diagonal $`2\times 2`$ matrices of operators and the same diagonal terms appear in $`dqdq^{}`$ and $`dq^{}dq`$ as we saw in the proof of (120), the first two lines only contribute by,
$$\left(t\frac{1}{2}\right)(dqdq^{}dqdq^{}dq^{}dqdq^{}dq).$$
(127)
Similarly the two last lines only contribute by,
$`q^{}(dtdqdq^{}dqdqdtdq^{}dq+dqdq^{}dtdqdqdq^{}dqdt)`$
$``$ $`q(dtdq^{}dqdq^{}dq^{}dtdqdq^{}+dq^{}dqdtdq^{}dq^{}dqdq^{}dt).`$
The direct computation of (127) then gives
$$\epsilon (\sigma )\left(t\frac{1}{2}\right)da_{\sigma (1)}^0da_{\sigma (2)}^0da_{\sigma (3)}^0da_{\sigma (4)}^0$$
(129)
where $`a_1^0=\alpha `$, $`a_2^0=\overline{\alpha }`$, $`a_3^0=\beta `$, $`a_4^0=\overline{\beta }`$.
The direct computation of (78) gives
$$\underset{i,\sigma }{}\epsilon (\sigma )a_0^ida_{\sigma (1)}^ida_{\sigma (2)}^ida_{\sigma (3)}^ida_{\sigma (4)}^i$$
(130)
where $`i\{1,2,3,4\}`$ and,
$`a_0^1=\alpha ,a_1^1=t,a_2^1=\overline{\alpha },a_3^1=\overline{\beta },a_4^1=\beta `$
$`a_0^2=\overline{\alpha },a_1^2=t,a_2^2=\alpha ,a_3^2=\beta ,a_4^2=\overline{\beta }`$
$`a_0^3=\beta ,a_1^3=t,a_2^3=\overline{\beta },a_3^3=\overline{\alpha },a_4^3=\alpha `$
$`a_0^4=\overline{\beta },a_1^4=t,a_2^4=\beta ,a_3^4=\alpha ,a_4^4=\overline{\alpha }.`$
We thus obtain the required formula for the cycle $`c`$. When $`dx=[D,x]`$ with $`D`$ the Dirac operator associated to a Riemannian metric $`g`$ on $`S^4`$ we get as above, using the Clifford algebra, that
$$\pi (c)=\rho \gamma $$
(131)
where $`\rho `$ is the smooth function such that
$$\omega =\rho \sqrt{g}d^4x$$
(132)
where $`\omega `$ is the differential form associated to $`c`$. Now, up to normalization one has,
$`\omega `$ $`=\left(t{\displaystyle \frac{1}{2}}\right)d\alpha d\overline{\alpha }d\beta d\overline{\beta }\alpha dtd\overline{\alpha }d\beta d\overline{\beta }`$
$`+\overline{\alpha }dtd\alpha d\beta d\overline{\beta }\beta dtd\overline{\beta }d\alpha d\overline{\alpha }`$
$`+\overline{\beta }dtd\beta d\alpha d\overline{\alpha },`$
which using $`t^2t+\alpha \overline{\alpha }+\beta \overline{\beta }=0`$ gives up to a factor 2,
$$\omega =\frac{1}{2t1}d\alpha d\overline{\alpha }d\beta d\overline{\beta }.$$
(133)
Thus by hypothesis on $`g`$ we get $`\rho =1`$ and $`\pi (c)=\gamma `$ which by the above computation means,
$$\left(e\frac{1}{2}\right)[D,e]^4=\gamma .$$
(134)
This shows that any Riemannian structure, with the given volume form on $`M=S^4`$, does give us a solution to our basic equation. Conversely exactly as in the two dimensional case we get, provided that $`ds=D^1`$ is of order $`\frac{1}{4}`$,
$$ds^4=\text{Index}D_e^+.$$
(135)
In particular the 4-dimensional volume, taken in suitable units, is “quantized” by this equation since the index is always an integer.
Let $`\pi =(e,D,\gamma )`$ be a solution of equations (108) (117) (118) above and let us assume (100), together with harmless regularity conditions, . Then there exists a unique Riemannian structure $`g`$ on $`M`$ such that the geodesic distance is given by
$$d(x,y)=Sup\{|a(x)a(y)|;a𝒜,[D,a]1\}.$$
The metric $`g=g(\pi )`$ depends only on the unitary equivalence class of $`\pi `$. The fiber of the map $`\{`$unitary equivalence classes$`\}g(\pi )`$ is an affine space $`𝒜`$ on which the functional $`ds^2`$ is a positive quadratic form with a unique real minimum $`\pi _0`$ which is the representation described above in $`L^2(S^4,S)`$ given by multiplication operators and the Dirac operator associated to the Levi–Civita connection of the metric $`g`$.
The value of $`ds^2`$ on $`\pi _o`$ is the Hilbert–Einstein action of the metric $`g`$,
$$ds^2=(48\pi ^2)^1r\sqrt{g}d^4x,.$$
We use the convention that the scalar curvature $`r`$ is positive for the round sphere $`S^4`$, in particular, the sign of the action $`ds^2`$ is the correct one for the Euclidean formulation of gravity. We refer to , , for detailed computations.
## Noncommutative Spectral Manifolds
The main nuance in passing to the noncommutative case is that, since the diagonal in the square no longer corresponds to an algebra homomorphism (the map $`xyxy`$ is no longer an algebra homomorphism), the algebra $`𝒜𝒜^0`$ now plays a central role.
The fundamental class of a noncommutative space (cf ), is a class $`\mu `$ in the $`KR`$–homology of the algebra $`𝒜𝒜^0`$ equipped with the involution
$$\tau (xy^0)=y^{}(x^{})^0x,y𝒜$$
(136)
where $`𝒜^0`$ denotes the algebra opposite to $`𝒜`$. The $`KR`$-homology cycle representing $`\mu `$ is given by a spectral triple, as above, equipped with an anti-linear isometry $`J`$ on $``$ which implements the involution $`\tau `$,
$$JwJ^1=\tau (w)w𝒜𝒜^0,$$
(137)
Instead of giving the action of the algebra $`𝒜𝒜^0`$ in $``$ one can equivalently give an action of $`𝒜`$ satisfying the commutation rule, $`[a,b^0]=0a,b𝒜`$ where
$$b^0=Jb^{}J^1b𝒜$$
(138)
$`KR`$-homology ( ) is periodic with period $`8`$ and the dimension modulo $`8`$ is specified by the following commutation rules. One has $`J^2=\epsilon `$, $`JD=\epsilon ^{}DJ`$, $`J\gamma =\epsilon ^{\prime \prime }\gamma J`$ where $`\epsilon ,\epsilon ^{},\epsilon ^{\prime \prime }\{1,1\}`$ and with $`n`$ the dimension modulo 8,
| n | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`\epsilon `$ | 1 | 1 | -1 | -1 | -1 | -1 | 1 | 1 |
| $`\epsilon ^{}`$ | 1 | -1 | 1 | 1 | 1 | -1 | 1 | 1 |
| $`\epsilon ^{\prime \prime }`$ | 1 | | -1 | | 1 | | -1 | |
The anti-linear isometry $`J`$ is given in Riemannian geometry by the charge conjugation operator and in the noncommutative case by the Tomita-Takesaki antilinear conjugation operator . Given an involutive algebra of operators $`𝒜`$ on the Hilbert space $``$, the Tomita-Takesaki theory associates to all vectors $`\xi `$, cyclic for $`𝒜`$ and for its commutant $`𝒜^{}`$
$$\overline{𝒜\xi }=,\overline{𝒜^{}\xi }=$$
(139)
an anti-linear isometric involution $`J:`$ obtained from the polar decomposition of the operator
$$Sa\xi =a^{}\xi a𝒜.$$
(140)
It satisfies the following commutation relation:
$$J𝒜^{\prime \prime }J^1=𝒜^{}.$$
(141)
In particular $`[a,b^0]=0a,b𝒜`$ where
$$b^0=Jb^{}J^1b𝒜$$
(142)
so $``$ becomes an $`𝒜`$-bimodule using the representation of the opposite algebra. The class $`\mu `$ specifies only the stable homotopy class of the spectral triple $`(𝒜,,D)`$ equipped with the isometry $`J`$ (and $`/2`$–grading $`\gamma `$ if $`n`$ is even). The non-triviality of this homotopy class shows up in the intersection form
$$K_{}(𝒜)\times K_{}(𝒜)$$
which is obtained from the Fredholm index of $`D`$ with coefficients in $`K_{}(𝒜𝒜^0)`$. Note that it is defined without using the diagonal map $`m:𝒜𝒜𝒜`$, which is not a homomorphism in the noncommutative case. This form is quadratic or symplectic according to the value of $`n`$ modulo $`8`$.
The Kasparov intersection product allows to formulate the Poincaré duality in terms of the invertibility of $`\mu `$,
$$\beta KR_n(𝒜^0𝒜),\beta _𝒜\mu =id_{𝒜^0},\mu _{𝒜^0}\beta =id_𝒜.$$
It implies the isomorphism $`K_{}(𝒜)\stackrel{\mu }{}K^{}(𝒜)`$.
The condition that D is an operator of order one becomes
$$[[D,a],b^0]=0a,b𝒜.$$
(Notice that since $`a`$ and $`b^0`$ commute this condition is equivalent to $`[[D,a^0],b]=0a,b𝒜`$.)
One can show that the von Neumann algebra $`𝒜^{\prime \prime }`$ generated by $`𝒜`$ in $``$ is automatically finite and hyperfinite and there is a complete list of such algebras up to isomorphism as we saw in section 2. The algebra $`𝒜`$ is stable under smooth functional calculus in its norm closure $`A=\overline{𝒜}`$ so that $`K_j(𝒜)K_j(A)`$, i.e. $`K_j(𝒜)`$ depends only on the underlying topology (defined by the $`C^{}`$ algebra $`A`$). The integer $`\chi =\mu ,\beta `$ gives the Euler characteristic in the form
$$\chi =RangK_0(𝒜)RangK_1(𝒜)$$
and the general operator theoretic index formula of section 8 gives a local formula for $`\chi `$.
The group $`Aut^+(𝒜)`$ of automorphisms $`\alpha `$ of the involutive algebra $`𝒜`$, which are implemented by a unitary operator $`U`$ in $``$ commuting with $`J`$,
$$\alpha (x)=UxU^1x𝒜,$$
plays the role of the group $`Diff^+(M)`$ of diffeomorphisms preserving the K-homology fundamental class for a manifold $`M`$.
In the general noncommutative case, parallel to the normal subgroup $`Int𝒜Aut𝒜`$ of inner automorphisms of $`𝒜`$,
$$\alpha (f)=ufu^{}f𝒜$$
(143)
where $`u`$ is a unitary element of $`𝒜`$ (i.e. $`uu^{}=u^{}u=1`$), there exists a natural foliation of the space of spectral geometries on $`𝒜`$ by equivalence classes of inner deformations of a given geometry. To understand how they arise we need to understand how to transfer a given spectral geometry to a Morita equivalent algebra. Given a spectral triple $`(𝒜,,D)`$ and the Morita equivalence between $`𝒜`$ and an algebra $``$ where
$$=End_𝒜()$$
(144)
where $``$ is a finite, projective, hermitian right $`𝒜`$–module, one gets a spectral triple on $``$ by the choice of a hermitian connection on $``$. Such a connection $``$ is a linear map $`:_𝒜\mathrm{\Omega }_D^1`$ satisfying the rules ()
$$(\xi a)=(\xi )a+\xi da\xi ,a𝒜$$
(145)
$$(\xi ,\eta )(\xi ,\eta )=d(\xi ,\eta )\xi ,\eta $$
(146)
where $`da=[D,a]`$ and where $`\mathrm{\Omega }_D^1()`$ is the $`𝒜`$–bimodule of operators of the form
$$A=\mathrm{\Sigma }a_i[D,b_i],a_i,b_i𝒜.$$
(147)
Any algebra $`𝒜`$ is Morita equivalent to itself (with $`=𝒜`$) and when one applies the above construction in the above context one gets the inner deformations of the spectral geometry.
Such a deformation is obtained by the following formula (with suitable signs depending on the dimension mod 8) without modifying neither the representation of $`𝒜`$ in $``$ nor the anti-linear isometry $`J`$
$$DD+A+JAJ^1$$
(148)
where $`A=A^{}`$ is an arbitrary selfadjoint operator of the form 147. The action of the group $`Int(𝒜)`$ on the spectral geometries is simply the following gauge transformation of $`A`$
$$\gamma _u(A)=u[D,u^{}]+uAu^{}.$$
(149)
The required unitary equivalence is implemented by the following representation of the unitary group of $`𝒜`$ in $``$,
$$uuJuJ^1=u(u^{})^0.$$
(150)
The transformation (148) is the identity in the usual Riemannian case. To get a nontrivial example it suffices to consider as we did in section 11, the product of a Riemannian triple by the unique spectral geometry on the finite-dimensional algebra $`𝒜_F=M_N()`$ of $`N\times N`$ matrices on $``$, $`N2`$. One then has $`𝒜=C^{\mathrm{}}(M)𝒜_F`$, $`Int(𝒜)=C^{\mathrm{}}(M,PSU(N))`$ and inner deformations of the geometry are parameterized by the gauge potentials for the gauge theory of the group $`SU(N)`$. The space of pure states of the algebra $`𝒜`$, $`P(𝒜)`$, is the product $`P=M\times P_{N1}()`$ and the metric on $`P(𝒜)`$ determined by the formula (6.3) depends on the gauge potential $`A`$. It coincide with the Carnot metric on $`P`$ defined by the horizontal distribution given by the connection associated to $`A`$. The group $`Aut(𝒜)`$ of automorphisms of $`𝒜`$ is the following semi–direct product
$$Aut(𝒜)=𝒰>Diff^+(M)$$
(151)
of the local gauge transformation group $`Int(𝒜)`$ by the group of diffeomorphisms. In dimension $`n=4`$, the Hilbert–Einstein action functional for the Riemannian metric and the Yang–Mills action for the vector potential $`A`$ appear with the correct signs in the asymptotic expansion for large $`\mathrm{\Lambda }`$ of the number $`N(\mathrm{\Lambda })`$ of eigenvalues of $`D`$ which are $`\mathrm{\Lambda }`$ (cf. ),
$$N(\mathrm{\Lambda })=\mathrm{\#}\text{eigenvalues of }D\text{ in}[\mathrm{\Lambda },\mathrm{\Lambda }].$$
(152)
This step function $`N(\mathrm{\Lambda })`$ is the superposition of two terms,
$$N(\mathrm{\Lambda })=N(\mathrm{\Lambda })+N_{\mathrm{osc}}(\mathrm{\Lambda }).$$
The oscillatory part $`N_{\mathrm{osc}}(\mathrm{\Lambda })`$ is the same as for a random matrix, governed by the statistic dictated by the symmetries of the system and does not concern us here. The average part $`N(\mathrm{\Lambda })`$ is computed by a semiclassical approximation from local expressions involving the familiar heat equation expansion. Other nonzero terms in the asymptotic expansion are cosmological, Weyl gravity and topological terms. As we saw above in our characterization of section 11 we are only dealing with metrics with a fixed volume form so that the bothering cosmological term does not enter in the variational equations associated to the spectral action $`N(\mathrm{\Lambda })`$. It is tempting to speculate that the phenomenological Lagrangian of physics, combining matter and gravity appears from the solution of an extremely simple operator theoretic equation along the lines described above. As a starting point for such investigations see .
## Noncommutative Tori
A more sophisticated example of a spectral manifold is provided by the noncommutative torus $`𝕋_\theta ^2`$. The parameter $`\theta /`$ defines the following deformation of the algebra of smooth functions on the torus $`𝕋^2`$, with generators $`U,V`$. The relations
$$VU=\mathrm{exp}2\pi i\theta UV\text{and }UU^{}=U^{}U=1,VV^{}=V^{}V=1$$
(153)
define the presentation of the involutive algebra $`𝒜_\theta =\{\mathrm{\Sigma }a_{n,m}U^nV^n;a=(a_{n,m})𝒮(^2)\}`$ where $`S(^2)`$ is the Schwartz space of sequences with rapid decay. We shall first describe a completely canonical procedure for constructing the $`K`$-cycle $`(\text{H},D,\gamma )`$ over $`𝒜_\theta `$ from the fundamental class in cyclic cohomology, i.e., the choice of orientation, and the formal positive element
$$G=dU(dU)^{}+dV(dV)^{}\mathrm{\Omega }_+^2(𝒜_\theta ),$$
(154)
which specifies the metric in the naive classical sense.
This transition from the $`g_{\mu \nu }`$ to the spectral triple extends in principle to arbitrary formal metrics $`G\mathrm{\Omega }_+^2(𝒜_\theta )`$ but we stick to this specific flat example for simplicity. The construction will be possible thanks to the noncommutative analogue of the Polyakov action of string theory.
We need first to explain briefly how this works in the commutative case. The basic data is the fundamental class in cyclic cohomology, and the formal positive element
$$G=\underset{\mu ,\nu =1}{\overset{d}{}}g_{\mu \nu }dx^\mu (dx^\nu )^{}\mathrm{\Omega }_+^2(𝒜),$$
(155)
The first key notion is that of positivity in Hochschild cohomology. By definition (cf. ) a Hochschild cocycle $`\psi `$ on a $``$-algebra $`𝒜`$ is positive if it has even dimension $`n=2m`$ and the following equality defines a positive sesquilinear form on the vector space $`𝒜^{(m+1)}`$:
$$a^0a^1\mathrm{}a^m,b^0b^1\mathrm{}b^m=\psi (b^0a^0,a^1,\mathrm{},a^m,b^m,\mathrm{},b^1)$$
(156)
for any $`a^j,b^j𝒜`$.
In general the positive Hochschild cocycles form a convex cone
$$Z_+^n(𝒜,𝒜^{})Z^n(𝒜,𝒜^{})$$
(157)
in the vector space $`Z^n`$ of Hochschild cocycles on $`𝒜`$.
Let $`M`$ be a 2-dimensional oriented compact manifold, $`𝒜`$ be the $``$-algebra of smooth functions on $`M`$ and take for the class $`C`$ the fundamental class, i.e. the class of the de Rham current $`C`$
$$C,f^0df^1df^2=\frac{1}{2\pi i}_Mf^0𝑑f^1df^2f^jC^{\mathrm{}}(M).$$
(158)
There is a natural correspondence between conformal structures on $`M`$ and extreme points of $`Z_+^2C`$. Thus, let $`g`$ be a conformal structure on $`M`$ or equivalently, since $`M`$ is oriented, a complex structure. Then, to the Lelong notion of positive current corresponds the positivity in the above sense of the following Hochschild 2-cocycle:
$$\phi _g(f^0,f^1,f^2)=\frac{i}{\pi }_Mf^0f^1\overline{}f^2,$$
(159)
where $``$ and $`\overline{}`$ are inherited from the complex structure. The mapping $`g\phi _g`$ is an injection, since one can read off from $`\phi _g`$ what it means for a function $`f`$ to be holomorphic in a given small open set $`UM`$. Each $`\phi _g`$ is an extreme point of the convex set $`Z_+^2C`$, and, conversely, the exposed points of this convex set can be determined as follows: for any element of the dual cone $`(Z_+^2)^{}`$ of $`Z_+^2`$, of the form
$$G=\underset{\mu ,\nu =1}{\overset{d}{}}g_{\mu \nu }dx^\mu (dx^\nu )^{}\mathrm{\Omega }^2(𝒜),$$
(160)
where $`g_{\mu \nu }`$ is a positive element of $`M_d(𝒜)`$, one can show, assuming a suitable condition of nondegeneracy, that the linear form
$$G,\phi =\phi (g_{\mu \nu },x^\mu ,(x^\nu )^{})$$
(161)
attains its minimum at a unique point in $`Z_+^2C`$, and that this point is equal to $`\phi _g`$, where $`g`$ is the conformal structure on $`M`$ associated with the classical Riemannian metric
$$g=g_{\mu \nu }dx^\mu (dx^\nu )^{}.$$
(162)
This allows us to understand the complex structures on $`M`$ as the solutions of a variational problem involving the fundamental class of $`M`$ and positivity in Hochschild cohomology. This problem is by no means restricted in its formulation to the commutative case, but it requires the notion of fundamental class in cyclic cohomology. It can be taken as a starting point for developing complex geometry in the noncommutative case.
Let us now show that the previous considerations extend without change to the noncommutative case and treat the noncommutative torus from a metric point of view.
The cyclic cohomology group $`HC^0(𝒜_\theta )`$ is 1-dimensional and is generated by the unique trace $`\tau _0`$ of $`𝒜_\theta `$,
$$\tau _0\left(a_{n,m}U^nV^m\right)=a_{0,0},$$
(163)
whereas the cyclic cohomology $`HC^2(𝒜_\theta )`$ is two dimensional and besides $`S\tau _0HC^2`$ (where $`S`$ is the periodicity operator in cyclic cohomology), is generated by the class of the cyclic 2-cocycle
$$\tau _2(a^0,a^1,a^2)=2\pi i\underset{\genfrac{}{}{0pt}{}{n_0+n_1+n_2=0}{m_0+m_1+m_2=0}}{}(n_1m_2n_2m_1)a_{n_0,m_0}^0a_{n_1,m_1}^1a_{n_2,m_2}^2.$$
(164)
Note that only the class of this cocycle matters, not the above specific representative. This nuance is very important since the above class only involves the smooth algebra $`𝒜_\theta `$; we shall now fix the metric.
$$G=dU(dU)^{}+dV(dV)^{}\mathrm{\Omega }_+^2(𝒜_\theta ).$$
(165)
On the intersection of the cyclic cohomology class $`\tau _2+b(\mathrm{Ker}B)`$ with the positive cone $`Z_+^2`$ in Hochschild cohomology, the functional $`G`$ defined by
$$\phi Z^2G,\phi =\phi (1,U,U^{})+\phi (1,V,V^{})$$
(166)
reaches its minimum at a unique point $`\phi _2`$ given by
$$\phi _2(a^0,a^1,a^2)=2\pi \underset{\genfrac{}{}{0pt}{}{n_0+n_1+n_2=0}{m_0+m_1+m_2=0}}{}(n_1im_1)(n_2im_2)a_{n_0,m_0}^0a_{n_1,m_1}^1a_{n_2,m_2}^2.$$
(167)
We then use the noncommutative analogue of a conformal structure, i.e., the positive cocycle $`\phi _2`$ together with the trace $`\tau _0`$, to construct the analogue of the Dirac operator for $`𝒜_\theta `$, that is, we shall obtain a $`(2,\mathrm{})`$-summable $`K`$-cycle $`(\text{H},D)`$ on $`𝒜_\theta `$. The Hilbert space H is the direct sum $`\text{H}=\text{H}^+\text{H}^{}`$ of the Hilbert space $`\text{H}^{}=L^2(𝒜_\theta ,\tau _0)`$ of the G.N.S. construction of $`\tau _0`$, and a Hilbert space $`\text{H}^+`$ of forms of type $`(1,0)`$ on the noncommutative torus which is obtained canonically from $`\phi _2`$ as follows: Let $`𝒜`$ be a $``$-algebra and let $`\phi _2Z_+^2(𝒜,𝒜^{})`$ be a positive Hochschild $`2`$-cocycle on $`𝒜`$. Let $`\text{H}^+`$ be the Hilbert space completion of $`\mathrm{\Omega }^1(𝒜)`$ equipped with the inner product
$$a^0da^1,b^0db^1=\phi _2(b^0a^0,a^1,b^1).$$
(168)
Then the actions of $`𝒜`$ on $`\text{H}^+`$ by left and right multiplications are unitary. They are automatically bounded if $`𝒜`$ is a pre-$`C^{}`$-algebra.
Thus, $`\text{H}^+`$ is a bimodule over $`𝒜`$ and the differential $`d:𝒜\mathrm{\Omega }^1(𝒜)`$ gives a derivation which, for reasons that will become clear, we shall denote by $`:𝒜\text{H}^+`$.
In our specific example, the computation is straightforward and gives $`\text{H}^+=L^2(𝒜_\theta ,\tau _0)`$ as an $`𝒜_\theta `$-bimodule and $`:𝒜\text{H}^+`$ given by $`=\frac{1}{\sqrt{2\pi }}(\delta _1i\delta _2)`$, where $`\delta _1`$, $`\delta _2`$ are the standard derivations of $`𝒜_\theta `$.
$$\delta _1=2\pi iU\frac{}{U},\delta _2=2\pi iV\frac{}{V}$$
(169)
so that $`\delta _1\left(b_{nm}U^nV^m\right)=2\pi inb_{nm}U^nV^m`$ and similarly for $`\delta _2`$. One has of course
$$\delta _1\delta _2=\delta _2\delta _1$$
(170)
and the $`\delta _j`$ are derivations of the algebra $`𝒜_\theta `$,
$$\delta _j(bb^{})=\delta _j(b)b^{}+b\delta _j(b^{})b,b^{}𝒜_\theta .$$
(171)
One can immediately check the following: Let $`𝒜=𝒜_\theta `$ act on the left on both $`\text{H}^{}=L^2(𝒜_\theta ,\tau _0)`$ and $`\text{H}^+`$. Then, the operator
$$D=\left[\begin{array}{cc}0& \\ ^{}& 0\end{array}\right]$$
(172)
in $`\text{H}=\text{H}^+\text{H}^{}`$ defines a $`(2,\mathrm{})`$-summable $`K`$-cycle over $`𝒜_\theta `$. The $`/2`$-grading $`\gamma `$ is given by the matrix $`\gamma =\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]`$ and the real structure $`J`$ is given simply in terms of the Tomita-Takesaki antilinear isometry (cf. ).
Translation invariant geometries on $`𝕋_\theta ^2`$ are parameterized by complex numbers $`\tau `$ with positive imaginary part like in the case of elliptic curves. Up to isometry the geometry depends only on the orbit of $`\tau `$ under the action of $`PSL(2,)`$. However, a new phenomenon appears in the noncommutative case, namely, the Morita equivalence which relates the algebras $`𝒜_{\theta _1}`$ and $`𝒜_{\theta _2}`$ if $`\theta _1`$ and $`\theta _2`$ are in the same orbit of the $`PSL(2,)`$ action on $``$ . We first need to give a concrete description of the finite projective modules over $`𝒜_\theta `$, it is obtained by combining the results of . The finite projective modules are classified up to isomorphism by a pair of integers $`(p,q)`$ such that $`p+q\theta 0`$. Let us describe the simplest example of the modules $`_{p,q}^\theta `$. The underlying linear space is the usual Schwartz space,
$$𝒮()=\{\xi ,\xi (s)s\}$$
(173)
of smooth function on the real line whose all derivatives are of rapid decay.
The right module structure is given by the action of the generators $`U,V`$
$$(\xi U)(s)=\xi (s+\theta ),(\xi V)(s)=e^{2\pi is}\xi (s)s.$$
(174)
One of course checks the relation (1), and it is a beautiful fact that as a right module over $`𝒜_\theta `$ the space 173 is finitely generated and projective (i.e. complements to a free module). It follows that it has the correct algebraic attributes to deserve the name of “noncommutative vector bundle” over $`𝕋_\theta ^2`$ according to the first line of the dictionary of section 4,
$$\begin{array}{cc}\text{Space }𝕋_\theta ^2& \text{Algebra }𝒜_\theta \\ \\ \text{Vector bundle}& \text{Finite projective module.}\end{array}$$
The algebraic counterpart of a vector bundle $`E`$ on a space $`X`$ is its space of smooth sections $`C^{\mathrm{}}(X,E)`$ and one can in particular compute its dimension by computing the trace of the identity endomorphism of $`E`$. If one applies this method in the above noncommutative example, one finds
$$\mathrm{dim}_{𝒜_\theta }(𝒮)=\theta .$$
(175)
The appearance of non integral dimension displays a basic feature of von Neumann algebras of type II. The dimension of a vector bundle is the only invariant that remains when one looks from the measure theoretic point of view (Section 2). The von Neumann algebra which describes the noncommutative torus $`𝕋_\theta ^2`$ from the measure theoretic point of view is the well known hyperfinite factor $`R`$ of type II<sub>1</sub>. In particular the classification of finite projective modules over $`R`$ is given by a positive real number, the Murray and von Neumann dimension,
$$\mathrm{dim}_R()_+.$$
(176)
The next point () is that even though the dimension of the above module is irrational, when we compute the analogue of the first Chern class, i.e. of the integral of the curvature of the vector bundle, we obtain an integer. We first need to determine the connections (in the sense of Section 12, equation 10) on the finite projective module $`𝒮`$. It is not hard to see (using 17) that they are characterized by a pair of covariant differentials
$$_j:𝒮()𝒮()$$
(177)
such that
$$_j(\xi b)=(_j\xi )b+\xi \delta _j(b)\xi 𝒮,b.$$
(178)
One checks that, as in the usual case, the trace of the curvature $`\mathrm{\Omega }=_1_2_2_1`$, is independent of the choice of the connection. Now the remarkable fact here is that (up to the correct powers of $`2\pi i`$) the integral curvature of $`𝒮`$ is an integer. In fact for the following choice of connection the curvature $`\mathrm{\Omega }`$ is constant, equal to $`\frac{1}{\theta }`$ so that the irrational number $`\theta `$ disappears in the integral curvature, $`\theta \times \frac{1}{\theta }`$
$$(_1\xi )(s)=\frac{2\pi is}{\theta }\xi (s)(_2\xi )(s)=\xi ^{}(s).$$
(179)
Whith this integrality, one could get the wrong impression that the noncommutative torus $`𝕋_\theta ^2`$ looks very similar to the ordinary 2-torus. A striking difference is obtained by looking at the range of Morse functions. These are of course connected intervals for the 2-torus. For the above noncommutative torus the spectrum of a real valued function such as
$$h=U+U^{}+\mu (V+V^{})$$
(180)
can be a Cantor set, i.e. have infinitely many disconnected pieces. This shows that the one dimensional shadows of our space $`𝕋_\theta ^2`$ are considerably different from the commutative case. The above noncommutative torus is the simplest example of noncommutative manifold, it arises naturally not only from foliations but also from the Brillouin zone in the Quantum Hall effect as understood by J. Bellissard, and in M-theory as we shall see in section 14.
We shall now describe the natural moduli space (or more precisely, its covering Teichmüller space) for the noncommutative tori, together with a natural action of $`SL(2,)`$ on this space. The discussion parallels the description of the moduli space of elliptic curves but we shall find that our moduli space is the boundary of the latter space.
We first observe that as the parameter $`\theta /`$ varies from $`1`$ to $`0`$ in the above labelling of finite projective modules $`_{p,q}^\theta `$ one gets a monodromy, using the isomorphism $`𝕋_\theta ^2𝕋_{\theta +1}^2`$. The computation shows that this monodromy is given by the transformation $`\left[\begin{array}{cc}1& 1\\ 0& 1\end{array}\right]`$ i.e., $`xxy`$, $`yy`$ in terms of the $`(x,y)`$ coordinates in the $`K`$ group. This shows that in order to follow the $`\theta `$-dependence of the $`K`$ group, we should consider the algebra $`𝒜`$ together with a choice of isomorphism,
$$K_0(𝒜)\stackrel{\rho }{}^2,\rho \text{(trivial module)}=(1,0).$$
(181)
Exactly as the Jacobian of an elliptic curve appears as a quotient of the $`(1,0)`$ part of the cohomology by the lattice of integral classes, we can associate canonically to $`𝒜`$ the following data:
1. The ordinary two dimensional torus $`𝕋=HC_{\mathrm{even}}(𝒜)/K_0(𝒜)`$ quotient of the cyclic homology of $`𝒜`$ by the image of $`K`$ theory under the Chern character map.
2. The foliation $`F`$ (of the above torus) given by the natural filtration of cyclic homology (dual to the filtration of $`HC^{\mathrm{even}}`$).
3. The transversal $`T`$ to the foliation given by the geodesic joining $`0`$ to the class $`[1]K_0`$ of the trivial bundle.
It turns out that the algebra associated to the foliation $`F`$, and the transversal $`T`$ is isomorphic to $`𝒜`$, and that a purely geometric construction associates to every element $`\alpha K_0`$ its canonical representative from the transversal given by the geodesic joining $`0`$ to $`\alpha `$. (Elements of the algebra associated to the transversal $`T`$ are just matrices $`a(i,j)`$ where the indices $`(i,j)`$ are arbitrary pairs of elements $`i,j`$ of $`T`$ which belong to the same leaf. The algebraic rules are the same as for ordinary matrices. Elements of the module associated to another transversal $`T^{}`$ are rectangular matrices, and the dimension of the module is the transverse measure of $`T^{}`$.)
This gives the correct description of the modules $`_{p,q}`$. The above is in perfect analogy with the isomorphism of an elliptic curve with its Jacobian. The striking difference is that we use the even cohomology and $`K`$ group instead of the odd ones.
It shows that, using the isomorphism $`\rho `$, the whole situation is described by a foliation $`dx=\theta dy`$ of $`^2`$ where the exact value of $`\theta `$ (not only modulo $`1`$) does matter now.
Now the space of translation invariant foliations of $`^2`$ is the boundary $`N`$ of the space $`M`$ of translation invariant conformal structures on $`^2`$, and with $`^2^2`$ a fixed lattice, they both inherit an action of $`SL(2,)`$. We now describe this action precisely in terms of the pair $`(𝒜,\rho )`$. Let $`g=\left[\begin{array}{cc}a& b\\ c& d\end{array}\right]SL(2,)`$. Let $`=_{p,q}`$ where $`(p,q)=\pm (d,c)`$, we define a new algebra $`𝒜^{}`$ as the commutant of $`𝒜`$ in $``$, i.e. as
$$𝒜^{}=\mathrm{End}_𝒜().$$
(182)
It turns out (this follows from Morita equivalence) that there is a canonical map $`\mu `$ from $`K_0(𝒜^{})`$ to $`K_0(𝒜)`$ (obtained as a tensor product over $`𝒜^{}`$) and the isomorphism $`\rho ^{}:K_0(𝒜^{})^2`$ is obtained by
$$\rho ^{}=g\rho \mu .$$
(183)
This gives an action of $`SL(2,)`$ on pairs $`(𝒜,\rho )`$ with irrational $`\theta `$ (the new value of $`\theta `$ is $`(a\theta +b)/(c\theta +d)`$ and for rational values one has to add a point at $`\mathrm{}`$).
Finally another group $`SL(2,)`$ appears when we discuss the moduli space of flat metrics on $`𝕋_\theta ^2`$. Provided we imitate the usual construction of Teichmüller space by fixing an isomorphism,
$$\rho _1:K_1(𝒜)^2$$
(184)
of the odd $`K`$ group with $`^2`$, the usual discussion goes through and the results of show that for all values of $`\theta `$ one has a canonical isomorphism of the moduli space with the upper half plane $`M`$ divided by the usual action of $`SL(2,)`$. Moreover, one shows that the two actions of $`SL(2,)`$ actually commute. The striking fact is that the relation between the two Teichmüller spaces,
$$N=M$$
(185)
is preserved by the diagonal action of $`SL(2,)`$. Finally note that the above action of $`SL(2,)`$ on the parameter $`\theta `$ lies beyond the purely formal realm of deformation theory in which $`\theta `$ is treated as a formal deformation parameter. This is a key point in which noncommutative geometry should be distinguished from formal atempts to deform standard geometry.
## Noncommutative gauge Theory and String Theory
The analogue of the Yang-Mills action functional and the classification of Yang-Mills connections on the noncommutative tori was developped in , with the primary goal of finding a ”manifold shadow” for these noncommutative spaces. These moduli spaces turned out indeed to fit this purpose perfectly, allowing for instance to find the usual Riemannian space of gauge equivalence classes of Yang-Mills connections as an invariant of the noncommutative metric. The next surprise came from the natural occurence (as an unexpected guest) of both the noncommutative tori and the components of the Yang-Mills connections in the classification of the BPS states in M-theory . In the matrix formulation of M-theory the basic equations to obtain periodicity of two of the basic coordinates $`X_i`$ turns out to be the following variant of equation 1 of section 11,
$$U_iX_jU_i^1=X_j+a\delta _i^j,i=1,2$$
(186)
where the $`U_i`$ are unitary gauge transformations.
The multiplicative commutator $`U_1U_2U_1^1U_2^1`$ is then central and in the irreducible case its scalar value $`\lambda =\mathrm{exp}2\pi i\theta `$ brings in the algebra of coordinates on the noncommutative torus. The $`X_j`$ are then the components of the Yang-Mills connections. It is quite remarkable that the same picture emerged from the other information one has about M-theory concerning its relation with 11 dimensional supergravity and that string theory dualities could be interpreted using Morita equivalence. The latter relates as we saw above in section 13 the values of $`\theta `$ on an orbit of $`SL(2,)`$ and this type of relation would be invisible in a purely deformation theoretic perturbative expansion like the one given by the Moyal product.
In their remarkable paper, Nekrasov and Schwarz showed that Yang-Mills gauge theory on noncommutative $`^4`$ gives a conceptual understanding of the nonzero B-field desingularization of the moduli space of instantons obtained by perturbing the ADHM equations. In their paper , Seiberg and Witten exhibited the unexpected relation between the standard gauge theory and the noncommutative one.
The question of renormalizability of quantum field theories on noncommutative spaces which was the basis of is generating remarkable similarities with string theory which hopefully should yield a better formulation of $`M`$-theory than what is currently available. The rate at which progress is occuring in this interplay between noncommutative geometry and physics makes it rather futile to try and foresee what will happen even in the near future but there are a few issues on which I cant help to make brief comments (as a non-expert). The first has to do with locality, the expressions discussed in section 8 which involve the residue applied to multiple products of elements of the algebra and the operator $`D`$ do generate the natural candidate for local cochains in the general case. This was the basic procedure used in to generate local interactions.
Also the transformation from one standard gauge theory to the noncommutative one in has the basic feature of respecting the foliations of gauge potentials by gauge equivalence and since gauge transformations are isospectral deformations of the corresponding Dirac operators (with potential) it is natural to wonder wether the Seiberg-Witten transformation can be interpreted in spectral terms.
String theory is a generalization of ordinary geometry whose onshell formulation is understood via conformal field theory. The corresponding mathematical question of existence of $`\sigma `$-models should benefit from the investigation of the Riemann-Hilbert problem attached to the renormalization of such a theory as in section 10.
Finally, one should probably also look for an offshell formulation of string-geometry. It is well known that the spectral information on a homogeneous Riemannian space can be grasped using Lie group representations but what we showed in section 11 is that even the nonhomogeneous metrics are accessible to such a Hilbert space representation treatment. The new feature is that the basic equations are no longer related to Lie group representations but to algebraic K-theory considerations. It is tempting to speculate that a similar adaptation of the Lie algebra representation theoretic approach to conformal field theory could yield the desired offshell formulation of stringy geometry. |
warning/0003/astro-ph0003109.html | ar5iv | text | # Lopsided Galaxies, Weak Interactions and Boosting the Star Formation Rate
## 1 INTRODUCTION
Galaxies do not live isolated lives, but exist in the tidal fields of their environment. Arp (1966), in his Atlas of Peculiar Galaxies, lay the observational groundwork for the modern study of interacting galaxy systems by identifying many ”peculiar” systems, later interpreted as various stages of major galaxy mergers. Strong galaxy-galaxy interactions may dramatically alter the stellar populations (e.g. Larson & Tinsley 1978; Kennicutt et al. 1987; Turner 1998; Kennicutt 1998), morphology (e.g. Toomre & Toomre 1972; Hernquist, Heyl & Spergel 1993) and kinematics of galaxies (e.g. Toomre & Toomre 1972; Barnes & Hernquist 1992) driving evolution along the Hubble sequence. Massive mergers are also capable of funneling gas into the center of galaxies causing nuclear starbursts (Barnes & Hernquist 1991; Mihos, Richstone & Bothun 1992; Barnes & Hernquist 1996) and QSO activity (e.g. Sanders et al. 1988). At the present epoch, however, major mergers are fairly rare events (e.g. Kennicutt et al. 1987) and their broad evolutionary importance is unclear.
Minor mergers and, in general, weak tidal interactions between galaxies occur with much higher frequency than major ones (e.g. Lacey & Cole 1993). By weak interactions we mean those which do not destroy the disk of the “target” spiral. Hierarchical structure formation models (e.g. cold dark matter) predict that the merging histories for high mass objects today contained multiple low mass accretion events in their past (e.g. Lacey & Cole 1993). The specific roles which weak interactions play in the evolution of galaxies, however, is uncertain. Weak interactions may cause disk heating (e.g. Toth & Ostriker 1992; Quinn, Hernquist & Fullagar 1993) and satellite remnants may build up the stellar halo (e.g. Searle & Zinn 1978; Johnston, Hernquist & Bolte 1996). Kennicutt et al. (1987) studied the relation between interaction strength and star formation by making a comparison between isolated galaxies, close pairs, and galaxies from the Arp Atlas. They found that close pairs have larger values of $`EW(H\alpha _{em})`$, i.e. higher star formation rates (SFR) than isolated galaxies. While pair spacing is weakly correlated with the SFR, they could not determine the specific role of interaction strength on the SFR. Hashimoto et al. (1998) and Allam et al. (1999) both studied the Hubble type specific effects of environment on the SFR in galaxies. They found that the SFR/mass of existing stars was inversely proportional to the local galaxy density. They postulate that the anti-correlation is due partly to gas stripping and due partly to the anti-correlation of the merger cross-section with the galaxy-galaxy velocity dispersion.
There is also evidence that interactions excite nuclear activity. In their close pair and strongly interacting sample Kennicutt et al. (1987) found a strong correlation between $`H\alpha `$ emission in the disk and that in the nucleus. Such a correlation between disk and nuclear emission is supported by theoretical work; Mihos & Hernquist (1994) and Hernquist & Mihos (1995) demonstrated that minor interactions form bar instabilities in the disk which in turn funnel large amounts of gas into the nucleus. The effectiveness of this process is suppressed by the presence of a dense bulge, which prevents bar formation. Due to the numerical expense in computing high resolution N-body/SPH (collisionless particle/smoothed particle hydrodynamics) models, the exact interaction parameters which result in such activity are uncertain.
Weak interactions may also manifest themselves as kinematic or structural irregularities. Roughly $`50\%`$ of all spiral galaxies have asymmetric HI profiles and rotation curves (Baldwin, Lynden-Bell & Sancisi 1980; Richter & Sancisi 1994; Haynes et al. 1998). Baldwin et al. (1980) postulated that these asymmetries are caused by weak interactions in the galaxy’s past or by lopsided orbits. Barton et al. (1999) examined the optical rotation curves of a set of observed and simulated interacting disk galaxies. They showed that interactions can cause large scale, time dependent asymmetries in the rotation curves of their sample galaxies. Swaters et al. (1999) studied the kinematic asymmetries present in two galaxies lopsided in their optical and HI distributions. They qualitatively reproduced the kinematic asymmetries by placing closed orbits in mildly lopsided potential.
A dynamical indicator of weak interactions may be “lopsidedness.” In the context of this paper (following Rudnick & Rix 1998; hereafter RR98), lopsidedness is defined as a bulk asymmetry in the mass distribution of a galactic disk. Surveys for lopsidedness in the stellar light of galaxies were first carried out by Rix & Zaritsky (1995; hereafter RZ95) and Zaritsky & Rix (1997; hereafter ZR97). Using near-IR surface photometry of face-on spiral galaxies (spanning all Hubble types) they examined the magnitude of the $`m=1`$ azimuthal Fourier component of the I and K-band surface brightness, thus characterizing the global asymmetry of the stellar light. RZ95 and ZR97 found that a quarter of the galaxies in their sample were significantly lopsided. Using a larger, magnitude limited sample restricted to early type disks (S0 to Sab) and imaged in the R-band, RR98 found that the fraction of significantly lopsided early type disks is identical to that for late-type disks. RR98 convincingly demonstrated that lopsidedness is not an effect of dust, but is in fact the asymmetric distribution of the light from old stars and hence from the stellar mass in the disk.
Some theoretical work has been done in examining long lived $`m=1`$ modes (Syer & Tremaine 1996; Zang & Hohl 1978; Sellwood & Merritt 1994), little convincing evidence however has been put forth to show that isolated galaxies will form stable $`m=1`$ modes without external perturbations or significant counter-rotating populations. Without invoking the special cases above, long lived lopsidedness is possible if the disk resides in a lopsided potential. The question remains however: how is a lopsided potential created/maintained? Numerical simulations of hyperbolic encounters between disk galaxies fail to produce $`m=1`$ modes of amplitude $`>10\%`$ without destroying the pre-existing stellar disk (Naab, T.; private communication). Minor mergers and possibly some weak interactions therefore remain as the most probable cause of lopsidedness (RR98). Recent work has shown that perturbations in the outer halo of a galaxy may be amplified and even transmitted down into the disk (Weinberg 1994). Work by Walker, Mihos & Hernquist (1996) and ZR97 showed that the type and magnitude of lopsidedness seen in RZ95, ZR97 and RR98 is comparable to the result of the accretion of a small satellite, if the mass ratio with the main galaxy is $`1/10`$. In a preliminary study (i.e. a rigid halo with no dynamical friction) Levine and Sparke (1998) showed that lopsided galaxies may be formed by disks orbiting off center and retrograde in a flat-cored, dark matter dominated halo. They postulated that a galaxy may be pushed off center by a satellite accretion.
Using phase mixing arguments (Baldwin et al. 1980; RZ95) and analysis of N-body simulations (Walker et al. 1996; ZR97) the lifetime of lopsided features has been estimated at $`t_{lop}1`$ Gyr. That lopsidedness is transient ($`t_{lop}t_{Hubble}`$) yet common, requires that it must be recurring and therefore lopsidedness may have significant evolutionary consequences.
The current paper focuses on the impact that minor mergers (observed as lopsidedness) may have on boosting the SFR and the recent star formation history (SFH) of disk galaxies. For the purpose of this discussion, we will assume that lopsidedness is caused by minor mergers. Regardless of what causes lopsidedness however, the perturbation in the gravitational potential manifestly exists and therefore may affect the gas in the galaxy to such a degree as to boost the SFR. Indeed, ZR97 find that lopsidedness is correlated (at $`96\%`$ confidence) with the “excess” of blue luminosity (over what is predicted by the Tully-Fisher relation). Modeling the integrated spectral evolution of starbursts using evolutionary population synthesis (EPS) codes has been been well studied (e.g. Couch and Sharples 1987; Barger et al. 1996; Turner 1998) and despite its limitations, is a useful tool in determining the relative SFH over the past $`1`$ Gyr. The same techniques used to probe the SFH in massive starbursts should also work to probe the recent SFH in the putative mini-bursts which we seek to study. By comparing measured indicators of recent SF (e.g. $`EW(H\delta _{abs})`$, $`4000\mathrm{\AA }`$ break strength, A star content), to the same indicators derived from the EPS models, we will place limits on the mini-burst mass and duration.
We have obtained spatially integrated spectra of a sample of 40 late type spiral galaxies (Sab-Sbc) of varying degrees of lopsidedness with the intent of using their relative stellar populations (as determined from stellar template fitting and EPS models) to determine their recent SF histories. Unlike the mass-normalized blue light excess, $`\mathrm{\Delta }B`$ used in ZR97, our method operates independently of assumptions about a galaxy’s mass, inclination or luminosity. In addition to probing the recent ($`1`$ Gyr) SFH with studies of the stellar continuum we probe the current SFR by measuring the integrated Balmer line emission strengths (e.g. Kennicutt et al. 1994).
The layout of the paper is as follows. In §2 we discuss the sample selection, observations, data reduction and determination of galaxy lopsidedness; In §3 we examine our methods for determining the current SFR and recent SFH via the measurement of emission and stellar continuum properties as a function of lopsidedness. The discussion of the significance of these results, including the correlation of the boost parameters with other galaxy characteristics and the impact of our results on previous works (i.e. RZ95,ZR97 & RR98) is contained in §4. In §5 we present a summary and possible directions for future work.
## 2 THE DATA
### 2.1 Sample Selection
To build a sample of galaxies with varying degrees of lopsidedness, we imaged a large number of galaxies ($`N_{gal}100`$) taken from the RC3 catalog (De Vaucouleurs et al. 1991), selected according to the following criteria: apparent blue magnitude m$`{}_{B}{}^{}`$14, redshift cz$``$10,000 km/s, axis ratio $`b/a`$0.64 ($`50^{}i0^{}`$), de Vaucouleurs type $`abbc`$, and a maximum diameter of 4$`\mathrm{}`$. The median diameter of the galaxies in our sample was 1.8$`\mathrm{}`$. The magnitude and redshift limits were chosen to minimize the required exposure times. The axis ratio of the galaxies was constrained because it is hard to measure an azimuthal $`m=1`$ component in a highly inclined galaxy. Once imaging was obtained and lopsidedness determined for each galaxy (see §2.2.1), we constructed a sample for spectroscopy consisting of 40 of our imaged galaxies (see Table 1). These were selected to give the sample equal numbers of lopsided and symmetric targets.
Our sample is partially selected according to Hubble type, and we must explore the effects which morphological evolution induced by lopsidedness may have on our conclusions. Walker et al. (1996) suggested that minor mergers increase bulge size, heat the galactic disk vertically and consume a large fraction of the galaxy’s gas, eventually resulting in a low post-merger SFR. These two effects may drive galaxies towards earlier Hubble type after they experience minor mergers. Evolutionary processes such as these however become dominant either after the expected lifetime of lopsidedness ($`t1`$ Gyr), or after repeated merger events (Walker et al. 1996). During the interaction itself the irregularity in structure made manifest by lopsidedness as well as the creation of spiral arms via tidal perturbations will temporarily move a galaxy later in Hubble type. This will effectively push lopsided early type disk galaxies into our sample while pushing those of later type out of it. Due to their lower gas masses (Roberts & Haynes 1994), early type spirals have less potential for a large absolute increase in their SFR than late types. Small boosts, however, may be easily noticeable against the typically older stellar population of early type disks. The exact interplay of these two effects may bias our measurement of the relation between SFR and lopsidedness.
### 2.2 Observations
#### 2.2.1 Imaging
Our imaging data were obtained during runs at Steward Observatory’s 2.3-m Bok reflector on Kitt Peak (1997 November 6-7 and 1998 February 1-2) and at its 61-inch (1.5-m) reflector on Mt. Bigelow (1998 May 15-18). The CCD pixel scales at the Bok reflector and Bigelow reflector were $`0\stackrel{}{\mathrm{.}}4\mathrm{p}\mathrm{i}\mathrm{x}\mathrm{e}\mathrm{l}^1`$ with fields of view $`6\stackrel{}{\mathrm{.}}8\times 6\stackrel{}{\mathrm{.}}8`$ and $`3\stackrel{}{\mathrm{.}}4\times 3\stackrel{}{\mathrm{.}}4`$ respectively. The median seeing at the 2.3-m for the November and February runs were 1$`\stackrel{}{\mathrm{.}}`$3 and 1$`\stackrel{}{\mathrm{.}}`$5, respectively, while the median seeing at the 1.5-m in May was 1$`\stackrel{}{\mathrm{.}}`$3.
A Nearly-Mould $`R`$band filter ($`\lambda _{center}=650nm`$) was used at the 2.3-m and a Kron-Cousins R-band filter with $`\lambda _{center}=650nm`$ was used at the 1.5-m. The effects of wavelength on observed lopsidedness are discussed in §2.1 of RR98, and found not to be critical.
To determine the lopsidedness of a galaxy we perform an azimuthal Fourier decomposition of the R-band surface brightness, as in RR98:
$$I(R_m,\varphi )=a_o\{1+\underset{j=1}{\overset{N}{}}a_je^{i[j(\varphi _j\varphi _j^o)]}\},$$
(1)
where for each radius, $`\left|a_o\right|(R)`$ is the average intensity and $`\left|a_1\right|(R)`$ describes the lopsidedness. We define the luminosity normalized quantities $`A_1a_1/a_0`$. Instead of $`A_1(R)`$, we use $`\stackrel{~}{A}_1(R)`$ (the error corrected value which accounts for the positive definite nature of our measurements and the presence of errors; see RR98 for details) as our measure of asymmetry. We calculate the mean asymmetry of each galaxy, $`\stackrel{~}{A}_1`$ (see Table 2), from 1.5 to 2.5 disk scale lengths using the weighted average described in RR98.
#### 2.2.2 Spectroscopy
Spectra were obtained with the Bollers & Chivens Spectrograph at the 2.3-m Bok reflector during the nights, 1998 March 22-25, 1998 May 25-28, and 1998 June 29. We used a $`400gmm^1`$ in $`2^{nd}`$ order, blazed at $`3753\mathrm{\AA }`$, and a $`2.5\mathrm{}`$ slit, resulting in a resolution of $`1500`$ and a wavelength range of $`3600\mathrm{\AA }\lambda 5300\mathrm{\AA }`$. This range includes the entire Balmer series redward to $`H\beta `$, the $`4000\mathrm{\AA }`$ break, Ca H+K doublet, \[O II\]$`\lambda \lambda 3726,3729\mathrm{\AA }`$ and \[O III\]$`\lambda \lambda 4959,5007\mathrm{\AA }`$. To reduce read noise, the CCD was binned in the spatial direction; on 22 March we binned by 2 for a resultant pixel scale of $`1.67\mathrm{}\mathrm{pixel}^1`$ while on all other nights we binned by 4 for a pixel scale of $`3.33\mathrm{}\mathrm{pixel}^1`$. A CuSO<sub>4</sub> filter was used to block $`1^{st}`$ order light. Aside from using standard stars to calibrate the relative spectral response of the instrument, no absolute flux calibration was attempted. This was done partly because of the non-photometric conditions of some of our nights and partly due to the independence of our analysis methods on absolute flux levels.
Following Kennicutt (1992), we obtained spatially integrated spectra of the galaxies by repeatedly scanning the slit across the galaxy between $`2.5R_{exp}x2.5R_{exp}`$. For each galaxy we obtained 2 exposures of 25 minutes each. To isolate the disk contributions to the integrated spectra, we also obtained a 5 minute exposure of the nucleus for each galaxy for later subtraction.
### 2.3 Reduction
#### 2.3.1 Images
The basic image reduction was carried out with standard IRAF<sup>3</sup><sup>3</sup>3IRAF is distributed by the National Optical Astronomical Observatories, which are operated by AURA, Inc. under contract to the NSF. routines. The total dark current was $`1\mathrm{e}^{}/\mathrm{exposure}/\mathrm{pixel}`$ and so it was ignored. The images were flat-fielded with a combination of twilight and smoothed night sky flats. High $`S/N`$ twilight flats ($`N5`$) were used to take out small-scale variations and the lower $`S/N`$ smoothed night sky flats ($`N6`$) were used to take out large-scale sensitivity fluctuations. The large-scale flat-field quality was determined by measuring variance in the median sky level at the four corners of the images. In all images where the galaxy was small in the field of view, the images were found to be flattened to better than $`1\%`$. In some cases, the large size of the galaxies in comparison to the field of view at the 1.5-m telescope precluded such an estimate.
Point sources in the images were selected using DAOFIND, and surrounding pixels were excised in the subsequent analysis to a radius where the stellar point spread function declined to the level of the sky.
#### 2.3.2 Spectra
The basic spectral reduction was carried out with IRAF routines. The measured dark current from multiple 25 minute dark exposures was $`2\mathrm{e}^{}/\mathrm{exp}/\mathrm{pixel}`$ and so was not accounted for. Small-scale variations were removed with a combined series of 2 min. dome flats. Twilight flats were fit in the spatial direction with a $`57^{th}`$ order cubic spline at several different positions in the spectral direction to construct a map of the slit response as a function of wavelength. Non-linear pixels were interpolated over using a bad pixel mask generated from the ratio between a combined series of 2 minute and 10 second dome flats.
Removing cosmic rays from spectra without accidently removing emission lines is best automated by using the shape of the spectrum itself in the cleaning process: for a given column (spatial direction), we medianed together the target column with the eight columns on either side to construct a slit profile, $`I_{slit}`$. Using $`\chi ^2`$ minimization, we fit the target column with the the following function:
$$I_{model}(y)=a_1I_{slit}(y)+a_2y+a_3.$$
(2)
where the $`a_3`$ term accounts for the sky background. Pixels deviating by more than 7 $`\sigma `$ from the best fit are replaced with the value of the best fit model at that point. Flagged segments larger than the spectral resolution of the instrument were not removed so as to avoid the accidental cleaning of emission lines. This algorithm is very efficient and, once a suitable set of parameters (i.e. $`N_{med}`$, threshold) has been chosen, can remove almost all of the cosmic rays on the chip.
After cosmic ray removal, the spectra were rectified using He-Ar lamps taken at various times during the night. Background contributions were then subtracted. To extract the spectra, we summed the number of center rows which corresponded to $`\pm 2.5R_{exp}`$ for the disk spectra and extracted the center row only for the nuclear spectra. Using standard stars taken during the night, we then removed the spectral response of the system from the 1D, extracted spectra. Finally, the wavelength scale of each spectra was shifted to the rest frame of the galaxy.
The guided nuclear exposures were scaled by the effective exposure times on the nucleus during the drift scanning. These scaled exposures were then subtracted from the drift spectra. In this way, we separated the disk and nuclear contributions. Finally, all spectra were scaled to their median flux level.
### 2.4 Errors
The error calculation for $`\stackrel{~}{A}_1`$ is described in RR98; it accounts for both photon statistics and systematic flat-fielding uncertainties.
Using Poisson statistics to describe the error in a raw spectrum we created an “error spectrum” and performed on it all of the standard reduction steps discussed in §2.3.2 except for the cosmic ray cleaning. After normalizing the “science spectra,” the “error spectra” were scaled by the median of the “science spectra” to form a detailed representation of the error at each pixel. The “error spectra” were used in all of the following analysis steps.
## 3 SPECTRAL ANALYSIS & STAR FORMATION HISTORY DIAGNOSTICS
By examining spatially integrated spectral characteristics as a function of $`\stackrel{~}{A}_1`$, we can determine whether lopsidedness affects the SF histories of galaxies. It is difficult, even in major mergers, to invert optical spectrophotometry into a SFH estimate (e.g. Turner 1998). However, the relative strength of the current and recent SFR in different galaxies may be studied with moderate $`S/N`$, non flux-calibrated spectra (e.g. Couch & Sharples 1987; Barger et al. 1996). The equivalent width of $`H\alpha `$ in emission ($`EW(H\alpha _{em})`$) integrated over the whole disk is a robust measure of the current global SFR in terms of the previously formed stars (Kennicutt et al. 1994). Much of the Balmer emission inn a galactic disk comes from the HII regions seen in SFR regions. In this regime, the Balmer decrement relates the emission flux in $`H\beta `$ to that in $`H\alpha `$. If the continuum level at these two wavelengths is similar, then the decrement will also relate the $`EWs`$ of the two lines. Therefore, the absorption corrected $`EW(H\beta _{em})`$ (see §3.2) serves as an indirect indicator of the current SFR in the disk.
The likely lifetime of lopsidedness has been estimated as $`t_{lop}1`$ Gyr (Baldwin et al. 1980; RZ95; ZR97). To examine SF on these timescales, we need a SF tracer with a comparable lifetime. Main sequence A-stars have lifetimes of $`0.5`$ Gyr (Clayton 1983), have strong spectral signatures (e.g. strong Balmer lines, a blue continuum, and a weak $`4000\mathrm{\AA }`$ break), and so serve as appropriate probes of the recent SFH.
### 3.1 Fitting Stellar Templates
Fitting population synthesis models (e.g. Bruzual & Charlot 1993) to our spectra can give us a measure of the recent SFH. However for four reasons we choose to simply fit with empirical stellar templates (Jacoby et al. 1984): 1) The spectral resolution of available population synthesis models ($`10\mathrm{\AA }`$) is significantly less than that of our spectra ($`3\mathrm{\AA }`$). 2) The empirical templates of Jacoby et al. (1984) have a spectral resolution ($`4.5\mathrm{\AA }`$) slightly less than that of our spectra. 3) An adequate fit to the spectra can be obtained with a small number of high S/N stellar templates. To estimate the recent, relative SFH, we therefore need not deal with the complexities of the IMF and metallicity of the stellar populations.
As a qualitative measure of the relative contributions to our spectra from young and old stars, we synthesize the global absorption spectra of the galaxies in our sample with a linear combination of two stellar templates, an A0V and a G0III spectra from the Jacoby et al. (1984) stellar library. We also individually fit and subtract a $`3^{rd}`$ order polynomial of zero mean from the spectra and from each stellar template to correct for color terms (e.g. from calibration errors, from approximating the spectra with only two stellar templates). The spectra (with the polynomial subtracted) are fit by $`\chi ^2`$ minimization with the following model:
$$I_{model}(\lambda )=(C_{A0V}I_{A0V}(\lambda ))G(\sigma )+(C_{G0III}I_{G0III}(\lambda ))G(\sigma )I_{poly},$$
(3)
where $`G(\sigma )`$ and the weights, C were determined iteratively. $`C_{A0V}`$ and $`C_{G0III}`$ are the weights for the normalized stellar template spectra, $`I_{A0V}`$ and $`I_{G0III}`$. $`I_{poly}`$ is the sum of the weighted template polynomial components, $`G(\sigma )`$ represents the Doppler broadening of the stars which is convolved with the stellar templates, and $``$ is the convolution operator. For more details on the fitting procedure see Rix et al. (1995) and Turner (1998).
Independent of continuum slope, there is a unique set of line shapes and strengths for stars of each spectral type. The polynomial fit minimizes the effects of a global continuum slope on our best fit solution so that we are instead performing a global fit to the spectral features ($`4000\mathrm{\AA }`$ break, Ca H$`+`$K, Balmer series, etc.) We find that with only two, polynomial subtracted, stellar templates, we are able to consistently achieve fits with $`\chi _\nu ^22`$.
### 3.2 Emission Lines
Our template fit to the galaxy spectra should reflect only the stellar populations in the galaxy, not interstellar emission. Therefore, we first fit all portions of spectra, omitting expected emission regions, and use the residual of the best model fit to construct an emission spectrum. We remove the large scale variations in the residual by fitting it with a high order polynomial ($`50`$) at all locations where no emission is expected. We then fit each emission feature with a Gaussian. To isolate the stellar continuum, we subtract these gaussian components from the original spectra. We then re-fit our $`I_{model}(\lambda )`$ to this cleaned, pure absorption spectrum to determine $`I_{model}^{best}(\lambda )`$. This process is illustrated in Figure 1.
We measure $`EW(H\beta _{em})`$ (see Table 2) from the complementary absorption corrected emission line spectra for all of our galaxies. A linear continuum was fit on either side of $`H\beta `$ ($`4720\mathrm{\AA }\lambda _B4800\mathrm{\AA };4900\mathrm{\AA }\lambda _R4940`$). We then measured the equivalent width of the emission line integrating from $`4843\mathrm{\AA }\lambda _{line}4883\mathrm{\AA }`$. The boundaries of our integration were chosen by visual inspection to minimize noise contributions from the continuum while maximizing the amount of line flux.
### 3.3 Quantifying the A-star Fraction
We can quantify the relative A star abundances by simply using the value of $`C_{A0V}`$ (see Table 2) in $`I_{model}^{best}(\lambda )`$. Because the basis templates are normalized to their median fluxes (as are the data,) the scaling factor of the individual templates gives a measure of how much A-stars contribute to the integrated spectra.
Balmer absorption lines are strongest in A-stars, the $`4000\mathrm{\AA }`$ break is weak, and so we may also use these two features to measure the recent SFH (Couch & Sharples 1987; Barger et al. 1996). We use the equivalent width of $`H\delta `$ in absorption ($`EW(H\delta _{abs})`$) as our indicator of Balmer line strength in order to minimize emission contamination and sample a relatively isolated region of the spectrum. A linear continuum was fit on either side of $`H\delta `$ ($`4000\mathrm{\AA }\lambda _B4050\mathrm{\AA };4150\mathrm{\AA }\lambda _R4250\mathrm{\AA }`$). We measure the $`EW`$ itself from $`4080\mathrm{\AA }\lambda _{line}4120\mathrm{\AA }`$.
Instead of measuring $`EW(H\delta _{abs})`$ directly from the spectrum, we decide to measure $`EW_{mod}(H\delta _{abs})`$ (see Table 2) of the best matching template spectrum, $`I_{model}^{best}(\lambda )`$. These template spectra are in general a good match to the data and provide an essentially noiseless estimate of $`EW(H\delta _{abs})`$ with a value derived from the best fit to a broad wavelength range. When measuring $`EW(H\delta _{abs})`$ directly from the spectra, the relation between $`EW(H\delta _{abs})`$ and $`\stackrel{~}{A}_1`$ (see §4.3) clearly remains, but the scatter in $`EW(H\delta _{abs})`$ at a given $`\stackrel{~}{A}_1`$ is slightly larger.
To further parameterize the recent SFH, we use the strength of the $`4000\mathrm{\AA }`$ break ($`D_{4000}`$) (see Table 2). Our measure of the break strength is defined as:
$$D_{4000}=\frac{_{4050}^{4250}f_\lambda 𝑑\lambda }{_{3750}^{3950}f_\lambda 𝑑\lambda }.$$
(4)
The suppression of the continuum blueward of $`4000\mathrm{\AA }`$ manifested in the break is caused by the combined absorption from the Ca H$`+`$K and $`Hϵ`$ absorption line. As young, massive stars (with blue continua and weak metal lines) contribute more to the overall spectrum, $`D_{4000}`$ decreases.
## 4 RESULTS AND DISCUSSION
### 4.1 Overall Spectral Characteristics
To examine qualitatively the link between lopsidedness and SFR, we constructed composite disk spectra from the median of our nine most symmetric and nine most lopsided galaxies, and these we show in Figure 2. The lopsided composite spectrum has a much bluer continuum, stronger Balmer lines, stronger emission and a shallower $`4000\mathrm{\AA }`$ break than its symmetric counterpart regardless of exactly how many galaxies we include in the median spectrum. This result indicates that lopsidedness is in fact correlated with the recent SFH. However, a quantitative treatment of the relation is still needed.
### 4.2 Is the Current Star Formation Correlated with Lopsidedness?
Figure 3 shows the relation between $`EW(H\beta _{em})`$ and $`\stackrel{~}{A}_1`$ for the entire spectroscopic sample. Using a Spearman-rank test, we find that the distribution of $`EW(H\beta _{em})`$ vs. $`\stackrel{~}{A}_1`$ deviates from a random one at the $`99.9\%`$ level. The median $`EW(H\beta _{em})`$ of our nine most symmetric galaxies is $`3.2\mathrm{\AA }`$ while the median for our nine most lopsided galaxies is $`6.3\mathrm{\AA }`$. The correlation of $`EW(H\beta _{em})`$, and hence the current SFR, with $`\stackrel{~}{A}_1`$ is clearly shown in Fig. 3. This implies that an elevation in the instantaneous SFR occurs over the same timescales as lopsidedness, lasting maybe as long as $`1`$ Gyr after the onset of lopsidedness.
### 4.3 Is Recent Star Formation Correlated with Lopsidedness?
Figure 4 shows the relation of A-star content, $`EW_{mod}(H\delta _{abs})`$, and $`D_{4000}`$ with $`\stackrel{~}{A}_1`$. They deviate from random distributions at the $`99.997\%`$, $`99.99\%`$, and $`99.68\%`$ levels respectively. Our sample is not large enough to study these correlations separately for different Hubble types; instead we analyze the change in spectral properties of the sample as a whole. Figure 5 shows that the data are differentiated by lopsidedness in the $`EW_{mod}(H\delta _{abs})`$ vs. $`D_{4000}`$ plane. The large scatter is not surprising, as even within a single Hubble type there is a considerable variation in the specific SFHs (Kennicutt et al. 1994). Between the most lopsided and symmetric $`1/3`$ of our galaxies the median values of $`EW_{mod}(H\delta _{abs})`$ and $`D_{4000}`$ differ by, $`\mathrm{\Delta }EW_{mod}(H\delta _{abs})=2.1\pm 1.0\mathrm{\AA }`$ and $`\mathrm{\Delta }D_{4000}=0.24\pm 0.01`$, respectively. This defines a boost vector (Fig. 5) whose direction and magnitude characterize the difference in spectral indices between a symmetric and lopsided galaxy in our sample.
### 4.4 Constraining the Boost Mass
A number of efforts have shown that the detailed SFH, even over the last $`10^9`$ years, cannot be determined unambiguously from integrated spectra (e.g. Turner 1998 and references therein; Leonardi & Rose 1996). Instead, we restrict ourselves to a set of model SFHs consisting of a “normal” or “underlying” spiral galaxy SFH, which in lopsided galaxies has been boosted in the past by a factor $`C_{boost}`$. We then attempt to characterize the boost strength ,$`C_{boost}`$, and an associated timescale. Using the EPS code of Bruzual & Charlot, (GISSEL 1995; in preparation) we construct SEDs representing the light from the underlying stellar populations of galaxies in our sample. We specify the SFH by the birthrate parameter $`b`$ (Scalo 1986), where
$$b=\frac{\mathrm{SFR}_{current}}{\mathrm{SFR}_{past}}.$$
(5)
Assuming an exponential SFH, i.e. $`SFR(t)\mathrm{exp}\left[t/\tau (b)\right]`$, each $`b`$ has an associated timescale, $`\tau (b)`$ defined through
$$\mathrm{exp}\left[\frac{t_{gal}}{\tau (b)}\right]1=\frac{t_{gal}}{\tau (b)}\frac{1}{b},$$
(6)
with $`t_{gal}`$ as the age of the galaxy. We adopt an age of $`t_{gal}10`$ Gyr and a metallicity of $`Z=Z_{}`$. We also use the Salpeter (1955) IMF with a mass range of $`0.1M_{\mathrm{}}125M_{\mathrm{}}`$. This IMF provides a better fit than does the Scalo IMF (1986) to the global photoionization rates and to the colors of galactic disks (Kennicutt et al. 1994). Applying the Scalo IMF results in slightly lower fractional boost masses than the Salpeter IMF, but the IMF choice does not qualitatively affect our conclusions. The model SED is then given by the integral:
$$F(\tau (b),\lambda )=_0^{t_{gal}}S(t^{},\lambda )r(\tau (b),t^{})𝑑t^{},$$
(7)
where
$$r(\tau (b),t)=\beta (b)\mathrm{exp}\left[\frac{t}{\tau (b)}\right],$$
(8)
where $`\beta (b)`$ is the Hubble type dependent initial SFR of the galaxy used to normalize the present day masses to the same value, and where $`S(t^{},\lambda )`$ is the SED for a population $`t^{}`$ years after a delta function burst (Bruzual & Charlot’s simple starburst model).
We account for the change in the spectral properties by adding an exponentially declining increase ,or “boost” in SF on top of the underlying galactic SFH. We assume that the boost originated about $`5\times 10^8`$ years ago, comparable to the presumed dynamical age of the boost and the A-stars seen in Figure 2. The model SED for a galaxy with a boost population is given by:
$$F(\tau (b),\lambda )=_0^{t_{gal}}S(t^{},\lambda )\beta (b)\mathrm{exp}\left[\frac{t^{}}{\tau (b)}\right]𝑑t^{}+_0^{t_{boost}}S(t^{},\lambda )C_{boost}\mathrm{exp}\left[\frac{t^{}}{\tau _{boost}}\right]𝑑t^{}$$
(9)
.
For each Hubble type, i.e. for each $`\tau (b)`$ we construct model tracks by varying $`C_{boost}`$ and $`\tau _{boost}`$ in Equation (9), effectively adding differing boosts to the unperturbed galaxy SFH. As shown in Fig. 6, moving the underlying population later in Hubble type, or to a larger $`\tau (b)`$, shifts the tracks to greater values of $`EW_{mod}(H\delta _{abs})`$ and smaller values of $`D_{4000}`$. For a given Hubble type, varying the boost timescale changes the slope of the track of increasing boost mass.
It is interesting to note that our empirical boost vector cannot be matched directly with any of the models in the $`EW_{mod}(H\delta _{abs})`$ vs. $`D_{4000}`$ plane, suggesting that the EPS models may have systematic errors. Assuming, however, that the correct models will differ from these by a simple shift in the $`EW_{mod}(H\delta _{abs})`$ vs. $`D_{4000}`$ plane, we choose the underlying population model which minimizes the distance (in the $`EW_{mod}(H\delta _{abs})`$ vs. $`D_{4000}`$ plane) to the beginning of our observed boost vector. We then match the slope of the boost vector by varying the boost timescale. The best match we find was for $`b=0.33`$ (Hubble type Sb; Kennicutt et al. 1994) and $`\tau _{boost}=500`$ Myr. For reference, we also show the tracks for underlying galaxy spectra with $`b`$ parameters corresponding to Hubble types of Sab and Sbc ($`b=0.17`$ and $`0.84`$ respectively; Kennicutt et al. 1994). For each of these Hubble types, we plot tracks corresponding to boosts with $`\tau _{boost}=100`$ Myr and $`\tau _{boost}=500`$ Myr. We also plot the $`\tau _{boost}=100`$ Myr boost for the $`b=0.33`$ model. Measuring the projected length of the boost vector on the best model, and assuming the galaxy disk has a final stellar mass of $`10^{10}M_{}`$, we find that $`1\times 10^9M_{}`$ of stars have been formed over the the past $`0.5`$ Gyr in addition to the “underlying” SFH. This corresponds to an factor of $`8`$ increase in the SFR over the past $`5\times 10^8`$ years.
Extinction will have two primary effects on our data. It will redden the continuum from the galaxy. By allowing for a polynomial element to our template fits, we minimize the effect of the continuum slope on our results. Extinction may also affect regions of current SF; obscuring both emission lines and the continuum from massive young stars. We do not make a further account of reddening but acknowledge that our boost masses may increase if reddening is included.
### 4.5 Correlations in Disguise
It is important to verify that our result is not an artifact of correlations between the underlying SFH and lopsidedness. To determine whether mass, gas content or Hubble type effects mimic our relation between lopsidedness and SFR, we examined the distributions of $`EW(H\beta _{em})`$, $`EW_{mod}(H\delta _{abs})`$, $`D_{4000}`$, A star content and $`\stackrel{~}{A}_1`$ vs. $`L_B`$, $`M_{HI}/L_B`$ and T-type (De Vaucouleurs et al. 1991). A hidden correlation exists if any of the latter three parameters are correlated both with a SF indicator and with $`\stackrel{~}{A}_1`$.
None of the above distributions, except for $`EW(H\beta _{em})`$ and $`D_{4000}`$ vs. T-type and $`M_{HI}/L_B`$, have more than a $`1.7\sigma `$ probability of being correlated, with a $`2.6\sigma `$ significance for the four remaining correlations. Given the known correspondence between SFR and Hubble type, mean stellar population and Hubble type, and SFR and gas surface density (Kennicutt et al. 1994), these correlations with $`EW(H\beta _{em})`$ and $`D_{4000}`$ are expected. However, none of our canonical galaxy parameters show a correlation with lopsidedness at above the $`1.2\sigma `$ level and we can therefore be confident that we are seeing a true correlation between our SF indicators and $`\stackrel{~}{A}_1`$.
### 4.6 Nuclear Properties and Contributions
As an initial diagnostic of the nuclear contributions, we examine the fraction of galactic light originating in the nucleus. Even without absolute spectrophotometry, we can still examine what percentage of the total galaxy light is made up of light from the nucleus as long as the spectra are calibrated with respect to each other.
The fractional luminosity of the nucleus ranged from $`0.16.4\%`$ with a median value of $`1.7\%`$. The fractional luminosity of the nucleus is uncorrelated with $`\stackrel{~}{A}_1`$. A Spearman-Rank test showed that there was a $`52\%`$ probability that the data was drawn from a random sample. Figure 7 shows that while the nuclear $`EW(H\beta _{em})`$ vs. $`\stackrel{~}{A}_1`$, and $`EW_{mod}(H\delta _{abs})`$ vs. $`\stackrel{~}{A}_1`$ are correlated, the relations show more scatter than in the disk. The Spearman-Rank test shows that these two relations are less significantly correlated in the nucleus than in the disk by $`0.4\sigma `$ and $`0.7\sigma `$, respectively. Only for $`D_{4000}`$ vs. $`\stackrel{~}{A}_1`$ was the correlation greater than in the disk.
Our data indicate that the enhanced SF in lopsided galaxies occurs in the disk with equal or greater strength than in the nucleus. This result appears to run contrary to what numerical simulations suggest about the reactions of gas in the disk to tidal perturbations from external sources. Mihos & Hernquist (1994) and Hernquist & Mihos (1995) show that even high mass ratio mergers ($`M_{gal}/M_{sat}>10`$) can form a bar which torques gas in the disk and funnels much of it into the nucleus. However the presence of a dense bulge easily inhibits bar formation and hence gas is not funneled as effectively. Another explanation is that star formation is occurring in large amounts at the nucleus, but is partly masked by extinction. High extinction is often found to be coincident with high star formation regions in mergers and is easily capable of hiding the presence of young populations (Turner 1998).
### 4.7 The Frequency of Lopsidedness– Revisited
To estimate the volume density of lopsided galaxies from a magnitude limited sample requires the understanding of any systematic differences in luminosity between lopsided and symmetric galaxies (ZR97). Our best fit boost brightens the galaxies by 1.3, 1.0 and 0.8 magnitudes (in B, V, and R, respectively) over the brightness of the underlying population. Lopsided galaxies will hence be overrepresented in our B-band selected sample, and their volume density will be $`25\%`$ of that derived from any magnitude limited survey. To compare the frequency of lopsidedness for different Hubble types (e.g. RR98), we need to understand how this brightening depends on the “underlying” SFH. Unfortunately, we have too few galaxies in our sample to discuss Hubble type dependent effects on the boost strength and luminosity evolution.
At present, the main constraint on the frequency of lopsidedness is from its frequency in imaging samples ($`20\%`$; RR98). Coupled with our brightening estimates, we find that $`5\%`$ of the galaxies in a volume limited sample will be lopsided. Lopsidedness is likely caused by minor mergers, the frequency of which increases with look back time. Given that lopsidedness lasts for $`1`$ Gyr, we can estimate that the average galaxy has been lopsided at least once in its lifetime.
## 5 SUMMARY & CONCLUSION
To quantify the correlation between the recent SF histories of present-day spiral galaxies and their global asymmetry, we compare the integrated spectral properties of late-type spirals of varying lopsidedness. We find that the recent ($`0.5`$ Gyr) SFH and current ($`10^7`$ years) SFR are both strongly correlated with $`\stackrel{~}{A}_1`$ although there is appreciable scatter in the individual galaxy-to-galaxy properties. For $`EW(H\beta _{em})`$, reflecting the current SFR, we find a $`3.2\sigma `$ Spearman-rank correlation with $`\stackrel{~}{A}_1`$. We fit a combination of A0V and G0III stellar spectra to our galaxy spectra to quantify the relative abundance of A-stars in the disk (which traces the SFR within 0.5 Gyrs). From these best fit model spectra, $`I_{model}^{best}(\lambda )`$, we measure a number of spectral indices, and find that $`EW_{mod}(H\delta _{abs})`$, $`D_{4000}`$, and $`C_{A0V}`$ are correlated with $`\stackrel{~}{A}_1`$ at the $`3.9\sigma `$, $`3.0\sigma `$, and $`4.2\sigma `$ levels, respectively.
We measure the same spectral indices in the nucleus, and find them less correlated with $`\stackrel{~}{A}_1`$ (except $`D_{4000}`$). Unless a nuclear starburst is obscured, the disk and not the nucleus is the primary site of the SF increase we see in lopsided galaxies. This is in contrast to numerical simulations where minor mergers funnel gas into the nucleus of galaxies, causing intense starbursts (Mihos & Hernquist 1994; Hernquist & Mihos 1995). Only by the presence of a dense bulge can the formation of a bar, and the subsequent funneling of gas, be prevented.
To quantify the mass of additional stars formed in lopsided galaxies, we defined a boost vector in $`EW_{mod}(H\delta _{abs})`$ vs. $`D_{4000}`$ space, by comparing the median values of these properties for the most symmetric third and the most lopsided third of our sample. We find $`\mathrm{\Delta }EW_{mod}(H\delta _{abs})=2.1\pm 1.0\mathrm{\AA }`$ and $`\mathrm{\Delta }D_{4000}=0.024\pm 0.01`$. We fit this vector with an “underlying population $`+`$ boost” EPS model corresponding to a progenitor galaxy with $`b=0.33`$, $`\tau _{boost}=500`$ Myr, and boost age of $`0.5`$ Gyr. Using this best fit EPS model, we find that $`1\times 10^9M_{}`$ is formed in the boost in addition to the “underlying” SFH (assuming a final stellar mass of $`10^{10}M_{}`$). This is a considerable fraction ($`10\%`$) of the final stellar mass of the galaxy and corresponds to a factor of 8 increase in the SFR over the past $`5\times 10^8`$ years. Given the increasing merger rates and increasing gas fractions towards higher redshifts, minor merger induced SF boosts of short duration played an important role in assembling the present day stellar content of galaxies.
Finally, we address by how much the frequency of lopsidedness from a magnitude limited sample is increased by the corresponding luminosity boost. Our best fit EPS boost model corresponds to a $`1`$ magnitude brightening when galaxies becomes lopsided, increasing their presence four-fold in magnitude limited samples. We lack the statistics however, to examine any Hubble type dependent differences in the luminosity boost.
It is obvious that more work needs to be done to fully understand the cause of lopsidedness as well as the SFH of lopsided galaxies. To quantify the Hubble type specific boost in the recent SFH, a large sample should be obtained with significant numbers of galaxies in each Hubble type bin. Since imaging and spectroscopy will be needed for this project, a volume limited sample may be constructed which bypasses many of the problems encountered when selecting galaxies according to an apparent magnitude limit. Companion searches to sufficiently faint magnitudes will help to study the possible link between environment and lopsidedness (as caused by weak tidal interactions). With the recent commissioning of large area imaging and spectroscopy surveys such as Sloan Digital Sky Survey, constructing such a sample will become relatively straightforward.
Numerical simulations have shown to be a useful tool in studying the evolution of the stellar and gas distributions in minor mergers. High resolution simulations with a live halo are crucial for studying the detailed response of the disk to the merger (Walker et al. 1996). A thorough exploration of interaction parameter space is needed to quantify the structural and kinematic response in the stellar and gas components. High resolution N-body studies are also needed to explore the global stability of isolated galactic disks.
This work was completed with partial support from NSF grants AST9870151, AST9421145 and AST9900789. Greg Rudnick and H.-W. Rix would like to thank Nelson Caldwell for many valuable discussions on measuring SFHs from integrated spectra. Greg Rudnick would like to thank Craig Kulesa and Christopher Gottbrath for many useful discussions in the early hours of the morning. Greg Rudnick would also like to thank Megan Sosey and Chris Gottbrath for assisting with our observing program. Finally, we would like to thank the Steward Observatory 2.3-m telescope Operators for their assistance in the completion of this project. |
warning/0003/cond-mat0003051.html | ar5iv | text | # Interacting Neural Networks
## I Mutual Learning, symmetric case
In this section we investigate a system of interacting neural networks as follows: several identical networks are arranged on an oriented ring. All networks receive an identical input and produce different output according to their weight vectors. Each network is trained by the output of its neighbour on the ring. This process is iterated until a stationary state is reached in which the norms and angles between the weight vectors no longer change. We are interested in the properties of this stationary state.
We consider the simplest feed-forward networks, an ensemble of $`K`$ simple perceptrons, which are represented by $`N`$-dimensional weight vectors $`𝐰_i`$ $`(i=1,\mathrm{},K)`$ and which map a common input vector $`𝐱`$ onto binary outputs $`\sigma _i=\text{sign}(𝐱𝐰_i)`$. As order parameters we use the norms $`w_i=|𝐰_i|`$ and the respective overlaps $`R_{ij}=𝐰_i𝐰_j`$ or $`\mathrm{cos}(\theta _{ij})=𝐰_i𝐰_j/w_iw_j`$. When only two perceptrons are considered, the subscript is dropped: $`\mathrm{cos}(\theta )=𝐰_1𝐰_2/w_1w_2`$. The components of the input vector (or pattern) are Gaussian with mean 0 and variance 1, yielding $`𝐱𝐱=O(N)`$.
The updates are of the form
$$𝐰_i^+=𝐰_i+(\eta _i/N)f(\sigma _i,s)s𝐱$$
(1)
for unnormalized weights or
$$𝐰_i^+=\frac{𝐰_i+(\eta _i/N)f(\sigma _i,s)s𝐱}{|𝐰_i+(\eta _i/N)f(\sigma _i,s)s𝐱|}$$
(2)
for normalized $`𝐰_i`$. The $`+`$ denotes a quantity after one learning step, $`\eta _i`$ is the learning rate, $`s`$ is the desired output, and $`f(\sigma _i,s)`$, the so-called weight function, defines the learning algorithm. We mostly use $`f=1`$ (the Hebbian rule, called $`H`$ from now on) and $`f=\mathrm{\Theta }(\sigma _is)`$ (the perceptron learning rule, abbreviated $`P`$ ), and the respective variations where the $`𝐰_i`$ are kept normalized, denoted as $`HN`$ and $`PN`$ respectively.
We derive differential equations for the order parameters in the thermodynamic limit $`N\mathrm{}`$ by taking the scalar product of the update rules and introducing a time variable $`\alpha =p/N`$, where $`p`$ is the number of patterns shown so far. We use the analytic tools which were previously developed for the teacher/student scenario . If the order parameters are self-averaging (see for criteria of self-averaging in this context), integrating over the distribution of patterns gives deterministic differential equations for the order parameters as $`N\mathrm{}`$. The required averages are listed in the appendix.
### A Perceptron learning rule
We first restrict ourselves to two perceptrons that try to come to an agreement by learning the output of the respective other perceptron.
For rule $`P`$ with identical learning rates $`\eta _1=\eta _2=\eta `$, the update rule is
$`𝐰_1^+`$ $`=`$ $`𝐰_1+{\displaystyle \frac{\eta }{N}}𝐱\sigma _2\mathrm{\Theta }(\sigma _1\sigma _2);`$ (3)
$`𝐰_2^+`$ $`=`$ $`𝐰_2+{\displaystyle \frac{\eta }{N}}𝐱\sigma _1\mathrm{\Theta }(\sigma _1\sigma _2).`$ (4)
The sum of both vectors is conserved under this rule: if a learning step takes place, it has the same direction and absolute value, but different signs for the two vectors. This conservation can be used to link $`w_1`$ and $`w_2`$ to $`\mathrm{cos}(\theta )`$: assuming that $`w_1=w_2=w`$ and starting from $`\theta _0=\pi /2`$, simple geometry gives $`w_0/\sqrt{2}=\mathrm{cos}(\theta /2)w`$. The conservation is also visible in the differential equations that can be derived using the described formalism:
$`{\displaystyle \frac{dw_1}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{\eta }{\sqrt{2\pi }}}(1\mathrm{cos}(\theta ))+{\displaystyle \frac{\eta ^2\theta }{2w_1\pi }};`$ (5)
$`{\displaystyle \frac{dw_2}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{\eta }{\sqrt{2\pi }}}(1\mathrm{cos}(\theta ))+{\displaystyle \frac{\eta ^2\theta }{2w_2\pi }};`$ (6)
$`{\displaystyle \frac{dR}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{\eta }{\sqrt{2\pi }}}(1\mathrm{cos}(\theta ))(w_1+w_2)\eta ^2{\displaystyle \frac{\theta }{\pi }}.`$ (7)
If the right-hand side of 5 and 6 vanish, so does 7. There is a curve of fixed points of the system given by the equation
$$w=\frac{\eta }{\sqrt{2\pi }}\frac{\theta }{1\mathrm{cos}(\theta )}.$$
(8)
Using the relation $`w=w_0/(\sqrt{2}\mathrm{cos}(\theta /2))`$, this can be solved numerically to give the fixed point of $`\mathrm{cos}(\theta )`$ as a function of the scaled learning rate $`\eta /w_0`$, as shown in Fig. 1. For small learning rates, the perceptrons come to good agreement, while large $`\eta `$ leads to antiparallel vectors.
Geometrically, this can be understood as follows: each learning step has a component parallel to the plane spanned by $`𝐰_1`$ and $`𝐰_2`$, which decreases the distance between the vectors, and a perpendicular component, which increases the distance (see Fig. 2). Equilibrium is reached when a typical learning step no longer changes the angle, i.e. the vectors stay on a cone around $`𝐰_1+𝐰_2`$. The radius of this cone increases with growing $`\eta `$.
### B Perceptron learning with normalized weights
A similar calculation can be done for the perceptron learning rule with normalized weights ($`PN`$), where the length $`w_i`$ of the weight vectors is set to $`1`$ after each step. The perceptrons move on a hypersphere of radius 1; in equilibrium, the average learning step leads back onto that sphere before the vectors are normalized again.
We derive the following differential equation for $`R=\mathrm{cos}(\theta )`$:
$$\frac{dR}{d\alpha }=(R+1)\left(\sqrt{\frac{2}{\pi }}\eta (1R)\eta ^2\frac{\theta }{\pi }\right).$$
(9)
Fixed points are $`R=1`$, $`R=1`$ and
$$\frac{\eta }{\sqrt{2\pi }}\frac{\theta }{1\mathrm{cos}(\theta )}=1.$$
(10)
It is not a coincidence that this is equivalent to (8) if $`w`$ is set to 1. The fixed point of (10) at $`R=1`$ is repulsive; the one at $`R=1`$ is unstable for $`\eta <4/\sqrt{2\pi }1.60`$. A solution of (10) can only be found for $`\eta \eta _c1.816`$, which corresponds to $`\mathrm{cos}(\theta )0.689`$.
Simulations show that the system relaxes to the fixed point given by Eq. (10) for $`\eta <\eta _c`$ and jumps to $`R=1`$ for larger $`\eta `$ (see Fig. 3). This behaviour shows the characteristics of a first-order phase transition.
Hence for small learning rates the two perceptrons relax to a state of nearly complete agreement, $`\theta 0`$. Increasing $`\eta `$ leads to a nonzero angle between the two vectors up to $`\theta 133^{}`$. At this rate the system jumps to complete disagreement, $`\theta =180^{}`$.
### C Mutual learning on a ring
The mutual learning-scenario can be generalized to $`K`$ perceptrons: perceptron $`i`$ learns from perceptron $`i+1`$ if they disagree, with cyclic boundary conditions. Under rule $`P`$, the total sum of vectors is conserved again: as many perceptrons take a step in one direction as in the opposite.
Performing the necessary averages for the equations of motion would involve Gaussian integrals over $`K1`$ correlated variables with $`\mathrm{\Theta }`$-functions – it is not clear to us whether this can be done analytically in general cases. However, we find in simulations that the fixed point for rule $`P`$ is completely symmetric: there is only one angle $`\theta `$ between all pairs of perceptrons. Assuming that relation (8) still holds, and using the conservation of $`𝐰_i`$, one can derive
$$\frac{\eta }{\sqrt{2\pi }}\frac{\theta }{1\mathrm{cos}(\theta )}=\frac{w_0}{\sqrt{1+(K1)\mathrm{cos}(\theta )}}.$$
(11)
The largest angle that the perceptrons can take is $`\mathrm{cos}(\theta )=1/(K1)`$, corresponding to a $`K`$-cornered hypertetrahedron. This happens when $`|𝐰_i|`$ is negligible w.r.t $`w_i`$. Simulations confirm that (11) holds, as can be seen in Fig. 1
Similar to the case of two networks, all perceptrons agree with each other for small learning rate $`\eta 0`$. For larger rates the system relaxes to a state of high symmetry where all mutual angles between the $`K`$ weight vectors are identical, $`\theta _{ij}=\theta `$. Note that the symmetry is higher than the topology of the flow of information (the ring). For high rates $`\eta \mathrm{}`$ the system relaxes to a state of maximal disagreement, i.e. the largest possible mutual angle $`\theta `$ that is still compatible with a symmetric arrangement.
For rule $`PN`$, the sum of the weights is not preserved. The fixed point of the dynamics follows the curve for two normalized weights described by Eq. (10) in a completely symmetric configuration. When the hypertetrahedron angle is reached and $`𝐰_i`$ vanishes, the symmetry is partly broken. There are now different angles to nearest neighbours, next-nearest neighbours etc., so the angles split up into $`(K1)/2`$ different branches for odd $`K`$ and $`K/21`$ for even $`K`$. Note that the system still has the symmetry of the ring.
With odd $`K`$, increasing $`\eta `$ increases the angle between nearest neighbours, up to some limit value. This angle is not the maximum nearest-neighbour angle allowed for by the geometric constraints, but seems to decrease with increasing K.
In the case of even $`K`$, simulations show a second transition at some higher value of $`\eta `$, where the vectors split into two antiparallel clusters, thus maximizing the nearest-neighbour angle. The learning rate at which this transition typically appears during the run of the program increases with $`N`$. The conclusion is that the antiparallel fixed point is not stable in the $`N\mathrm{}`$ limit, but de facto stable in simulations because the self-averaging property of the ODEs breaks down at this point.
One may ask which symmetries survive if the perceptrons are allowed to have different individual learning rates. A close look reveals that for rule $`P`$, there is a more general conserved quantity: $`_i^K𝐰_i/\eta _i`$. Simulations show that the angles $`\theta _{ij}`$ again relax to a completely symmetric configuration depending on the average $`\eta `$ and the initial value of the new conserved quantity, while the norms $`w_i`$ are proportional to the respective learning rates $`\eta _i`$. For rule $`PN`$, variations in the learning rates not only lead to slightly different curves for each of the angles with individually different $`\eta _c`$, they also suppress the transition to the antiparallel state that is observed for even $`K`$.
### D Hebbian learning
The reason why $`P`$ and $`PN`$ lead to antiparallel orientation of the weight vectors for larger learning rates is that they concentrate on cases where the networks disagree. Algorithms that reinforce what both networks agree on are more successful, as can be seen for rule $`H`$ for two perceptrons.
The differential equations are
$`{\displaystyle \frac{dw_i}{d\alpha }}`$ $`=`$ $`\eta \sqrt{{\displaystyle \frac{2}{\pi }}}\mathrm{cos}(\theta )+{\displaystyle \frac{\eta ^2}{2w_i}};`$ (12)
$`{\displaystyle \frac{dR}{d\alpha }}`$ $`=`$ $`\eta \sqrt{{\displaystyle \frac{2}{\pi }}}(w_1+w_2)+\eta ^2(1{\displaystyle \frac{2\theta }{\pi }}).`$ (13)
This system has no common fixed point, which means that the $`w_i`$ grow without bounds. The asymptotic behaviour can be seen from the equation for $`\mathrm{cos}(\theta )`$. Assuming that $`w_1=w_2=w`$, we find
$$\frac{d\mathrm{cos}(\theta )}{d\alpha }=\frac{\eta }{w}\frac{4}{\sqrt{2\pi }}(1\mathrm{cos}(\theta )^2)+\frac{\eta ^2}{w^2}(1\mathrm{cos}(\theta )\frac{2\theta }{\pi }).$$
(14)
By taking $`w\sqrt{2/\pi }\eta \alpha `$, the ODE leads to $`1\mathrm{cos}(\theta )\alpha ^4`$ for $`\alpha \mathrm{}`$. This means that $`\theta \alpha ^2`$.
Simulations agree with the numerical integration of Eqs. (13), with the exception of very large $`\alpha `$ and correspondingly small $`\theta `$ (see Fig. 4). This is not surprising, since the $`\alpha ^2`$-decay is an effect of patterns that are classified differently. As long as the perceptrons give the same output on all patterns, $`w_1`$ and $`w_2`$ grow linearly, but the difference $`𝐰_1𝐰_2`$ does not change, leading to $`\theta \alpha ^1`$. This is observed in simulations for small angles, where no patterns happened to be classified differently on the considered timescale. Mathematically, this is related to a breakdown of the self-averaging properties of Eqs. (13) at the point $`\theta =0`$.
## II Mutual learning, competition
In the previous section, all of the neural networks behave in the same way. Each perceptron tries to learn the output of its neighbour, and only the initial weight vectors are chosen randomly and differ from each other. Now we investigate a scenario where two networks behave differently. Network 1 is trying to simulate network 2 while 2 ist trained on the opposite of the opinion of 1. This scenario describes a competition between two adaptive algorithms. If 2 is completely successful, the overlap is $`\mathrm{cos}(\theta )=1`$, and perceptron 1 always fails in its prediction, and vice versa. A motivation from game theory can be drawn from the game of penny matching, where both players make a binary decision simultaneously. One player wins if the decisions are the same, the other if they are different.
### A Rule $`P`$
If both perceptrons use rule $`P`$ for their respective learning aim, the update rules are
$`𝐰_1^+`$ $`=`$ $`𝐰_1+(\eta _1/N)𝐱\sigma _2\mathrm{\Theta }(\sigma _1\sigma _2);`$ (15)
$`𝐰_2^+`$ $`=`$ $`𝐰_2(\eta _2/N)𝐱\sigma _1\mathrm{\Theta }(\sigma _1\sigma _2).`$ (16)
The corresponding differential equations for the order parameters are
$`{\displaystyle \frac{dw_1}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{\eta _1}{\sqrt{2\pi }}}(1\mathrm{cos}(\theta ))+{\displaystyle \frac{\eta _1^2}{2w_1}}{\displaystyle \frac{\theta }{\pi }}`$ (17)
$`{\displaystyle \frac{dw_2}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{\eta _2}{\sqrt{2\pi }}}(1+\mathrm{cos}(\theta ))+{\displaystyle \frac{\eta _2^2}{2w_2}}(1{\displaystyle \frac{\theta }{\pi }});`$ (18)
$`{\displaystyle \frac{dR}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{\eta _1w_2}{\sqrt{2\pi }}}(1\mathrm{cos}(\theta )){\displaystyle \frac{\eta _2w_1}{\sqrt{2\pi }}}(1+\mathrm{cos}(\theta )).`$ (19)
The common fixed point for these equations is $`w_i=\sqrt{2\pi }\eta _i/4`$, $`\mathrm{cos}(\theta )=0`$. This is hardly surprising, since none of the perceptrons has a better algorithm than the other. The learning rate only rescales the weight vectors; the ratio $`\eta _i/w_i`$, which determines how fast the direction of $`𝐰_i`$ in weight space can change, is independent of $`\eta `$ at the fixed point.
### B Rule $`H`$
The picture is slightly different if both perceptrons learn from every pattern they see. The resulting differential equations are
$`{\displaystyle \frac{dw_1}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta _1\mathrm{cos}(\theta )+{\displaystyle \frac{\eta _1^2}{2w_1}};`$ (20)
$`{\displaystyle \frac{dw_2}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta _2\mathrm{cos}(\theta )+{\displaystyle \frac{\eta _2^2}{2w_2}};`$ (21)
$`{\displaystyle \frac{dR}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta _1w_2\sqrt{{\displaystyle \frac{2}{\pi }}}\eta _2w_1\eta _1\eta _2(\pi 2\theta ).`$ (22)
The fixed point of $`R`$ is reached if $`\theta =\pi /2`$ and $`\eta _1/w_1=\eta _2/w_2`$, i.e. the vectors are perpendicular and the scaled learning rates $`\eta _i/w_i`$ are the same for both perceptrons. Under these conditions, the equations for $`w_i`$ can be solved: $`w_i=\eta _i(\alpha +(w_{i,0}/\eta _1)^2)^{1/2}`$, so $`w_i`$ shows the $`\sqrt{\alpha }`$-scaling typical for random walks. Geometrically, the Hebb rule adds corrections to the weight vector that are on average parallel to the teacher vector. Since the teacher is moving at the same angular velocity as the student, the movement of both vectors resembles a random walk. Again, $`\eta `$ only sets the temporal and spatial scale.
### C Rule $`P`$ vs. rule $`H`$
The result of the competition becomes more interesting when both perceptrons use different algorithms. For example, we let perceptron 1 use rule $`P`$, while 2 uses $`H`$. The derivation of the differential equations is again straightforward:
$`{\displaystyle \frac{dw_1}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{\eta _1}{\sqrt{2\pi }}}(1\mathrm{cos}(\theta ))+{\displaystyle \frac{\eta _1^2}{2w_1}}{\displaystyle \frac{\theta }{\pi }};`$ (23)
$`{\displaystyle \frac{dw_2}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta _2\mathrm{cos}(\theta )+{\displaystyle \frac{\eta _2^2}{2w_2}};`$ (24)
$`{\displaystyle \frac{dR}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta _2w_1+{\displaystyle \frac{\eta _1w_2}{\sqrt{2\pi }}}(1\mathrm{cos}(\theta ))+{\displaystyle \frac{\eta _1\eta _2\theta }{\pi }}.`$ (25)
They have a common fixed point defined by
$`\theta {\displaystyle \frac{\mathrm{cos}(\theta )^2}{(1\mathrm{cos}(\theta ))^2}}`$ $`=`$ $`{\displaystyle \frac{\pi }{4}};`$ (26)
$`w_1`$ $`=`$ $`{\displaystyle \frac{\eta _1}{\sqrt{2\pi }}}{\displaystyle \frac{\theta }{1\mathrm{cos}(\theta );}}`$ (27)
$`w_2`$ $`=`$ $`{\displaystyle \frac{\sqrt{2\pi }}{4}}{\displaystyle \frac{\eta _2}{\mathrm{cos}(\theta )}}.`$ (28)
These equations can be solved numerically and yield $`\mathrm{cos}(\theta )0.459`$, $`w_10.806\eta _1`$ and $`w_21.37\eta _2`$. Although perceptron 1 makes less use of the provided information, it wins the competition: the perceptron using rule $`H`$ has a smaller $`\eta /w`$-ratio and is thus less flexible.
### D Normalized weights
By setting the weights to 1 after each learning step, a new length scale is introduced, leading to a more complex dependence of the solution on the learning rates. For brevity, we only give the differential equations for the different learning rules and explain some common features. If both networks use rule $`PN`$, the ODE is
$$\frac{dR}{d\alpha }=\frac{1}{\sqrt{2\pi }}(\eta _1(1R)\eta _2(1+R))+\frac{R}{\sqrt{2\pi }}(\eta _1(1R)+\eta _2(1+R))\frac{R}{2\pi }(\eta _1^2\theta +\eta _2^2(\pi \theta ));$$
(29)
for rule $`HN`$ we find
$$\frac{dR}{d\alpha }=\sqrt{\frac{2}{\pi }}(\eta _2\eta _1)(R^21)\frac{R}{2}(\eta _1^2+\eta _2^2)\eta _1\eta _2(1\frac{2\theta }{\pi });$$
(30)
and if rule $`PN`$ is used by perceptron 1 and $`HN`$ by 2, the equation is
$`{\displaystyle \frac{dR}{d\alpha }}`$ $`=`$ $`R\left(\sqrt{{\displaystyle \frac{2}{\pi }}}\eta _2R{\displaystyle \frac{\eta _2^2}{2}}+{\displaystyle \frac{\eta _1}{\sqrt{2\pi }}}(1R){\displaystyle \frac{\eta _1^2}{2}}{\displaystyle \frac{\theta }{\pi }}\right)`$ (32)
$`+{\displaystyle \frac{\eta _1}{\sqrt{2\pi }}}(1R)\sqrt{{\displaystyle \frac{2}{\pi }}}\eta _2+\eta _1\eta _2{\displaystyle \frac{\theta }{\pi }}.`$
The behaviour of the fixed point is similar in all cases (see Fig. 5):
* if, say, $`\eta _2`$ is fixed and $`\eta _10`$, $`R`$ goes to a value $`R1`$. This is expected, since both $`PN`$ and $`HN`$ only achieve finite values of $`R`$ for fixed teachers.
* if both perceptrons use the same algorithm with the same learning rate, the result is $`R=0`$, as expected.
* if $`\eta _i\mathrm{}`$ for either $`i`$, $`R0`$. Infinite learning rate means that in every time step the perceptron discards all the information it previously had, replacing it with the current $`\pm 𝐱`$. Theoretically, that makes it predictable for the other network; in practice, both agents are confused. The notable exception is the case of $`PN`$ vs. $`HN`$, where a non-vanishing $`R`$ results if both $`\eta _i\mathrm{}`$ with a finite ratio $`\eta _1/\eta _2`$.
## III Confused Teacher
For any prediction algorithm there is a bit sequence for which this algorithm fails completely, with $`100\%`$ error . In fact, such a sequence is easily constructed: Just take the opposite of the predicted bit at each time step. In Ref. a perceptron was used for the prediction algorithm.
Here we do not consider bit sequences. However, it turns out that many statistical properties of the prediction algorithm are similar when random inputs are used instead of a window of the antipredictable bit sequence. Hence we consider the following scenario: Preceptron 1 is trained on the negative of its own output. Perceptron 2 is trained on the output of perceptron 1.
This is similar to the teacher/student model where the teacher weight vector performs a random walk . But here the teacher is “confused”, it does not believe its own opinion and learns the opposite of it.
The update rule of perceptron 1 now only depends on its own output:
$$𝐰_1^+=𝐰_1(\eta /N)𝐱\sigma _1.$$
(33)
Geometrically speaking, the vector performs a directed random walk in which every learning step has a negative overlap with the current vector. An equilibrium length is reached when a typical learning step leads back onto the surface of an $`N`$-dimensional hypersphere. This fixed point of $`w_1`$ is easily calculated to be
$$w_1=\sqrt{2\pi }\eta /40.6267\eta ,$$
(34)
and the weight vector typically moves on the surface of a hypersphere of that radius.
### A Rule $`H`$
What happens if a second perceptron tries to follow the output of the confused teacher? Again, the results depend entirely on the used algorithm. The simplest case, the Hebb rule, also has a geometrical interpretation that is revealed by a look at the update rule:
$`𝐰_1^+`$ $`=`$ $`𝐰_1(\eta /N)𝐱\sigma _1;`$ (35)
$`𝐰_2^+`$ $`=`$ $`𝐰_2+(\eta /N)𝐱\sigma _1.`$ (36)
As in section I A, the sum of both vectors is constant, so there is a class of solutions to the ODEs
$`{\displaystyle \frac{dw_1}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta +{\displaystyle \frac{\eta ^2}{2w_1}};`$ (37)
$`{\displaystyle \frac{dw_2}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta \mathrm{cos}(\theta )+{\displaystyle \frac{\eta ^2}{2w_2}};`$ (38)
$`{\displaystyle \frac{dR}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta (w_1w_2\mathrm{cos}(\theta ))+\eta ^2`$ (39)
defined by $`w_{1,f}=\sqrt{2\pi }\eta /4`$ and $`w_{2,f}=\sqrt{2\pi }\eta /(4\mathrm{cos}(\theta ))`$. The solution is given by the initial condition, i.e. the initial sum $`|𝐰_1+𝐰_2|`$. The fixed point angle can be calculated by applying the cosine theorem to a triangle with side lengths $`w_{1,f}`$, $`w_{2,f}`$ and $`|𝐰_1+𝐰_2|`$; starting from perpendicular vectors of norm $`w_0`$, one finds
$$\mathrm{cos}(\theta )=\left(1+\frac{16}{\pi }\left(\frac{w_0}{\eta }\right)^2\right)^{1/2}.$$
(40)
Geometrically, for large learning rate $`\eta `$ both norms become much larger than $`w_0`$; the only way to achieve this while keeping the sum constant is a large angle. For small $`\eta `$, $`w_1`$ becomes very small compared to the sum, and thus to $`w_2`$. So the direction of $`𝐰_2`$ stays nearly unchanged while $`𝐰_1`$ performs its random walk, leading to nearly perpendicular vectors on average.
### B Rule $`P`$
If perceptron 2 uses rule $`P`$, the sum of the vectors is not conserved, and a simple geometrical interpretation is not possible. However, the equations of motion can still be solved:
$`{\displaystyle \frac{dw_1}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta +{\displaystyle \frac{\eta ^2}{2w_1}};`$ (41)
$`{\displaystyle \frac{dw_2}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{\eta }{\sqrt{2\pi }}}(1\mathrm{cos}(\theta ))+{\displaystyle \frac{\eta ^2}{2w_2}}{\displaystyle \frac{\theta }{\pi }};`$ (42)
$`{\displaystyle \frac{dR}{d\alpha }}`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}\eta \mathrm{cos}(\theta ){\displaystyle \frac{w_1\eta }{\sqrt{2\pi }}}(1c)\eta ^2{\displaystyle \frac{\theta }{\pi }}.`$ (43)
The fixed point of $`\mathrm{cos}(\theta )`$ is given by the solution of $`4\theta /\pi =(1+\mathrm{cos}(\theta ))^2`$, independent from $`\eta `$. The numerical solution is $`\theta 0.777\pi `$, $`\mathrm{cos}(\theta )=0.761`$, $`w_2=0.552\eta `$ (in accordance with Ref. , where a special case of this problem was solved). Remarkably, the generalization error is larger than 50% - even the “smarter” perceptron learning rule predicts the behaviour of the confused teacher with less success than random guessing would.
### C Optimal learning rule
This raises an interesting question: is there any “reasonable” algorithm for perceptrons that allows them to track the confused teacher? If there are algorithms that achieve a positive overlap, one of them has to be the rule that optimizes student-teacher overlap in each time step – the optimal weight function derived by Kinouchi and Caticha :
$$f_{\text{opt}}=\frac{w_2\mathrm{tan}(\theta )}{\sqrt{2\pi }}\mathrm{exp}\left[\frac{(𝐱𝐰_2)^2}{2\mathrm{tan}(\theta )^2w_2^2}\right]\frac{1}{\mathrm{\Phi }(\sigma _1𝐱𝐰_2/(w_2\mathrm{tan}(\theta )))},$$
(44)
where $`\mathrm{\Phi }(x)=_{\mathrm{}}^x\mathrm{exp}(z^2/2)/\sqrt{2\pi }𝑑z`$. If $`w_1`$ is set to its fixed point for simplicity’s sake, calculation yields the following ODEs for $`\mathrm{cos}(\theta )`$ and $`w_2`$:
$`{\displaystyle \frac{d\mathrm{cos}(\theta )}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{1}{4\pi }}{\displaystyle \frac{\mathrm{sin}(\theta )^2}{\mathrm{cos}(\theta )}}I{\displaystyle \frac{2}{\sqrt{2\pi }w_1\mathrm{cos}(\theta )}};`$ (45)
$`{\displaystyle \frac{dw_2}{d\alpha }}`$ $`=`$ $`{\displaystyle \frac{w_2}{4\pi }}\mathrm{tan}(\theta )^2I,\text{ where}`$ (46)
$`I`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{1}{\sqrt{2\pi }}}\mathrm{exp}\left({\displaystyle \frac{1+\mathrm{cos}(\theta )^2}{2\mathrm{sin}(\theta )^2}}x^2\right){\displaystyle \frac{1}{\mathrm{\Phi }(x\mathrm{cot}(\theta ))\mathrm{\Phi }(x\mathrm{cot}(\theta ))}}𝑑x.`$ (47)
Calculating whether $`\mathrm{cos}(\theta )=0`$ is in fact a fixed point of the confused teacher/optimal student scenario is problematic, since the optimal weight function (44) diverges at $`\theta =\pi /2`$. However, the numerical solution of Eqs. (45) and (46) shows clearly that even starting from $`\mathrm{cos}(\theta )=1`$, the system evolves towards $`\theta =\pi /2`$, which indeed seems to be the upper limit for success. Simulations of the learning process again agree weel with our theory (see Fig. 6).
### D Rule $`HN`$
There is a way of achieving a positive overlap with the confused teacher with simple learning rules: if the teacher perceptron is “slowed down” by keeping its weights normalized and setting $`\eta `$ to some small value, a student using $`PN`$ or $`HN`$ can track the teacher nearly perfectly for very small learning rates. For simplicity’s sake, let us consider $`HN`$ with identical learning rates. The differential equation for $`R`$ is
$$\frac{dR}{d\alpha }=(R+1)\left(\sqrt{\frac{2}{\pi }}(1R)\eta \eta ^2\right),$$
(48)
the fixed points are $`R=1`$ or $`R=\sqrt{2/\pi }\eta +1`$. This result is again confirmed by simulations, as seen in Fig. 7. The fixed point goes to $`1`$ as $`\eta 0`$.
## IV Perceptrons in the Minority Problem
The concept of interacting neural networks can be applied to a problem that has received much attention recently: the El Farol Bar Problem . The problem was originally inspired by a popular bar that has a limited capacity: if too many people attend, it becomes crowded, and patrons don’t enjoy the evening. In a more special formulation, each agent out of a population of $`K`$ decides in each time step (each Saturday evening) to take one of two alternatives (go to the bar or stay at home). Those agents who are in the minority win, the others lose. Decisions are made independently; the only information available to agents is the decision of the minority was in the last $`N`$ time steps.
Many papers (see e.g. ) investigated a specific realisation of the model called the Minority Game. In this model each agent has a small number of randomly chosen decision tables (Boolean functions) that prescribe an action based on the previous history, and which of the tables is used is decided according to how successful each one was in the course of the game. It turned out that the success of the game depends on the ratio between the number of players and the size of the history window, and general conclusions on the behaviour of crowded markets were drawn .
We will discuss a different approach that yields different behaviour: Each agent $`i`$ is represented by a perceptron $`𝐰_i`$ that uses the time series $`𝐒_t=(S_t,S_{t1},\mathrm{},S_{tN+1})`$ of past minority decisions to make a prediction on the next time step. It then learns the output of the minority according to some learning rule.
In our approach, all of the agents are flexible in their decisions. Each agent uses an identical adaptive algorithm which is trained by the history of the game, the only information available to each of the agents. However, each agent uses a different randomly chosen initial state of its network. If all weight vectors of the networks would collapse, all agents would make the same decision, and all would lose. If all weights remained in the random initial state, each agent would make a random guess which yields a reasonable performance of the system. Our calculation shows that training can improve the performance of the system compared to the random state.
Following Ref. , we replace the history $`𝐒_t`$ by a random vector $`𝐱`$. Simulations show that this changes the results only quantitatively, if at all.
This strategy fulfills the restrictions that the original problem posed: the agents do not communicate except through majority decisions, and individual decisions are based on experience (induction or learning) rather than perfect knowledge of the system (deduction). However, since each player uses only one strategy whose parameters can be fine-tuned to the current environment rather than a set of completely different strategies, no quenched bias in the players’ behaviour is to be expected.
### A General notes on performance
The commonly used measure of collaboration in the minority problem is the average standard deviation of the sum of outputs of all agents:
$$\frac{\sigma ^2}{K}=\frac{1}{K}(\underset{i=1}{\overset{K}{}}\sigma _i)^2.$$
(49)
If each agent makes random decisions, one gets $`\sigma ^2/K=1`$. The probability of two perceptrons $`i`$ and $`j`$ giving the same output on a random pattern is $`1\theta _{ij}/\pi `$. Any ensemble of vectors $`𝐰_i`$ can be thought of as centered around a center of mass $`𝐂=_{i=1}^K𝐰_i/K`$ with a norm $`C`$ (for random vectors of length 1, $`C`$ would be of order $`\sqrt{K}`$). The weights can then be written as $`𝐰_i=𝐠_i+𝐂`$, with $`_{i=1}^K𝐠_i=\mathrm{𝟎}`$. For the sake of simplicity, we will assume a symmetrical configuration with $`g_i=1`$ and $`𝐠_i𝐠_j=1/(K1)`$ for $`ij`$. (An ensemble of randomly chosen vectors of norm 1 would give $`g_i^2=11/K\pm O(1/\sqrt{N})`$ and $`𝐠_i𝐠_j=1/K\pm O(1/\sqrt{N})`$.)
The average overlap between different weights is now $`R=C^21/(K1)`$, their average norm $`w_i=\sqrt{𝐂^2+1}`$. With this, Eq. (49) can be evaluated:
$`{\displaystyle \frac{\sigma ^2}{K}}`$ $`=`$ $`{\displaystyle \frac{1}{K}}{\displaystyle \underset{i=1}{\overset{K}{}}}1+{\displaystyle \underset{i=1}{\overset{K}{}}}{\displaystyle \underset{ji}{\overset{K}{}}}\text{sign}(𝐱𝐰_i)\text{sign}(𝐱𝐰_j)_𝐱`$ (50)
$`=`$ $`1+(K1)\left(1{\displaystyle \frac{2}{\pi }}\mathrm{arccos}\left({\displaystyle \frac{C^21/(K1)}{C^2+1}}\right)\right).`$ (51)
If $`C`$ is set to 0 and $`K`$ is large, a linear expansion of the arccos term in Eq. (51) gives $`\sigma _{opt}^2/K12/\pi 0.363`$. The small anticorrelations (of order $`1/K`$) between the vectors suffice to change the prefactor in the standard deviation.
If $`C`$ is much larger than $`g`$, there is a strong correlation between the perceptrons. Most perceptrons will agree with the classification by the center of mass $`\text{sign}(𝐱𝐂)`$. As $`C\mathrm{}`$, $`\sigma ^2/K`$ saturates at $`K`$.
### B Hebbian Learning
Now each perceptron is trying to learn the decision of the minority according to rule $`H`$. $`S`$ denotes the majority decision:
$$𝐰_i^+=𝐰_i\frac{\eta }{M}𝐱\text{sign}(\underset{j=1}{\overset{N}{}}\text{sign}(𝐱𝐰_j))=𝐰_i\frac{\eta }{M}𝐱S.$$
(52)
As the same correction is added to each weight vector, their mutual distances remain unchanged. Only the center of mass is shifted. We now treat $`C`$ as an order parameter:
$`𝐂^+`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{K}{}}}{\displaystyle \frac{𝐰_i}{N}}{\displaystyle \frac{\eta }{M}}𝐱s.`$ (53)
$`C_{}^{2}{}_{}{}^{+}`$ $`=`$ $`C^2{\displaystyle \frac{2\eta }{N}}𝐱𝐂S+{\displaystyle \frac{\eta ^2}{N}}.`$ (54)
To average over $`𝐱𝐂S`$ in the thermodynamic limit, we introduce a field $`h=𝐱𝐂`$ and average over $`𝐱`$ for fixed $`h`$:
$$𝐱𝐂S=|h|\text{sign}(\underset{i=1}{\overset{K}{}}\text{sign}(h)\text{sign}(𝐱𝐠_i+h)).$$
(55)
The quantity $`\text{sign}(h)\text{sign}(𝐱𝐠_i+h)`$ is a random variable with mean $`\text{erf}(|h|/\sqrt{2})`$ and variance $`1\text{erf}(|h|/\sqrt{2})^2`$. In a linear approximation for small $`|h|`$, we replace this by mean $`\sqrt{2/\pi }|h|`$ and variance 1.
For sufficiently large $`K`$, one can use the Central Limit Theorem to show that $`_{i=1}^K\text{sign}(h)\text{sign}(𝐱𝐠_i+h)`$ becomes a Gaussian random variable with mean $`\sqrt{2/\pi }K|h|`$. Since the terms of the sum in (55) are anticorrelated rather than independent, the variance turns out to be $`(12/\pi )K`$ rather than $`K`$, analogously to Eq. (51). This yields
$$\text{sign}(\underset{i=1}{\overset{K}{}}\text{sign}(h)\text{sign}(𝐱𝐠_i+h))=\text{erf}(\sqrt{K/(\pi 2)}|h|).$$
(56)
Since $`h`$ is a Gaussian variable with mean $`0`$ and variance $`C^2`$, the average over $`hS`$ can now be evaluated. We find the following differential equation for the norm of the center of mass:
$$\frac{dC^2}{d\alpha }=\frac{4\eta }{\sqrt{2\pi }}\sqrt{\frac{2K/(\pi 2)}{1+2K(\pi 2)C^2}}C^2+\eta ^2.$$
(57)
The fixed point of $`C`$, which can be plugged into Eq. (51) to get $`\sigma ^2/K(\eta ,K)`$, is
$$C=\frac{\sqrt{\pi }}{4}\eta \sqrt{1+\sqrt{1+\frac{16(\pi 2)}{\pi K\eta ^2}}}$$
(58)
(see Figs. 8 and 9).
If $`C`$ is large, the majority of perceptrons will usually make the same decision as $`𝐂`$, which then behaves like the single confused perceptron: $`C\sqrt{2\pi }\eta /4`$ if $`K\eta ^2\mathrm{}`$ – compare to Eq. (34).
For small $`C`$, the majority may not coincide with $`\text{sign}(𝐱𝐂)`$. In that case, the learning step has a positive overlap with $`𝐂`$, leading to $`C\sqrt{\eta }`$ as $`\eta 0`$.
The derivation given is only correct if $`N\mathrm{}`$ and $`K`$ is large. However, simulations show very good agreement even for $`K=21`$ and $`N=100`$ (see Fig. 9). For a smaller number of dimensions $`N`$, there is even a tendency towards smaller $`\sigma ^2/K`$. This can be understood in the extreme case of $`N=1`$: Each perceptron is characterized by one number; the outcome is decided by whether the majority of numbers is smaller than $`0`$ or larger, regardless of the “pattern”. The learning step consists of shifting all numbers up or down by the same amount. In the case of small $`\eta `$, the fixed point is characterized by $`(N1)/2`$ players firmly on one side of the origin, $`(N1)/2`$ on the other side, and one unfortunate loser who changes sides at every step.
Interestingly, if the time series generated by the minority decisions is used as patterns, the functions $`\sigma ^2(C)`$ and $`C(\eta )`$ are quantitatively different from those found for random patterns. However, in the final result $`\sigma ^2(\eta )`$ no disagreement can be noticed (see Fig. 9).
The presented Hebb algorithm may appear too simplistic and the chosen initial conditions too artificial. It must therefore be emphasized that there are other learning algorithms that lead to the same anticorrelated state. In particular, a variation of rule $`PN`$ has proven successful in simulations (see Fig. 10): all perceptrons that are on the minority side take a learning step, and weights are kept normalized. The regular rule $`P`$ where perceptrons on the majority side move, however, leads to strong clustering and $`\sigma ^2/KK`$.
The absence of scaling behaviour if $`N>K`$ and the fact that smaller dimensions (corresponding to smaller memory of the time series) even improve the results show that the conclusions drawn from the “conventional” Minority Game do not apply to all conceivable strategies for the Bar Problem. We think that the dependence of $`\sigma ^2/K`$ on the ratio between available strategies and players is caused by the use of quenched strategies and will not arise in any scenario in which agents stick to one strategy which is fine-tuned by some learning process.
The case of $`N=1`$ implies that there are strategies that give $`\sigma ^2/K1/K`$. We will elaborate this point in another publication.
## V Summary
We have investigated several scenarios of mutually interacting neural networks. Using perceptrons with well-known on-line training algorithms in the limit of infinite system size, we derived exact equations of motion for the dynamics of order parameters which describe the properties of the system. In the first scenario a system of $`K`$ perceptrons is placed on a ring. All perceptrons receive the same input and each perceptron is trained by the output of its neighbour on the ring. We have used two well-known training algorithms: the perceptron rule which concentrates on examples where the networks disagree, and the Hebbian rule where each example changes the weights. We find that with unnormalized weights the system relaxes to a stationary state of high symmetry: each perceptron has the same overlap with all others. The overlap depends on the learning rate: with increasing $`\eta `$ the perceptrons increase their mutual angle as much as possible.
For the perceptron learning rule with normalized weights we find phase transitions with increasing learning rate $`\eta `$. For large values of $`\eta `$, the symmetry is broken, but the symmetry of the ring is still conserved. For the Hebbian rule we find a different behaviour. The lengths of the weights diverge, the mutual angles shrink to zero and the perceptrons eventually come to perfect agreement in the limit of infinitely many training examples.
We furthermore study the behaviour of perceptrons that pursue competing learning aims for different learning algorithms. If two perceptrons follow mutually exclusive learning aims using the same algorithm, a draw results. If they use different rules, the outcome depends on factors like the rescaled learning rate $`\eta /w`$. We find that a perceptron that learns the opposite of its own prediction cannot be tracked by a student perceptron that learns the positive output of the confused teacher: all rules achieve a negative overlap.
Finally an ensemble of interacting perceptrons is used to solve a model of a closed market. Each agent uses a perceptron which is trained on the decision of the minority. Our analytic solution shows that the system relaxes to a stationary state which yields a good performance of the system for small learning rates $`\eta `$. In contrast to the minority game of Refs. our approach leads to identical profits for all agents in the long run. In addition, the performance of the algorithm is insensitive to the size of the history window used for the decision.
This paper is a first step towards more complex models of interacting neural networks. We have presented analytically accessible cases which may open the road to a general understanding of interacting adaptive systems with possible applications in biology, computer science and economics.
## VI Acknowledgement
All authors are grateful for support by the GIF. This paper also benefitted from a seminar at the Max-Planck-Institut komplexer Systeme, Dresden. We thank Johannes Berg, Michael Biehl, Liat Ein-Dor, Andreas Engel, Georg Reents, and Robert Urbanczik for helpful discussions.
## VII Appendix
The following averages are used in our calculations to derive deterministic differential equations from the update rules. The angled brackets denote averages over isotropically distributed pattern vectors. In the limit $`N\mathrm{}`$, $`𝐰_1𝐱`$ and $`𝐰_2𝐱`$ are correlated gaussian random variables, and the averages can be calculated by integrating over their joint probability distribution with appropriate boundaries. In many cases, simple geometrical calculations give the same result with less effort.
$`𝐱𝐰_1\sigma _2\mathrm{\Theta }(\sigma _1\sigma _2)`$ $`=`$ $`{\displaystyle \frac{w_1}{\sqrt{2\pi }}}(1\mathrm{cos}(\theta ));`$ (59)
$`𝐱𝐱\mathrm{\Theta }(\sigma _1\sigma _2)`$ $`=`$ $`N{\displaystyle \frac{\theta }{\pi }};`$ (60)
$`𝐱𝐰_1\sigma _1\mathrm{\Theta }(\sigma _1\sigma _2)`$ $`=`$ $`{\displaystyle \frac{w_1}{\sqrt{2\pi }}}(1+\mathrm{cos}(\theta ));`$ (61)
$`𝐱𝐱\mathrm{\Theta }(\sigma _1\sigma _2)`$ $`=`$ $`N(1{\displaystyle \frac{\theta }{\pi }});`$ (62)
$`𝐱𝐰_1\sigma _1`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}w_1;`$ (63)
$`𝐱𝐰_1\sigma _2`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi }}}w_1\mathrm{cos}(\theta );`$ (64)
$`f_{\text{opt}}`$ $`=`$ $`{\displaystyle \frac{2w_2}{\sqrt{2\pi }}}{\displaystyle \frac{\mathrm{sin}(\theta )^2}{\mathrm{cos}(\theta )}};`$ (65)
$`f_{\text{opt}}𝐱𝐰_2\sigma _1`$ $`=`$ $`0;`$ (66)
$`I`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{1}{\sqrt{2\pi }}}\mathrm{exp}\left({\displaystyle \frac{1+\mathrm{cos}(\theta )^2}{2\mathrm{sin}(\theta )^2}}x^2\right)\left(\mathrm{\Phi }(x\mathrm{cot}(\theta ))\mathrm{\Phi }(x\mathrm{cot}(\theta ))\right)^1𝑑x;`$ (67)
$`f_{\text{opt}}^2`$ $`=`$ $`{\displaystyle \frac{w_2^2}{2\pi }}\mathrm{tan}(\theta )^2I;`$ (68)
$`f_{\text{opt}}𝐱𝐰_1\sigma _1`$ $`=`$ $`{\displaystyle \frac{w_1w_2}{2\pi }}{\displaystyle \frac{\mathrm{sin}(\theta )^2}{\mathrm{cos}(\theta )}}I.`$ (69) |
warning/0003/astro-ph0003247.html | ar5iv | text | # UV Light Curves of Thermonuclear Supernovae
## 1 Introduction
Early ultraviolet (UV) emission from Type Ia supernovae (SNe Ia) is poorly known by now. Only a very few brightest events have been observed a decade ago with the International Ultraviolet Explorer (IUE). Although observational data were sometimes quite fascinating, as in the case of SN 1990N (Leibundgut et al. leietal (1991), Jeffery et al. jefetal (1992)), an amount of observed UV light curves and spectra remained too small to reveal what is typical for the UV emission from SNe Ia and how individual features of the explosion can be displayed.
The Hubble Space Telescope (HST) has slightly improved the situation. More data of better quality were obtained. This allowed theorists to make a comparison between the predictions of explosion models and the observational results. Reproduction of supernova UV emission is a good test for an explosion model because this spectral region reflects more directly the distribution of $`{}_{}{}^{56}\mathrm{Ni}_{}^{}`$ synthesized during the explosion and the conditions in the exploding star. Several models were already used to fit the observed spectra. Analyses of early and late emission from SNe Ia by Kirshner et al. kiretal (1993), Ruiz-Lapuente et al. rulaetal (1995), Nugent et al. nugetal (1997) show that Chandrasekhar-mass models DD4 (Woosley & Weaver 1994b ) and W7 (Nomoto et al. nthyo (1984)) and sub-Chandrasekhar-mass helium detonation models (see Livne liv (1990); Livne & Glasner ligl (1991); Woosley & Weaver 1994a ; Höflich & Khokhlov hoekho (1996)) can reproduce some features of UV spectra of SNe Ia quite well.
In this Letter we calculate the light curves of the similar models and discuss how they differ from each other in several UV wavelength ranges. It is quite probable that more observational data will soon be available with the HST, and that the Far-Ultraviolet Spectroscopic Explorer (FUSE), operating at shorter wavelengths (Sembach sem (1999)), will be able to obtain light curves and spectra of SNe Ia in the range where they were not observed so far. The analysis proposed here can help to distinguish which mode of explosion is actually realized in the SNe Ia.
## 2 Models and method of calculations
In our analysis we have studied two Chandrasekhar-mass models: the classical deflagration model W7 (Nomoto et al. nthyo (1984)) and the delayed detonation model DD4 (Woosley & Weaver 1994b ), as well as two sub-Chandrasekhar-mass models: helium detonation model 4 of Livne & Arnett (liar (1995)) (hereafter, LA4) and low-mass detonation model with low $`{}_{}{}^{56}\mathrm{Ni}_{}^{}`$ production (hereafter, WD065; Ruiz-Lapuente et al. rula91bg (1993)). Main parameters of these models are gathered in the Table 1, as well as the values of rise time and maximum UV fluxes in the standard IUE range (just as it is the most typical form of representation; e.g., Pun et al. punetal (1995)) and in the FUSE range. For definiteness the distance to a supernova is supposed to be 10 Mpc.
The method used here for light curve modeling is multi-energy group radiation hydrodynamics. Our code stella (Blinnikov & Bartunov blba (1993); Blinnikov et al. bebpw (1998)) solves simultaneously hydrodynamic equations and time-dependent equations for the angular moments of intensity averaged over fixed frequency bands, using up to $`200`$ zones for the Lagrangean coordinate and up to 100 frequency bins (i.e., energy groups). This allows us to have a reasonably accurate representation of non-equilibrium continuum radiation in a self-consistent calculation when no additional estimates of thermalization depth are needed. Local Thermodynamic Equilibrium (LTE) for ionization and atomic level populations is assumed in our modeling. In the equation of state, LTE ionization and recombination are taken into account. The effect of line opacity is treated as an expansion opacity according to the prescription of Eastman & Pinto (easpi (1993)) and Blinnikov (blinn96 (1996, 1997)).
## 3 Results and discussion
The main results of our calculations are presented in Figs. 12. The light curves of our models are shown in near and far UV ranges. The fluxes are plotted as they would be seen for supernovae at distance of 10 Mpc. Declared sensitivity of FUSE is $`310^{15}`$ ergs s<sup>-1</sup> cm<sup>-2</sup> Å<sup>-1</sup>, and sensitivity of HST (at working range of wavelengths almost equal to that of IUE) is roughly $`10^{16}`$ ergs s<sup>-1</sup> cm<sup>-2</sup> Å<sup>-1</sup>. This allows us to estimate that SNe Ia could be observed up to 300 Mpc in the near UV and up to 30 Mpc in the far UV (with HST and FUSE, respectively). Yet it should be noticed that in the far UV SNe Ia are bright enough only during several days, and their flux declines very quickly after the maximum light, so the probability of discovering for them is quite low, unless they are very close to us (a few Mpc or even less).
Differences in the shapes and absolute brightness of SNe Ia light curves become more pronounced in this spectral range, especially in far UV, for different modes of explosion. As we have found in our paper (Sorokina et al. sbb (2000)), the light curves of W7 and DD4 are very similar in B band close to the maximum light and differ drastically several days after it. We can see likely behaviour in near-UV wavelengths (Fig. 1). In far UV, differences grow up, so that even at maximum light one can distinguish between two Chandrasekhar-mass models. Such a behaviour can most probably be explained by different distribution of $`{}_{}{}^{56}\mathrm{Ni}_{}^{}`$ inside debris of exploded star. In our case, the fraction of $`{}_{}{}^{56}\mathrm{Ni}_{}^{}`$ decreases more sharply in the model W7, and this perhaps leads to the faster decline of its light curve.
The shape of the UV light curve of WD065 is conformal to those of Chandrasekhar-mass models, though it shows much lower absolute flux (due to an order-of-magnitude lower $`{}_{}{}^{56}\mathrm{Ni}_{}^{}`$ mass). The light curve maxima of the Chandrasekhar-mass models are shifted progressively towards later epochs for longer wavelengths remaining almost equal in brightness, while the WD065 maxima occur at nearly the same epoch for all of the IUE and FUSE ranges, and the emission virtually disappears at the shorter edge of the spectrum (see Table 1 and Fig. 1), since such a small amount of $`{}_{}{}^{56}\mathrm{Ni}_{}^{}`$ as present in this model is not able to maintain high temperature inside the ejecta. The emission of WD065 in far UV becomes so weak that it could be detected only if supernova exploded in our neighbourhood (not farther than a few hundreds of parsecs from us).
The model LA4 (helium detonation in outer layers) is apparently distinguished by its rise time in the shortest spectral bands. The most interesting feature of this light curve is its clear two-maxima structure in the short LWP range. The earliest spike of the far-UV radiation is due to the outer $`{}_{}{}^{56}\mathrm{Ni}_{}^{}`$ layer specific for this model. It is well known that those helium detonation models are too blue near visual maximum (Höflich & Khokhlov hoekho (1996); Ruiz-Lapuente et al. rulaproc (1999)). This is also confirmed by our UBVRI computations (Sorokina et al. sbb (2000)). In far UV, LA4 looks out not so hot, yet it can be detected in far UV earlier than in visual light.
One should be cautious applying our results directly to observations in UV range. A large fraction of SNe Ia shows a significant correlation with star-forming regions (Bartunov et al. btsf (1994); McMillan & Ciardullo mcmici (1996); Bartunov & Tsvetkov batsv (1997)). The circumstellar medium in those regions can absorb radiation, especially in UV band. In this case our predictions should be used as an input for calculations of reprocessing of UV photons to redder wavelengths.
Certainly, more thorough investigation of the UV emission from SNe Ia has to be done. It is still necessary to calculate UV light curves of wider range of SNe Ia models and to predict their UV spectra. This work is worth doing because, as it is seen from this Letter, near-UV and far-UV observations with modern UV space telescopes, when combined with standard UBVRI study, could be used as an efficient means to distinguish modes of explosion of thermonuclear supernovae leading us to better understanding of these phenomena.
###### Acknowledgements.
We would like to thank W. Hillebrandt, who has encouraged us to do this work. We are grateful to E. Livne, K. Nomoto, P. Ruiz-Lapuente, and S. Woosley for constructing the SNe Ia models used in our analysis, and we are thankful to P. Lundqvist for his comments on the possibilities of modern space telescopes. The work of SB is supported in Russia by ISTC grant 370–97, in Germany by MPA, and in Japan by ILE, Osaka University. |
warning/0003/astro-ph0003379.html | ar5iv | text | # Detection of polarized mm and sub-mm emission from Sgr A⋆
## 1 Introduction
Observations of stellar proper motions in the vicinity of Sgr A (Ghez et al. 1998), the non-thermal radio source at the apparent centre of the Galaxy, show that its mass of $`2.5\times 10^6`$M is highly compact on a scale $`<0.01`$pc, reinforcing its claim to be a massive black-hole candidate. The spectrum of Sgr A extends as a rough power law $`\nu ^\alpha `$, where $`\alpha `$ lies between $`\frac{1}{4}`$ and $`\frac{1}{3}`$, from a low-frequency turn-over at a few GHz up to $``$ 100GHz, (e.g. Mezger, Duschl and Zylka 1996 and references therein), and above this frequency there is evidence of a mm to sub-mm excess over this power law extending almost to the atmospheric cut-off near 1000 GHz (Serabyn et al. 1997, Falcke et al. 1998). The nature of the mm/sub-mm excess has been discussed by Serabyn et al. (1997) and the possibility that it is due to dust is effectively eliminated: at 1.3mm its size is less than 1 arcsecond with an implied brightness temperature in excess of 100 K, inconsistent with the shorter wavelength data, and the implication is that there are two separate synchrotron components.
The radio spectrum of Sgr A has for some years been modelled in various ways in terms of synchrotron radiation but searches for linear polarization, the characteristic signature of this mechanism, have only recently been reported. Bower et al. (1999a,b) used the VLA at 4.8, 8.4, 22 and 43 GHz in its spectropolarimetric mode and the BIMA array at 86 GHz finding upper limits of $`<`$0.2% at the lower frequencies and $`<`$1% at 86 GHz. These small upper limits on linear polarization, given the sensitivity of the observations to large rotation measures due to Faraday rotation (RM $`10^7`$ rad m<sup>-2</sup> at 8.4 GHz), are difficult to account for and this is discussed at length by Bower et al. (1999a). Apart from this there has been only a marginal detection at 800 µm of $`4.9\pm 3.2\%`$ at 129 $`\pm 19^{}`$ (Flett and Murray 1991) and a report of circular polarization at 4.8 GHz by Bower et al. (1999c).
In mm to sub-mm polarimetric imaging studies of the central 15 pc of the Galaxy we have detected linear polarization in Sgr A at 750, 850, 1350 and 2000 µm which we present here. Discussion of the field distribution in the circum-nuclear disk (CND) and neighbouring molecular clouds as revealed by these observations will be presented separately.
## 2 Observations and Results
The observations were made using the SCUBA camera (Holland et al. 1999) and its polarimeter (Greaves et al. 1999) on the 15m James Clerk Maxwell Telescope (JCMT) in Hawaii. SCUBA observes simultaneously with arrays of 37 and 91 bolometers, at either 850 and 450 $`\mu `$m respectively, or 750 and 350 $`\mu `$m with a change of filters. Polarimetric observations were made in imaging mode at all four submillimetre wavelengths, and single-point polarimetry towards Sgr A\* was also done at 1.35 and 2mm, using single bolometers and supplementing the data with small grid maps. The arrays have fields of view of 2.3 arcmin, and full-width half-maximum beam sizes range from 7–8<sup>′′</sup> at the shortest wavelengths to 34<sup>′′</sup> at 2 mm (Table 1).
Observations were made by chopping the secondary mirror at 7.8 Hz to remove the mean sky level, and nodding between left and right beams at slower rates to take out sky gradients. For polarimetry, a rotating half-wave plate and fixed etched grid were used to modulate the signal seen by the detectors. Polarimetry at 750 and 850 $`\mu `$m resulted in a Nyquist-sampled image after 32s of integration, after which the waveplate was stepped by 22.5. The single-pixel data were observed in a similar manner but with 8s of integration per waveplate angle. Total integration times were 5–10 complete waveplate cycles, or 15–85 minutes. The data were reduced using the SURF (Jenness and Lightfoot 1998) and POLPACK (Starlink user note 223) software packages; for each pixel a sinusoidal modulation is fitted to deduce the source percentage and direction ($`p,\theta `$) of the polarization.
From observations of Saturn and Uranus, during the run and from archival data, the instrumental polarization is found to be close to 1% and known to $`\pm 0.1`$$`0.2`$ % at a given wavelength and elevation. At 450 and 1350$`\mu `$m the instrumental polarizations are larger (3.5 and 2.7% respectively due to wind blind structure) but still known to 0.2 to 0.8% respectively. The planets were assumed to be unpolarized and a limited check was made on this for Saturn where sky rotation can be used to separate instrumental from any planetary polarization. While we cannot exclude the possibility that Saturn is slightly polarized this introduces an error of less than 0.3% in the instrumental polarization subtraction. The instrumental polarization changes by only 0.1% over the elevation range used and the uncertainties in the corrected polarization is considered good to $`\pm 0.10.2`$%. Errors in the absolute orientation of the waveplate are one degree or less.
Intensity calibration at 1350 and 2000$`\mu `$m was obtained using Uranus taking the brightness temperatures as 96K and 110K respectively, and at 850$`\mu `$m from Mars taking $`T_B`$ = 208K. Since the emphasis is on the polarimetric results, we did not perform extremely accurate calibration observations, but typical uncertainties are only about $`\pm 10`$ %. For the 750 $`\mu `$m data Saturn was used for calibration taking $`T_B`$=123K and a greater error was introduced since this planet is larger than the beam; we note that the 750 $`\mu `$m flux for Sgr A appears anomalously low. Atmospheric tranmission was measured with skydips.
A fundamental limitation is the maximum chop throw of 180<sup>′′</sup> (in this case at a position angle of 145 east of north), which proved to be insufficient at 350 and 450 $`\mu `$m. Surrounding dust emission has a steep spectral index (Pierce-Price et al.. 2000) resulting in increasing off-beam contamination at these shortest wavelengths where Sgr A was not detected above background. At 850 $`\mu `$m the off-beam signals are only 15% of the Sgr A flux and these uncertainties should be similar or smaller at 750, 1350 and 2000 $`\mu `$m at all of which Sgr A was clearly detected. During the observation period the field of view rotates but the chop orientation on the sky is maintained; thus every spatial point in the map retains the same chop positions in $`\alpha `$ and $`\delta `$ although different array bolometers may be involved. The results of polarimetric imaging in the vicinity of Sgr A are presented in Table 1 and the total flux at the position of Sgr A is given in column 3.
There will in general be two contributions to the background, namely from cool dust in the inner cavity and the edges of the CND and free-free emission from the HII region Sgr A West, predominantly from the central ionized filaments.
Dust and free-free emission contribute significantly to the observed flux but in this wavelength region there is no independent evidence at sufficient resolution to determine the contribution of the dust emission component in the central beams. In the sub-mm the flux in annuli from one to two beam radii were determined from the images and at 1350 and 2000$`\mu `$m observations in eight beams circumferentially distributed about SgrA sampled the ambient flux between 1.7–3.7 beam radii from SgrA. Estimates of the free-free contribution in the central beams and these peripheral areas were made from the 2 arcsecond resolution 3.6cm radio continuum maps of Roberts and Goss (1993), after removing the Sgr A point source itself. These fluxes were then scaled to the present wavelengths as $`\nu ^{0.1}`$, appropriate for optically thin free-free emission, and the approximation is made that the central dust contribution is just that derived from the peripheral regions after subtraction of the free-free component (column 7 in the table). The flux from SgrA after subtraction of the dust and free-free contributions (column 8) will also include any local dust excess (or deficit) over background. Since at 2mm the correction for free-free is large due to the large beam size the reliability of the non-thermal flux estimate at this wavelength will be affected.
The observed E vector polarizations and position angles (north through east) are listed in Table 1 in columns 9 and 10. At 450$`\mu `$m there is general polarization over the central 30 arcseconds with no significant change at the position of Sgr A and its average $``$ 3% at a position angle of 100 (E vector) is attributed to dust. At wavelengths between 750 and 2000$`\mu `$m the polarization at the position of Sgr A is well detected at the 10-$`\sigma `$ level. In the sub-mm it stands out from its surroundings and is clearly associated with the flux peak at the position of Sgr A. This can be seen in Fig 1 which shows the central 80 $`\times `$ 100<sup>′′</sup> region about Sgr A at 850$`\mu `$m. At 1350 and 2000$`\mu `$m the observations are from single pointed observations where the HII region Sgr A West and Sgr A are the dominant sources in the beam and Sgr A is the only significant contributor to polarization.
## 3 Interpretation
At 2000$`\mu `$m, and to a lesser extent at 1350$`\mu `$m, there are only two contributors to the flux: diffuse free-free emission and Sgr A itself. Since the former is unpolarized the polarization observed confirms the synchrotron nature of Sgr A; it also suggests that the polarization is only associated with the mm/sub-mm excess since it has not been observed at wavelengths of 3.5mm and longer where the power-law spectrum dominates. Note that at 2000$`\mu `$m the largest contributer is from free-free emission and therefore the intrinsic polarization of SgrA must be large.
A large difference of position angle between the mm and sub-mm observations is evident in Table 1 and various ways of understanding this are next investigated.
Although the mm position angles themselves are very close the three shortest wavelength position angles define a straight line when plotted against $`\lambda ^2`$; if a rotation of $`\pi `$ is subtracted from the 2mm angle a rough linear relation is maintained. However the rotation measure then indicated is so large ($`1.44\times 10^6`$ rad/m<sup>2</sup> = $`\pi /(\lambda _2^2\lambda _{1.35}^2)`$) that polarization within the band pass of 40GHz at 2mm would be undetectable. Therefore the closeness of the mm position angles indeed indicate that there is little if any Faraday rotation.
Changes in polarization and position angle with time are well documented in the compact cores of blazars and some AGN (e.g. Saikia and Salter 1988) and also have been observed at mm wavelengths (Nartallo et al. 1998). Sgr A is known to be variable on a time scale of months. While there is a separation of some months between the observations presented here, those at 750, 1350 and 2000$`\mu `$m were all made in August 1999 within a week and these do show the large position angle difference between the mm and sub-mm polarizations. Furthermore, the 850$`\mu `$m observation in March also lies on the same trend of position angle with wavelength. Variability is therefore unlikely to have been a significant influence on the position angle, although it seems possible that Sgr A may have increased in brightness during this period.
Since the sub-mm dust emission is polarized it will affect the observed polarization and it is tempting to try to attribute the whole position angle change to dust.
Denoting by $`I_o,I_d`$ and $`I_s`$ the observed central flux, the estimated dust and Sgr A components respectively and by $`q_o,q_d`$ and $`q_s`$ their Stokes $`q`$ parameters, the synchrotron component $`q_s`$ can be found from
$$I_oq_o=I_dq_d+I_sq_s$$
and similarly for the Stokes $`u`$ components.
It is difficult to estimate the dust contribution to polarization from the 750 and 850$`\mu `$m observations because of the proximity of Sgr A and its high intrinsic polarization. Instead we use the 450$`\mu `$m observations in which Sgr A is not detected, and the polarization is $``$ 3% at a position angle of $`100^{}`$. There are some problems here with polarized flux in the reference beams but other indicators of dust alignment in the central parsec are consistent with this estimate: observations at 350$`\mu `$m (Novak et al. 2000) yield 1.9% at 87 near the position of SgrA, and at 100$`\mu `$m the results are similar (Hildebrand et al. 1993) though both these are in a larger beam; the only information at high resolution is from the mid-infrared (Aitken et al. 1996) where integrations over a central 14 arcsecond diameter yield 2.4% at 125, admittedly weighted to warm dust.
The intrinsic polarization resulting from removing the unpolarized free-free contribution and adopting 3% at 100 for dust emission is shown in Table 1 columns 11 and 12. At mm wavelengths these numbers are insensitive to the uncertainties in the adopted dust polarization and even in the submm the range of dust polarizations given in the previous paragraph changes the intrinsic polarization by $`\pm `$3% at most and less than $`\pm `$3. Similarly uncertainties in the relative fractions of the three flux components has little effect on the intrinsic position angle but changing dilution does introduce errors to the intrinsic polarization fraction and at 750 and 2000$`\mu `$m the upper bounds are poorly defined. In Table 1 the errors to the intrinsic polarization are derived by assuming $`\pm `$15% errors to the free-free and $`\pm `$20% errors to the dust contributions except at 2000$`\mu `$m where the adopted error is $`\pm `$40% (these dust contribution errors are derived from the observed variation in the peripheral flux). It is clear that dust polarization cannot explain the abrupt change of position angle between the mm and sub-mm, rather it is increased and it approaches $`\pi `$/2.
We can turn the above argument around to determine what dust polarization is needed in the sub-mm to cause the position angle shift given the intrinsic synchrotron polarization of the mm results, say $`10`$% at 85. With the 850$`\mu `$m fluxes and polarization given in Table 1 the required dust polarization is then 12.0% at 163, larger than any dust emission polarization yet observed. It might be argued that this could be due to unusual conditions close to Sgr A and an additional dust component in the central beam. Any such emission is already included in the Table 1 entry for Sgr A and because the same polarized intensity ($`450`$mJy) has to be supplied from much less dust emission the required polarization would be in excess of 25%.
An assumption made so far is that throughout the wavelength region of these observations the synchrotron radiation remains either optically thin or self-absorbed. Most models of the synchrotron SED in Sgr A consider the mm/sub-mm excess to be self absorbed but it may well be that at the shortest wavelengths it becomes optically thin. In that case the synchrotron position angle shifts through 90 and becomes close to 175 in the sub-mm. The dust corrected sub-mm points are now discrepant by 6 to 14 and this could be attributed to the crudeness of the correction.
Approximating the synchrotron emitting region by a simple slab with constant energy density and magnetic field we find (e.g. Ginzburg and Syrovatskii 1965, 1969, Pacholczyk and Swihart 1967) that for the polarization to change sign near 1mm then $`\nu _{ssa}`$, the synchrotron self absorption frequency, lies in the range 0.6–0.7mm, close to the peak of the mm/sub-mm excess, and the required fields, densities and energies are similar to those in the compact source models of Beckert and Duschl (1997). The polarization transition of this simple model is quite sharp, as is observed, with the polarization quickly approaching saturation on the long wavelength side, changing sign near 1mm and steadily rising at shorter wavelengths.
A variant on this explanation is that self-absorption results in different optical depths with possibly different field distributions being probed as a function of wavelength. In that case, however, one should encounter diminishing polarization towards shorter wavelengths due to the superposition of a range of field orientations, and this is not observed.
## 4 Discussion
Synchrotron radiation from Sgr A has been detected between 750$`\mu `$m and 2000$`\mu `$m with an intrinsic polarization $``$ 10%. This polarization arises from the spectral region of the mm/sub-mm excess. The present observations are compounded by the presence of polarized emission from dust and dilution by free-free emission in these large beams. Although this has led to a number of possible explanations for the large position angle shift between the mm and sub-mm results all but one of these can be eliminated: the closeness of the position angles at 1.35 and 2mm require that there is little or no Faraday rotation, dust polarization can only produce a small shift which is in the wrong sense, and the shift is not due to variability. Only a transition between optically thin synchrotron radiation in the sub-mm and self-absorption in the mm region appears as a plausible explanation of the shift. Future small beam observations will be needed to confirm this result.
A puzzle re-emphasized here is the lack of polarization at 3.5mm (Bower et al.. 1999a,b) and longer wavelengths in view of the large intrinsic polarization of Sgr A at 1.35 and 2mm. Sufficient dilution of the mm/sub-mm excess to bring it below 1% requires not only a steeply falling spectrum with decreasing frequency for the excess, where $`\nu ^{5/2}`$ is expected for self absorption, but also that the cut-off frequency of the power-law section is below 2mm to ensure a fast rising flux of the power law region. It is possible, as well, that a flare in Sgr A at 100GHz in March 1998 (Tsuboi, Miyazaki and Tsutsumi 1998) has affected the result of Bower et al. 1999b.
The synchrotron self-absorption transition in the sub-mm proposed here is consistent with the model of the excess by Beckert and Duschl (1997), and requires the excess to arise from a compact source $``$ 2 Schwarzschild radii in size. Also because of the transition the $`\pi `$/2 ambiguity of the relationship between position angle and magnetic field is removed and hence the field orientation is close to east-west, since at mm wavelengths the emission is considered self-absorbed.
## 5 Conclusions
We report the observation of millimetre and sub-millimetre polarization from Sgr A, confirming the role of synchrotron radiation. The polarization is a property of the mm/sub-mm excess, demonstrating that the excess is real and not an artefact of variability and that it and the power-law spectrum arise in distinct structures. There is a large position angle shift between the mm and sub-mm observations. This can be explained by a transition between optically thin and self absorbed synchrotron radiation near 1 mm. Such a high self-absorption frequency implies a very compact source $``$ 2R<sub>S</sub>.
We are grateful to the UK Panel for Allocation of Telescope Time for the award of observing time for this project. JSR acknowledges a Royal Society Fellowship and DP-P a PPARC Research Studentship. The JCMT is operated by the Joint Astronomy Centre on behalf of PPARC of the UK, the Netherlands OSR, and NRC Canada. The authors would like to thank the NCSA Astronomy Digital Image Library (ADIL) for providing images for this paper. We thank an unknown referee for useful comments. |
warning/0003/cs0003061.html | ar5iv | text | # dcs: An Implementation of DATALOG with Constraints
## General Info
$`\mathrm{DC}`$ is an answer set programming (ASP) system (?) similar to propositional logic but extended to include Horn clauses. The semantics of $`\mathrm{DC}`$ is a natural extension of the semantics of propositional logic. The $`\mathrm{DC}`$ system is implemented in two modules, ground and dcs, which are written in the ’C’ programming language and compiled with gcc. There are approximately 2500 lines of code for ground and approximately 1500 for dcs. $`\mathrm{DC}`$ has been implemented on both SUN SPARC and a PC running linux.
A $`\mathrm{DC}`$ theory (or program) consists of constraints and Horn rules (DATALOG program). This fact motivates our choice of terminology — DATALOG with constraints. We start a discussion of $`\mathrm{DC}`$ with the propositional case. Our language is determined by a set of atoms $`At`$. We will assume that $`At`$ is of the form $`At=At_CAt_H`$, where $`At_C`$ and $`At_H`$ are disjoint.
A $`\mathrm{DC}`$ theory (or program) is a triple $`T_{dc}=(T_C,T_H,T_{PC})`$, where
1. $`T_C`$ is a set of propositional clauses $`\neg a_1\mathrm{}\neg a_mb_1\mathrm{}b_n`$ such that all $`a_i`$ and $`b_j`$ are from $`At_C`$,
2. $`T_H`$ is a set of Horn rules $`a_1\mathrm{}a_mb`$ such that $`bAt_H`$ and all $`a_i`$ are from $`At`$,
3. $`T_{PC}`$ is a set of clauses over $`At`$.
By $`At(T_{dc})`$, $`At_C(T_{dc})`$ and $`At_{PC}(T_{dc})`$ we denote the set of atoms from $`At`$, $`At_C`$ and $`At_{PC}`$, respectively, that actually appear in $`T_{dc}`$.
With a $`\mathrm{DC}`$ theory $`T_{dc}=(T_C,T_H,T_{PC})`$ we associate a family of subsets of $`At_C(T_{dc})`$. We say that a set $`MAt_C(T_{dc})`$ satisfies $`T_{dc}`$ (is an answer set of $`T_{dc}`$) if
1. $`M`$ satisfies all the clauses in $`T_C`$, and
2. the closure of $`M`$ under the Horn rules in $`T_H`$, $`M^c=LM(T_HM)`$ satisfies all clauses in $`T_{PC}`$ ($`LM(P)`$ denotes the least model of a Horn program $`P`$).
Intuitively, the collection of clauses in $`T_C`$ can be thought of as a representation of the constraints of the problem, Horn rules in $`T_H`$ can be viewed as a mechanism to compute closures of sets of atoms satisfying the constraints in $`T_C`$, and the clauses in $`T_{PC}`$ can be regarded as constraints on closed sets (we refer to them as post-constraints). A set of atoms $`MAt_C(T_{dc})`$ is a model if it (propositionally) satisfies the constraints in $`T_C`$ and if its closure (propositionally) satisfies the post-constraints in $`T_{PC}`$. Thus, the semantics of $`\mathrm{DC}`$ retains much of the simplicity of the semantics of propositional logic. $`\mathrm{DC}`$ was introduced by the authors in (?).
## Applicability of the System
$`\mathrm{DC}`$ can be used as a computational tool to solve search problems. We define a search problem $`\mathrm{\Pi }`$ to be determined by a set of finite instances, $`D_\mathrm{\Pi }`$, such that for each instance $`ID_\mathrm{\Pi }`$, there is a finite set $`S_\mathrm{\Pi }(I)`$ of all solutions to $`\mathrm{\Pi }`$ for the instance $`I`$. For example, the problem of finding a hamilton cycle in a graph is a search problem: graphs are instances and for each graph, its hamilton cycles (sets of their edges) are solutions.
A $`\mathrm{DC}`$ theory $`T_{dc}=(T_C,T_H,T_{PC})`$ solves a search problem $`\mathrm{\Pi }`$ if solutions to $`\mathrm{\Pi }`$ can be computed (in polynomial time) from answer sets to $`T_{dc}`$. Propositional logic and stable logic programming (??) are used as problem solving formalisms following the same general paradigm.
We will illustrate the use of DC to solve search problems by discussing the problem of finding a hamilton cycle in a directed graph. Consider a directed graph $`G`$ with the vertex set $`V`$ and the edge set $`E`$. Consider a set of atoms $`\{hc(a,b):(a,b)E\}`$. An intuitive interpretation of an atom $`hc(a,b)`$ is that the edge $`(a,b)`$ is in a hamilton cycle. Include in $`T_C`$ all clauses of the form $`\neg hc(b,a)\neg hc(c,a)`$, where $`a,b,cV`$, $`bc`$ and $`(b,a),(c,a)E`$. In addition, include in $`T_C`$ all clauses of the form $`\neg hc(a,b)\neg hc(a,c)`$, where $`a,b,cV`$, $`bc`$ and $`(a,b),(a,c)E`$. Clearly, the set of propositional variables of the form $`\{hc(a,b):(a,b)F\}`$, where $`FE`$, satisfies all clauses in $`T_C`$ if and only if no two different edges in $`F`$ end in the same vertex and no two different edges in $`F`$ start in the same vertex. In other words, $`F`$ spans a collection of paths and cycles in $`G`$.
To guarantee that the edges in $`F`$ define a hamilton cycle, we must enforce that all vertices of $`G`$ are reached by means of the edges in $`F`$ if we start in some (arbitrarily chosen) vertex of $`G`$. This can be accomplished by means of a simple Horn program. Let us choose a vertex, say $`s`$, in $`G`$. Include in $`T_H`$ the Horn rules $`hc(s,t)vstd(t)`$, for every edge $`(s,t)`$ in $`G`$. In addition, include in $`T_H`$ Horn rules $`vstd(t),hc(t,u)vstd(u)`$, for every edge $`(t,u)`$ of $`G`$ not starting in $`s`$. Clearly, the least model of $`FT_H`$, where $`F`$ is a subset of $`E`$, contains precisely these variables of the form $`vstd(t)`$ for which $`t`$ is reachable from $`s`$ by a nonempty path spanned by the edges in $`F`$. Thus, $`F`$ is the set of edges of a hamilton cycle of $`G`$ if and only if the least model of $`FT_H`$, contains variable $`vstd(t)`$ for every vertex $`t`$ of $`G`$. Let us define $`T_{PC}=\{vstd(t):tV\}`$ and $`T_{ham}(G)=(T_C,T_H,T_{PC})`$. It follows that hamilton cycles of $`G`$ can be reconstructed (in linear time) from answer sets to the $`\mathrm{DC}`$ theory $`T_{ham}(G)`$. In other words, to find a hamilton cycle in $`G`$, it is enough to find an answer set for $`T_{ham}(G)`$.
This example illustrates the simplicity of the semantics of $`\mathrm{DC}`$ — it is only a slight adaptation of the semantics of propositional logic to the case when in addition to propositional clauses we also have Horn rules in theories. It also illustrates the power of $`\mathrm{DC}`$ to generate concise encodings. All known propositional encodings of the hamilton-cycle problem require that additional variables are introduced to “count” how far from the starting vertex an edge is located. Consequently, propositional encodings are much larger and lead to inefficient solutions to the problem.
## Description of the System
In this section we will discuss general features of $`\mathrm{DC}`$. First, we will discuss the language for encoding problems and give an example by showing the encoding of the hamiltonicity problem. Second, we will describe how we execute the $`\mathrm{DC}`$ system. Third, we will give some details concerning the implementation of dcs, our solver. Last we discuss the expressitivity of $`\mathrm{DC}`$.
In the previous section we described the logic of $`\mathrm{DC}`$ in the propositional case. Our definitions can be extended to the predicate case (without function symbols). Each constraint is treated as an abbreviation of a set of its ground substitutions. This set is determined by the set of constants appearing in the theory and by additional conditions associated with the constraint ( we illustrate this idea later in this section). When constructing predicate $`DC`$-based solutions to a problem $`\mathrm{\Pi }`$ we separate the representation of an instance to $`\mathrm{\Pi }`$ from the constraints that define $`\mathrm{\Pi }`$. The representation of an instance of $`\mathrm{\Pi }`$ is a collection of facts or the extensional database (EDB). The constraints and rules that define $`\mathrm{\Pi }`$ will be referred to as the intensional database (IDB) and the language for writing the problem descriptions that constitute the IDB will be referred to as $`L_{dc}`$. The separation of IDB and EDB means only one predicate description of $`\mathrm{\Pi }`$ is needed.
The modules of $`\mathrm{DC}`$, ground and dcs provide a complete system for describing and finding solutions to problems. An IDB in $`L_{dc}`$ along with a specific EDB are the input to ground. A grounded propositional theory $`T_{dc}`$ is output by ground and used as input to dcs.
A problem description in $`L_{dc}`$ defines predicates, declares variables, and provides a description of the problem using rules. The predicates are defined using types from the EDB. Similarly, the variables are declared using types from the EDB. The rules consist of constraints, Horn rules and post-constraints. The constraints and post-constraints use several constructs to allow a more natural modeling. These constructs could be directly translated to clauses. (We use them as shorthands to ensure the conciseness of encodings.)
We present here a brief discussion of the constraints, Horn rules and post-constraint in $`L_{dc}`$ and their meanings. Let $`PRED`$ be the set of predicates occurring in the IDB. For each variable $`X`$ declared in the IDB the range $`R(X)`$ of $`X`$ is determined by the EDB.
where $`n,m`$ are nonnegative integers such that $`nm,qPRED`$ and $`p_1,\mathrm{},p_i`$ are EDB predicates or logical conditions (logical conditions can be comparisons of arithmetic expressions or string comparisons). The interpretation of this constraint is as follows: for every $`\stackrel{}{x}R(\stackrel{}{X})`$ at least $`n`$ atoms and at most $`m`$ atoms in the set $`\{q(\stackrel{}{x},\stackrel{}{y}):\stackrel{}{y}R(\stackrel{}{Y})\}`$ are true.
where $`n,m`$ are nonnegative integers such that $`nm,qPRED`$. The interpretation of this constraint is as follows: for every $`\stackrel{}{x}R(\stackrel{}{X})`$ at least $`n`$ atoms and at most $`m`$ atoms in the set $`\{q(\stackrel{}{x},\stackrel{}{y}):\stackrel{}{y}R(\stackrel{}{Y})\}`$ are true.
where $`n,m`$ are nonnegative integers such that $`nm,q_1,\mathrm{},q_jPRED`$. The interpretation of this constraint is as follows: for every $`\stackrel{}{x}R(\stackrel{}{X})`$ at least $`n`$ atoms and at most $`m`$ atoms in the set $`\{q_1(\stackrel{}{x}),\mathrm{},q_j(\stackrel{}{x})\}`$ are true.
Certain choices of $`n,m`$ in any of the Select constraints allow construction of even more specific constraints in $`T_{dc}`$:
$`n=m`$ exactly $`n`$ atoms must be true.
$`n=0`$ at most $`m`$ atoms must be true.
$`m=999`$ at least $`n`$ atoms must be true.
where $`q_1,\mathrm{},q_iPRED`$. For every $`\stackrel{}{x}R(\stackrel{}{X})`$ at least one atom in the set $`\{q_1(\stackrel{}{x}),\mathrm{},q_i(\stackrel{}{x})\}`$ must be false.
where $`q_1,\mathrm{},q_iPRED`$. For every $`\stackrel{}{x}R(\stackrel{}{X})`$ at least one atom in the set $`\{q_1(\stackrel{}{x}),\mathrm{},q_i(\stackrel{}{x})\}`$ must be true.
where $`p_1,\mathrm{},p_i,q_1,\mathrm{},q_jPRED`$. For every $`\stackrel{}{x}R(\stackrel{}{X})`$ if all atoms in the set $`\{p_1(\stackrel{}{x}),\mathrm{},p_i(\stackrel{}{x})\}`$ are true then at least one atom in the set $`\{q_1(\stackrel{}{x}),\mathrm{},q_j(\stackrel{}{x})\}`$ must be true. (This constraint represents standard propositional logic clauses.)
where $`p_1,\mathrm{},p_i,q_1,\mathrm{},q_jPRED`$. For every $`\stackrel{}{x}R(\stackrel{}{X})`$ if all atoms in the set $`\{p_1(\stackrel{}{x}),\mathrm{},p_i(\stackrel{}{x})\}`$ are true then all atoms in the set $`\{q_1(\stackrel{}{x}),\mathrm{},q_j(\stackrel{}{x})\}`$ must be true.
where $`p_1,\mathrm{},p_i,q_1,\mathrm{},q_jPRED`$.
### Methodology
Here we show the encoding of a problem in $`\mathrm{DC}`$. We will use the example of hamiltonicity of a graph which we discussed previously. Figure 1 shows the EDB format used by ground. This format is compatible to that used by smodels and others. However, we require that the EDB be in separate files from the IDB. The format for the EDB allows data to be entered as sets, ranges, or individual elements and constant values can be entered on the command line.
The IDB provides a definition of the problem in $`L_{dc}`$. The IDB file has three parts. First a definition of the predicates, next the declaration of variables and last a set of constraints and Horn clauses. The types used in the IDB must be in the data file(s). For example, the only data types in the graph file (Fig. 1) are vtx and edge. Thus the only data types which can be used in the IDB are vtx and edge.
In the idbpred section of Fig. 2, we define two predicates, the vstd predicate and the hc predicate. The vstd predicate has one parameter of type vtx and hc has two parameters both of type vtx.
The idbvar section of Fig. 2 declares two variables $`X`$, $`Y`$ both of type vtx.
The section containing the constraints, Horn clauses, and post-constraints is proceeded by the keyword idbrules (see Fig.2). The order in which the rules are entered is not important. The first constraint, $`Select(1,1,Y;edge(X,Y))hc(X,Y).`$, ensures that each vertex has exactly one outgoing edge. The second constraint, $`Select(1,1,X;edge(X,Y))hc(X,Y).`$, requires that each vertex has exactly one incoming edge.
The first Horn rule ranges over all $`X,Y`$ such that $`edge(X,Y)EDB`$ and $`Xi`$ where $`i`$ is a constant used to initialize $`vstd(i)`$. This rule requires both $`vstd(X)`$ and $`hc(X,Y)`$ to be true before we can assign the value true to $`vstd(Y)`$. The second Horn rule only requires $`hc(X,Y)`$ to be true before $`vstd(Y)`$ is assigned value true; however, in this rule $`X`$ is restricted to the value of $`i`$.
The last line is a post-constraint that requires $`vstd(X)`$ to be true for all $`XR(X)`$ ensuring that the cycle is closed.
### Running the system
Here we will describe the steps for execution of the $`\mathrm{DC}`$ system. The first module of $`\mathrm{DC}`$, ground, has as input a data file(s), a rule file and command line arguments. Output is the theory file which will be used as input to dcs. The theory is written to a file whose name is the catenation of the constants and file names given as command line arguments. The extension .tdc is then appended to the output file name. The command line arguments for ground:
ground -r rf -d df \[-c label=v\] \[-V\]
Required arguments
* rf is the file describing the problem. There must be exactly one rule file.
* df must be one or more files containing data that will be used to instantiate the theory. It is often convenient to use more than one file for data.
Optional arguments
* This option allows use of constants in both data and rule files. When label is found while reading either file it is replaced by v. v can be any string that is valid for the data type. If label is to be used in a range then v must be an integer. For example, if the data file contains the entry queens\[1..q\]. then we can define the constant q with the option -c q=8. If more than one constant is needed then -c b=3 n=14 defines both constants b,n using the -c option.
* The verbose options sends output to stdout during the execution of ground. This output may be useful for debugging of the data or rule files.
For the example of hamiltonicity we could have a data file (see Fig. 1) named 1.gph and an IDB file (see Fig. 2) named hcp. The constant i is needed in the IDB file to initialize the first vertex in the graph. The command line argument would be:
ground -r hcp -d 1.gph -c i=1
The theory file produced would be named 1\_1.gph\_hcp.tdc.
The second module of the $`\mathrm{DC}`$ system, dcs, has as input the theory file produced by ground. A file named dcs.stat is created or appended with statistics concerning the results of executing dcs on the theory. The command line arguments are:
dcs -f filename \[-A\] \[-P\] \[-C\] \[-V\]
Required arguments
* filename is the name of the file containing a theory produced by ground.
Optional arguments
* Prints the positive atoms for solved theories in readable form.
* Prints the input theory and then exits.
* Counts the number of solutions. This information is recorded in the statistics file.
* Prints information during execution (branching, backtracking, etc). Useful for debugging.
### Discussion of dcs
The $`\mathrm{DC}`$ solver uses a Davis-Putnam type approach, with backtracking, propagation and LookAHead to deal with constraints represented as clauses, select constraints and Horn rules, and to search for answer sets. The LookAHead in $`\mathrm{DC}`$ is similar to local processing performed in csat (?). However, we use different methods to determine how many literals to consider in the LookAHead phase. Other techniques, especially propagation and search heuristics, were designed specifically for the case of $`\mathrm{DC}`$ as they must take into account the presence of Horn rules in programs.
Propagation consists of methods to reduce the theory. Literals which appear in the heads of Horn rules, $`l_hAt_H`$ require different interpretations. A literal is a $`l_h`$ if and only if it appears in the head of a Horn rule. We can not guess an assignment for $`l_h`$ rather it must be computed. We can only assign value true to $`l_h`$ if it appears in the head of a Horn rule for which all literals in the body of the Horn rule have been assigned the value true. If one or more literals in the body of a Horn rule have been assigned the value false then that rule is removed. If $`l_h`$ has not been assigned a value and all Horn rules in which $`l_h`$ appeared in the head have been removed then $`l_h`$ is assigned the value false. If we have a post-constraint that required a value be assigned to $`l_h`$ and the value cannot be computed then we must backtrack.
Non Horn rules are constraints which must be satisfied. These rules are identified by tags which indicate which method is needed to evaluate the constraint during propagation.
The LookAHead procedure tests a number of literals not yet assigned a value. The LookAHead procedure is similar to the local processing procedure used for csat (?). A literal is chosen, assigned the value true, then false using propagation to reduce and evaluate the resulting theories. If both evaluations of assignments result in conflicts then we return false and backtracking will result. If only one evaluation results in conflict then we can assign the literal the opposite value and continue the LookAHead procedure. If neither evaluation results in conflict we cannot make any assignments, but we save information (number of forced literals and satisfied constraints) computed during the evaluations of the literal.
The number of literals to be tested has been determined empirically. It is obvious that if all unassigned literals were tested during each LookAHead it would greatly increase the time. However, if only a small number of literals are to be tested during each LookAHead then they must be chosen to provide the best chance of reducing the theory. To choose the literals with the best chance of reducing the theory, we order the unassigned literals based on a sum computed by totaling the weights of the unsatisfied constraints in which they appear. The constraint weight is based on both the length and type of constraint. The shorter the constraint the larger the weight and when literals are removed the weight is recomputed. The literals are tested in descending order of the sum of constraints weights. Using this method we need to test only a very small number of literals during each LookAHead to obtain good results.
At the completion of the LookAHead procedure, we use the information computed during the evaluation of literals to choose the next branching literal and its initial truth assignment.
### Expressitivity
The expressive power of $`\mathrm{DC}`$ is the same as that of logic programming with the stable-model semantics. The following theorem is presented in (?)
###### Theorem 1
The expressive power of $`\mathrm{DC}`$ is the same as that of stable logic programming. In particular, a decision problem $`\mathrm{\Pi }`$ can be solved uniformly in $`\mathrm{DC}`$ if and only if $`\mathrm{\Pi }`$ is in the class NP.
## Evaluating the System
The $`\mathrm{DC}`$ system provides a language, $`L_{dc}`$, which facilitates writing problem descriptions. Once an IDB is written in $`L_{dc}`$ it can be used for any instance of the problem for which data, in the EDB format, is available or can be generated. It is possible to add constraints to IDB for a given problem when new requirements or constraints occur. The constructs used in $`L_{dc}`$ allow for a natural description of constraints. Users need only know enough about a specific problem to be able to describe the problem in $`L_{dc}`$ (there is a user’s manual with examples). To help with programming in $`\mathrm{DC}`$ ground provides error messages and compiling information that are useful for debugging the IDB.
### Benchmarks
The $`\mathrm{DC}`$ system has been executed using problems from NP, combinatorics, and planning. In particular, it has been used to compute hamilton cycles and colorings in graphs, to solve the $`N`$-queens problem, to prove that the pigeonhole problem has no solution if the number of pigeons exceeds the number of holes, and to compute Schur numbers.
The instances for computing hamilton cycles were obtained by randomly generating one thousand directed graphs with $`v=30,40\mathrm{},80`$ and density such that $`50\%`$ of the graphs contained hamilton cycles.
The graphs for instances of coloring were one hundred randomly generated graphs for $`v=50,100,\mathrm{},300`$ with density such that $`50\%`$ had solutions when encode as 3-coloring.
### Comparison
We have used the benchmarks to compare $`\mathrm{DC}`$ with systems based on stable model semantics and satisfiability. The performance of $`\mathrm{DC}`$ solver dcs was compared with smodels, a system for computing stable models of logic programs (?), and csat, a systems for testing propositional satisfiability (?). In the case of smodels we used version 2.24 in conjunction with the grounder lparse, version 0.99.41 (?). The extended rules (?) allowed in the newer versions of smodels and lparse were used where applicable. The programs were all executed on a Sun SPARC Station 20. For each test we report the cpu user times for processing the corresponding propositional program or theory. We tested all three system using the benchmarks discussed in the previous section.
Note that csat performs comparable to $`\mathrm{DC}`$ for pigeonhole (see Fig. 6), N-queens (see Fig. 5) and coloring (see Fig. 4). These are problems where $`\mathrm{DC}`$ encodings do not use Horn rules (in the examples here only encodings for computing hamilton cycles use Horn rules). The closure properties of Horn rules allow for much smaller theories as shown in Table 2. The satisfiability theories for computing hamilton cycles are so large that they were not practical to execute for over $`40`$ vertices (see Fig. 3). smodels performs much better than satisfiability solvers for computing the hamilton cycles although not as well as $`\mathrm{DC}`$. The results for computing Schur numbers (see Table 1) also show much better results for $`\mathrm{DC}`$.
Experimental results show that dcs often outperforms systems based on satisfiability as well as systems based on non-monotonic logics, and that it constitutes a viable approach to solving problems in AI, constraint satisfaction and combinatorial optimization. We believe that our focus on short encodings (see Fig. 2) is the key to the success of dcs. |
warning/0003/hep-lat0003005.html | ar5iv | text | # A new approach to Ginsparg-Wilson fermions
## Abstract
We expand the most general lattice Dirac operator $`D`$ in a basis of simple operators. The Ginsparg-Wilson equation turns into a system of coupled quadratic equations for the expansion coefficients. Our expansion of $`D`$ allows for a natural cutoff and the remaining quadratic equations can be solved numerically. The procedure allows to find Dirac operators which obey the Ginpsparg-Wilson equation with arbitrary precision.
In the last two years we have witnessed exciting developments for chiral fermions on the lattice (see for some reviews). It was found that the perfect action approach leads to the fixed point Dirac operator which obeys the Ginsparg-Wilson equation . Also the overlap formalism for chiral fermions on the lattice has led to an explicit expression for a Dirac operator which solves the Ginsparg-Wilson equation, the so-called overlap operator . It was realized, that based on the Ginsparg-Wilson relation it is possible to redefine the chiral symmetry for lattice fermions and this new construction has spurred a wealth of interesting work .
Unfortunately the numerical implementation of the two known solutions of the Ginsparg-Wilson equation seems to be rather troublesome. Computing the inverse square root which appears in the overlap operator is a numerically challenging and expensive task. For the perfect action the situation is even worse, since already the blocking procedure necessary for its computation is extremely costly and so far has been implemented only in two dimensions .
In this letter we present a new approach to the Ginsparg-Wilson equation: The first step is to expand the most general Dirac operator $`D`$ in a basis of simple operators on the lattice. In the second step we insert the expanded $`D`$ into the Ginsparg-Wilson equation which then turns into a system of coupled quadratic equations for the expansion coefficients of $`D`$. Solving the Ginsparg-Wilson equation is equivalent to finding solutions for this system of quadratic equations. Any solution of the Ginsparg-Wilson equation, such as the overlap operator or the perfect action, corresponds to a solution of our system of quadratic equations. Many interesting aspects can be addressed in this new framework, such as e.g. the question if there is a continuous manifold of solutions in the space of coefficients of $`D`$.
Here we concentrate on exploring the possibilities for solving our system of quadratic equations numerically and using this approach for constructing new solutions of the Ginsparg-Wilson equation. We show that our expansion of $`D`$ provides for a natural cutoff which turns the quadratic equations into a simple finite system. Solving this system numerically gives the values for the expansion coefficients. This procedure allows to construct approximate solutions which obey the Ginsparg-Wilson equation with arbitrary precision.
Before we outline the details of our construction, let us begin with some remarks: Usually the derivative term on the lattice is discretized by the following nearest neighbor term ($`U_\mu (x)`$ SU(N) and we set the lattice spacing to 1):
$$\frac{1}{2}\underset{\mu =1}{\overset{4}{}}\gamma _\mu \left[U_\mu (x)\delta _{x+\widehat{\mu },y}U_\mu (x\widehat{\mu })^1\delta _{x\widehat{\mu },y}\right].$$
(1)
However, it is perfectly compatible with all the symmetries to instead discretize the derivative term using
$`{\displaystyle \frac{1}{4}}{\displaystyle \underset{\mu =1}{\overset{4}{}}}\gamma _\mu [U_\mu (x)U_\mu (x+\widehat{\mu })\delta _{x+2\widehat{\mu },y}`$ (2)
$`U_\mu (x\widehat{\mu })^1U_\mu (x2\widehat{\mu })^1\delta _{x2\widehat{\mu },y}],`$ (3)
and there are many more terms that are eligible. Thus an ansatz for the most general Dirac operator $`D`$ must allow for a superposition of all of the possible discretizations for the derivative term. In order to remove the doublers we also have to allow for terms that come with 1I in spinor space and again we allow for all possible terms. We generalize our $`D`$ further, by including all terms also for the remaining elements $`\mathrm{\Gamma }_\alpha `$ of the Clifford algebra, i.e. we include also tensor, pseudovector and pseudoscalar terms. From our examples (1), (3) it can be seen that the terms are characterized by the orientations of the link variables $`U_\mu (x)`$. The ensemble of links supporting the gauge transporters can be viewed as paths and we can denote our ansatz for the most general lattice Dirac operator in the following form:
$$D_{x,y}=\underset{\alpha =1}{\overset{16}{}}\mathrm{\Gamma }_\alpha \underset{p𝒫_{x,y}^\alpha }{}c_p^\alpha \underset{lp}{}U_l.$$
(4)
To each generator $`\mathrm{\Gamma }_\alpha `$ of the Clifford algebra and to each pair of points $`x,y`$ on the lattice we assign a set $`𝒫_{x,y}^\alpha `$ of paths $`p`$ connecting the two points. Each path is weighted with some complex weight $`c_p^\alpha `$ and the construction is made gauge invariant by including the ordered product of the gauge transporters $`U_l`$ $``$ SU(N) for all links $`l`$ of $`p`$. The action is then given by $`S=_{x,y}\overline{\psi }_xD_{x,y}\psi _y`$, where $`x`$ and $`y`$ run over all of the lattice.
The next step is to impose on $`S`$ the symmetries which we want to maintain: Translation and rotation invariance and invariance under C and P. In addition we require our $`D`$ to be $`\gamma _5`$-hermitian, i.e. we require $`\gamma _5D\gamma _5=D^{}`$. This property can be seen to correspond to what leads to the CTP theorem in Minkowski space, i.e. the vector generators $`\gamma _\mu `$ come with a derivative term etc. (compare Equation (13) below).
Translation invariance requires the sets $`𝒫_{x,y}^\alpha `$ of paths and their coefficients to be invariant under simultaneous shifts of $`x`$ and $`y`$, while rotation invariance implies that a term and its rotated image have the same weight. Parity implies that for each path $`p`$ (with coefficient $`c_p^\alpha `$) we must include the parity-reflected copy with coefficient $`s_{parity}^\alpha c_p^\alpha `$ where the signs $`s_{parity}^\alpha `$ are defined by $`\gamma _4\mathrm{\Gamma }_\alpha \gamma _4=s_{parity}^\alpha \mathrm{\Gamma }_\alpha `$.
More interesting are the symmetries C and $`\gamma _5`$-hermiticity. It is easy to see that both of them imply a relation between the coefficient for a path $`p`$ and the coefficient of the inverse path $`p^1`$. Implementing both these symmetries firstly restricts all coefficients $`c_p^\alpha `$ to be either real or purely imaginary. Secondly we find that the coefficient for a path $`p`$ and the coefficient for its inverse $`p^1`$ are equal up to a sign $`s_{charge}^\alpha `$ defined by $`C\mathrm{\Gamma }_\alpha C=s_{charge}^\alpha \mathrm{\Gamma }_\alpha ^T`$, where $`T`$ denotes transposition and $`C`$ is the charge conjugation matrix. To be specific, we use the chiral representation for the $`\gamma _\mu `$ and $`C=i\gamma _2\gamma _4`$. When implementing all these symmetries we find that paths in our ansatz become grouped together where, up to sign factors, all paths in a group come with the same coefficient.
Before we present the explicit form of the most general Dirac operator with the above symmetries, let us first introduce an abbreviated notation: The terms in the Dirac operator are determined by the corresponding element of the Clifford algebra and the paths with their relative signs. For the two examples (1), (3) the paths are given by one (two) steps in positive $`\mu `$-direction and one (two) steps in negative $`\mu `$-direction with a relative minus sign. We now denote an arbitrary path of length $`n`$ on the lattice by $`<l_1,l_2,\mathrm{}l_n>`$, with the $`l_i`$ giving the directions of the subsequent hops, i.e. $`l_i\{\pm 1,\pm 2,\pm 3,\pm 4\}`$. So for example the terms occurring in (3) are denoted by $`<\mu ,\mu >`$ and $`<\mu ,\mu >`$, where the minus in front of the second path is due to the relative minus sign in (3).
Using this notation we can now write the most general Dirac operator on the lattice in the form:
$`D`$ $``$ $`\text{1I}[s_1+s_2{\displaystyle \underset{l_1}{}}<l_1>+s_3{\displaystyle \underset{l_2l_1}{}}<l_1,l_2>`$ (6)
$`+s_4{\displaystyle \underset{l_1}{}}<l_1,l_1>\mathrm{}]`$
$`+`$ $`{\displaystyle \underset{\mu }{}}\gamma _\mu {\displaystyle \underset{l_1=\pm \mu }{}}s(l_1)[v_1<l_1>+v_2{\displaystyle \underset{l_2\pm \mu }{}}[<l_1,l_2>`$ (8)
$`+<l_2,l_1>]+v_3<l_1,l_1>\mathrm{}]`$
$`+`$ $`{\displaystyle \underset{\mu <\nu }{}}\gamma _\mu \gamma _\nu {\displaystyle \underset{\genfrac{}{}{0pt}{}{l_1=\pm \mu }{l_2=\pm \nu }}{}}s(l_1)s(l_2){\displaystyle \underset{i,j=1}{\overset{2}{}}}ϵ_{ij}\left[t_1<l_i,l_j>\mathrm{}\right]`$ (9)
$`+`$ $`{\displaystyle \underset{\mu <\nu <\rho }{}}\gamma _\mu \gamma _\nu \gamma _\rho {\displaystyle \underset{\genfrac{}{}{0pt}{}{l_1=\pm \mu ,l_2=\pm \nu }{l_3=\pm \rho }}{}}s(l_1)s(l_2)s(l_3){\displaystyle \underset{i,j,k=1}{\overset{3}{}}}ϵ_{ijk}\times `$ (11)
$`\left[a_1<l_i,l_j,l_k>\mathrm{}\right]`$
$`+`$ $`\gamma _5{\displaystyle \underset{\genfrac{}{}{0pt}{}{l_1=\pm 1,l_2=\pm 2}{l_3=\pm 3,l_4=\pm 4}}{}}s(l_1)s(l_2)s(l_3)s(l_4){\displaystyle \underset{i,j,k,n=1}{\overset{4}{}}}ϵ_{ijkn}\times `$ (13)
$`\left[p_1<l_i,l_j,l_k,l_n>\mathrm{}\right].`$
The variables $`l_i`$ run over all of $`\{\pm 1,\pm 2,\pm 3,\pm 4\}`$ unless restricted otherwise. We use the abbreviation $`s(l_i)=`$ sign$`(l_i)`$. By $`ϵ`$ we denote the totally anti-symmetric tensors with 2,3 and 4 indices. We choose the normalization of the elements of the Clifford algebra such that the elements appear as all possible products of the $`\gamma _\mu `$ without any extra factors of $`i`$. For this normalization the symmetries C and $`\gamma _5`$-hermiticity render all coefficients $`s_i,v_i,t_i,a_i`$ and $`p_i`$ real.
The above mentioned structure of paths appearing in groups is obvious in (13) and we ordered the terms according to the length of the paths they contain. The paths in each group are related by symmetries and up to the sign factors have to come with the same real coefficient. All paths within a group have the same length. It has to be stressed, that in (13) for each element of the Clifford algebra we show only the leading terms, i.e. the terms with the shortest paths, of an infinite series of groups of paths. The dots indicate that we omitted groups with paths that are longer than the terms we show.
Let us now insert our expanded Dirac operator $`D`$ into the Ginsparg-Wilson equation. To that purpose we define:
$$E=D\gamma _5D\gamma _5+\gamma _5D\gamma _5D,$$
(14)
and finding a solution $`D`$ of the Ginsparg-Wilson equation corresponds to having $`E=0`$. Computing the linear part of $`E`$ is straightforward: When evaluating the sandwich $`\gamma _5D\gamma _5`$ we find that the terms of (13) with an odd number of $`\gamma _\mu `$, i.e. vector- and pseudovector terms, pick up a minus sign, while the other terms remain unchanged. Thus when adding the two linear terms we find that the terms with an odd number of $`\gamma _\mu `$ cancel while the other terms pick up a factor of 2.
The next step is to compute the quadratic term $`\gamma _5D\gamma _5D`$. Here we have to multiply the various terms appearing in $`D`$. Each term is made of two parts, a generator of the Clifford algebra and a group of paths. The multiplication of two of these terms proceeds in two steps: Firstly the two elements of the Clifford algebra are multiplied giving again an element of the algebra. In the second step we have to multiply the paths of our two terms. This multiplication can be noted very conveniently in our notation, where multiplication of two paths simply consists of writing the paths into one long path: $`<l_1,l_2\mathrm{}l_n>\times <l_1^{},l_2^{}\mathrm{}l_n^{}^{}>=<l_1,l_2\mathrm{}l_n,l_1^{},l_2^{}\mathrm{}l_n^{}^{}>`$. It is straightforward to establish this rule by translating back to the algebraic expression of our examples (1),(3) and performing the multiplication in this notation. It can happen that, after multiplying two paths, a hop in some direction $`l_i`$ is immediately followed by its inverse $`l_i`$. These two terms cancel each other and we find $`<l_1\mathrm{}l_{i1},l_i,l_i,l_{i+1}\mathrm{}l_n>=<l_1\mathrm{}l_{i1},l_{i+1}\mathrm{}l_n>`$. This rule is used to reduce all products of paths appearing in $`\gamma _5D\gamma _5D`$ to their true length. In a final step we decompose the sums over the product terms into groups of paths related by the symmetries in the same way as we did above when constructing the most general ansatz for $`D`$. Adding the linear and quadratic terms of (14) we end up with the following expansion for $`E`$:
$`E`$ $``$ $`\text{1I}[e_1^s+e_2^s{\displaystyle \underset{l_1}{}}<l_1>+e_3^s{\displaystyle \underset{l_2l_1}{}}<l_1,l_2>`$ (16)
$`+e_4^s{\displaystyle \underset{l_1}{}}<l_1,l_1>\mathrm{}]`$
$`+`$ $`{\displaystyle \underset{\mu }{}}\gamma _\mu {\displaystyle \underset{l_1=\pm \mu }{}}s(l_1)\left[e_1^v{\displaystyle \underset{l_2\pm \mu }{}}[<l_1,l_2><l_2,l_1>]\mathrm{}\right]`$ (17)
$`+`$ $`{\displaystyle \underset{\mu <\nu }{}}\gamma _\mu \gamma _\nu {\displaystyle \underset{\genfrac{}{}{0pt}{}{l_1=\pm \mu }{l_2=\pm \nu }}{}}s(l_1)s(l_2){\displaystyle \underset{i,j=1}{\overset{2}{}}}ϵ_{ij}\left[e_1^t<l_i,l_j>\mathrm{}\right]`$ (18)
$`+`$ $`{\displaystyle \underset{\mu <\nu <\rho }{}}\gamma _\mu \gamma _\nu \gamma _\rho {\displaystyle \underset{\genfrac{}{}{0pt}{}{l_1=\pm \mu ,l_2=\pm \nu }{l_3=\pm \rho }}{}}s(l_1)s(l_2)s(l_3){\displaystyle \underset{i,j,k=1}{\overset{3}{}}}ϵ_{ijk}\times `$ (23)
$`[{\displaystyle \underset{l_4\pm \mu ,\nu ,\rho }{}}\{e_1^a[<l_i,l_j,l_k,l_4><l_4,l_i,l_j,l_k>]`$
$`+e_2^a[<l_i,l_4,l_j,l_k><l_i,l_j,l_4,l_k>]\}`$
$`+e_3^a[<l_i,l_j,l_k,l_j><l_i,l_j,l_i,l_k>]`$
$`+e_4^a<l_i,l_j,l_k,l_i>\mathrm{}]`$
$`+`$ $`\gamma _5{\displaystyle \underset{\genfrac{}{}{0pt}{}{l_1=\pm 1,l_2=\pm 2}{l_3=\pm 3,l_4=\pm 4}}{}}s(l_1)s(l_2)s(l_3)s(l_4){\displaystyle \underset{i,j,k,n=1}{\overset{4}{}}}ϵ_{ijkn}\times `$ (25)
$`\left[e_1^p<l_i,l_j,l_k,l_n>\mathrm{}\right].`$
Again the expansion of $`E`$ is an infinite series and we display only the leading groups of paths. The coefficients $`e_i^\alpha `$ are now quadratic polynomials in the original coefficients $`s_i,v_i,t_i,a_i`$ and $`p_i`$ given by
$`e_1^s`$ $`=`$ $`2s_1+s_1^2+8s_2^2+48s_3^2+8s_4^2+8v_1^2+96v_2^2+8v_3^2`$ (27)
$`+48t_1^2+192a_1^2+384p_1^2\mathrm{},`$
$`e_2^s`$ $`=`$ $`2s_2+2s_1s_2+12s_2s_3+2s_2s_4+12v_1v_2`$ (29)
$`+2v_1v_3\mathrm{},`$
$`e_3^s`$ $`=`$ $`2s_3+2s_1s_3+s_2^2+4s_3^2+2s_3s_4+4v_2^2+2v_2v_3\mathrm{},`$ (30)
$`e_4^s`$ $`=`$ $`2s_4+2s_1s_4+s_2^2+6s_3^2v_1^26t_1^224a_1^2`$ (32)
$`48p_1^2\mathrm{},`$
$`e_1^v`$ $`=`$ $`s_2v_14s_3v_22s_4v_2s_3v_3v_3t_14v_2t_1\mathrm{},`$ (33)
$`e_1^t`$ $`=`$ $`2t_1+2s_1t_12s_4t_1v_1^24v_2^22v_2v_34t_1^2`$ (35)
$`+8v_1a_18a_1^2+16t_1p_1\mathrm{},`$
$`e_1^a`$ $`=`$ $`s_2a_1+v_2t_1v_3p_1\mathrm{},`$ (36)
$`e_2^a`$ $`=`$ $`v_2t_1\mathrm{},`$ (37)
$`e_3^a`$ $`=`$ $`s_2a_12v_2p_1\mathrm{},`$ (38)
$`e_4^a`$ $`=`$ $`2s_2a_1+2v_2t_14v_2p_1\mathrm{},`$ (39)
$`e_1^p`$ $`=`$ $`2p_1+2s_1p_12s_4p_12v_1a_1+t_1^2\mathrm{}.`$ (40)
Each coefficient $`e_i^\alpha `$ is an infinite series of terms.
The crucial step in our construction is to realize that the groups of paths together with the corresponding element of the Clifford algebra, appearing in the expansion (25), are linearly independent for arbitrary background gauge field. These groups of paths together with the products of $`\gamma _\mu `$ have to be viewed as basis elements for the expansion of $`E`$. For a solution of the Ginsparg-Wilson equation we must have $`E=0`$ and thus all coefficients $`e_i^\alpha `$ have to vanish simultaneously. Hence we have rewritten the problem of finding a solution of the Ginsparg-Wilson equation to the problem of solving the system of coupled quadratic equations (40).
Before we discuss solving (40) let us first analyze our Dirac operator $`D`$ for trivial background gauge field. In this case we can compute the Fourier transform $`\widehat{D}(p)`$ of $`D`$. For small momenta we have to implement the condition $`\widehat{D}(p)=i\mathit{}+𝒪(p^2)`$, which leads to two more equations for the coefficients, the first one makes the constant term vanish and the second one requires the slope to be equal to 1.
$`0`$ $`=`$ $`s_1+8s_2+48s_3+8s_4\mathrm{},`$ (41)
$`1`$ $`=`$ $`2v_1+24v_2+4v_3\mathrm{}.`$ (42)
These two equations have to be implemented as well, and together with (40) entirely describe Dirac operators which obey the Ginsparg-Wilson equation and have the correct behavior at small momenta.
The quadratic system of equations (40), (42) is completely equivalent to the Ginsparg-Wilson equation. The known solutions of the Ginsparg-Wilson equation, i.e. the overlap operator as well as the perfect action, correspond to two different solutions of (40) and (42). Many interesting questions, such as the existence of a connected continuum of solutions of (40), (42) can be addressed in our new framework.
In the remaining part of this letter we now explore the possibility of finding solutions of our system of quadratic equations numerically. It has to be stressed, that (40), (42) is a system of infinitely many quadratic equations for infinitely many coefficients. Thus in order to be able to apply numerical methods for solving these equations one has to introduce a cutoff, i.e. we set the coefficients for terms with longer paths to zero. This cutoff comes in a natural way for proper lattice Dirac operators, since in order to maintain universality of our lattice fermions, the coefficients $`s_i,v_i,t_i,a_i,p_i`$ have to decrease in size exponentially as the length of the corresponding paths increases.
In addition, the following qualitative argument (we use only the leading order, i.e. we ignore effects of back-folding paths) shows that the structure of the Ginsparg-Wilson equation strongly supports this exponential suppression of longer paths: When setting $`E=0`$, the Ginsparg-Wilson equation (14) turns to $`D+\gamma _5D\gamma _5=\gamma _5D\gamma _5D`$. Short paths contained in the original $`D`$ combine to longer paths when evaluating the product $`\gamma _5D\gamma _5D`$ on the right-hand side. The linear term on the left-hand side requires also these longer paths with their coefficients to be contained in $`D+\gamma _5D\gamma _5`$. The coefficients of the combined longer paths are products of the coefficients of the short paths, and since the coefficients of the shortest paths are already smaller than 1 we find an exponential suppression of the coefficients for longer paths. This behavior was confirmed numerically .
Thus the physical requirement of universality as well as the structure of the Ginsparg-Wilson equation imply the coefficients to decrease exponentially in size as the length of the paths in $`D`$ increases. We now consider a truncated expansion of $`D`$ which contains only terms with paths up to a certain length. Introducing this cutoff is equivalent to what is done when computing numerically the fixed point action using a finite parametrization for the Dirac operator. The resulting Dirac operator will only be an approximate solution of the Ginsparg-Wilson equation, but the error, i.e. the norm of $`E`$, can be made arbitrarily small by including more terms in $`D`$.
After one has introduced the cutoff discussed above, e.g. one includes only terms with a maximal length of 3, the system (40), (42) turns into a finite system. Each coefficient $`e_i^\alpha `$ in the system (40) as well as the two equations (42) now consist of only finitely many terms. We numerically solve the equations corresponding to the leading terms in the expansion of $`E`$, and so determine the coefficients $`s_i,v_i,t_i,a_i,p_i`$ in the expansion (13) of $`D`$.
This approach has been implemented and carefully tested in two dimensions . We found that adding only a few simple terms to the Dirac operator improves the chiral properties tremendously, at only a small increase of the cost of a numerical treatment. The resulting Dirac operator was tested in a dynamical simulation of the 2-flavor Schwinger model and it was found that the mass of the $`\pi `$-particle in this model can be brought down, close to the values obtained with the perfect action and the overlap operator.
We are currently implementing the quadratic system in 4 dimensions and have begun to construct approximate solutions of the Ginsparg-Wilson equation along the lines outlined in this letter. Preliminary results are very encouraging, in particular we find a considerable reduction of the additive fermion mass renormalization. These results will be presented elsewhere .
Acknowledgement: The author would like to thank Ivan Hip, Christian Lang and Uwe-Jens Wiese for active interaction during the course of this work. |
warning/0003/astro-ph0003306.html | ar5iv | text | # Goodness-of-Fit Analysis of Radial Velocities Surveys
## 1 Introduction
Surveys of radial velocities of galaxies have played a major role in the study of the large scale structure. The analysis of such surveys has been conducted in two main directions, the mapping of the local cosmography and the estimation of the cosmological parameters (cf. Dekel 1994 for a review). The Bayesian framework provides one with very elegant and powerful tools for conducting both the mapping and parameter estimation, where the recovery of the large scale structure is done by means of the Wiener filter and the parameters are estimated by maximum likelihood (MaxLike) analysis (Zaroubi et al. 1995, hereafter ZHFL). In the case where the deviations from a homogeneous and isotropic universe constitute a Gaussian random field the Wiener filter and the MaxLike are the optimal tools for performing such an analysis (ZHFL). Indeed, the MARK III catalog of radial velocities (Willick et al. 1995, 1996, 1997a) have been recently analyzed by Wiener filtering (Zaroubi, Hoffman and Dekel 1999) and by MaxLike (Zaroubi et al. 1997). The SFI survey of da Costa et al. (1996) has been studied by MaxLike analysis by Freudling et al. (1999) and by Wiener filtering (Hoffman and Zaroubi, unpublished). Both surveys seem to yield similar results.
In the Bayesian MaxLike analysis one calculates the posterior probability of a model to be correct given the data (ZHFL, Vogeley and Szalay 1996). Thus the model that maximizes the likelihood function, over a given parameter (or model) space, is the most likely model in that space. The MaxLike analysis cannot guarantee, however, that the most probable model is indeed consistent with the data. It provides only a relative measure for models to be correct. It is common to adopt an independent measure for the goodness-of-fit, which is often given by the requirement that the reduced $`\chi ^2`$ is close to unity. Often, when the most likely model (given the data) passes also the goodness-of-fit test one assumes that the ‘correct’ model has been nailed down. Here, the $`\chi ^2`$ test is expanded and a much more critical test is suggested and then applied to the Mark III and SFI surveys.
The $`\chi ^2`$ ’goodness-of-fit’ is based on the assumptions that all the random variables that affect the observables are normally distributed. In the cosmological context this applies to both the underlying dynamical (e.g. density and velocity) field and the statistical errors. Thus for a survey containing $`N`$ data observables (e.g. radial velocities) the $`\chi ^2`$ of the system of $`N`$ degrees of freedom (DOF) is calculated given the model that maximizes the likelihood function. The goodness-of-fit is measured by how close is the $`\chi ^2/\mathrm{DOF}`$ to unity. This provides a global measure for the consistency of the data with the model, as it includes all the observables. A situation might occur of some ‘conspiracy’ where different parts of the data deviate from the predictions of the model, but when combined together they ‘conspire’ to yield a reasonable $`\chi ^2`$. A much stronger test on the model is to decompose the data into statistical independent eigenmodes and observe the $`\chi ^2`$ behavior of the independent modes. Eigenmode analysis, also known as principal component analysis (PCA) and the Karhunen-Loeve transform, is not a new tool in the field. It has been applied to studies of redshift surveys (Vogeley and Szalay 1996),the cosmic microwave background (Bunn 1997, Bond 1995) and more recently radial velocities surveys (Hoffman, 1999). The later study is extended here to perform the ’goodness-of-fit’ test on a mode-by-mode basis. The basic formalism is presented in § 2, and its application to the Mark III and SFI surveys is given in § 3. Our results are discussed and the conclusions are summarized in § 4.
## 2 Eigenmode Analysis of Radial Velocities
Consider a data base of radial velocities $`\{u_i\}_{i=1,\mathrm{},N}`$, where
$$u_i=𝐯(𝐫_i)\widehat{𝐫}_i+ϵ_i,$$
(1)
$`𝐯`$ is the three dimensional velocity, $`𝐫_i`$ is the position of the i-th data point and $`ϵ_i`$ is the statistical error associated with the i-th radial velocity. The assumption made here is of a cosmological model that well describes the data, that systematic errors have been properly dealt with and that the statistical errors are well understood. The data auto-covariance matrix is then written as:
$$R_{ij}<u_iu_j>=\widehat{𝐫}_j<𝐯(𝐫_i)𝐯(𝐫_j)>\widehat{𝐫}_j+\sigma {}_{ij}{}^{2}.$$
(2)
(Here $`<\mathrm{}>`$ denotes an ensemble average.) The last term is the error covariance matrix. The velocity covariance tensor that enters this equation was derived by Górski (1988, see also Zaroubi, Hoffman and Dekel 1999) and it depends on the power spectrum and cosmological parameters.
The eigenmodes of the data covariance matrix provides a natural representation of the data:
$$R\eta ^{(i)}=\lambda _i\eta ^{(i)}$$
(3)
The set of $`N`$ eigenmodes $`\{\eta ^{(i)}\}`$ constitutes an orthonormal basis and the eigenvalues $`\lambda _i`$ are arranged in decreasing order (in absolute values). A new representation of the data is given by:
$$\stackrel{~}{a}_i=\eta {}_{j}{}^{(i)}u_{j}^{}$$
(4)
This provides a statistical orthogonal representation, namely:
$$\stackrel{~}{a}_i\stackrel{~}{a}_j=\lambda _i\delta _{ij}$$
(5)
The normalized transformed variables are defined by:
$$a_i=\frac{\stackrel{~}{a}_i}{\sqrt{\lambda _i}}$$
(6)
Eq. 5 is written now as:
$$a_ia_j=\delta _{ij}$$
(7)
Note that as the modes are statistically independent one can measure the $`\chi ^2`$ of a given mode, $`\chi {}_{i}{}^{2}=a_i^2`$, and the cumulative reduced $`\chi ^2`$ is given by:,
$$\chi {}_{M}{}^{2}=\frac{1}{M}\underset{i=1}{\overset{M}{}}a_i^2$$
(8)
For normally distributed errors and a Gaussian random velocity field the $`a_i`$’s are normally distributed with zero mean and a variance of unity.
In addition the probability of finding such $`\chi _M^2`$ is calculated as well. The probability is defined by
$`P(\chi {}_{M}{}^{2})`$ $`=`$ $`P_{\chi ^2}(M\chi {}_{M}{}^{2},M)\mathrm{for}P_{\chi ^2}(M\chi {}_{M}{}^{2},M)<0.5`$ (9)
$`=`$ $`\mathrm{\hspace{0.33em}\hspace{0.33em}1}P_{\chi ^2}(M\chi {}_{M}{}^{2},M)\mathrm{otherwise},`$ (10)
where $`P_{\chi ^2}(x,M)`$ is the probability that a random variable drawn from a $`\chi ^2`$ distribution with $`M`$ degrees of freedom is less than a given value $`x`$.
## 3 Differential $`\chi ^2`$ Analysis
Here the goodness-of-fit of the Mark III and SFI surveys is studied. The models studied here are the MaxLike solutions for these surveys, which are slightly different from one another. The most likely model given Mark III is a tilted-CDM (T-CDM) of $`\mathrm{\Omega }_0=1,h=0.75`$ and $`n=0.8`$ where $`\mathrm{\Omega }_0`$ is the cosmological density parameter, $`h`$ is Hubble’s constant in units of $`100km/s/Mpc`$ and $`n`$ is the power spectrum index (Zaroubi et al. 1977). The most likely model given SFI is an open CDM (OCDM) of $`\mathrm{\Omega }_0=0.79,h=0.6`$ and $`n=0.92`$ (Fruedling et al. 1999). For both cases the MaxLike best model has a total $`\chi _{M=N}^2`$ very close to unity. Thus, from the point of view of the integral $`\chi ^2`$ the MaxLike solutions seem to be very consistent with the data. This is extended to perform a differential $`\chi ^2`$ analysis, namely to study the $`\chi ^2`$ behavior across the modes spectrum.
To study the robustness of this probe it is first applied to a linear mock catalog of Mark III, constructed from an unconstrained realization of the velocity field. This field is sampled at the location of the Mark III data points, to which normally distributed errors are added according to Mark III’s error covariance matrix. The cumulative $`\chi ^2`$ of such a catalog should oscillate around unity, given that the model used to generate the catalog is known. Indeed, this has been confirmed by an analysis of a few linear mock catalogs of Mark III. The probabilities of obtaining such $`\chi ^2`$ distribution lies comfortably within the $`90\%`$ confidence level. The non-trivial result of this test is that the very poor sampling of the long wavelength Fourier waves, i.e. cosmic variance, does not affect the goodness-of-fit test.
The differential $`\chi ^2`$and its associated probability of the Mark III and SFI surveys are presented in Fig. 1, each case analyzed in its maximum likelihood solution. A clear trend is noticed, namely over almost the entire mode spectrum the cumulative $`\chi ^2`$ increases monotonically. When all modes are included the total $`\chi ^2`$/DOF is indeed close to $`1`$, but if we had to take half the modes, starting from the top or the bottom, a very different $`\chi ^2`$ would have obtained.
The differential $`\chi ^2`$ analysis is repeated for the currently popular model of $`\mathrm{\Lambda }`$-CDM($`\mathrm{\Omega }_0=0.4,h=0.6`$ and $`n=1`$; Fig. 1). Indeed, the same trend is found in this case as well but the total $`\chi ^2`$converges to a value outside the $`90\%`$ confidence level.
The conclusions that follows is that for both data sets, Mark III and SFI, and for a variety of theoretical models the differential $`\chi ^2`$ increase monotonically with the mode number (with the exception of the first 10 modes of the Mark III). The theoretical expectation is that if indeed the data is consistent with the assumed model then $`\chi _M^2`$ will fluctuate around unity. The probability of observing such a trend given a model is very small across most of the mode number range.
## 4 Discussion
What have we learned from the differential $`\chi ^2`$ analysis? It has been found that even the most probable CDM-like model, the one that maximizes the likelihood function given the data, is not fully consistent with the data. The cumulative $`\chi ^2`$ has been calculated both downwards and upwards (namely starting from the modes with the largest and smallest eigenmode, respectively). Over more than $`90\%`$ of the modes the cumulative $`\chi ^2`$ lies well outside the $`90\%`$ confidence level, indicating a very small probability of measuring such data given the assumed model. Over most of the mode number range $`\chi _M^2`$ increases monotonically. It is this behavior of the $`\chi ^2`$ which indicates a systematic inconsistency of the model with the data. The assumed model actually contains two ingredient, the theoretical power spectrum and the error model. However, the present analysis cannot indicate which one is to be ‘blamed’ for the systematic trend. It should be noted here that apart from the first few ($`1020`$) modes there is a clear correlation of the eigenvalues with its weighted mean distance (of data points of the given mode). Namely, the variance associated with a mode (i.e. its eigenvalue) increases with its mean distance (Zehavi, private communication, Silverman et al. in preparation). It follows that the $`\chi ^2`$ trend seen here is closely correlated with the distance and that the data ‘asks’ for less power on large scales than the model (power spectrum and noise) provides. A detailed study of the power spectrum and error model possible modifications is to be given elsewhere (Silverman et al. in preparation). (Note that these first $`1020`$ modes are the ones dominated by the underlying velocity field and not the noise, Hoffman 1999.)
The cosmological implications of the present findings are that either the error and/or the theoretical model need to be modified. The theoretical model assumed in the analysis of large scale radial velocity surveys is that the velocities are drawn from a Gaussian random field defined by a given power spectrum. The present study might indicate the inconsistency of the power spectrum with the data. A less likely possibility is that it indicates a departure from the Gaussian statistics. Alternatively, the present work might indicate a systematic error that has not been accounted for that causes this trend. Still another possibility is that of an indication for a velocity bias.
The conclusions reached here should not be taken as a contradiction of the results of Zaroubi et al. (1997) and Fruedling et al. (1999), but rather as extending and complementing them. The Bayesian MaxLike analysis can be performed only within the assumed parameter/model space. The differential $`\chi ^2`$ allows one to go beyond this and analyze the nature of the agreement, or the lack of it, between a given model and the data on a mode by mode basis.
The PCA transforms the data to a statistically independent representation and enables the study of the compatibility of the data with the model on a mode by mode basis. This differential analysis is in contrast to the more traditional approach where a data set is analyzed as a whole. The differential $`\chi ^2`$ analysis should be performed together with the Bayesian MaxLike analyze and complement it. This should be useful in fields where the MaxLike is the basic tool of analysis such as the mapping of the CMB angular fluctuations and the study of redshift surveys as well as all radial velocity surveys. The present analysis can prove to be very useful and powerful in those fields where systematic errors play a crucial roles, such as redshift and radial velocities surveys.
We have benefited from many interesting discussions with Avishai Dekel, Zafrir Kolatt, Ofer Lahav, Lior Silverman, Simon White and Idit Zehavi. The hospitality of the Racah Inst. Physics and the Max Planck Institut fur Astrophysik is gratefully acknowledged. This research has been partially supported by a Binational Science Foundation grant 94-00185 and an Israel Science Foundation grant 103/98. |
warning/0003/cond-mat0003290.html | ar5iv | text | # Binary hard-sphere fluids near a hard wall
## I Introduction
A variety of experimental techniques has emerged which allow one to resolve the inhomogeneous density distributions of fluids at interfaces, a subject which enjoys broad scientific interest. In this context the ability to manufacture highly monodisperse colloidal suspensions has turned out to be particular useful as they provide the possibility to tune the effective interactions in these systems such that, e.g., the colloidal particles closely resemble hard-sphere fluids . Since many of these experimental probes are indirect scattering techniques there is a substantial demand to guide them theoretically. Computer simulations and integral theories are important tools of statistical physics to address these issues. Density functional theory (DFT) has emerged as an additional approach which is capable to capture interfacial phase transitions and to sweep the thermodynamic and interaction parameter space of the system under consideration. The potential to combine these two possibilities poses already a major challenge for the other techniques. If DFT acquires in addition the same accuracy as the other two techniques, it could gain a clear competitive edge.
Although there is no recipe for constructing systematically a reliable DFT in spatial dimensions $`d2`$, the constant flux of developments over many years has led to a rather high level of sophistication. Among these theories for hard-sphere fluids, which act as paradigmatic systems and stepping-stones for more complicated models, the Rosenfeld functional has emerged as a particular powerful theory which resorts to the fundamental geometrical measures of the individual sphere . For the standard test case of the highly inhomogeneous density distribution of a one-component hard-sphere fluid near a planar hard wall, the predictions of the Rosenfeld functional are very close to those of numerical simulations serving as benchmarks. For this case the mean square deviations \[see, c.f., Eq. (22)\] of Rosenfeld DFT results from the simulation data from Ref. are at most $`1\times 10^3`$ at high packing fractions, otherwise less than $`3\times 10^4`$.
Another virtue of the Rosenfeld functional is that is easily lends itself to the generalization to multi-component hard-sphere fluids. This opens the door to investigate rich new physical phenomena as particles of different size compete for interfacial positions . Even for the simplest multi-component system, the binary hard-sphere fluid, there are only relatively few theoretical studies which determine their structural properties near a planar hard wall, using Monte Carlo simulations , integral equation theories , and various kinds of density functional theory , as well as in spherical pores . Here we analyze this problem by using the corresponding Rosenfeld functional both in its original version as well as for sophistications thereof . By comparing these results with published simulation data we assess to which extent the quantitative reliability of the Rosenfeld functional for the one-component hard-sphere fluid remains valid for the corresponding binary system. Moreover we determine concentration profiles, the excess coverage, and the surface tension of the binary hard-sphere fluids at a hard wall. We describe the DFT in Sec. II and report our results in Sec. III followed by a summary and our conclusions in Sec. IV. The Appendix contains important technical details.
## II Density functional theory
The Rosenfeld functional for the excess (over the ideal gas) Helmholtz free energy of a mixture of hard spheres with number density profiles $`\{\rho _i(𝐫)\},i=1,\mathrm{},N`$, can be written as
$$\beta _{ex}[\{n_\alpha \}]=\mathrm{d}^3r\mathrm{\Phi }(\{n_\alpha (𝐫)\})$$
(1)
which is a functional of the four scalar weighted densities $`n_\alpha (𝐫)`$ for the $`N`$-component mixture
$$n_\alpha (𝐫)=\underset{i=1}{\overset{N}{}}\mathrm{d}^3r^{}\rho _i(𝐫^{})\omega _i^{(\alpha )}(𝐫𝐫^{}),\alpha =0,\mathrm{},3,$$
(2)
with $`4N`$ scalar weight functions $`\omega _i^{(\alpha )}`$ and two three-component vector weighted densities $`𝐧_\alpha (𝐫)`$
$$𝐧_\alpha (𝐫)=\underset{i=1}{\overset{N}{}}\mathrm{d}^3r^{}\rho _i(𝐫^{})𝝎_i^{(\alpha )}(𝐫𝐫^{}),\alpha =1,2,$$
(3)
with $`2N`$ vector weight functions $`𝝎`$$`{}_{}{}^{(\alpha )}{}_{i}{}^{}`$. The weight functions contain only information about the fundamental geometrical measures of a single sphere of species $`i`$, namely its volume, surface area, and radius $`R_i`$, i.e., in particular they are independent of the density profiles. The explicit expressions for the weight functions are given in the Appendix. $`\mathrm{\Phi }(\{n_\alpha \})=\mathrm{\Phi }_1+\mathrm{\Phi }_2+\mathrm{\Phi }_3`$ is a function of the weighted densities with
$`\mathrm{\Phi }_1`$ $`=`$ $`n_0\mathrm{log}(1n_3),`$ (4)
$`\mathrm{\Phi }_2`$ $`=`$ $`{\displaystyle \frac{n_1n_2𝐧_1𝐧_2}{1n_3}},`$ (5)
and
$`\mathrm{\Phi }_3`$ $`=`$ $`{\displaystyle \frac{\frac{1}{3}n_2^3n_2𝐧_2𝐧_2}{8\pi (1n_3)^2}}={\displaystyle \frac{n_2^3}{24\pi (1n_3)^2}}(13\xi ^2),`$ (6)
where $`\xi (𝐫)|𝐧_2(𝐫)|/n_2(𝐫)`$, which is the ratio of the modulus of the vector weighted density $`𝐧_2(𝐫)`$ and the scalar weighted density $`n_2(𝐫)`$. We want to note that $`\xi (𝐫)0`$ in the bulk and is small for small inhomogeneities. While this original Rosenfeld functional describes very successfully the fluid phase of a one-component hard-sphere system , it fails to predict the freezing transition. This failure has been studied in detail in Refs. and . For the freezing transition it turns out that the zero-dimensional limit of the functional, in which a small cavity can accommodate only a single sphere, plays a key role. In a crystal the thermal vibrations around a lattice site can be interpreted as the motions in such a cavity formed by the neighboring spheres. Only if the statistical mechanics in such a cavity is described properly by the density functional, the freezing transition is predicted correctly. This is not the case for the original Rosenfeld functional. This problem can be fixed by modifying slightly the contribution $`\mathrm{\Phi }_3`$ \[see Eq. (6)\] such that the freezing transition is predicted by the modified functional while at the same time, for lower packing fractions, the accuracy of the original functional in describing the inhomogeneous fluid is kept. The following modifications have been suggested :
$$\mathrm{\Phi }_{3,q}=\frac{n_2^3}{24\pi }(1\xi ^2)^q$$
(7)
with $`q2`$ and
$$\mathrm{\Phi }_{3,int}=\frac{n_2^3}{24\pi }(13\xi ^2+2\xi ^3).$$
(8)
The first suggestion, $`\mathrm{\Phi }_{3,q}`$, is an antisymmetrized version of $`\mathrm{\Phi }_3`$ in Eq. (6) and the second, $`\mathrm{\Phi }_{3,int}`$, interpolates between $`\mathrm{\Phi }_3`$ of Eq. (6) and $`\mathrm{\Phi }_{3,0D}`$ in the exact zero-dimensional limit
$$\mathrm{\Phi }_{3,0D}=\frac{n_2^3}{24\pi (1n_3)^2}\xi (1\xi )^2.$$
(9)
While the modified Rosenfeld function with $`\mathrm{\Phi }_{3,0D}`$ does successfully predict the freezing transition of the one-component system, it leads to modified bulk properties and hence cannot describe the hard-sphere fluid as accurate as the original Rosenfeld functional. We note that the difference between $`\mathrm{\Phi }_3`$ of the original Rosenfeld functional and both $`\mathrm{\Phi }_{3,q}`$ with $`q=3`$ and $`\mathrm{\Phi }_{3,int}`$ is of the order of $`𝒪(\xi ^3)`$. Therefore we expect the biggest differences between the various versions of the Rosenfeld DFT to occur close to the wall where $`\xi `$ is largest.
Both the original Rosenfeld functional and the modifications corresponding to Eqs. (7) and (8), i.e., the functionals that share common bulk properties, are very successful and accurate for the one-component fluid. But far less is known for binary mixtures. While in this latter respect in Ref. very good agreement between the density profiles obtained from the Rosenfeld functional and simulation data from Ref. has been mentioned, in a recent study significant deviations between the Rosenfeld DFT results and simulations have been found. In this latter study an alternative but equivalent formulation of the original Rosenfeld functional has been applied.
Here we are interested in the equilibrium density profiles $`\rho _{s,0}(𝐫)`$ and $`\rho _{b,0}(𝐫)`$ of both the small and big components of binary hard-sphere mixtures close to a planar hard wall. To this end we freely minimize the functional
$$\mathrm{\Omega }[\rho _s(𝐫),\rho _b(𝐫)]=[\rho _s(𝐫),\rho _b(𝐫)]+\underset{i=s,b}{}\mathrm{d}^3r^{}\rho _i(𝐫^{})\left(V_{ext}^i(𝐫^{})\mu _i\right)$$
(10)
which is written in terms of the functional
$$[\rho _s(𝐫),\rho _b(𝐫)]=_{id}[\rho _s(𝐫),\rho _b(𝐫)]+_{ex}[\rho _s(𝐫),\rho _b(𝐫)]$$
(11)
with the exactly known ideal gas contribution $`_{id}`$,
$$\beta _{id}=\underset{i=s,b}{}\mathrm{d}^3r^{}\rho _i(𝐫^{})\left(\mathrm{ln}(\lambda _i^3\rho _i(𝐫^{}))1\right),$$
(12)
with $`\lambda _i`$ the thermal wave length of species $`i`$. For the equilibrium density profiles $`\rho _{i,0}(𝐫)`$, $`i=s,b`$, the functionals $``$ and $`\mathrm{\Omega }`$ reduce to the Helmholtz free energy and the grand canonical potential of the mixture, respectively; $`\mu _s`$ and $`\mu _b`$ are the chemical potentials of the two species. The external potentials entering into Eq. (10) model the planar hard wall at $`z=0`$:
$$V_{ext}^i(z)=\{\begin{array}{cc}\hfill \mathrm{},& z<R_i,\hfill \\ \hfill 0,& \text{otherwise},\hfill \end{array}$$
(13)
$`i=s,b`$, with $`z`$ the normal distance from the wall. The external potentials prevent the centers of spheres of species $`i`$ to approach the wall, located at $`z=0`$, closer than $`R_i`$ in which case they are in contact.
In the absence of spontaneous symmetry breaking due to freezing, which we do not consider here, the profiles $`\rho _{i,0}(z)`$, $`i=s,b`$, depend only on the normal coordinate $`z`$ which simplifies the minimization of the functional.
Far away from the wall, i.e., in the bulk system, the vector weighted densities $`𝐧_1`$ and $`𝐧_2`$ and thus $`\xi `$ vanish. In this limit both the original Rosenfeld functional and the two modifications corresponding to Eqs. (7) and (8) reduce to the same bulk expression given by
$$\mathrm{\Phi }_{bulk}=n_0\mathrm{ln}(1n_3)+\frac{n_1n_2}{1n_3}+\frac{n_2^3}{24\pi (1n_3)^2}$$
(14)
and hence they share the same bulk properties. We want to emphasize that as a consequence of this feature all versions of the Rosenfeld functional predict density profiles which show the same asymptotic decay towards the bulk value . The weighted densities in the bulk limit are obtained by inserting the bulk densities $`\rho _{i,bulk}:=\rho _{i,0}(z=\mathrm{})`$ into Eq. (2) yielding
$`n_3`$ $`=`$ $`{\displaystyle \frac{4\pi }{3}}{\displaystyle \underset{i=s,b}{}}R_i^3\rho _{i,bulk}{\displaystyle \underset{i=s,b}{}}\eta _i,`$ (15)
$`n_2`$ $`=`$ $`4\pi {\displaystyle \underset{i=s,b}{}}R_i^2\rho _{i,bulk},`$ (16)
$`n_1`$ $`=`$ $`{\displaystyle \underset{i=s,b}{}}R_i\rho _{i,bulk},`$ (17)
and
$`n_0`$ $`=`$ $`{\displaystyle \underset{i=s,b}{}}\rho _{i,bulk}.`$ (18)
The equation of state following from Eq. (14),
$$\beta p=\frac{n_0}{1n_3}+\frac{n_1n_2}{(1n_3)^2}+\frac{1}{12\pi }\frac{n_2^3}{(1n_3)^3},$$
(19)
is the Percus-Yevick compressibility equation of state of the mixture . This expression is related to the contact values of the density profiles according to the sum rule
$$\beta p=\underset{i=s,b}{}\rho _i(z=R_i+0).$$
(20)
This sum rule is respected by the Rosenfeld functional as by any weighted-density DFT and therefore provides a test for the numerical accuracy of the calculations. In the following we suppress the subscript 0 which indicates equilibrium profiles as opposed to variational functions.
## III Structural and thermodynamic properties
### A Density profiles
The number density profiles of both components of binary hard-sphere mixtures close to a planar hard wall are obtained by a free minimization of the functional given in Eq. (10). We use the original Rosenfeld functional as well as the modified versions corresponding to Eq. (8) and to Eq. (7) with $`q=2`$ and $`q=3`$. The systems considered here have two different size ratios, $`R_b:R_s=5:3`$ and $`R_b:R_s=3:1`$, and various packing fractions $`\eta _s`$ and $`\eta _b`$ of the small and big spheres , respectively. The resulting density profiles are compared with simulation data published in Ref. . In addition we calculate the local concentrations $`\mathrm{\Phi }_s(z)`$ and $`\mathrm{\Phi }_b(z)`$ of the small and big spheres, respectively, defined as
$$\mathrm{\Phi }_i(z)=\frac{\rho _i(z)}{\rho _s(z)+\rho _b(z)},i=s,b.$$
(21)
We find excellent agreement between the density profiles of both components obtained by density functional theory and the simulation data for all systems under consideration. This holds for all versions of the Rosenfeld functional analyzed here. While at low total packing fractions $`\eta =\eta _s+\eta _b`$ the density functional theory results for all versions of the Rosenfeld functional are practically equivalent, small deviations among the results from different versions of the functional appear at larger values of $`\eta `$, i.e., for $`\eta 0.3`$. We quantify the degree of agreement between our DFT density profiles $`\rho _i(z)`$ and the simulation data from Ref. , available as data points $`(z_j,\rho _i^{sim}(z_j))`$, $`j=1\mathrm{}N_i^{sim}`$ and $`i=s,b`$, by determining the mean square deviations $`\overline{E}_i`$, $`i=s,b`$, defined as
$$\overline{E}_i=\frac{1}{N_i^{sim}}\underset{j=1}{\overset{N_i^{sim}}{}}\left(\frac{\rho _i^{sim}(z_j)\rho _i(z_j)}{\rho _{i,bulk}}\right)^2.$$
(22)
We find that $`\overline{E}_s`$ and $`\overline{E}_b`$ are at most $`5\times 10^4`$ and $`6\times 10^3`$, respectively, for all versions of the Rosenfeld functional. However, since the statistical errors in the simulation data are comparable with or even larger than the differences between the density profiles obtained by different version of the Rosenfeld DFT this approach does not enable us to determine which of the various versions is the most accurate one.
To illustrate the agreement between the DFT and the simulation data, in Fig. 1 we show the density profiles of the small spheres (a) and of the big spheres (b) for $`\eta _s=0.0607`$ and $`\eta _b=0.3105`$ and a size ratio $`R_b:R_s=5:3`$. The symbols ($`\mathrm{}`$) denote the simulation data from Ref. and the solid line is obtained by the original Rosenfeld functional. The dotted lines denote coarse grained densities $`\overline{\rho }_i^{(j)}`$, $`i=s,b`$ and $`j=0,1`$, defined as
$$\overline{\rho }_i^{(j)}=\frac{1}{z_i^{(j+1)}z_i^{(j)}}\underset{z_i^{(j)}}{\overset{z_i^{(j+1)}}{}}dz\frac{\rho _i(z)}{\rho _{i,bulk}},$$
(23)
with $`z_i^{(0)}0`$ and $`z_i^{(j>0)}`$ the position of the $`j`$th minimum . All details of the density profiles found in the simulations are reproduced very accurately by the density functional theory. The oscillatory behavior, i.e., the amplitudes, the phases, and the decay of the oscillations as obtained by DFT agree excellently with the simulations. The total packing fraction of the system $`\eta =\eta _s+\eta _b=0.3712`$ is already rather high, giving rise to the pronounced structure of the density profiles. The corresponding concentration profiles $`\mathrm{\Phi }_s(z)`$ and $`\mathrm{\Phi }_b(z)`$ of the small and big spheres, respectively, are shown in Fig. 2. These concentration profiles demonstrate that, apart from the purely geometrical constraints, near the wall the big particles are enriched and the small particles depleted. This is in line with the expectation based on the attractive depletion potential near a hard wall of a single big sphere immersed in a fluid of small spheres . Correlation effects reverse this relative distribution in the second layer and restore it in the third.
As mentioned above, small differences between the DFT results corresponding to the various versions of the Rosenfeld functional can be found for these values of $`\eta `$. In order to be able to resolve these small differences magnified parts of the density profiles from Fig. 1 are shown in Fig. 3 together with the simulation data from Ref. ($`\mathrm{}`$). The solid lines in Fig. 3 correspond to the original Rosenfeld functional, the dotted lines correspond to the interpolated version \[Eq. (8)\] whereas the dashed and dashed-dotted lines correspond to the antisymmetrized version \[Eq. (7)\] with $`q=2`$ and $`q=3`$, respectively. All DFT results are very close to the simulation data. However, the deviations between the simulations and the antisymmetrized functional with $`q=2`$ seem to be systematically the largest.
In Figs. 4 and 5 we show the density profiles of a binary mixture with size ratio $`R_b:R_s=3:1`$. The packing fraction of the small spheres is $`\eta _s=0.0047`$ and that of the big spheres is $`\eta _b=0.3859`$ so that the total packing fraction $`\eta =\eta _s+\eta _b=0.3906`$ is again rather high. Therefore there is a strong spatial variation of the density profiles. The agreement between DFT (solid line) and simulations ($`\mathrm{}`$) is again found to be excellent for both the density profile of the small spheres (a) and of the big spheres (b). The dotted lines denote the coarse grained densities as defined in Eq. (23). The concentrations profiles of the small and big spheres, corresponding to these density profiles are shown in Fig. 6. For this larger size ratio the anti-correlated behavior of $`\mathrm{\Phi }_s(z)`$ and $`\mathrm{\Phi }_b(z)`$ is even more pronounced than for the smaller ratio discussed in Fig. 2 and strongly locked in without additional fine structure such as the double peak appearing in Fig. 2.
In addition we test the numerical accuracy of our calculations by means of the sum rule given in Eq. (20). In Table I the sum of the contact values of the binary mixture with size ratio $`R_b:R_s=5:3`$ for various packing fractions is compared with two equations of state. $`\beta p_{PY}^c`$ denotes the Percus-Yevick compressibility equation of state \[Eq. (19)\], to which the Rosenfeld functional reduces in the bulk limit, and $`\beta p_{MCSL}`$ corresponds to the more accurate Mansoori-Carnahan-Starling-Leland equation of state , which represents a generalization to a mixture of the very accurate Carnahan-Starling equation of state for the one-component fluid. The very good agreement between the contact values and $`\beta p_{PY}^c`$ demonstrates the high accuracy of our numerical procedure. However, at higher packing fractions, $`\beta p_{PY}^c`$ deviates from the more accurate equation of state $`\beta p_{MCSL}`$. The same analysis of our results for a binary mixture with size ratio $`R_b:R_s=3:1`$ is summarized in Table II.
Equation (20) represents a sum rule which must be fulfilled by the density profiles as obtained by any of the density functionals considered here. However, no corresponding rules are available for the individual contact values. We find that for all versions of the Rosenfeld functional under consideration here, the sum rule is respected equally well. However, the individual contact values may differ. This statement is in line with the expectation that the biggest differences between the various versions of the Rosenfeld functional occur in a region where $`\xi `$ is large, i.e., close to the wall, and is substantiated in Table III for the binary mixture with size ratio $`R_b:R_s=5:3`$ and in Table IV for the size ratio $`R_b:R_s=3:1`$.
Vested with this confidence in our numerical procedures we are now able to comment on the comparison of our DFT results for the original Rosenfeld functional with those obtained by earlier DFT studies . We find that DFT approaches other than the Rosenfeld DFT predict the structure of the density profiles of a binary hard-sphere mixture near a planar hard wall only qualitatively , whereas the Rosenfeld functional yields quantitatively reliable predictions. This is demonstrated by Figs. 1 and 35 and the very small values of the mean square deviations $`\overline{E}_s`$ and $`\overline{E}_b`$ \[Eq. (22)\]. The predictions of the Rosenfeld functional agree excellently in all details with the simulation results from Ref. . The deviations between the DFT results of Ref. , calculated with the Rosenfeld functional, and their own simulations originate most likely from numerically problems of the iteration procedure used in Ref. . This suspicion is further supported by the fact that the DFT density profiles shown in Ref. seem to violate the sum rule Eq. (20). Thus we are led to the conclusion that the deviations between the Rosenfeld DFT results and the simulation data reported in Ref. most likely are artifacts generated by numerical problems in implementing the iteration procedure used in Ref. for solving the Euler-Lagrange equations corresponding to the Rosenfeld density functional. Therefore the doubts raised in Ref. about the performance of the Rosenfeld DFT for binary hard-sphere mixtures are not justified. We conclude that the Rosenfeld DFT exhibits the same high accuracy in predicting density distributions of binary hard-sphere mixtures as for the one-component hard-sphere fluid.
### B Excess adsorption and surface tension
One of the virtues of DFT is that based on the knowledge of the local structural properties $`\rho _i(z)`$, $`i=s,b`$, it is straightforward to calculate also thermodynamic properties such as the excess adsorptions $`\mathrm{\Gamma }_i`$ and the surface tension $`\gamma `$. Here we determine these quantities near a hard wall for a binary hard-sphere fluid whose components exhibits a size ratio $`R_b:R_s=3:1`$. Our analysis is confined to the fluid phase of the mixture; the phase boundary for freezing is estimated from the bulk phase diagrams presented in Refs. .
The excess adsorption of species $`i`$, $`i=s,b`$, is defined as
$`\mathrm{\Gamma }_i`$ $`=`$ $`{\displaystyle \underset{0}{\overset{\mathrm{}}{}}}dz(\rho _i(z)\rho _{i,bulk})`$ (24)
$`=`$ $`{\displaystyle \underset{\sigma _i/2}{\overset{\mathrm{}}{}}}dz(\rho _i(z)\rho _{i,bulk}){\displaystyle \frac{\sigma _i}{2}}\rho _{i,bulk}.`$ (25)
This definition of the excess adsorption differs from the definition used in Ref. as well as from that used in Ref. . To recover the results for the excess adsorption of a one-component hard-sphere fluid in Ref. and Ref. one has to subtract from and add to our results, respectively, the constant $`\frac{\sigma _i}{2}\rho _{i,bulk}`$. These differences originate from different choices for the position of the wall. While for the one-component fluid there is no preference for any of these definitions, our choice used here appears to be particularly suited for a mixture because independent of the diameter $`\sigma _i`$ of species $`i`$ the integrals in Eq. (24) start at the same lower bound for all species, namely at the position of the physical wall. We use the same definition for the position of the wall to determine the surface tension. Because the excess adsorptions follow from integrating over oscillatory functions, they depend very sensitively on the precise structure of the density profiles and require very accurate calculations. Moreover, near the phase boundary for freezing the original Rosenfeld functional yields values for $`\mathrm{\Gamma }_s`$ and $`\mathrm{\Gamma }_b`$ which differ significantly from those obtained from the modifications of the Rosenfeld functional. Thus, unless stated otherwise, we have determined the excess adsorption by using the modified Rosenfeld functional corresponding to Eq. (8), which is known to capture the freezing transition of the one-component hard-sphere fluid.
In Fig. 7 we show the excess adsorptions $`\mathrm{\Gamma }_s`$ of the small spheres as function of the packing fractions $`\eta _s`$ and $`\eta _b`$; note that $`\mathrm{\Gamma }_s(\eta _s=0,\eta _b)0`$. For a fixed packing fraction of the small spheres $`\eta _s`$, the excess adsorption of the small spheres increases upon increasing $`\eta _b`$. The reason for this is, that the increasing packing effects of the big spheres associated with large values of $`\eta _b`$ enforces also the packing of the small spheres, giving rise to a strongly enhanced contact value and very pronounced structures in the density profile of the small spheres. For very small $`\eta _s`$, i.e., $`\eta _s0.1`$, and large $`\eta _b`$ we find that $`\mathrm{\Gamma }_s`$ can become positive. As function of $`\eta _b`$, $`\mathrm{\Gamma }_s`$ exhibits a turning point for any fixed value of $`\eta _s`$.
The excess adsorption $`\mathrm{\Gamma }_b`$ of the big spheres is shown in Fig. 8. For constant $`\eta _s`$ and increasing $`\eta _b`$, $`\mathrm{\Gamma }_b`$ decreases. But due to the same mechanism as described above, $`\mathrm{\Gamma }_b`$ increases upon increasing $`\eta _s`$ for constant packing fractions of the big spheres $`\eta _b`$. The square symbols ($`\mathrm{}`$) in Fig. 8 denote simulation data for the excess adsorption of a one-component hard-sphere fluid near a hard wall at packing fractions $`\eta _s`$=0.3680, 0.4103, and 0.4364, respectively, taken from Ref. . Whereas the simulation data for the two smaller packing fractions agree very well with our DFT results, that for the highest value of $`\eta _s`$ clearly deviates from the DFT prediction. Thus from this comparison it remains unclear to which extent the simulation data and the DFT results are reliable close to freezing.
In order to illustrate the large differences between the excess adsorptions calculated from different functionals, in Fig. 9 we show the excess adsorption $`\mathrm{\Gamma }_s`$ and $`\mathrm{\Gamma }_b`$ calculated from the original Rosenfeld functional together with those calculated by using the modified functional corresponding to Eq. (8). The packing fraction of the small spheres is $`\eta _s=0.15`$. Only for small packing fractions of the big spheres, i.e., far away from freezing, we find good agreement between the results obtained from the different functionals. Close to the phase boundaries the differences become large, which is indeed surprising at first glance. However, the main contribution of the excess adsorption stems from the vicinity of the wall where $`\xi `$ is large and the different versions of the Rosenfeld functional are expected to differ. These differences were already indicated in the previous subsection by the behavior of the individual contact values $`\sigma _i^3\rho _i(R_i+0)`$, $`i=s,b`$, of the density profiles of the small and big spheres, respectively. For large distances from the wall the density profiles exhibit decaying oscillations and therefore are not expected to contribute essentially to the excess adsorption. Moreover, all versions of the Rosenfeld functional display a common characteristic decay because they share the same bulk properties so that the contributions to the excess adsorption far away from the wall are very similar for the original Rosenfeld functional and its modifications.
The grand potential $`\mathrm{\Omega }`$ of a system in contact with a wall
$$\mathrm{\Omega }=\mathrm{\Omega }_{bulk}+\mathrm{\Omega }_{surf}$$
(26)
decomposes into a bulk contribution, $`\mathrm{\Omega }_{bulk}=pV`$, given by the bulk pressure $`p`$ in the system times the volume $`V`$ occupied by fluid particles, and a surface contribution, $`\mathrm{\Omega }_{surf}=\gamma A`$, which is the surface tension $`\gamma `$ times the surface area $`A`$ of the wall. Scaled particle theory (SPT) provides an approximate expression for $`\gamma `$ for a one-component hard-sphere fluid as well as a generalization to hard-sphere mixtures close to a planar hard wall. The surface tension of a one-component hard-sphere fluid within SPT is well tested and turns out to provide reliable results as compared with both DFT calculations and simulations . In Ref. a fit to simulation results of the surface tension of a one-component hard-sphere fluid at a planar hard wall is presented, which gives quasi-exact results and closely resembles the SPT expression.
In terms of the weighted bulk densities $`n_0,\mathrm{},n_3`$, defined in Eqs. (15)-(18), the SPT approximation for the surface tension of a hard-sphere mixture close to a planar hard wall can be written as
$$\beta \gamma _{SPT}=\frac{n_1}{1n_3}+\frac{1}{8\pi }\frac{n_2^2}{(1n_3)^2}.$$
(27)
This expression reduces to the one-component SPT approximation of the surface tension when it is evaluated for the one-component bulk weighted densities. In this latter case the surface tension can be expressed solely in terms of the packing fraction $`\eta `$ of this single component.
Within the Rosenfeld functional the surface tension $`\gamma `$ of a binary hard-sphere mixture at a planar hard wall follows from the equilibrium density profiles $`\rho _i(z)`$, $`i=s,b`$, as obtained in the previous subsection:
$`\beta \gamma `$ $`=`$ $`{\displaystyle \frac{\beta \mathrm{\Omega }+\beta pV}{A}}`$ (28)
$`=`$ $`\underset{L\mathrm{}}{lim}\left[\beta pL+{\displaystyle \underset{0}{\overset{L}{}}}dz\left\{\mathrm{\Phi }[\rho _s(z),\rho _b(z)]+{\displaystyle \underset{i=s,b}{}}\rho _i(z)(V_{ext}^i(z)\mu _i)\right\}\right].`$ (29)
In contrast to the strong dependence of the results of the excess adsorption on the choice of functional, we find that the original Rosenfeld functional as well as its modifications predict very similar results for the surface tension in the whole range of packing fractions studied here. Our results for the surface tension of a binary hard-sphere mixture with size ratio $`R_b:R_s=3:1`$, calculated within the original Rosenfeld functional, are shown in Fig. 10. The deviation between these results and those obtained by the modifications of the Rosenfeld functional are at most 3% and deviate from the predictions of the SPT at most by 10%.
## IV Summary and conclusions
Based on the Rosenfeld density functional we have analyzed structural and thermodynamic properties of binary hard-sphere mixtures near a hard wall with the following main results:
1. Figures 1 and 4 demonstrate that the structure of both density profiles $`\rho _i(z)`$, $`i=s,b`$, of a binary hard-sphere mixture close to a planar hard wall as obtained by the original Rosenfeld functional is in excellent agreement with simulation results for size ratios $`R_b:R_s=5:3`$ and $`R_b:R_s=3:1`$, respectively. The high level of agreement between our DFT results and the simulation data from Ref. is confirmed quantitatively by small mean square deviations $`\overline{E}_s`$ and $`\overline{E}_b`$ defined in Eq. (22). In terms of these quantities all versions of the Rosenfeld functional are practically of the same quality. Only at high packing fractions rather small differences between the results obtained by different versions of the Rosenfeld DFT become visible like those shown in Figs. 3 and 5.
2. The concentration profiles (see Fig. 2 and Fig. 6) calculated from the density profiles confirm the depletion picture: the small spheres are depleted from regions close to the wall while the big spheres are enriched.
3. The numerical accuracy of our calculations is demonstrated in Tables I and II by the high degree at which the sum rule Eq. (20), which relates the sum of the contact values of both density profiles with the equation of state, is respected. The sum rule, however, makes no prediction for the individual contact values and we find in Tables III and IV that each version of the Rosenfeld functional takes a different route to satisfy the sum rule.
4. Using the modified Rosenfeld functional corresponding to Eq.(8) we have calculated the excess adsorption of the small spheres $`\mathrm{\Gamma }_s(\eta _s,\eta _b)`$ (see Fig. 7) and of the big spheres $`\mathrm{\Gamma }_b(\eta _s,\eta _b)`$ (see Fig. 8) as functions of the packing fractions $`\eta _s`$ and $`\eta _b`$ for a binary hard-sphere mixture with size ratio $`R_b:R_s=3:1`$. We find that these quantities depend very sensitively on the accuracy of the numerical calculations and, as can be seen in Fig. 9, they differ significantly from the excess adsorption calculated by the original Rosenfeld functional.
5. The surface tension of a binary hard-sphere mixture with size ratio $`R_b:R_s=3:1`$ close to a planar hard wall is shown in Fig. 10. We find that all versions of the Rosenfeld functional give results which are in good agreement with the prediction of scaled particle theory \[Eq. (27)\].
From these results we conclude that the class of Rosenfeld functionals yields quantitatively reliable descriptions of interfacial structures in binary hard-sphere fluids. We expect that the same level of reliability also holds for multi-component hard-sphere fluids.
The excess adsorptions $`\mathrm{\Gamma }_s`$ and $`\mathrm{\Gamma }_b`$ of the small and big spheres emphasize the differences between the various versions of the Rosenfeld functional most. In order to decide whether the original Rosenfeld functional or whether its modifications predict these quantities more accurately, additional simulation data of the excess adsorption in a binary hard-sphere fluid are needed.
###### Acknowledgements.
It is a pleasure to thank Bob Evans for many stimulating discussions. We want to thank Dr. Noworyta for providing us with his simulation data.
## A Calculation of the weighted densities
Within the minimization procedure of the Rosenfeld functional the weighted densities $`n_\alpha `$ and $`𝐧_\alpha `$ have to be calculated repeatedly. Therefore it is necessary to optimize these calculations with respect to both computational speed and numerical accuracy.
The weight functions of the Rosenfeld functional are given by
$`\omega _i^{(3)}(𝐫)`$ $`=`$ $`\mathrm{\Theta }(|𝐫|R_i),`$ (A1)
$`\omega _i^{(2)}(𝐫)`$ $`=`$ $`\delta (|𝐫|R_i),`$ (A2)
and
$`𝝎_i^{(2)}(𝐫)`$ $`=`$ $`{\displaystyle \frac{𝐫}{|𝐫|}}\delta (|𝐫|R_i)`$ (A3)
with the Heaviside function $`\mathrm{\Theta }`$ and the Dirac delta function $`\delta `$. The remaining scalar weight functions are proportional to $`\omega _i^{(2)}`$: $`\omega _i^{(1)}=\omega _i^{(2)}/(4\pi R_i)`$ and $`\omega _i^{(0)}=\omega _i^{(2)}/(4\pi R_i^2)`$. The first vector weight function is collinear with $`𝝎`$$`{}_{}{}^{(2)}{}_{i}{}^{}`$: $`𝝎`$$`{}_{}{}^{(1)}{}_{i}{}^{}`$=$`𝝎`$$`{}_{i}{}^{(2)}/(4\pi R_i)`$.
In order to calculate the weighted densities integrals $`I_i^{(\alpha )}`$ of the type
$$I_i^{(\alpha )}=\mathrm{d}^3r^{}\rho _i(𝐫)\omega _i^\alpha (𝐫𝐫^{})$$
(A4)
have to be evaluated. For these convolution type integrals one can exploit the symmetry properties of the density profiles. For the present geometry the weighted densities can be written as
$$n_\alpha (z)=\underset{i=s,b}{}_{R_i}^{R_i}dz^{}\rho _i(z+z^{})\overline{\omega }_i^{(\alpha )}(z^{})$$
(A5)
with $`s`$ and $`b`$ for small and big, respectively, and with reduced weight functions $`\overline{\omega }_i^{(\alpha )}`$ which are functions of $`z`$ only:
$`\overline{\omega }_i^{(3)}(z)`$ $`=`$ $`\pi (R_i^2z^2),`$ (A6)
$`\overline{\omega }_i^{(2)}(z)`$ $`=`$ $`2\pi R_i,`$ (A7)
and
$`\overline{𝝎}_i^{(2)}(z)`$ $`=`$ $`2\pi z𝐞_z`$ (A8)
with the unit vector $`𝐞_z`$ in $`z`$-direction. The relations between these and the remaining weight functions are the same as for the original weight functions. The integrals in Eq. (A5) are one-dimensional convolutions which can be calculated faster and more accurate in Fourier space than in real space. By introducing the Fourier transforms of the density profiles,
$$\widehat{\rho }_i(k)=𝒯(\rho _i(z)),$$
(A9)
and those of the weight functions,
$$\widehat{\omega }_i^{(\alpha )}(k)=𝒯(\overline{\omega }_i^{(\alpha )}(z)),$$
(A10)
the weighted densities can be expressed as
$$n_\alpha (z)=𝒯^1\left(\underset{i=s,b}{}\widehat{\rho }_i(k)\widehat{\omega }_i^\alpha (k)\right).$$
(A11)
This route of calculation offers the important advantage that the numerical calculation of these convolutions can be speed up significantly by applying Fast-Fourier-Transform (FFT) methods. Moreover it turns out that calculations of convolutions in real space depend more sensitively on the grid size $`\mathrm{\Delta }z`$ to be used for discretization than those in Fourier space. We expect that the reason for this is that the FFT algorithm interpolates between data points with trigonometrical functions. To overcome this problem in real space a sophisticated integration scheme would have be applied or a very small grid size would have to be chosen. Both remedies additionally slow down the numerical calculation in real space.
The results presented in this appendix are applicable if the density profiles depend on the $`z`$ coordinate only. However, similar results can be obtained if the density profiles have radial symmetry. |
warning/0003/cond-mat0003046.html | ar5iv | text | # Mechanism of spin-triplet superconductivity in Sr2RuO4
\[
## Abstract
The unique Fermi surfaces and their nesting properties of Sr<sub>2</sub>RuO<sub>4</sub> are considered. The existence of unconventional superconductivity is shown microscopically, for the first time, from the magnetic interactions (due to nesting) and the phonon-mediated interactions. The odd-parity superconductivity is favored in the $`\alpha `$ and $`\beta `$ sheets of the Fermi surface, and the various superconductivities are possible in the $`\gamma `$ sheet. There are a number of possible odd-parity gaps, which include the gaps with nodes, the breaking of time-reversal symmetry and $`\stackrel{}{d}\widehat{z}`$.
\]
The nature of superconductivity discovered in Sr<sub>2</sub>RuO<sub>4</sub> has been the subject of intense theoretical and experimental activity. Although Sr<sub>2</sub>RuO<sub>4</sub> has the same layered perovskite structure as La<sub>2</sub>CuO<sub>4</sub>, the prototype of the high $`T_\mathrm{c}`$ cuprates superconductors, the electronic structures are very different and the nature of superconductivity seems to be totally different.
The normal state in Sr<sub>2</sub>RuO<sub>4</sub> is characterized as essentially a Fermi liquid below $`50`$K. The resistivities in all directions show $`T^2`$ behavior for $`T50`$K. The effective mass is about $`34m_{\mathrm{electron}}`$ and the susceptibility is also about $`34\chi _0`$ where $`\chi _0`$ is the Pauli spin susceptibility. In contrast to the conventional normal state (below $`50`$K), there are considerable experimental evidences that the superconducting state (below about $`1.5`$K) is unconventional. The nuclear quadrupole resonance(NQR) does not show the Hebel-Slichter peak. The transition temperature is very sensitive to non-magnetic impurities. The <sup>17</sup>O NMR Knight shift shows that the spin susceptibility has no change across $`T_\mathrm{c}`$ but stays just the same as in the normal state for the magnetic field parallel to the Sr<sub>2</sub>RuO<sub>4</sub> plane . In addition, spontaneous appearance of an internal magnetic field below the transition temperature is reported by muon spin rotation measurements ($`\mu SR`$) .
Shortly after the discovery of the superconductivity in Sr<sub>2</sub>RuO<sub>4</sub>, it was proposed that the odd-parity(spin -triplet) Cooper pairs are formed in the superconducting state in analogy with <sup>3</sup>He . The existence of ferromagnetic interaction is crucial in this proposal. In stead, incommensurate antiferromagnetic(AF) fluctuations were found by inelastic neutron scattering experiment . Ferromagnetic interaction is almost negligible compared with AF one.
Earlier specific heat and NQR measurements show a large residual density of states (DOS), $`5060`$% of DOS of the normal state, in the superconducting phase. A possible explanation, so called, orbital dependent superconductivity was proposed . Since four 4d electrons in Ru<sup>4+</sup> partially fill the $`t_{2g}`$ band, the relevant orbitals are $`d_{xy}`$, $`d_{xz}`$ and $`d_{yz}`$ which determine the electronic bands. The gap of of one class of bands is substantially smaller than that of the other class of band. The presence of gapless excitations for temperatures greater than the smaller gap would account for the residual DOS. Recent specific measurements on high quality compounds, however, suggest the absence of residual DOS.
Sigrist et al. proposed the following order parameter which is claimed to be compatible with most of the present experimental data,
$`\stackrel{}{d}=\widehat{z}(k_x\pm ik_y),`$ (1)
where $`\widehat{z}`$ is parallel to the $`\widehat{c}`$ axis and the gap is described as the tensor represented by $`\stackrel{}{d}`$ as $`\mathrm{\Delta }(k)=i(\stackrel{}{d}(k)\stackrel{}{\sigma })\sigma _y,`$ where $`\stackrel{}{\sigma }`$ is the Pauli matrix . Notice that the direction of the order parameter is frozen along the $`\widehat{c}`$ direction due to the crystal field and there is a full gap on the whole Fermi surface. The experiment of Josephson coupling between In and Sr<sub>2</sub>RuO<sub>4</sub> also supports the gap directing $`\widehat{c}`$ axis . Based on the gap (1), the effects of impurity scattering, spin wave excitations and collective modes and sound propagation are studied theoretically.
However, the recent experiments on high quality compounds of Ru NQR and specific heat strongly suggest the existence of nodes and the absence of residual DOS. It is proposed a number of gap order parameters consistent with the existence of nodes phenomenologically . See also Ref..
Details of the Fermi surface have been observed by quantum oscillations . The Fermi surface consists of three sheets, which is consistent with the electronic band calculations . The Fermi sheets are labeled by $`\alpha `$, $`\beta `$, and $`\gamma `$. See Fig. 1. While the $`\gamma `$ sheet of the Fermi surface can be attributed solely to $`d_{xy}`$ Wannier function, the $`\alpha `$ and $`\beta `$ sheets are due to the hybridization of the $`d_{xz}`$ and $`d_{yz}`$ Wannier functions. The $`\gamma `$ band is quasi-isotropic two-dimensional, on the other hand the $`\alpha `$ and $`\beta `$ sheets are quasi-one dimensional which can be visualized as a set of parallel planes separated by $`Q=4\pi /3`$ running both $`k_x`$ and $`k_y`$ directions. Therefore it is natural to expect a sizable nesting effects at the wave vectors $`(\pm Q,k_y,k_z)`$ and ($`k_x,\pm Q,k_z`$). The nesting vectors $`(\pm Q,\pm Q,k_z)`$ and $`(Q,\pm Q,k_z)`$ have the maximum effects since they connect the one dimensional Fermi surfaces in both directions. Collective modes in the spin dynamics are studied based on these nesting effects . In fact the neutron scattering experiment shows peaks at $`(\pm 0.6\pi ,\pm 0.6\pi )`$ close to the nesting vectors (up to $`(\pm 2\pi ,\pm 2\pi )`$) .
– Band structure and the pairing interactions
We denote the annihilation operators for electrons in the three $`4d`$-$`t_{2g}`$ orbitals $`d_{xz}`$, $`d_{yz}`$ and $`d_{xy}`$ of Ru-ions as $`a_{k,s}`$, $`b_{k,s}`$ and $`c_{k,s}`$. The kinetic term of the $`\alpha `$ and $`\beta `$ bands is given by
$`H_{\mathrm{kin}}^{\alpha \beta }`$ $`=`$ $`{\displaystyle \underset{k,s}{}}(\epsilon ^{\alpha \beta }(k_x)\mu )a_{k,s}^{}a_{k,s}`$ (2)
$`+`$ $`{\displaystyle \underset{k,s}{}}(\epsilon ^{\alpha \beta }(k_y)\mu )b_{k,s}^{}b_{k,s}`$ (3)
$`+`$ $`{\displaystyle \underset{k,s}{}}t(k)a_{k,s}^{}b_{k,s}+{\displaystyle \underset{k,s}{}}t^{}(k)b_{k,s}^{}a_{k,s},`$ (4)
and that of the $`\gamma `$ band is given by
$`H_{\mathrm{kin}}^\gamma ={\displaystyle \underset{k,s}{}}(\epsilon ^\gamma (k_x,k_y)\mu )c_{k,s}^{}c_{k,s}.`$ (5)
The $`\alpha `$ and $`\beta `$ bands are quasi one-dimensional and the Fermi surfaces are well-approximated by four sheets $`k_x\pm Q/2`$ and $`k_y\pm Q/2`$. The mixing coefficient $`t(k)`$ can be neglected except around $`k_x=\pm k_y`$. The $`\gamma `$ band is quasi-isotropic two-dimensional one. The mixing term between the $`\alpha `$ or $`\beta `$ band and the $`\gamma `$ one is suppressed by the reflection symmetry of $`z`$.
The nesting of the $`\alpha `$ and $`\beta `$ bands leads to the following AF fluctuations:
$`H_{\mathrm{AF}}^{\alpha \beta }`$ (6)
$`=`$ $`{\displaystyle \underset{k,q,s_i}{}}J_{}^{\alpha \beta }(q_x)(\sigma _z)_{s_1s_3}(\sigma _z)_{s_2s_4}a_{k,s_1}^{}a_{k,s_2}^{}a_{k+q,s_3}a_{kq,s_4}`$ (7)
$`+`$ $`{\displaystyle \underset{k,q,s_i}{}}J_{}^{\alpha \beta }(q_y)(\sigma _z)_{s_1s_3}(\sigma _z)_{s_2s_4}b_{k,s_1}^{}b_{k,s_2}^{}b_{k+q,s_3}b_{kq,s_4}`$ (8)
$`+`$ $`{\displaystyle \underset{k,q,s_i}{}}J_{}^{\alpha \beta }(q_x)\left[(\sigma _x)_{s_1s_3}(\sigma _x)_{s_2s_4}+(\sigma _y)_{s_1s_3}(\sigma _y)_{s_2s_4}\right]`$ (10)
$`\times a_{k,s_1}^{}a_{k,s_2}^{}a_{k+q,s_3}a_{kq,s_4}`$
$`+`$ $`{\displaystyle \underset{k,q,s_i}{}}J_{}^{\alpha \beta }(q_y)\left[(\sigma _x)_{s_1s_3}(\sigma _x)_{s_2s_4}+(\sigma _y)_{s_1s_3}(\sigma _y)_{s_2s_4}\right]`$ (12)
$`\times b_{k,s_1}^{}b_{k,s_2}^{}b_{k+q,s_3}b_{kq,s_4},`$
where $`J_{}^{\alpha \beta }(q_i)>0`$ and $`J_{}^{\alpha \beta }(q_i)>0`$ have peaks at $`q_i\pm Q`$. In general, $`J_{}^{\alpha \beta }(q_i)J_{}^{\alpha \beta }(q_i)`$. The magnetic field generated by the AF fluctuations above induces the interaction $`H_{\mathrm{AF}}^\gamma `$ in the $`\gamma `$ band:
$`H_{\mathrm{AF}}^\gamma `$ (13)
$`=`$ $`{\displaystyle \underset{k,q,s_i}{}}J_{}^\gamma (q_x,q_y)(\sigma _z)_{s_1s_3}(\sigma _z)_{s_2s_4}c_{k,s_1}^{}c_{k,s_2}^{}c_{k+q,s_3}c_{kq,s_3}`$ (14)
$`+`$ $`{\displaystyle \underset{k,q,s_i}{}}J_{}^\gamma (q_x,q_y)\left[(\sigma _x)_{s_1s_3}(\sigma _x)_{s_2s_4}+(\sigma _y)_{s_1s_3}(\sigma _y)_{s_2s_4}\right]`$ (16)
$`\times c_{k,s_1}^{}c_{k,s_2}^{}c_{k+q,s_3}c_{kq,s_3},`$
where $`J_{}^\gamma (q_x,q_y)>0`$ and $`J_{}^\gamma (q_x,q_y)>0`$ have peaks at $`(q_x,q_y)=(\pm Q,\pm Q)`$ and $`(q_x,q_y)=(Q,\pm Q)`$.
In addition to the AF fluctuations, we consider the electron-phonon interaction. Due to the low dimensionality of the system, this interaction is weakly screened. We assume that the phonon-mediated interaction for the $`\alpha `$ and $`\beta `$ bands $`H_{\mathrm{ph}}^{\alpha \beta }`$ is quasi-one dimensional,
$`H_{\mathrm{ph}}^{\alpha \beta }`$ $`=`$ $`{\displaystyle \underset{k,k^{},s,s^{}}{}}f^{\alpha \beta }(q_x)a_{k,s}^{}a_{k,s^{}}^{}a_{k+q,s}a_{kq,s^{}}`$ (17)
$`+`$ $`{\displaystyle \underset{k,k^{},s,s^{}}{}}f^{\alpha \beta }(q_y)b_{k,s}^{}b_{k,s^{}}^{}b_{k+q,s}b_{kq,s^{}},`$ (18)
where $`f^{\alpha \beta }(q_i)>0`$ has a peak at $`q_i=0`$. For the $`\gamma `$ band, the phonon-mediated interaction we consider is
$`H_{\mathrm{ph}}^\gamma ={\displaystyle \underset{k,k^{},s,s^{}}{}}f^\gamma (q_x,q_y)c_{k,s}^{}c_{k,s^{}}^{}c_{k+q,s}c_{kq,s^{}}`$ (19)
where $`f^\gamma (q_x,q_y)>0`$ has a peak at $`(q_x,q_y)=(0,0)`$.
– Superconductivity in the $`\alpha `$ and $`\beta `$ bands
First, we examine the superconductivity in the $`\alpha `$ and $`\beta `$ bands. As we will show immediately, the odd-parity superconductivity is realized due to the quasi-one dimensionality . For simplicity, we put $`f^{\alpha \beta }(q_i)f^{\alpha \beta }(0)\delta _{q_i,0}`$, $`J_{}^{\alpha \beta }(q_i)J_{}^{\alpha \beta }(Q)\delta _{q_i,\pm Q}`$ and $`J_{}^{\alpha \beta }(q_i)J_{}^{\alpha \beta }(Q)\delta _{q_i,\pm Q}`$ in the following.
In the lowest order approximation, we neglect the mixing term $`t(k)`$, so the gap equation is separated for $`a_{k,s}`$ and $`b_{k,s}`$. For $`a_{k,s}`$ electrons, the gap $`\mathrm{\Delta }^{(a)}`$ is defined by
$`\mathrm{\Delta }_{s_2,s_1}^{(a)}(k)={\displaystyle \underset{k^{},s_3,s_4}{}}V_{s_1,s_2,s_3,s_4}(k_x,k_x^{})a_{k^{},s_3}a_{k^{},s_3},`$ (20)
where
$`V_{s_1,s_2,s_3,s_4}(k_x,k_x^{})`$ (21)
$`=f^{\alpha \beta }(k_x+k_x^{})\delta _{s_1s_3}\delta _{s_2s_4}f^{\alpha \beta }(k_xk_x^{})\delta _{s_1s_4}\delta _{s_2s_3}`$ (22)
$`+J_{}^{\alpha \beta }(k_x+k_x^{})(\sigma _z)_{s_1s_3}(\sigma _z)_{s_2s_4}`$ (23)
$`J_{}^{\alpha \beta }(k_xk_x^{})(\sigma _z)_{s_1s_4}(\sigma _z)_{s_2s_3}.`$ (24)
$`+J_{}^{\alpha \beta }(k_x+k_x^{})\left[(\sigma _x)_{s_1s_3}(\sigma _x)_{s_2s_4}+(\sigma _y)_{s_1s_3}(\sigma _y)_{s_2s_4}\right]`$ (25)
$`J_{}^{\alpha \beta }(k_xk_x^{})\left[(\sigma _x)_{s_1s_4}(\sigma _x)_{s_2s_3}+(\sigma _y)_{s_1s_4}(\sigma _y)_{s_2s_3}\right].`$ (26)
If the gap is unitary (as we assume), the gap equation becomes
$`\mathrm{\Delta }_{s_2,s_1}^{(a)}(k)`$ $`={\displaystyle \underset{k^{},s_3,s_4}{}}V_{s_1,s_2,s_3,s_4}(k_x,k_x^{})`$ (28)
$`\times {\displaystyle \frac{\mathrm{\Delta }_{s_3,s_4}^{(a)}(k^{})}{2E_k^{}^{(a)}}}\mathrm{tanh}\left({\displaystyle \frac{\beta E_k^{}^{(a)}}{2}}\right),`$
where $`E_k^{(a)}=\sqrt{ϵ^{\alpha \beta }(k_x)+\mathrm{tr}(\mathrm{\Delta }^{(a)}\mathrm{\Delta }^{(a)})/2}`$. The gap $`\mathrm{\Delta }^{(a)}`$ is parameterized as
$`\mathrm{\Delta }^{(a)}(k)=\{\begin{array}{cc}i\psi ^{(a)}(k)\sigma _y\hfill & \text{for even-parity gap}\hfill \\ i(\stackrel{}{d}^{(a)}(k)\stackrel{}{\sigma })\sigma _y\hfill & \text{for odd-parity gap}\hfill \end{array},`$ (31)
and we obtain the solutions as $`\psi ^{(a)}|_{k_x\pm Q/2}=\mathrm{const}.`$ for even-parity gap and $`\stackrel{}{d}^{(a)}|_{k_x\pm Q/2}=\pm \mathrm{const}.`$ for odd-parity gap. For these solutions, the critical temperature is $`k_\mathrm{B}T_\mathrm{c}=1.13\mathrm{}\omega _\mathrm{D}e^{1/N(\mu )\lambda },`$ where $`\omega _\mathrm{D}`$ is the cut-off for the interactions, $`N(\mu )`$ is the DOS at the Fermi surface for $`a_{k,s}`$ electrons and $`\lambda `$ is
$`\lambda =\{\begin{array}{cc}f^{\alpha \beta }(0)J_{}^{\alpha \beta }(Q)2J_{}^{\alpha \beta }(Q)\hfill & \text{for even-parity gap}\hfill \\ f^{\alpha \beta }(0)+J_{}^{\alpha \beta }(Q)2J_{}^{\alpha \beta }(Q)\hfill & \text{for }\stackrel{}{d}^{(a)}\stackrel{}{z}\hfill \\ f^{\alpha \beta }(0)J_{}^{\alpha \beta }(Q)\hfill & \text{for }\stackrel{}{d}^{(a)}\stackrel{}{z}\hfill \end{array}.`$ (35)
Therefore, the odd-parity superconductivity is realized in $`a_{k,s}`$, and $`\stackrel{}{d}^{(a)}`$ vector becomes
$`\stackrel{}{d}^{(a)}|_{kyQ/2}=\stackrel{}{d}^{(a)}|_{kyQ/2}=\mathrm{const}.,`$ (36)
and
$`\begin{array}{cc}\stackrel{}{d}^{(a)}\widehat{z}\hfill & \text{if }J_{}^{\alpha \beta }(Q)>J_{}^{\alpha \beta }(Q)\text{ }\hfill \\ \stackrel{}{d}^{(a)}\widehat{z}\hfill & \text{if }J_{}^{\alpha \beta }(Q)<J_{}^{\alpha \beta }(Q)\text{ }\hfill \end{array}.`$ (39)
Note that if $`J_{}^{\alpha \beta }(Q)`$ is large enough, the superconductivity is realized even when $`f^{\alpha \beta }(0)=0`$.
In a similar manner, the superconductivity in $`b_{k,s}`$ is odd-parity and characterized by (36) and (39) when every $`a`$’s, $`k_x`$ and $`k_y`$ are replaced by $`b`$’s, $`k_y`$ and $`k_x`$. The critical temperature is same as $`a_{k,s}`$. Due to $`D_{4h}`$ symmetry, it holds that $`|\stackrel{}{d}^{(a)}|=|\stackrel{}{d}^{(b)}|`$.
The mixing term $`t(k)`$ affects the gaps $`\stackrel{}{d}^{(a)}`$ and $`\stackrel{}{d}^{(b)}`$ at $`k=\pm K(\pm Q/2,\pm Q/2,k_z)`$ and $`k=\pm \stackrel{~}{K}(Q/2,\pm Q/2,k_z)`$. If $`t(k)`$ is large enough, the Fermi surface at $`k=\pm K`$ and $`k=\pm \stackrel{~}{K}`$ is largely deformed. Because of the weakness of the screening, the attractive force for the electron at those points decreases. This causes nodes at those points. The relative phase of $`\stackrel{}{d}^{(a)}`$ and $`\stackrel{}{d}^{(b)}`$ is not determined in this case, so the time-reversal symmetry is broken in general.
If $`t(k)`$ is small, $`\stackrel{}{d}^{(a)}`$ and $`\stackrel{}{d}^{(b)}`$ do not disappear at $`k=\pm K`$ and $`k=\pm \stackrel{~}{K}`$. In this case, the relative phase of $`\stackrel{}{d}^{(a)}`$ and $`\stackrel{}{d}^{(b)}`$ is determined. Since $`t(k)`$ at $`k_x=\pm k_y`$ is real, it can be easily shown that $`\mathrm{arg}\stackrel{}{d}^{(a)}=\mathrm{arg}\stackrel{}{d}^{(b)}`$ or $`\mathrm{arg}\stackrel{}{d}^{(a)}=\mathrm{arg}\stackrel{}{d}^{(b)}+\pi `$ at $`k=\pm K`$ and $`k=\pm \stackrel{~}{K}`$. The superconductivity in this case is illustrated in Fig. 2. The existence of nodes in the small $`t(k)`$ case depends on the symmetry of the system. If the system is invariant under the following transformation,
$`\begin{array}{cc}a_{k,s}b_{k,s},b_{k,s}a_{k,s}\hfill & \text{for }k_x=\pm k_y\hfill \\ a_{k,s}a_{k,s},b_{k,s}b_{k,s}\hfill & \text{for }k_x\pm k_y\hfill \end{array},`$ (42)
the following pairing interaction is added.
$`\mathrm{\Delta }H_{\mathrm{int}}^{\alpha \beta }=`$ $`{\displaystyle \underset{s,s^{}}{}}\mathrm{\Delta }_{ss^{}}^{(a)}(K)b_{K,s}b_{K,s}+(\text{h.c.})`$ (45)
$`{\displaystyle \underset{s,s^{}}{}}\mathrm{\Delta }_{ss^{}}^{(a)}(\stackrel{~}{K})b_{\stackrel{~}{K},s}b_{\stackrel{~}{K},s}+(\text{h.c.})`$
$`+(ab).`$
This interaction changes the spectrum at $`k=\pm K`$ and $`k=\pm \stackrel{~}{K}`$. If $`\stackrel{}{d}^{(a)}\stackrel{}{d}^{(b)}`$ holds, nodes exists at either $`k=\pm K`$ or $`k=\pm \stackrel{~}{K}`$. If the system does not have the symmetry above or $`\stackrel{}{d}^{(a)}\parallel ̸\stackrel{}{d}^{(b)}`$, no node exists in the gap.
– Superconductivity in the $`\gamma `$ band
Next, we examine the superconductivity in the $`\gamma `$ band based on the phonon-mediated interaction and AF fluctuations. In contrast to the $`\alpha `$ and $`\beta `$ bands, both even-parity and odd-parity superconductivity can be realized. If $`J_{}^\gamma (q_x,q_y)`$ is large enough, the odd-parity superconductivity is also realized in the $`\gamma `$ band, but if not, the various superconductivity can be realized. To show this, we introduce the following AF interaction and phonon-mediated interaction:
$`J_{\left\{\genfrac{}{}{0pt}{}{}{}\right\}}^\gamma (q_x,q_y)`$ (46)
$`={\displaystyle \underset{Q_{\left\{\genfrac{}{}{0pt}{}{x}{y}\right\}}=\pm 4\pi /3}{}}{\displaystyle \frac{g_{\left\{\genfrac{}{}{0pt}{}{}{}\right\}}}{(q_xQ_x)^2+(q_yQ_y)^2+Q_{\mathrm{AF}}^2}},`$ (47)
and
$`f^\gamma (q_x,q_y)={\displaystyle \frac{g_{\mathrm{ph}}}{q_x^2+q_y^2+Q_{\mathrm{ph}}^2}},`$ (48)
where $`g_{\left\{\genfrac{}{}{0pt}{}{}{}\right\}}`$, $`g_{\mathrm{ph}}`$, $`Q_{\mathrm{ph}}`$ are constants. From the experimental data in , we take $`Q_{\mathrm{AF}}=0.0797\times 2\pi `$.
To solve the gap equation, we approximate the Fermi surface of the $`\gamma `$ band as a cylinder with $`k_\mathrm{F}=2\pi /3`$ and use the mean-field approximation. We have solved the gap equation analytically when the phonon-mediated interaction dominates. The gap in this case is given by
$`\psi (k)\{\begin{array}{cc}\mathrm{const}.\hfill & \text{for }s\text{ gap}\hfill \\ k_x^2k_y^2\hfill & \text{for }d\text{ gap}\hfill \end{array},`$ (51)
$`\stackrel{}{d}(k)(k_x\pm ik_y)\text{for }p\text{ gap}.`$ (52)
In Fig.3, we show the phase diagram in the case where the phonon-mediated interaction dominates and $`g_{}=g_{}`$.
– Discussions
To conclude, we have studied the superconductive properties of Sr<sub>2</sub>RuO<sub>4</sub>based on AF fluctuations and phonon-mediated interactions. Due to the quasi-one dimensionality, the odd-parity superconductivity is realized in the $`\alpha `$ and $`\beta `$ bands. The existence of nodes in the gaps of these bands depends on the symmetry of the system (see Eq.(42)) or the strength of the hybridization of $`d_{xz}`$ and $`d_{yz}`$. The direction of $`\stackrel{}{d}`$ vector is determined by the anisotropy of the AF fluctuations.
The $`\gamma `$ band has many possibilities of superconductivity. If the AF fluctuations direct to the $`\widehat{c}`$ axis and are strong enough, the odd-parity superconductivity with $`\stackrel{}{d}\widehat{z}`$ is realized, however, if not, $`s`$\- or $`d`$-wave superconductivity is also possible.
There are various possibilities to explain the experimental results of Sr<sub>2</sub>RuO<sub>4</sub> from our theory. Here we give one of them which is not in the previous theories. The NMR data and the experiment of the Josephson coupling between In and Sr<sub>2</sub>RuO<sub>4</sub> support the odd-parity superconductivity with $`\stackrel{}{d}\widehat{z}`$. This is easily realized in our theory with large $`J_{}^{\alpha \beta }(Q)`$ and/or $`J_{}^\gamma (Q,Q)`$. The nodes suggested by NQR and specific heat are likely to be those in the $`\alpha `$ and $`\beta `$ bands. The square vortex lattice observed by the neutron scattering is consistent with the odd-parity superconductivity in the $`\alpha `$ and $`\beta `$ bands since the superconductivity occurs in the orthogonal quasi-one dimensional systems. The $`\mu `$SR data can be explained by the superconductivity in the $`\alpha `$ and $`\beta `$ bands with the large hybridization of $`d_{xz}`$ and $`d_{yz}`$ or/and the superconductivity in $`\gamma `$ bands with the gap (52). Since larger $`J_{}^{\alpha \beta }(Q)`$ induces larger $`J_{}^\gamma (Q,Q)`$, it is likely that the odd-parity superconductivity in the $`\alpha `$ and $`\beta `$ band is followed by that in $`\gamma `$ band, so the absence of residual DOS also can be explained. The transition temperatures could be different between $`\gamma `$ and $`\alpha `$, $`\beta `$ Fermi surfaces, but there will be a single $`T_\mathrm{c}`$ if the hybridization between $`d_{xy}`$ and $`d_{yz}`$,$`d_{xz}`$ on the Ru atoms is large enough.
Acknowledgments
It is a pleasure to thank P. Bourges, Y. Hasegawa, K. Ishida, H.-Y. Kee, Y. Maeno, K. Maki, Y. Matsuda, Y. Sidis, and M. Sigrist, for useful discussions. |
warning/0003/math-ph0003009.html | ar5iv | text | # Discrete spectral symmetries of low-dimensional differential operators and difference operators on regular lattices and two-dimensional manifolds
## §1. Introduction: history of the problem, the one-dimensional Schrödinger operator
Already in the 18th century (1742) Euler paid attention to the substitutions which transform solutions of a certain ordinary linear differential equation into solutions of another equation associated with it. Let
$$L=_x^2+u(x)$$
be the Sturm–Liouville operator defined on the real line $`(x)`$, and let $`\psi (x,\lambda )`$ be a solution of the linear equation $`L\psi =\lambda \psi `$. Let $`v(x)`$ be an arbitrary solution of the
Riccati equation
$$u(x)=v_x+v^2.$$
Under these conditions we have the following result.
###### Lemma 1
The new function $`\stackrel{~}{\psi }=\psi _xv(x)\psi `$ satisfies the new equation $`\stackrel{~}{L}\stackrel{~}{\psi }=\lambda \stackrel{~}{\psi }`$, where
$$\stackrel{~}{L}=_x^2+\stackrel{~}{u}(x),\stackrel{~}{u}(x)=v_x+v^2.$$
Proof This lemma easily follows from the fact that $`L`$ is in fact the product of two non-commuting factors of the first order:
$$L=_x^2+u=(_x+v)(_xv),$$
where $`u=v_x+v^2`$. Further, we have
$$\psi _{xx}+u\psi =\lambda \psi =(_x+v)(_xv)\psi =(_x+v)\stackrel{~}{\psi }.$$
Hence,
$$\stackrel{~}{L}\stackrel{~}{\psi }=(_xv)(_x+v)\stackrel{~}{\psi }=\lambda (_xv)\psi =\lambda \stackrel{~}{\psi },$$
where
$$\stackrel{~}{L}=(_xv)(_x+v)=_x^2+\stackrel{~}{u}(x).$$
We have proved Lemma 1. $`\mathrm{}`$
Hence, the Euler substitution is based on the representation of $`L`$ as the product of first-order operators which are formally adjoint to each other if $`u,v`$:
$$L=(_x+v)(_xv)=QQ^+,$$
and is given by interchanging them:
$$QQ^+=L\stackrel{~}{L}=Q^+Q,\psi \stackrel{~}{\psi }=Q^+\psi .$$
In the literature the Euler substitutions are often called ‘Darboux transformations’ or ‘Bäcklund–Darboux transformations’ because of the role they play in the theory of non-linear systems of Korteweg–de Vries (KdV) type for functions $`u(x,t)`$ that depend on the time as the parameter.
The classical mathematicians put it like this: the Darboux transformation $`L\stackrel{~}{L}`$ is determined by one chosen solution $`\phi `$ of the equation $`L\phi =0`$, and
$$\stackrel{~}{\psi }=\psi _x(\phi _x/\phi )\psi .$$
We have
$$v(x)=\phi _x/\phi ,\text{where}v_x+v^2=u(x).$$
The algebraic language, that is, the correspondence of this with the factorization $`\alpha +L=QQ^+`$, was used after the works of physicists of the 30s and 40s (Dirac, Schrödinger, Infeld).
¿From the point of view of formal spectral theory, that is, of local solutions of the equation $`L\psi =\lambda \psi `$, the Darboux–Euler transformation generates a transformation of the eigensubspaces for all $`\lambda `$:
$$B_\lambda \psi \stackrel{~}{\psi },$$
where $`L\psi =\lambda \psi `$, $`\stackrel{~}{L}\stackrel{~}{\psi }=\lambda \stackrel{~}{\psi }`$, $`\stackrel{~}{\psi }=\psi _xv\psi `$.
We should talk about the Darboux–Euler transformation of the whole linear hull (with respect to all $`\lambda `$)
$$Ba_i\psi _ia_i\stackrel{~}{\psi }_i,$$
where $`L\psi _i=\lambda _i\psi _i`$ and $`a_i`$ are arbitrary coefficients.
The transformation $`B`$ is not, strictly speaking, an isomorphism: for $`\lambda =0`$ and $`\psi =\phi `$ we have $`B\phi =0`$, $`v=\phi _x/\phi `$. Hence, the kernel of this transformation is one-dimensional.
Turning to the present ‘global’ spectral theory of the operator $`L`$ in the Hilbert space $`_2()`$, we see that if the function $`\phi (x)`$ does not belong to $`_2()`$, then the Darboux transformation $`B`$ can generate an isomorphism of the spectral theories of the operators $`L`$ and $`\stackrel{~}{L}`$; sometimes $`B`$ is a monomorphism on a part of the spectrum of $`L`$ that covers all but one eigenfunction in $`_2()`$; sometimes (if $`\phi (x)0`$ and $`\phi _2()`$) $`B`$ annihilates exactly one state of the spectrum $`\phi `$, the lowest one, the ‘basic’ one in $`_2()`$. We do not know the complete classification of all possible cases from the point of view of spectral theory in $`_2()`$.
Definition 1 We call the Darboux–Euler transformation the latent algebraic symmetry of the spectral theory of the one-dimensional Schrödinger operator, or the discrete symmetry.
We recall that the ‘discontinuous spectral symmetries’ or ‘isospectral deformations’ of the operator $`L`$ are the well-known systems of the theory of solitons like KdV, which generate a gigantic commutative continuous group and in which KdV is a one-parameter subgroup, see .
The Darboux–Euler transformation depends on a parameter, which appears in the factorization (3), that is, on the choice of a solution of the Riccati equation (2). We denote these transformations by $`B`$,
$$BL\stackrel{~}{L}=(_xv)(_x+v),\psi \stackrel{~}{\psi }=(_xv)\psi .$$
Moreover, there is still a one-parameter family of transformations of ‘energy shift’
$$T_\alpha LL+\alpha ,\alpha =\text{const}.$$
Let $`B_\alpha =BT_\alpha `$.
In the theory of solitons, by means of the transformations $`B_\alpha `$ we could create well-known non-reflecting potentials from the zero potential. By one application of the transformation $`B_\alpha `$ we can obtain ‘soliton potentials’ from the free operator $`L_0=_x^2`$, starting with $`v=ktanhk(xx_0)`$, $`\phi =coshk(xx_0)`$, $`\alpha =k^2`$.By multiple application of the transformations $`B_{\alpha _j}`$ we can obtain all rapidlydecreasing non-reflecting potentials, taking each time a non-vanishing solution $`\phi _j`$‘below the spectrum’:
$$u_{N+1}(x)=B_{\alpha _N}B_{\alpha _{N1}}\mathrm{}B_{\alpha _0}(0),$$
$$u_0(x)=0,u_1(x)=2k^2/cosh^2k(xx_0),\mathrm{}.$$
In the literature on the theory of solitons (see ), by means of Bäcklund–Darboux transformations the so-called ‘many-soliton potentials on the background of finite-zone potentials’ were obtained by applying the transformations $`B_\alpha `$ with suitable parameters to the known periodic and quasiperiodic finite-zone potentials. This class of potentials corresponds to the limits of finite-zone potentials for various degenerate Riemann surfaces. For a long time nobody succeeded in establishing the relation between the class of non-degenerate periodic (quasiperiodic) finite-zone potentials (one-dimensional Schrödinger operators) and the theory of Bäcklund–Darboux transformations. We recall that finite-zone potentials correspond to orbits of the infinite-dimensional continuous commutative group of isospectral symmetries, the so-called higher analogues of KdV, which have finite dimension (). These operators have remarkable spectral and algebro-geometric properties: their coefficients are calculated by means of theta-functions of hyperelliptic Riemann surfaces, and they generate remarkable fully integrable Hamiltonian systems.
An important idea on the connection between Bäcklund–Darboux transformations and the theory of finite-zone potentials was proposed in 1986 (). In this work the author considered ‘cyclic chains’ of the Bäcklund–Darboux transformations of length $`N+1`$, that is, equations of the form
$$u=B_{\alpha _N}B_{\alpha _{N1}}\mathrm{}B_{\alpha _0}(u)$$
under the condition $`\alpha _N=\alpha _{N1}=\mathrm{}=\alpha _0=0`$. Developing very interesting technical arguments in studying such chains as certain integrable non-linear systems, in the supposition was put forward that for $`N=2k`$ the potential $`u(x)`$ is always a finite-zone potential (Weiss’ hypothesis). This hypothesis was proved in in a more general form.
Suppose that $`N=2k`$ and the $`\alpha _j`$ may be non-zero. If $`_{j=0}^N\alpha _j=0`$, then the potential $`u(x)`$ in the cyclic chain of Bäcklund–Darboux transformations is finite-zone with a Riemann surface of genus no greater than $`N+1`$.
If $`_{j=0}^N\alpha _j=\alpha 0`$, then the potential $`u(x)`$ has an ‘oscillator-similar’ asymptotic form
$$u(x)=\frac{\alpha ^2x^2}{4(N+1)^2}+O(x),|x|\mathrm{}.$$
If the potential $`u(x)`$ is smooth and real (without singularities), then the spectrum of the operator $`L=_x^2+u(x)`$ is a combination of $`N+1`$ arithmetic progressions with the general difference $`\alpha `$. At the same time there is a differential operator $`A_{N+1}`$ of order $`N+1`$ such that $`[A_{N+1},L]=\alpha A_{N+1}.`$
V. E. Adler made a numerical investigation and established that the equation of cyclicity of the chains of transformations $`B_{\alpha _j}`$ for $`\alpha 0`$ has non-singular real solutions. In particular, for $`N=2`$ this equation is transformed into the Painlevé-IV equation.
Operators of the cyclic chain
$$L_0=L_{N+1},L_N,\mathrm{},L_1,L_0,\mathrm{}$$
have factorizations of the form $`\alpha _j+L_j=Q_jQ_j^+`$. ¿From the definition of Darboux–Euler substitutions we have:
$`L_j+\alpha _j=Q_jQ_j^+=(_x+v_j)(_xv_j),`$
$`L_{j+1}=\stackrel{~}{L}_j=Q_j^+Q_j.`$
For the operators $`L_j`$ we consider a set of ‘ground states’ defined by the equation
$$Q_{j1}\psi _{0,j}=0=L_j\psi _{0,j}.$$
We assume that $`\psi _{0,j}_2()`$, and all the translations $`\alpha _j`$ and the coefficients of the operators $`Q_j`$$`Q_j^+`$ are real. Then the sequence of ‘creation operators’ gives the whole spectrum in $`_2()`$ for operators $`L_M`$ ($`Mj`$) according to the formula
$`L_M\psi _{Mj,j}=\left({\displaystyle \underset{K=j}{\overset{M}{}}}\alpha _K\right)\psi _{Mj,j},`$
$`Q_M^+Q^+\mathrm{}Q_{j+1}^+Q_j^+\psi _{0,j}=\psi _{Mj,j},Mj.`$
Here $`L_M=L_{M+l(N+1)}`$, $`n=2k`$, $`l`$.
We note that the sequence of numbers $`_{K=j}^M\alpha _K`$ is a combination of finitely many arithmetic progressions with common difference $`_{K=0}^N\alpha _K=\alpha `$, since the set of numbers $`\alpha _j`$ is periodic. The spectrum of the operator $`L_M`$ has the form $`_{K=j}^M\alpha _K=\lambda _{j,M}`$, $`j=M,M1,M2,\mathrm{}`$ .
These are the basic results in the theory of cyclic Bäcklund–Darboux chains for the one-dimensional continuous Schrödinger operator ().<sup>1</sup><sup>1</sup>1For non-trivial cyclic chains of even length a solution of the problem of classification has not been obtained. It would not be superfluous to see in the literature a publication containing a full justification of this beautiful algebraic picture from the viewpoint of rigorous spectral theory, that is, functional analysis. So far, this problem can be considered ‘more or less solved’ on the level of the requirements of reasonable (not superrigorous) mathematical physics. An understanding of what happens and a set of rigorous formulae have already been obtained, but the rigorous completeness of the picture has not been proved in general.
Example 1 Let $`N=0`$ (a cyclic chain of period 1); we have
$$\mathrm{}=L_1=L_0=L_1=\mathrm{},$$
where $`L_0=Q^+Q`$, $`\alpha +L_1=QQ^+`$. Here there arises the Heisenberg algebra
$$QQ^+=Q^+Q+\alpha ,Q=_x+\frac{\alpha x}{2}.$$
Since $`Q=_x+v(x)`$, we have
$$u(x)=v_x+v^2=\frac{\alpha }{2}+\left(\frac{\alpha }{2}\right)^2.$$
The eigenfunctions $`\psi _{Mj,j}`$ (mentioned above) are the eigenfunctions of the quantum oscillator
$$L=_x^2+\frac{\alpha ^2x^2}{4}=\left(_x+\frac{\alpha x}{2}\right)\left(_x\frac{\alpha x}{2}\right)\frac{\alpha }{2}=\left(_x\frac{\alpha x}{2}\right)\left(_x+\frac{\alpha x}{2}\right)+\frac{\alpha }{2}.$$
Let $`\alpha >0`$. The equation $`Q\psi _0=0`$ has a solution $`\psi _0_2()`$, $`\psi _0=e^{\alpha x^2/4}`$. All eigenfunctions $`\psi _M=(Q^+)^M\psi _0`$, $`Q^+=(_x\alpha x/2)`$, belong to $`_2()`$ and generate the spectrum of the operator $`L=_x^2+\alpha ^2x^2/4`$,
$$\lambda _M=\frac{\alpha }{2}+M\alpha .$$
In this example everything is clear, but we should establish in the general case that the eigenfunctions $`\psi _{Mj,j}`$ belong to $`_2()`$.
## §2. The non-stationary one-dimensional Schrödinger equation
The Darboux substitution for the non-stationary Schrödinger equation
$$i\psi _t=\psi _{xx}+u\psi 0$$
is worthy of mention (see ). These substitutions have been used to construct some exact solutions of the KP equation. Starting from an exact solution
$$i\phi _t=\phi _{xx}+u(x,t)\phi ,1$$
we define the transformation
$$B\psi \stackrel{~}{\psi }=\psi _x\frac{\phi _x}{\phi }\psi .2$$
The function $`\stackrel{~}{\psi }`$ satisfies the equation
$$i\stackrel{~}{\psi }_t=\stackrel{~}{\psi }_{xx}+\stackrel{~}{u}(x,t)\stackrel{~}{\psi },3$$
where $`\stackrel{~}{u}=u2(\mathrm{log}\phi )_{xx}`$.
We express this in algebraic language.
###### Lemma 2
Suppose there is given a real connection of zero curvature
$$_t=_tw,_x=_xv,[_t,_x]=04$$
such that $`(10)`$ has the form
$$i_t\psi =_x^+_x\psi ,_x^+=(_x+v).5$$
Then the Darboux transformation has the form
$$\stackrel{~}{\psi }=_x\psi .6$$
The new function $`\stackrel{~}{\psi }(x,t)`$ satisfies the equation
$$i_t\stackrel{~}{\psi }=_x_x^+\stackrel{~}{\psi }.7$$
Proof ¿From (15) it follows that
$$i_t\psi =_x^+\stackrel{~}{\psi }.$$
We apply $`_x`$ to both sides of this equation and use the relation of zero curvature $`_x_t=_t_x`$. We obtain
$$i_x_t\psi =_x_x^+\stackrel{~}{\psi }=i_t(_x\psi )=i_t\stackrel{~}{\psi }.$$
We have proved Lemma 2. $`\mathrm{}`$
The representation (15) is determined by one solution $`\phi `$ of (11), since the connection of zero curvature has the form
$$_t=_t\frac{\phi _t}{\phi },_x=_x\frac{\phi _x}{\phi }\mathrm{\hspace{0.17em}.8}$$
Hence, a substitution of Euler–Darboux type here hardly differs from the case of the stationary Schrödinger operator.
We now consider cyclic chains of length 1. Let the Schrödinger equation and a chosen solution $`\phi `$ be such that in one step we come to this (gauge equivalent) equation:
$`\stackrel{~}{u}(x,t)=u2(\mathrm{log}\phi )_{xx}=uC(t),`$
$`i\phi _t=\phi _{xx}+u(x,t)\phi .`$
We obtain the relation
$$\phi (x,t)=\mathrm{exp}\left(\frac{Cx^2}{2}+Ax+B\right),9$$
where $`A`$, $`B`$, $`C`$ are arbitrary functions of $`t`$. For $`C0`$ we obtain an equation of the form
$$\phi (x,t)=\mathrm{exp}\left(\frac{C(xx_0(t))^2}{2}+D(t)\right).0$$
For $`C0`$ we have ($`A0`$):
$$\phi (x,t)=\mathrm{exp}\left(A(t)x+B\right)=\mathrm{exp}\left(A(t)(xx_0(t))\right).$$
It is easy to prove the following lemma.
###### Lemma 3
The Darboux chain for a non-stationary Schrödinger equation is cyclic of period $`1`$ if and only if the potential $`u(x,t)`$ has the form
$$u(x,t)=\left(\alpha (t)x\right)^2+\beta (t)x+\gamma (t),$$
and the chosen solution $`\phi (x,t)`$ is given in the form $`(19)`$, where the equalities
$$\begin{array}{cc}\hfill i\dot{C}& =2(\alpha ^2C^2),\hfill \\ \hfill i\dot{A}& =\beta 2AC,\hfill \\ \hfill i\dot{B}& =\gamma (A^2+C)\hfill \end{array}1$$
are satisfied. The dynamics of $`x_0`$ from $`(20)`$ is as follows:
$$\dot{x}_0=\frac{2\alpha ^2x_0\beta }{C}\mathrm{\hspace{0.17em}.2}$$
The proof is obtained by simple calculation.
We consider two cases.
Case 1. Let $`w+iv=C0`$. Then either $`\alpha ^20`$ or $`\alpha ^20`$. In the case $`\alpha 0`$ we have a ‘moving oscillator of variable form’.
###### Lemma 4
Solutions of the form $`(19)`$, where $`w(t_0)<0`$, are in $`_2()`$ with respect to the variable $`x`$ for any $`tt_0`$ if the condition
$$Im\alpha ^2(t)0$$
is satisfied.
Proof The domain $`ReC<0`$ is invariant for the system (21), since for $`C=w+iv`$ we have $`\dot{w}0`$ for $`w=0`$ by virtue of the equation
$$\dot{w}=2(Im\alpha ^22vw).$$
We have proved Lemma 4. $`\mathrm{}`$
Let $`\alpha ^2=\text{const}>0`$. Then the condition $`C=\alpha `$, $`\alpha `$, gives a stationary solution for $`C`$ in the domain $`u=ReC<0`$. All trajectories in this domain are periodic and the equation is easily integrated. The trajectories are given by the equation $`H(w,v)=\text{const}`$, where $`H=((\alpha w)^2+v^2)/w`$.
Remark $`1`$ A set of operators of the form $`Q_a=_x+ax`$ satisfies the commutation relations
$$[Q_a,Q_b]=(ab)1,Q^+=Q_{\overline{a}}3$$
for all complex $`a`$, $`b`$. The commutators of all these operators with $`H_\alpha =Q_\alpha Q_\alpha ^+`$, where $`Q_\alpha ^+=Q_{\overline{\alpha }}`$, have the form
$$[Q_a,H_\alpha ]=(a\alpha )Q_{\overline{\alpha }}+(a+\overline{\alpha })Q_\alpha .4$$
The usual coherent states are eigenvectors of the annihilation operators. The states we have studied are eigenvectors of the operators
$$Q_C\psi =\gamma \psi ,ReC<0.$$
Here $`\gamma `$, and in formula (20) above $`\gamma =Cx_0`$. Hence, the states (20) are determined purely algebraically by means of the Lie algebra (23), which contains the oscillator Hamiltonian $`H=H_\alpha `$.
Remark $`2`$ The well-known ‘coherent states’ of the oscillator are eigenvectors of the annihilation operator
$`Q\psi _\gamma =\gamma \psi _\gamma ,\gamma ,`$
$`Q=_x+\alpha x,\psi _\gamma (x)=e^{\alpha (x\gamma /\alpha )^2/2}.`$
Using these functions as the initial conditions, we obtain solutions of the Schrödinger equation in the particular form (20), where $`C=\alpha =\text{const}<0`$.For $`x_0(t)`$ we obtain motion along the imaginary straight line:
$`x_0(t)=x_0(0)+2it,`$
$`\psi _\gamma (x,t)=e^{\alpha (x\gamma /\alpha x_0(t))^2/2+B(t)}.`$
Question For $`t=0`$ suppose we have $`C(0)=q`$ for all solutions of (20). For which $`(x_{0,q},q)`$ is the set of functions $`\phi _q(x,0)`$, where $`\phi _q(x,0)=e^{q(xx_{0,q})^2/2},`$ complete in $`_2()`$, where $`q`$ runs along the curve from the point $`q_0=\alpha `$ to $`q=\mathrm{}`$ in such a way that $`Req<0`$? How can we choose a minimal complete basis if this set is overfull? Naturally we can choose a curve $`\mathrm{}<q\alpha `$ and translations $`x_{0,q}`$ such that the set will be complete, $`q`$.
Remark $`3`$ We also note that for the oscillator $`u=\alpha ^2x^2`$, $`\alpha ^2=\text{const}>0`$, we can perform the Darboux transformation, starting from the second solution in $`_2()`$:
$$\phi (x,t)|_{t=0}=P_n(x)e^{\alpha x^2/2},\alpha >0,$$
where $`P_n(x)`$ is an arbitrary polynomial in $`x`$ of degree $`n`$ (here $`\alpha =C=\text{const}`$). For the potential $`\stackrel{~}{u}(x,t)`$ we have
$$\stackrel{~}{u}=\alpha ^2x^22(\mathrm{log}\phi )_{xx}=\alpha ^2x^2\underset{j}{}\frac{2}{(xx_j(t))^2}.$$
We determine the dynamics of $`\phi (x,t)`$ by expanding $`P_n(x)`$ in terms of the Hermite polynomials, that is, in terms of eigenfunctions of the oscillator. There arises a motion of the poles of the potential $`\stackrel{~}{u}(x,t)`$ with respect to $`t`$ (or the motion with respect to $`t`$ of the polynomial $`P_n(x,t)`$, $`P_n(x,0)=P_n(x)`$) such that
$`P_n(x,t)={\displaystyle e^{i\lambda _jt}a_jH_j(x)}`$
$`P_n(x)={\displaystyle \underset{jn}{}}a_jH_j,\lambda _j=\alpha +2j\alpha .`$
$`H_j(x)`$ are the Hermite polynomials defined by the formulae
$$H_0=1,H_j(x)=\left((Q^+)^je^{\alpha x^2}\right)/e^{\alpha x^2},$$
where $`Q^+=_x\alpha x`$. We have
$$P_n(x,t)=a_ne^{i(2n+1)\alpha t}\underset{j=1}{\overset{n}{}}\left(xx_j(t)\right).$$
We can obtain a more extensive family of solutions of this type from initial conditions of the form
$$\phi (x,0)=P_n(x)e^{C(0)(xx_0)^2/2},$$
applying the operators $`Q_C^+=Q_{\overline{C}}`$ to the initial conditions for solutions of (20). The dynamics of their poles can also be of interest.
Case 2. Let $`C0`$. Then $`\alpha 0`$ and we have
$$u(x,t)=\beta x+\gamma ,\phi (x,t)=\mathrm{exp}\left(iA(t)(xx_0(t))\right).$$
This is a physically reasonable case when $`\beta =E(t)`$ and is periodic in $`t`$. In this case (known as the integrable case of L. V. Keldysh) we can also find a general solution.
We note with interest that even in the stationary case $`u(x)=\beta x+\gamma `$ when $`\beta ,\gamma `$ are constants, it is more convenient to use the basis of non-stationary solutions $`\phi (x,t)`$, which have the form of plane waves for fixed time.
Ultimately only the non-stationary Schrödinger equation is a law of nature. If the force $`u_x`$ does not depend on time, then we can consider the Hamiltonian $`\widehat{H}`$ in an appropriate gauge as stationary. As usual, in this case in place of the non-stationary equation
$$i\psi _t=\widehat{H}\psi $$
we can use the Fourier method and solve the stationary problem
$$\widehat{H}\psi =\lambda \psi $$
in the space $`_2()`$.
After this, we shall have the general solution in the form of finite or continuous linear combinations
$$\psi (x,t)=\underset{j}{}a_j\psi _je^{i\lambda _jt},\widehat{H}\psi _j=\lambda _j\psi _j.$$
However, this generally accepted approach makes sense only if the spectral problem is sufficiently well solved: it is only a mathematical trick, nothing more.
If the stationary problem is not sufficiently well solved, then the Fourier method is not appropriate. An example of such a situation is the constant electric field $`u(x)=Ex`$, where the basis of non-stationary solutions (plane waves) is much simpler, as we mentioned above, than the eigenfunctions of the stationary operator $`\widehat{H}`$ (Airey functions).
The advantage of this approach can be particularly important in the case when the potential has the form
$$u(x)=Ex+u_0(x),$$
where the function $`u_0(x)`$ is periodic (a constant electric field is applied to the crystal). Already in the perspectives of the non-stationary approach in this case were discussed. A physically reasonable general non-stationary case is
$$u(x,t)=E(x,t),$$
where $`E(x,t)`$ is the electric field, periodic in both variables. If the potential $`u(x,t)`$ is doubly periodic (‘zero electric current’), then we have an extensive family of algebro-geometric exactly soluble equations, where $`u(x,t)`$ is expressed in terms of theta-functions of Riemann surfaces (see ). We construct the Bloch solution $`\psi `$ of (10):
$$\begin{array}{cc}\hfill \psi (x+T_1,t)& =e^{ip_1T_1}\psi (x,t),\hfill \\ \hfill \psi (x,t+T_0)& =e^{ip_0T_0}\psi (x,t),\hfill \end{array}5$$
as the ‘Baker–Akhiezer function’ on some Riemann surface $`\mathrm{\Gamma }`$ of finite genus, that is, the parameters $`p_1`$$`p_0`$ are connected by the equation of the algebraic curve $`\mathrm{\Gamma }`$. These potentials were used for solutions of the ‘KP equation’, where $`t`$ is renamed as $`y`$, and time, denoted by $`t`$, is the third variable. Undoubtedly these solutions of Krichever are obtained from the cyclic chains of Darboux transformations, but a rigorous assertion has not been proved up to now.
More complex is the case of a non-zero ‘electric flux’. The case $`E(t)=\beta `$ is well known as the integrable case of Keldysh.
In these cases, to understand the problem it would be useful to broaden the number of known exact solutions. In under certain conditions (‘quantization of the electric flux’ using an elementary cell in the $`(x,t)`$ plane) Novikov introduced electric Bloch functions, which are eigenfunctions for ‘electric translations’ by periods, but their analytical properties for $`E0`$ have not been understood and formulated up to now.
We transform the Schrödinger equation (10) using the gauge transformation
$$\psi ^{}=e^{iF(x,t)}\psi ,L^{}=e^{iF}Le^{iF},6$$
so that
$$i\psi _t^{}=\left(_x+iA_1(x,t)\right)^2\psi ^{}.$$
For this we need to take $`F_t=u(x,t)`$, $`A_1=F_x`$.
Let $`u=\beta (t)x`$. We have $`F=A(t)`$, $`\dot{A}=\beta (t)`$. We have a solution of form
$$\psi _k^{}=e^{ikx}\phi _k(t),7$$
where
$$i\dot{\phi }_k=\left(kA(t)\right)^2\phi _k,\phi _k=e^{i^t(kA(\tau ))^2}d\tau .8$$
For
$$\beta (t)=E=\text{const}9$$
we have
$$\psi _k^{}=e^{ikx}e^{iE^2/3}(kE^1t)^3.0$$
This is a basis in the space of solutions.
The operators of ‘electric translations’, which preserve the Schrödinger equation,
$$i_t\psi =_x^2\psi ,1$$
where $`_t=_t+iA_0`$, $`_x=_x+iA_1`$, are fully analogous to magnetic translations. For a periodic force $`E(x,t)`$,
$$E(x,t)=A_{0x}A_{1t}=E(x+T_1,t)=E(x,t+T_0),$$
the electric translations that commute with (31) are given by
$$\widehat{T}_1\psi =e^{iF_0(x,t)}\psi (x+T_1,t),\widehat{T}_0\psi =e^{iF_1(x,t)}\psi (x,t+T_0),2$$
where $`dF_0=A(t+T_0)A(t)`$, $`dF_1=A(x+T_1)A(x)`$, $`A=A_0dt+A_1dx`$. We have the identity
$$\widehat{T}_1\widehat{T}_0\widehat{T}_0\widehat{T}_1=e^{i\mathrm{\Phi }_{01}},\mathrm{\Phi }_{01}=_0^{T_0}_0^{T_1}E(x,t)𝑑x𝑑t.3$$
If $`\mathrm{\Phi }_{01}=2\pi q`$, $`q`$, then the group of ‘electric translations’ is commutative (as in the magnetic case). In the particular case (29) we obtain
$$\widehat{T}_1\psi _k^{}=T_1\psi _k^{}=e^{ikT_1}\psi _k^{},\widehat{T}_0\psi _k^{}=\psi _{kET_0}^{}.4$$
The electric Bloch states, which are eigenstates for $`\widehat{T}_1,\widehat{T}_0`$ under the condition $`ET_0T_1=2\pi q`$, $`q`$, are formally written in the form
$$\psi (x,t,p_0,p_1)=\underset{m}{}e^{imp_0T_0}\widehat{T}_0^m\left(\psi _k^{}(x,t)\right).5$$
Obviously we have
$`\widehat{T}_0\psi =e^{ip_0T_0}\psi ,`$
$`\widehat{T}_1\psi =e^{ip_1T_1}\psi ,p_1=k(modET_0).`$
These are complicated generalized functions. For $`\beta (t)=E`$ we have, for example,
$$\psi (x,t,p_0,p_1)=e^{ikx}\left(\underset{m}{}e^{i(mT_0(p_0+Ex)+\frac{E^2}{3}(kE^1tmT_0)^3)}\right).6$$
(As opposed to the magnetic case, where the functions $`\phi _k`$ rapidly decrease as $`|x|\mathrm{}`$, and the analogue of the series (35) converges, giving analytic functions.)
As the simplest example we show that under the ‘strong integer’ condition $`q,s,l`$
$$ET_0T_1=2\pi q,E^2/3=2\pi s,T_0=l,7$$
we obtain for $`t_0=kE^1`$, $`0kET_0`$
$$\frac{1}{2\pi }\psi (x,t_0,p_0,p_1)=e^{ikx}\underset{m}{}\delta (T_0(p_0+Ex)+2\pi m),k=p_1(modET_0).8$$
Hence, under the condition (37) the electric Bloch states are obtained by solving the Cauchy problem with singular initial condition (38). In the survey , although the scheme and the ideology were correct, misprints and errors were made in the formulae of the text in the description of the electric Bloch states.
The function (36) is a sum of two functions $`\psi =\psi _++\psi _{},`$ where $`\psi _+`$ is the boundary value of the function which is analytic for $`Imp_0>0`$, and $`\psi _{}`$ is the boundary value of the function which is analytic for $`Imp_0<0`$. This property is also possibly true in the general case for electric Bloch states. Our series (36) converges for $`p_1=k_{}+i\kappa ^2`$, $`\kappa ^2>0`$, as we can easily see from (36), so in the variable $`k=p_1(modET_0)`$ the function (36) is the boundary value of an analytic function.
In the case of a constant field under the strong integer condition (37), starting from (38) the electric Bloch function can be completely calculated. We recall that a theta-function with zero characteristics has the form
$$\theta _\tau (z)=\underset{m}{}e^{i\tau m^2/2+mz}9$$
and is well defined for $`Im\tau >0`$.
Comparing with (36) we have finally
$$\psi (x,t,p_0,p_1)=\theta _\tau (z),\tau =2(ET_0)^2\left(\frac{k}{E}t\right)^2+i0,z=i(ET_0)\left(kEt\frac{p_0}{E}\right),0$$
that is, the limit of the theta function when $`Im\tau +0`$,
$`p_1=k(modET_0),`$
$`\widehat{T}_1\psi =e^{ip_0T_0}\psi ,\widehat{T}_2\psi =e^{ip_1T_1}\psi ,`$
$`T_0,E^2/32\pi ,ET_0T_12\pi .`$
In the remaining cases there arise generalized functions (distributions) given by infinite trigonometric sums that are cubic in $`m`$. Apparently, quantities of this type have not been studied.
Remark $`4`$ In the case $`\alpha =0`$ and $`\beta =\text{const}=E`$ we have a special family of solutions such that $`C0`$ for $`t`$. From (21) we find that $`i/(2t+ip)=C`$, $`p`$,
$$\phi (x,t)=\mathrm{exp}\left\{i\frac{(xx_0(t))^2}{4t+2ip}+D(t)\right\}.$$
We have
$$x_0(0)+x_0(t)=\beta t^2,$$
where $`p<0`$.
The motion of the centre $`x_0`$ for $`p=0`$ is the motion of a classical particle in the constant field $`E=\beta `$. In this solution we have coincidence of the classical and quantum pictures.
We have presented these elementary solutions here to demonstrate on the example $`N=1`$ how the problem on cyclic Darboux chains for the one-dimensional non-stationary Schrödinger operator should be correctly formulated.
Definition 2
We call the transformation
$$uu+\gamma (t),\psi e^{ig(t)}1$$
the gauge transformation of a one-dimensional non-stationary Schrödinger operator, where $`\gamma (t)`$ is an arbitrary function of time, and $`\dot{g}=\gamma `$.
Definition 3 The chain of Darboux transformations for a one-dimensional non-stationary Schrödinger operator is a sequence of equations
$$i\psi _t=L_j\psi =\psi _{xx}+u_j(x,t)\psi $$
such that the operator $`L_{j+1}`$ is obtained from $`L_j`$ by the Darboux transformation and then by the gauge transformation acting on the function $`\gamma _j(t)`$:
$$\begin{array}{c}u_{j+1}(x,t)=\stackrel{~}{u}_j(x,t)+\gamma _j(t),\\ \stackrel{~}{u}_i=u_i2(\mathrm{log}\phi _i)_{xx},\\ i\phi _{jt}=\phi _{jxx}+u_j(x,t)\phi .\end{array}2$$
Hence, the Darboux chain is determined by a choice of solutions $`\phi _j`$ and arbitrary functions of time $`\gamma _j(t)`$, $`j`$. In the case of a cyclic chain of period $`N`$ we have $`N`$ solutions $`\phi _j`$ and $`N`$ arbitrary functions $`\gamma _j(t)`$.
## §3. One-dimensional difference operators
We consider a difference operator $`L`$ of the second order
$$L\psi _n=c_{n1}\psi _{n1}+v_n\psi _n+c_n\psi _{n+1}.3$$
¿From the theory of solitons we know , , that for the conservation, in the discrete case, of the latent algebraic symmetries of Schrödinger operators (such as isospectral deformations of KdV type and the Bäcklund–Darboux–Euler transformation) it is necessary to introduce the covariant translations $`\mathrm{exp}(_x)`$ in place of the translations $`T=\mathrm{exp}(_x)`$, where $`\mathrm{exp}(_x)=c_nT\psi _n\psi _{n+1}c_n`$ (the ‘covariant translation’).
After this, the theory of the corresponding class of difference operators (43) is similar to the continuous case: there arise isospectral deformations of the type of the Toda chain for (43) or of the ‘discrete KdV’ for (43) under the condition $`v_n=0`$. There also arise analogues of Darboux transformations starting from factorization:
* $`\alpha +L=QQ^+`$, $`T^+=T^1`$, $`Q=a_n+b_nT`$, $`Q^+=a_n+T^1b_n`$;
* $`\alpha +L=\widehat{Q}^+\widehat{Q}`$, $`\widehat{Q}=\widehat{a}_n+\widehat{b}_nT`$.
For the factorization, as before, we have to solve a difference analogue of the Riccati equation
$$v_n+\alpha =a_n^2+b_n^2,a_{n+1}b_n=c_n,4$$
$$v_n+\alpha =\widehat{a}_n^2+\widehat{b}_{n1}^2,c_n=\widehat{a}_n\widehat{b}_n.5$$
Factorization is also possible in the general non-self-adjoint case (direct and inverse):
$$L=p_nT^1+q_n+r_nT=(a_n+b_nT)(x_n+y_nT^1)=(c_n+d_nT^1)(v_n+w_nT).$$
The Bäcklund–Darboux transformations are defined as in the continuous case ():
$$B_\alpha =BT_\alpha ,B_\alpha ^{}=B^{}T_\alpha ,6$$
$$\begin{array}{cccc}\hfill B& L\stackrel{~}{L}=Q^+Q,\hfill & & \psi \stackrel{~}{\psi }=Q^+\psi ,\hfill \\ \hfill B^{}& L\widehat{L}^{}=\widehat{Q}\widehat{Q}^+,\hfill & & \psi \widehat{\psi }^{}=\widehat{Q}\psi ,\hfill \end{array}7$$
$$T_\alpha LL+\alpha ,8$$
$`L=QQ^+`$ (in all cases we first perform factorization and then permute the non-commutative factors, that is, the first-order operators).
There naturally arises the problem on cyclic chains (), for example, on the direct transformations
$$B_{\alpha _N}B_{\alpha _{N1}}\mathrm{}B_{\alpha _0}(L)=L.9$$
It was established in that if $`\alpha _j=0`$, then $`L`$ is a finite-zone operator with Riemann surface (‘spectrum’) of genus $`g[N/2]`$. If the $`\alpha _j=\alpha 0`$, then we arrive at the difference analogues of the theory .
We should take into account that the transformations $`B_\alpha `$ and $`B_\alpha ^{}`$ depend also on a continuous parameter, that is, on the choice of a solution of the Riccati difference equation (44) or (45), which depends on the choice of the initial point $`b_0`$ (or $`\widehat{b}_0`$) and the set of signs ($`sgna_n`$). We denote this joint parameter by $`\kappa `$. The problem of an appropriate choice of signs is discussed, for example, in . In any case, we have the relation
$$B^\kappa ^{}B^\kappa L=L$$
for any $`L`$, where the choice of the parameter $`\kappa ^{}`$ is naturally consistent with $`\kappa `$.
In principle we can consider general words in the group generated by the transformations $`T_\alpha ^{}B^\kappa `$, $`B^\kappa `$ for all $`(\kappa ,\alpha )`$ and formulate the cyclicity condition with respect to them. Hence, we have the following conclusion.
Conclusion. The set of Darboux transformations for a one-dimensional difference operator is larger than for a continuous one because of the two different types of factorizations (above). In the cyclicity condition was studied for ‘direct’ chains of the form (49) only.
It is interesting to look at the case $`N=1`$, where we obtain an analogue of the oscillator, starting from the Heisenberg relation:
$$QQ^+=Q^+Q+\alpha .0$$
This equation has a solution
$$Q^+=1+\sqrt{a+bn}T.1$$
The coefficients of such an operator $`Q`$ cannot be real for all $`n`$.
###### Lemma 5
We assume that the ‘quantization condition’
$$a+bn=b(n_0+n)2$$
is satisfied, where $`n_0=a/b`$, $`a,b`$, $`b>0`$.
Then the operators $`Q`$ and $`Q^+`$ are well defined in the subspace of functions $`\psi `$ such that
$$\psi _n=0,n+n_00,\psi _2().$$
This lemma is easily proved by direct verification.
As was shown in , the ground state is defined according to Dirac:
$$Q^+\psi _0=0.$$
At the same time we have
$$\psi _{0n}^2=1/(\alpha ^{k1}(k1)!),k=n+n_0>0.3$$
Hence, $`\psi _{0n}^2`$ is the Poisson distribution. Applying creation operators we obtain the eigenfunctions
$$Q^k\psi _0=P_k(n+n_0)\psi _{0n},$$
where $`P_k(n)`$ are the well-known Charlier polynomials, orthogonal with respect to the weight $`(\psi _{0n}^2)`$, that is, the Poisson distribution on $`_+`$.
Hence, in the difference case the Heisenberg relation (50) is realized (among first-order operators) on the positive half-line $`_+`$.
There is also another analogue, the ‘$`q`$-analogue’ of the oscillator, where the operators $`Q`$$`Q^+`$ depend on two parameters:
$$Q_{c,a}=1+ca^nT.4$$
###### Lemma 6
We have the relation
$$Q_{c,a}Q_{c,a}^+1=a^2(Q_{c_1,a}^+Q_{c_1,a}1),c_1=ca^2.5$$
The proof consists of direct verification.
In the following theorem was proved (parts of it were already given in ).
###### Theorem 1
$`1)`$ The transformation
$$\tau n1n$$
acts on the operators $`Q_c`$, $`Q_c^+`$ in the following way:
$$\tau Q_{c,a}=Q_{c,a^1}^+\tau .6$$
$`2)`$ The equation $`Q_c\psi _0=0`$ has a solution in $`_2()`$ under the condition $`|a|>1`$. The equation $`Q_c^+\psi _0=0`$ has a solution in $`_2()`$ under the condition $`|a|<1`$.
$`3)`$ The spectrum of the operator $`L_{c,a}=Q_{c,a}Q_{c,a}^+`$ in the interval $`0\lambda <1`$ has the form
$$\begin{array}{cccc}& \lambda _n=1a^{2n},\hfill & & n0,\text{if}|a|>1,\hfill \\ & \lambda _n=1a^{2n},\hfill & & n1,\text{if}|a|<1.\hfill \end{array}7$$
Analogously, the spectrum of the operator $`\stackrel{~}{L}_{c,a}=Q_{c,a}^+Q_{c,a}`$ in the interval $`0\lambda <1`$ is given by
$$\begin{array}{cccc}& \lambda _n=1a^{2n},\hfill & & n1,\text{if}|a|>1,\hfill \\ & \lambda _n=1a^{2n},\hfill & & n0,\text{if}|a|<1.\hfill \end{array}8$$
Problem Investigate the spectrum of the operators $`L=QQ^+`$, $`\stackrel{~}{L}=Q^+Q`$ in $`_2()`$ in the domain $`\lambda 1`$, if $`Q`$ has the form (54). According to the hypothesis , this spectrum is continuous and Lebesgue (see §6 below).
## §4. The Laplace transformation and the two-dimensional Schrödinger operator in a magnetic field
We write a general $`n`$-dimensional stationary scalar Schrödinger operator in Euclidean space in the form
$$L=\frac{1}{2}\left(_\alpha +iA_\alpha (x)\right)^2+U(x),9$$
where $`x=(x^1,\mathrm{},x^n)`$, $`_\alpha =/x^\alpha `$. In particular, the 2-form of the ‘magnetic field’ has the form
$$\begin{array}{c}H=\underset{\alpha <\beta }{}H_{\alpha \beta }dx^\alpha dx^\beta ,dH=0,\\ H_{\alpha \beta }=_\alpha A_\beta _\beta A_\alpha =H_{\alpha \beta }.\end{array}0$$
The electric force has the form
$$\stackrel{}{E}=(_1U,\mathrm{},_nU),1$$
that is, the potential $`U`$ is defined up to a constant. By the gauge transformations
$$Le^fLe^f,\psi e^f\psi 2$$
we can change the ‘vector-potential’:
$$A_\alpha A_\alpha i_\alpha f.3$$
In the physically comprehensible self-adjoint case we have a real magnetic field and the potential: $`H_{\alpha \beta }`$, $`U`$, and we choose a real vector-potential $`A_\alpha `$. Then the gauge transformations have the form (62) (in the class of stationary operators), where $`f=i\phi (x)`$, $`\phi (x)`$. The operator obtained by this transformation possesses as before the formal self-adjoint property with respect to the standard scalar product in the Hilbert space $`_2()`$:
$$\psi _1,\psi _2=_^n\psi _1\overline{\psi }_2d^nx.4$$
An important class consists of operators with smooth real magnetic and electric fields, which are periodic with respect to some lattice $`\mathrm{\Gamma }^n`$ of rank $`n`$. In this case we introduce the following notions.
Definition 4 We say that an operator $`L`$ is topologically non-trivial if the magnetic field (60) is the non-zero cohomology class of a torus,
$$0[H]H^2(𝕋^n,).5$$
Definition 5 We say that an operator $`L`$ possesses a topologically non-trivial electric field if the 1-form of the electric field (in the non-relativistic stationary formalism) is the non-zero cohomology class of a torus:
$$0[E]H^1(𝕋^n,),6$$
$`E=E_\alpha dx^\alpha `$, $`E_\alpha =_\alpha U(x)`$. In other words, the potential has the form
$$U(x)=E_{0\alpha }x^\alpha +U_0(x),7$$
where the function $`U_0(x)`$ is periodic with respect to the lattice, $`E_0=\text{const}(^n0)`$.
A topologically non-trivial electric field has arisen already in the one-dimensional case. As we showed in §2, it was appropriate to study such a situation in the framework of the non-stationary $`(x,t)`$-formalism, even if the electric force is stationary.
For magnetic fields we use the stationary formalism. If $`\gamma _1,\mathrm{},\gamma _n\mathrm{\Gamma }`$ is the basis of the lattice, then $`H_{\alpha \beta }(x+\stackrel{}{\gamma }_j)=H_{\alpha \beta }(x)`$. For the vector-potential we have
$$A_\alpha (x+\stackrel{}{\gamma }_j)=A_\alpha (x)+_\alpha \phi _j(x),$$
where $`\phi _j`$ is some function of $`x=(x^1,\mathrm{},x^n)`$.
We define ‘magnetic translations’ (), which have been well known in the physical literature since the mid-60s:
$$\widehat{T}_j\psi (x)=\psi (x+\stackrel{}{\gamma }_j)e^{i\phi _j(x)}.8$$
We have the relation
$$\widehat{T}_j\widehat{T}_k=e^{i\mathrm{\Phi }_{jk}}\widehat{T}_k\widehat{T}_j,9$$
where $`\mathrm{\Phi }_{jk}`$ is the flux of a magnetic field through an elementary two-dimensional cell of the lattice in the plane $`(\stackrel{}{\gamma }_j,\stackrel{}{\gamma }_k)`$, or the scalar product (integral) of the cocycle $`[H]`$ with the basis 2-cycle $`Z_{jk}H_2(𝕋^n,)`$.
In the two-dimensional case $`n=2`$ we can present the operator $`L`$ in a factorized form, which we write as
$$2L=(\overline{}+B)(+A)+2W,0$$
where $`A(z,\overline{z})`$, $`B(z,\overline{z})`$, $`W(z,\overline{z})`$ are some functions of $`z`$$`\overline{z}`$,
$$z=x+iy,=_xi_y.$$
The magnetic field has the form
$$2H=B_zA_{\overline{z}}.1$$
If $`H`$, $`W`$ are real, then we can, by using the gauge transformation, bring the operator to the self-adjoint form, so we have
$`2L=Q_1Q_2+2W,`$
$`Q_1=\overline{}+B,Q_2=+A=Q_1^+,A=\overline{B}.`$
The operator $`L`$ is always written in the form (70), that is, it ‘admits factorization’ in our terminology.
The operator $`L`$ also admits an opposite factorization of the form
$$2L=Q_2Q_1+2V,$$
where $`V=W+H`$, $`Q_2=+A`$, $`Q_1=\overline{}+B`$.
Definition 6 The Laplace transformation is the transformation
$$L\stackrel{~}{L},2\stackrel{~}{L}=WQ_2W^1Q_1+2W,2$$
where $`2L=Q_1Q_2+2W`$.
###### Lemma 7
a) If $`\psi `$ is an arbitrary solution of the equation $`L\psi =0`$, then the function $`\stackrel{~}{\psi }=Q_2\psi `$ satisfies the equation $`\stackrel{~}{L}\stackrel{~}{\psi }=0`$.
b) If $`W=W_0=\text{const}`$, then for any solution $`\psi `$ of the equation $`L\psi =\lambda \psi `$ the function $`\stackrel{~}{\psi }=Q_2\psi `$ satisfies the equation $`\stackrel{~}{L}\stackrel{~}{\psi }=\lambda \stackrel{~}{\psi }`$. In this case $`2\stackrel{~}{L}=Q_2Q_1+2W_0`$.
Proof The proof of Lemma 7 is very simple. Since $`L\psi =0`$, we have
$$2W\psi =Q_1Q_2\psi =Q_1\stackrel{~}{\psi }.$$
¿From this it follows that
$`2\psi =W^1Q_1\stackrel{~}{\psi },`$
$`2\stackrel{~}{\psi }=2Q_2\psi =Q_2W^1Q_1\stackrel{~}{\psi },`$
$`2W\stackrel{~}{\psi }=WQ_2W^1Q_1\stackrel{~}{\psi }.`$
The lemma has been proved for the case a). In the case b), for $`W=W_0=\text{const}`$, the proof is exactly the same. $`\mathrm{}`$
###### Lemma 8
The Laplace transformation for the equation $`L\psi =0`$ commutes with the gauge transformation $`(62)`$. If $`W=W_0=\text{const}`$, then this applies to the whole family of equations $`L\psi =\lambda \psi `$.
In particular, this transformation is defined on the gauge invariant quantities $`H`$$`W`$:
$$\stackrel{~}{H}=H+\frac{1}{2}\overline{}\mathrm{log}W,\stackrel{~}{W}=W+\stackrel{~}{H}.3$$
The proof of this lemma is obtained by direct verification. We omit it.
Remark $`5`$ The classicists, beginning with Laplace, considered this transformation for the hyperbolic equation
1) $`2L=(_x+A)(_y+B)+2W`$,
$$2\stackrel{~}{L}=W(_y+B)W^1(_x+A)+2W;$$
2) $`2L=(_y+B)(_x+A)+2V,V=W+H`$,
$$2\stackrel{~}{L}^{}=V(_x+A)V^1(_y+B)+2V.$$
Here the pair of mutually inverse Laplace transformations acts on solutions of the equation $`L\psi =0`$:
1) $`L\stackrel{~}{L}`$, $`\psi \stackrel{~}{\psi }=(_y+B)\psi `$;
2) $`L\stackrel{~}{L}^{}`$, $`\psi \stackrel{~}{\psi }^{}=(_x+A)\psi `$.
As before, the gauge invariants have the form $`H`$$`W`$ or $`H`$$`V`$, where
$$2H=_xB_yA.$$
For these invariants the Laplace transformations have the form analogous to (73):
$$\begin{array}{cc}\hfill 1)& \stackrel{~}{H}=H+\frac{1}{2}_1_2\mathrm{log}W,\stackrel{~}{W}=W+\stackrel{~}{H};\hfill \\ \hfill 2)& \stackrel{~}{H}^{}=H+\frac{1}{2}_1_2\mathrm{log}V,\stackrel{~}{V}^{}=V+\stackrel{~}{H}^{}.\hfill \end{array}4$$
These transformations have been studied geometrically; they have found applications in the theory of congruences of surfaces in $`^3`$ (). The classicists of geometry (Darboux, Tzitzéica, Moutard, see ) studied chains of Laplace transformations and carried out a series of useful formal calculations. They observed that the infinite chain of Laplace transformations
$$\mathrm{},L_1,L_0,L_1,L_2,\mathrm{},$$
where $`L_{j+1}=\stackrel{~}{L}_j`$, is equivalent to a non-linear evolution system, which in the modern literature is called the ‘two-dimensionalized Toda chain’. Integrability of it by the methods of soliton theory was discovered in , and from the viewpoint of field theory and Lie algebras in . In fact, a series of formal results had already been obtained at the beginning of the 20th century.
We consider the following quantities, that is, the potentials of a chain:
$$W_n=\mathrm{exp}f_n.$$
¿From (74) we obtain
$$W_{n+1}W_n=e^{f_{n+1}}e^{f_n}=H_{n+1},H_{n+1}H_n=\frac{1}{2}_x_y(f_n).$$
By means of the substitution $`f_n=g_ng_{n1}`$ we obtain the ‘two-dimensionalized Toda chain’
$$\frac{1}{2}\mathrm{}g_n=e^{g_{n+1}g_n}e^{g_ng_{n1}},5$$
where $`\mathrm{}=_x_y`$.
Darboux posed the problem of classifying the cyclic Laplace chains $`L_N=L_0`$ of period $`N`$. For $`N=2`$ he deduced, in particular, that solutions of the equation
$$\mathrm{}g=sinhg6$$
define the class of cyclic chains.
His student Tzitzéica () posed the problem: let the operator $`L_0=L_N`$ be such that $`H_0=H_N=0`$. What kind of non-linear systems can arise in this case? For $`N=3`$ he showed that such cyclic chains are generated, in particular, by solutions of the equation
$$\mathrm{}g=e^ge^{2g}.$$
However, in the hyperbolic case, which was considered by the classicists, there are nothing but formal identities.
Well posed global problems arise only in the elliptic case $`_x\overline{}`$, $`_y`$, which we also consider in relation to the spectral theory of the Schrödinger operator in electric and magnetic fields for $`n=2`$, in particular when the fields are smooth non-singular real doubly periodic functions with some lattice of periods $`\mathrm{\Gamma }`$ on the plane $`^2`$.
Example 2 Let $`N=1`$. We have
$`L_1=L_0,H_1=H_0+{\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0,`$
$`e^{f_1}=e^{f_0}+H_1.`$
Since $`H_1=H_0`$, we obtain $`\mathrm{\Delta }f_0=0`$. If $`f_0`$ is a non-singular doubly periodic function (that is, $`W_00`$), then $`f_0=\text{const}`$.
Since $`f_1=f_0`$, we have $`H_1=0=H_0`$. Thus, we obtain only a free operator in the zero magnetic field, but this is a consequence of the global hypothesis.
Example 3 Let $`N=2`$. We have
$$f_2=f_0,H_2=H_0.$$
Since $`H_1=H_0+\frac{1}{2}\mathrm{\Delta }f_0`$, $`H_2=H_0=H_1+\frac{1}{2}\mathrm{\Delta }f_2`$, we obtain
$$\frac{1}{2}\mathrm{\Delta }(f_0+f_1)=0,$$
or
$$f_0=f_1+a.$$
Further, since
$`e^{f_2}=e^{f_0}=e^{f_1}+H_0,`$
$`e^{f_1}=e^{f_0}+H_1=e^{f_0}+H_0+{\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0,`$
we obtain
$`2(e^{f_1}e^{f_0})={\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0,`$
$`2(e^{f_0+a}e^{f_0})={\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0.`$
Let $`g=f_0a/2`$. We obtain
$`2(e^{g+a/2}+e^{g+a/2})={\displaystyle \frac{1}{2}}\mathrm{\Delta }g,`$
$`e^{a/2}sinhg={\displaystyle \frac{1}{8}}\mathrm{\Delta }g.`$
This shows that in the non-singular doubly periodic case all the cyclic Laplace chains of length $`N=2`$ are described by solutions of the equation
$$\gamma e^{a/2}sinhg=\mathrm{\Delta }g.$$
By stretching the axes $`(x,y)`$ we reduce it to the sinh-Gordon equation
$$\mathrm{\Delta }g=sinhg.7$$
Example 4 Let $`N=3`$, $`H_0=0=H_N`$. Since
$$H_0=H_3=H_2+\frac{1}{2}\mathrm{\Delta }f_2=H_0+\frac{1}{2}\mathrm{\Delta }(f_0+f_1+f_2),$$
we obtain for non-singular doubly periodic functions $`f_j`$:
$$f_0+f_1+f_2=C.$$
Since
$$e^{f_0}=e^{f_3}=e^{f_2}+H_3=e^{f_2}+H_0+\frac{1}{2}\mathrm{\Delta }(f_0+f_1+f_2)=e^{f_2}+H_0,$$
we come to the conclusion: if $`H_0=0`$, then we have
$$e^{f_3}=e^{f_2}=e^{f_0}.$$
Hence it follows that
$`2f_0+f_1=C,f_1=C_2f_0,`$
$`e^{f_2}=e^{f_1}+H_0+{\displaystyle \frac{1}{2}}\mathrm{\Delta }(f_0+f_1)=e^{f_0}=e^{f_1}+{\displaystyle \frac{1}{2}}\mathrm{\Delta }(f_0+f_1).`$
Finally, we have
$`e^{f_0}e^{f_1}={\displaystyle \frac{1}{2}}\mathrm{\Delta }(f_0+f_1),`$
$`e^{f_0}e^{C2f_0}={\displaystyle \frac{1}{2}}\mathrm{\Delta }(f_0+C2f_0)={\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0.`$
Let $`f_0=g+a`$. We obtain
$$e^ae^ge^{C2g2a}=\frac{1}{2}\mathrm{\Delta }g.$$
If $`a=C2a`$, that is, $`a=C/3`$, then we finally have
$$e^{C/3}(e^ge^{2g})=\frac{1}{2}\mathrm{\Delta }g,$$
or, after stretching the scales,
$$\mathrm{\Delta }g=e^{2g}e^g.$$
The following result was proved in .
###### Theorem 2
Suppose we are given a cyclic chain of Laplace transformations
$$\mathrm{},L_0,L_1,L_2,\mathrm{},L_N=L_0$$
such that the operators of the chain have smooth doubly periodic real magnetic field and potential. Then all the operators of the chain are topologically trivial and algebro-geometric with respect to the zero level: $`L_j\psi =0`$. The Bloch functions of zero level with all complex quasimomenta are defined and sweep out a Riemann surface of finite genus; the coefficients of the operators $`L_j`$ are expressed in terms of theta-functions of this surface (the ‘Fermi-curve’), which is common for all operators of the chain.
We recall that two-dimensional algebro-geometric operators were introduced and studied in ; for a review of their theory see .
The idea of the proof of this theorem can be seen for $`N=2`$, where the requirement of cyclicity of the chain under the conditions of the theorem reduces to the equation $`\mathrm{\Delta }g=sinhg`$.
In studying surfaces with the topology of a torus in $`^3`$ with constant mean curvature it was established that all doubly periodic non-singular solutions of this equation are finite-zone (algebro-geometric); see the survey . Finally, this follows from the fact that any manifold of solutions of a non-linear elliptic equation on a torus (compact manifold) has finite dimension at a non-singular point. Further arguments, which lead to the finite-zone property of solutions of the equation $`\mathrm{\Delta }g=sinhg`$, follow the standard scheme that was already developed in 1974 for the KdV equation (see the survey ); commutative fluxes are linearly dependent on a finite-dimensional invariant submanifold.
A natural generalization of this idea for all $`N2`$ also leads to the proof of the theorem, as was noted by the authors of .
Definition 7 A semicyclic Laplace chain is a set of Schrödinger operators
$$L_0,\mathrm{},L_N,L_{j+1}=\stackrel{~}{L}_j$$
such that $`L_N=L_0+C`$.
Definition 8 A quasicyclic Laplace chain is a set of Schrödinger operators
$$L_0,\mathrm{},L_N,L_{j+1}=\stackrel{~}{L}_j,$$
such that the extreme operators are completely factorized with a constant potential:
$`2L_0=Q_{01}Q_{02}+2V_0,2L_N=Q_{N1}Q_{N2}+2V_N,`$
$`V_0=\text{const},V_N=\text{const},`$
$`Q_{j2}=\overline{}+B_j(z,\overline{z}),Q_{j1}=+A_j(z,\overline{z}).`$
¿From the definition we have
$`2L_j=Q_{j2}Q_{j1}+2W_j,W_0=H_0,W_N=H_N,`$
$`2\stackrel{~}{L}_j=W_jQ_{j1}^+W_j^1Q_{j2}+2W_j.`$
Example 5 $`N=1`$. A semicyclic chain of length 1. We have
$`2L_1=2L_02C,H_1=H_0,`$
$`e^{f_1}=e^{f_0}+C=e^{f_0}+H_1,H_1=C.`$
Hence we obtain
$$\mathrm{\Delta }f_0=0.$$
In the non-singular doubly periodic case it follows that
$$f_0=\text{const},H=C.$$
Hence we arrive at a Landau operator in a homogeneous constant magnetic field $`H=C`$ and zero potential. Starting from the solutions (let $`C>0`$)
$$Q_{02}\psi _0=0,L_0\psi _0=0,V_0=0,$$
we obtain all the Landau levels
$`(Q_{01})\psi _0=\psi _1,`$
$`2L_1\psi _1=02L_0\psi _1=2C\psi _1,C=H.`$
Hence, from the lowest level $`Q_0\psi _0=0`$ we obtain the first level $`\psi _1=Q_0^+\psi _0`$. These spaces of solutions are infinite-dimensional on every level, called ‘Landau levels’ for the Landau operator
$$H+2L_0=Q_{02}Q_{01}+H,Q_{02}\psi _0=0.8$$
Its levels are
$$\lambda _n=H+2nH,\psi _n=(Q_{01})^n\psi _0.9$$
Example 6 $`N=2`$. A semicyclic chain of length 2. Here we have $`H_2=H_0`$. In the non-singular doubly periodic case, from the equality
$$H_2=H_0+\frac{1}{2}\mathrm{\Delta }(f_0+f_1)$$
it follows that
$$\mathrm{\Delta }(f_0+f_1)=0f_0+f_1=a.$$
Since
$`e^{f_2}=e^{f_0}+C=e^{f_1}+H_2=e^{f_1}+H_0,`$
$`e^{f_1}=e^{f_0}+H_0+{\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0,`$
we obtain
$`e^{f_0}e^{f_1}=H_0C,`$
$`e^{f_1}+e^{f_0}=H_0{\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0.`$
Finally, we have
$`2(e^{f_0}e^{f_1})=C{\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0,`$
$`f_1=af_0.`$
Thus we arrive at the equation
$$e^{f_0}e^{af_0}+\frac{C}{2}=\frac{1}{4}\mathrm{\Delta }f_0.$$
Let $`f_0=g+a/2`$. We have the equality
$`e^{g+a/2}e^{a/2g}+{\displaystyle \frac{C}{2}}={\displaystyle \frac{1}{4}}\mathrm{\Delta }g,`$
$`2C2e^{a/2}sinhg=\mathrm{\Delta }g.`$
This equation has an extensive family of doubly periodic non-singular real solutions.
Solutions of the equation $`L_N\psi _N=(L_0+C)\psi _N=0`$ are obtained from solutions of the equation $`L_0\psi _0=0`$ by the chain of algebraic ‘creation operators’
$$\psi _N=Q_{N1,2}\mathrm{}Q_{0,2}(\psi _0),L_0\psi _N=(C)\psi _N.$$
Hence, the levels of spectrum $`\epsilon =0`$ and $`\epsilon =C`$ for the operator $`L_0`$ are connected by these operators and are obtained from each other. It is easy to see that magnetic Bloch solutions are transformed into magnetic Bloch ones under the action of these operators if the magnetic flux is quantized (that is, the Bloch solutions are defined).
###### Lemma 9
If $`C0`$, then the magnetic flux of a non-singular semicyclic chain with potentials that do not vanish is different from zero.
Remark $`6`$ Since $`H_{j+1}=H_j+\frac{1}{2}\mathrm{\Delta }(f_j)`$ and $`f_j`$ is non-singular, we see that the fluxes of the operators $`L_j`$ coincide: $`[H_{j+1}]=[H_j]`$.
Proof of the lemma Since
$$e^{f_N}=e^{f_{N1}}+H_0+\frac{1}{2}\mathrm{\Delta }\left(\underset{j=0}{\overset{N1}{}}f_j\right)=e^{f_0}+C,$$
and $`[e^{f_N}]=[e^{f_{N1}}]+[H]`$, we obtain
$$[e^{f_N}]=[e^{f_0}]+[C]=[e^{f_0}]+KC=[e^{f_0}]+N[H].$$
Hence, for the flux $`N[H]=N[H_0]=KC`$ it follows that
$$C=\frac{N[H]}{K},0$$
where $`K`$ is the area of an elementary cell. We have proved Lemma 9. $`\mathrm{}`$
(For purely cyclic non-singular chains we have $`C=0`$, that is, the flux is always equal to zero.)
The greatest interest is in quasicyclic chains, with which the main results of are connected.
###### Theorem 3
Let the quasicyclic Laplace chain of the Schrödinger operators
$`L_0,\mathrm{},L_N,L_{j+1}=\stackrel{~}{L}_j,`$
$`2L_0=Q_{0,1}^+Q_{0,2},2L_N=Q_{N,1}^+Q_{N,2}+2C_N`$
be given with doubly periodic real quantities $`(f_j,H_j)`$, and suppose that the magnetic flux $`[H_j]`$ is positive, $`[H]>0`$. Then the operator $`L_N`$ has two strongly degenerate discrete levels, which are isomorphic to the Landau levels and one of them is the lowest:
* $`L_N\psi =C_N\psi `$, $`Q_N\psi =0`$, $`\lambda =C_N`$,
* $`L_N\psi =0`$, $`\lambda =0`$.
In addition, $`C_N=N[H]/K`$, where $`K`$ is the area of an elementary cell.
Here $`Q_{j2}=\overline{}+B_j(z,\overline{z})`$, $`Q_{j1}=+A_j(z,\overline{z})`$.
Proof First, solving the equation
$$Q_{N,2}\psi =0$$
according to the scheme of , we obtain the eigenvalue $`L_N\psi =C_N\psi `$. If $`[H]>0`$, such a solution exists and gives a huge space isomorphic to the Landau level according to , . A basis in this space, mentioned in , , is the magnetic Bloch basis, which is suitable only in the case of the integer-valued flux $`[H]=2\pi m>0`$; however, the result itself is true for all positive values of the flux.
We obtain the second level as follows. We solve the equation $`Q_{02}\psi =0`$ according to and obtain also for $`[H]>0`$ a huge space of functions $`\psi `$ isomorphic to the Landau level, which can be defined by using elliptic formulae for the integer-valued flux $`[H]`$, common for $`H_0`$ and $`H_N`$,
$$[H]=[H_0]=[H_N].$$
Next we factorize the operator $`L_0`$ in the form
$$2L_0=Q_{0,2}Q_{0,1}+2H_0=Q_{0,1}^+Q_{0,2}.$$
Then we apply the creation operator $`N`$ times to a solution of $`Q_0\psi =0`$:
$$\widehat{\psi }=Q_{N1,1}\mathrm{}Q_{0,1}\psi .$$
The new function $`\widehat{\psi }`$ satisfies the equation
$$L_N\widehat{\psi }=0=(Q_{N,1}Q_{N,2}+C_N)\widehat{\psi }.$$
The function $`\widehat{\psi }`$ belongs to $`_2()`$ if $`\psi `$ belongs to $`_2()`$.
Now we find the number $`C_N`$. We have for the Laplace chain the following boundary conditions:
$$e^{f_0}=H_0,e^{f_N}=H_N+C_N.$$
For the mean values we obtain
$$[e^{f_{j+1}}]=[e^{f_j}]+[H_{j+1}]=[e^{f_j}]+[H]=[e^{f_0}]+(j+1)[H]=(j+2)[H].$$
In particular, for $`j+1=N`$ we obtain
$$e^{f_N}=[H]+[C_N]=(N+1)[H],$$
where $`[C_N]=KC_N`$ or $`C_N=N[H]/K`$. $`\mathrm{}`$
Remark $`7`$ Let the operator $`L`$ be given in the self-adjoint form,
$`Q=Q_1=+A,Q_2=Q^+=\overline{}+B,`$
$`\overline{B}=A.`$
It is also convenient to choose the ‘Lorentz gauge’, for which all the formulae are written in :
$$Im(A_{\overline{z}})=0.1$$
Then we have
$$2L=Q^+Q+2W=QQ^++2V.$$
Further, if $`L\psi =0`$, we carry out the Laplace transformation together with the gauge transformation
$`Le^g\stackrel{~}{L}e^g,W=e^f,`$
$`\stackrel{~}{L}=WQW^1Q^++2W,`$
$`\stackrel{~}{\psi }e^gQ\psi =e^g\stackrel{~}{\psi }.`$
Choosing $`g=f/2`$, we obtain the operator $`\stackrel{~}{\stackrel{~}{L}}=e^{f/2}\stackrel{~}{L}e^{f/2}`$ in the form
$$\stackrel{~}{\stackrel{~}{L}}=e^{f/2}\stackrel{~}{L}e^{f/2}=\stackrel{~}{Q}\stackrel{~}{Q}^++2W,$$
where $`\stackrel{~}{Q}=e^{f/2}Qe^{f/2}`$, $`\stackrel{~}{Q}^+=e^{f/2}Q^+e^{f/2}`$. We obtain explicitly the self-adjoint operator in the Lorentz gauge transformation.
Example 7 Let $`N=1`$. From the definition for $`L_0`$ we have
$$2L_0=Q_0^+Q_0=Q_0Q_0^++2H_0,$$
where
$$Q_0=+A,Q_0^+\overline{}+B,H_0=W_0=e^{f_0}.$$
Therefore
$$2L_1=(e^{f_0/2}Q_0^+e^{f_0/2})(e^{f_0/2}Q_0e^{f_0/2})+2H_0=\stackrel{~}{\stackrel{~}{L}_0}=e^{f_0/2}\stackrel{~}{L}_0e^{f_0/2}.$$
¿From the quasicyclicity condition for $`N=1`$ we have
$$H_0=\text{const},L_1=L_0+2H_0.$$
We arrive at the Landau operator.
Example 8 Let $`N=2`$.
$`2L_0=Q_0^+Q_0=Q_0Q_0^++2H_0,H_0=e^{f_0},`$
$`2L_2=Q_2^+Q_2+2C_2=Q_2Q_2^++2C_2+2H_2,`$
$`e^{f_2}=C_2+H_2=e^{f_1}+H_2.`$
We see that $`e^{f_1}=\text{const}`$. Further,
$$C_2=e^{f_1}=e^{f_0}+H_1=e^{f_0}+H_0+\frac{1}{2}\mathrm{\Delta }f_0,$$
where $`H_0=e^{f_0}`$. Hence we have
$`C_22e^{f_0}={\displaystyle \frac{1}{2}}\mathrm{\Delta }f_0,`$
$`2C_2=4e^{f_0}=\mathrm{\Delta }f_0.`$
Putting $`f_0=g+a`$, where $`e^a=C_2/2`$, we obtain $`\mathrm{\Delta }g=e^a(1e^g)`$. By stretching the axes we bring this equation to the form
$$\mathrm{\Delta }g=1e^g.2$$
The last equation has an extensive family of smooth real doubly periodic solutions. Even its solutions, dependent on one variable (for example, on $`x`$), which can be found by quadrature, give us a continuous family belonging to this class, with an arbitrary period in $`y`$:
$`f_0^{\prime \prime }=2C_24e^{f_0},C_2>0,`$
$`\tau ^{\prime \prime }={\displaystyle \frac{}{\tau }}(U(\tau )),`$
$`U(\tau )=A\tau +4e^\tau ,\tau =f_0,A=2C_2.`$
The dependence of $`f_0(x)`$ is determined by quadrature:
$`xx_0={\displaystyle \frac{d\tau }{\sqrt{4e^\tau +A\tau +E}}},`$
$`EE_{\mathrm{min}}=A\left(1\mathrm{log}{\displaystyle \frac{A}{4}}\right).`$
Equation (82) has occurred in various physical problems—in plasma theory, superconductivity theory, and so on. Hector de Vega informed one of the authors (S. P. Novikov) that equation (82) arose as the Bogomol’nyi reduction for the Ginzburg–Landau equation for the critical value of the parameter which separates superconductors of types I and II (see ). However, the application of this form of the Ginzburg–Landau equation at a critical point is not clear. In addition, the magnetic field in this model differs from ours by a non-zero constant, and in the theory of superconductors solutions with singularities are necessary.
Hypothesis For periods $`N5`$ the quasiperiodicity condition does not give smooth real doubly periodic solutions, apart from a constant. In any case, the constant (the Landau operator) does not allow any variation—it is an isolated solution.
According to the hypothesis of Novikov, the resulting class of operators is ‘maximally exactly soluble’ among two-dimensional Schrödinger operators with smooth real doubly periodic electric potential and magnetic field with a non-zero (let it be integer-valued) flux $`[H]=2\pi m`$.
These operators have two discrete levels with infinite degeneration that are isomorphic to the Landau levels. One of those levels is basic (that is, the lowest one) and the second one is certainly not the nearest to the first one (at least one ‘magnetic zone’ lies between them). If there are three strongly degenerate discrete levels, then, as we suppose, the operator will coincide with the Landau operator.
The general arrangement of the spectrum of Schrödinger operators in a magnetic field with integer-valued flux, interesting aspects of the inverse spectral problem, and geometric and topological properties of the spectrum of these operators were discussed in , .
If the magnetic flux is quantized, then we have the ‘magnetic Bloch function’ of the dispersion law with number $`m0`$, $`m`$
$$\widehat{T}_1\psi =e^{ip_1T_1}\psi ,\widehat{T}_2\psi =e^{ip_2T_2}\psi ,L\psi _m=\epsilon _m(p_1,p_2)\psi _m,3$$
$`\psi =\psi _m(x,y,p_1,p_2)`$, $`[H]=\mathrm{\Phi }2\pi `$, $`\mathrm{\Phi }0`$.
For definiteness, let the function $`\psi `$ be normalized by analogy with finite-zone theory :
$$\psi _m(x,y,p_1,p_2)|_{x=x_0,y=y_0}=1.$$
The function $`\psi _m(x,y,x_0,y_0,p_1,p_2)`$ under the fixed normalization $`(x_0,y_0)`$ is defined on the ‘phase space’
$$(x,y,p_1,p_2)^2\times 𝕋^2.4$$
Moreover, the function $`\psi `$ is defined up to normalization
$$\psi _nf(p_1,p_2)\psi _n.5$$
The manifold of zeros of the function $`\psi _m`$, $`\psi _m=0`$, splits into two parts:
1) ‘topological zeros’, where $`\psi _m(x,y)0`$ for fixed $`p_1`$$`p_2`$; this part of the zeros changes under transformations (85), but the ‘algebraic number of zeros’ is the ‘Chern class’ $`c_1^{(m)}`$ of the dispersion law with number $`m`$ for the operator $`L`$ of general position, where the dispersion laws do not intersect,
$$\epsilon _{m_1}(p_1,p_2)\epsilon _{m_2}(p_1,p_2),m_1m_2;$$
2) ‘geometric zeros’, where locally at a general point we can assume that
$$(p_1,p_2)=\gamma _m(x,y),m0.$$
As $`m\mathrm{}`$ the dispersion laws converge to constants:
$$\epsilon _m(p_1,p_2)\text{const}=(2m+1)[H],$$
analogous to the forbidden zones of the one-dimensional periodic Schrödinger operator.
Problem Find the analogue of the ‘Dubrovin equations’ for the set $`\{\gamma _m(x,y)\}`$ for all $`m`$; solve the inverse problem of reconstructing the operator $`L`$, starting from the set of geometric data, in the phase space (84), analogous to the one-dimensional data, without entering into the complex domain of quasimomenta.
## §5. Difference equations and Laplace transformations. The hyperbolic case
The simplest hyperbolic difference equation on the standard quadratic lattice $`\mathrm{\Gamma }`$ in $`^2`$ is
$$a_n\psi _n+b_n\psi _{n+T_1}+c_n\psi _{n+T_2}+d_n\psi _{n+T_1+T_2}=0,6$$
where $`n=(n_1,n_2)`$ is the number of a lattice point, $`n_j`$, $`T_1=(1,0)`$, $`T_2=(0,1)`$ are the basis vectors of the lattice. We also denote by $`T_1`$$`T_2`$ the basis operators of translations which act on functions defined on $`^2`$:
$`T_1\psi (n_1,n_2)=\psi (n_1+1,n_2),`$
$`T_2\psi (n_1,n_2)=\psi (n_1,n_2+1).`$
We can write (86) in the following way:
$$L\psi =(a_n+b_nT_1+c_nT_2+d_nT_1T_2)\psi =0.7$$
It was already observed in that for an equation of this form there is a class of algebro-geometric exactly soluble cases with effective theta-function formulae.
The operator $`L`$ is defined up to multiplication by a function: $`Lf_nL`$. We also have gauge transformations, and so together we have a system of the following equivalences of (87):
$$Lf_nLg_n,\psi _ng_n^1\psi _n,8$$
where $`f_n`$, $`g_n`$ are non-vanishing functions on the lattice.
###### Lemma 10
0 The operator $`L`$ admits a unique representation in the form
$$\text{a)}L=f_n\left((1+u_nT_1)(1+v_nT_2)+w_n\right)9$$
or in the form
$$\text{b)}L=f_n^{}\left((1+v_n^{}T_2)(1+u_n^{}T_1)+w_n^{}\right).0$$
The coefficients $`u_n`$, $`v_n`$, $`w_n`$, $`f_n`$ for the factorization a) are determined from the formulae
$$a_n=f_n(1+w_n),f_nu_n=b_n,f_nv_n=c_n,f_nu_nv_{n+T_1}=d_n.1$$
The proof is based on direct verification.
Using both factorizations a) and b), we define two Laplace transformations of (87) $`L\psi =0`$.
Definition 9 A Laplace transformation of the first type is a transformation
$$\begin{array}{cc}\hfill L& \stackrel{~}{L}=\stackrel{~}{f}_n\left(w_n(1+v_nT_2w_n^1)(1+u_nT_1)+w_n\right),\hfill \\ \hfill \psi & \stackrel{~}{\psi }=(1+v_nT_2)\psi .\hfill \end{array}2$$
A Laplace transformation of the second type is a transformation
$$\begin{array}{cc}\hfill L& \stackrel{~}{L}^{}=\stackrel{~}{f}^{}\left(w_n^{}(1+u_n^{}T_1)(w_n^{})^1(1+v_n^{}T_2)+w_n^{}\right),\hfill \\ \hfill \psi & \stackrel{~}{\psi }^{}=(1+u_n^{}T_1)\psi .\hfill \end{array}3$$
###### Lemma 11
1 The Laplace transformations of the first and second types are mutually inverse.
The proof is obtained by simple substitution, as in the continuous case.
We denote a Laplace transformation of the first type by $`\mathrm{\Lambda }_{12}^{++}`$ and of the second type by $`\mathrm{\Lambda }_{21}^{++}`$. Lemma 11 claims that $`\mathrm{\Lambda }_{12}^{++}\mathrm{\Lambda }_{21}^{++}=1`$.
It is appropriate to note that in a completely analogous way we can define the Laplace transformation of (86), which corresponds to an arbitrary pair of orthonormal basis translations
$`(T_1,T_2),(T_2,T_1)`$ $`\mathrm{\Lambda }_{12}^{++},\mathrm{\Lambda }_{21}^{++},`$
$`(T_1^1,T_2),(T_2,T_1^1)`$ $`\mathrm{\Lambda }_{12}^+,\mathrm{\Lambda }_{21}^+,`$
$`(T_1,T_2^1),(T_2^1,T_1)`$ $`\mathrm{\Lambda }_{12}^+,\mathrm{\Lambda }_{21}^+,`$
$`(T_1^1,T_2^1),(T_2^1,T_1^1)`$ $`\mathrm{\Lambda }_{12}^{},\mathrm{\Lambda }_{21}^{}.`$
Thus, we have eight Laplace transformations $`\mathrm{\Lambda }_{12}^{\epsilon \sigma }`$, $`\mathrm{\Lambda }_{21}^{\epsilon \sigma }`$, $`\epsilon ,\sigma =\pm `$, and
$$\mathrm{\Lambda }_{12}^{\epsilon \sigma }\mathrm{\Lambda }_{21}^{\sigma \epsilon }=1.$$
Hence, the Laplace transformations here form a group with four generating elements $`\mathrm{\Lambda }_{12}^{\epsilon \sigma }`$, the structure of which we do not know so far.
Now we return to the transformation $`\mathrm{\Lambda }_{12}^{++}`$, described above. The Laplace transformations are well defined on the equivalence classes (88). Therefore, as in the continuous case, they should be registered on the invariants of the equivalence classes, which are analogous to the magnetic field and electric potential.
###### Lemma 12
2 The complete set of invariants with respect to the equivalence $`(88)`$ is given by the following basis functions:
$$K_{1n}=\frac{b_nc_{n+T_1}}{d_na_{n+T_1}},K_{2n}=\frac{c_nb_{n+T_2}}{d_na_{n+T_2}}\mathrm{\hspace{0.17em}.4}$$
All other invariants are expressed in terms of $`K_{1n}`$, $`K_{2n}`$ and their translations on the lattice.
Proof The proof of the invariance of these quantities is straightforward: it is a direct elementary verification. It is convenient to carry out the proof of their completeness for the factorized form (89). Expressing $`a_n,b_n,c_n,d_n`$ in terms of $`f_n,u_n,v_n,w_n`$, we obtain
$$K_{1n}=\frac{1}{1+w_n},K_{2n}=\frac{1}{1+w_{n+T_2}}\left(v_nu_{n+T_2}v_{n+T_1}^1u_n^1\right).$$
It is clear that the potential $`w_n`$ is gauge invariant, and the function $`f_n`$ can be shortened in equivalence classes. Hence, everything reduces to gauge equivalence for $`f_n=1`$. Here the ‘potential’ $`w_n`$ and the ‘curvature’
$$H=v_nu_n^1u_{n+T_2}v_{n+T_1}^1=\frac{v_nu_{n+T_2}}{u_nv_{n+T_1}}5$$
are gauge invariant. We need to find all gauge invariants of the operator
$$L_0=(1+u_nT_1)(1+v_nT_2)$$
with respect to the transformations
$$L_0\tau _n(1+u_nT_1)(1+v_nT_2)\tau _n^1=\left(1+u_n\frac{\tau _n}{\tau _{n+T_1}}T_1\right)\left(1+v_n\frac{\tau _n}{\tau _{n+T_2}}T_2\right).$$
This problem is easily reduced to the classification of gauge invariant monomials, that is, products of the quantities $`u_{n+kT_1+lT_2}`$, $`v_{n+kT_1+lT_2}`$ for different (finitely many) $`k`$$`l`$. Elementary combinatorics leads to the conclusion that all these monomials are products of monomials (94) and their translations in some powers. We omit these details. $`\mathrm{}`$
We use the basis of invariants $`w_n`$, $`H_n`$, by analogy with the continuous case.
###### Lemma 13
3 The Laplace transformation for invariants has the form
$$\begin{array}{c}1+\stackrel{~}{w}_{n+T_1}=(1+w_{n+T_2})\frac{w_nw_{n+T_1+T_2}}{w_{n+T_1}w_{n+T_2}}H_n^1,\\ \stackrel{~}{H}_n=\frac{1+w_{n+T_2}}{1+\stackrel{~}{w}_{n+T_2}}.\end{array}6$$
The proof is obtained by substituting the expressions for $`\stackrel{~}{K}_1`$, $`\stackrel{~}{K}_2`$ in terms of $`\stackrel{~}{a}`$, $`\stackrel{~}{b}`$, $`\stackrel{~}{c}`$$`\stackrel{~}{d}`$.
We now consider an infinite chain of Laplace transformations for the quantities
$$\stackrel{~}{H}_n^{(k)}=H_n^{(k+1)},\stackrel{~}{w}_n^{(k)}=w_n^{(k+1)},$$
where $`n=(n_1,n_2)`$, $`k`$, and $`H_n^{(k+1)}`$$`w_n^{(k+1)}`$ are obtained from the previous ones by a Laplace transformation. Eliminating $`H_n^{(k)}`$ from the equations, we obtain a ‘completely discrete analogue of the two-dimensionalized Toda chain’:
$$\frac{1+w_{n+T_1}^{(k+2)}}{1+w_{n+T_1}^{(k+1)}}\frac{1+w_{n+T_2}^{(k+1)}}{1+w_{n+T_2}^{(k)}}=\frac{w_{n+T_1}^{(k)}w_{n+T_2}^{(k)}}{w_n^{(k)}w_{n+T_1+T_2}^{(k)}}.$$
Problem Find a comparison between this completely discrete analogue of the two-dimensionalized Toda chain and the systems studied in , starting from the theory of the ‘Yang–Baxter’ equation.
Example 9 We consider a cyclic chain of length $`N=2`$. After simple calculations we obtain
$`w_n^{(1)}=C(w_n^{(0)})^1,`$
$`w_{n+T_1+T_2}=w_n^1(C+w_{n+T_1})(C+w_{n+T_2})(1+w_{n+T_1})^1(1+w_{n+T_2})^1.`$
Finally, we note that there is, as we showed above, the group of all hyperbolic Laplace transformations, generated by four basis transformations $`\mathrm{\Lambda }_{12}^{\epsilon \sigma }`$, $`\epsilon ,\sigma =\pm `$, which corresponds to the choice of the bases $`(T_1^\epsilon ,T_2^\sigma )`$. In principle, to each word in this group corresponds the condition for cyclicity of a chain. However, the structure of this group is not known so far.
## §6. The Laplace transformation for elliptic two-dimensional difference operators on a regular lattice. Equations of a triangle, curvature
It appears that in the elliptic case Laplace transformations cannot be defined for difference operators of the second order on a square lattice. To give their correct definition we should follow two principles.
I. In order that continuous and discrete ‘spectral symmetries’ of continuous Schrödinger operators be conserved in the difference case, all translations should be considered as covariant, as in the one-dimensional case (see §3).
II. In the two-dimensional case for elliptic operators we should replace the square lattice by an equilateral triangular lattice, where the following translations have the same length:
$$T_1^{\pm 1},T_2^{\pm 1},(T_1T_2)^{\pm 1}.7$$
We denote an arbitrary pair of neighbouring translations (97) (which go anticlockwise) by $`T_1^{}`$$`T_2^{}`$ and ascribe to this pair a Laplace transformation generated by factorization
$$L=(x_n+y_nT_1^{}+z_nT_2^{})(x_n+y_{nT_1^{}}T_{1}^{}{}_{}{}^{1}+z_{nT_2^{}}T_{2}^{}{}_{}{}^{1})+w_n,8$$
where the self-adjoint operator $`L`$ has the form
$$L=a_n+b_nT_1+c_nT_2+d_{nT_2}T_1T_2^1+b_{nT_1}T_1^1+c_{nT_2}T_2^1+d_{nT_1}T_2T_1^1,9$$
all the coefficients are real, and $`b_n,c_n,d_n_+`$. It is easy to see that the coefficient $`a_n`$ is a numerical function on $`^2`$, while the coefficients $`b_n,c_n,d_n`$ are a connection sitting on the corresponding edges of the lattice. As a result the operator contains an interaction of each vertex with all six nearest neighbours, and all edges have the same length. We start from the following observation ().
###### Lemma 14
4 A real self-adjoint operator of the form $`(99)`$ with non-zero coefficients $`b_n`$$`c_n`$$`d_n`$ always admits a unique representation in the form
$$L=QQ^++w_n,Q=x_n+y_nT_1+z_nT_2,00$$
where $`T_1^+=T_1^1`$, $`T_2^+=T_2^1`$, $`(AB)^+=B^+A^+`$. Moreover we have the equalities
$$\begin{array}{cc}\hfill x_ny_{nT_1}& =b_{nT_1},\hfill \\ \hfill x_nz_{nT_2}& =c_{nT_2},\hfill \\ \hfill y_{nT_1}z_{nT_2}& =d_{nT_1T_2}.\hfill \end{array}01$$
In all, following , we obtain six different factorizations, which correspond to the following six pairs of periods $`(T_1^{},T_2^{})`$: 1) $`(T_1,T_2)`$, 2) $`(T_2,T_1^1T_2)`$, 3) $`(T_1^1T_2,T_1^1)`$, 4) $`(T_1^1,T_2^1)`$, 5) $`(T_2^1,T_2^1T_1)`$, 6) $`(T_2^1T_1,T_1)`$,
$$L=Q_jQ_j^++w_{jn},$$
$`j=1,2,3,4,5,6`$. $`j`$ is determined modulo 6.
Definition 10 A Laplace transformation $`P_j`$ of type $`j`$ is as follows:
$$\begin{array}{cc}& P_jL\stackrel{~}{L}^{(j)}=w_{jn}^{1/2}Q_j^+w_{jn}^1Q_jw_{jn}^{1/2}+w_{jn},\hfill \\ & P_j\psi \stackrel{~}{\psi }=w_{jn}^{1/2}Q_j^+\psi .\hfill \end{array}02$$
It is defined if $`w_{jn}>0`$.
###### Lemma 15
5 The Laplace transformations $`P_j`$ and $`P_{j+3}`$ are mutually inverse. The others are connected by the following relations:
$$\begin{array}{c}w_{1n}=w_{3n}=w_{5n},\\ Q_1=Q_3T_1=Q_5T_2,\\ \stackrel{~}{L}^{(1)}=\sqrt{\frac{w_n}{w_{nT_1}}}T_1^1\stackrel{~}{L}^{(3)}T_1\sqrt{\frac{w_n}{w_{nT_1}}}=\sqrt{\frac{w_nw_{nT_2}}{}}T_2^1\stackrel{~}{L}^{(5)}T_2\sqrt{\frac{w_n}{w_{nT_2}}}.\end{array}03$$
In the equivalence class
$$Lf_nLf_n04$$
the operators $`\stackrel{~}{L}^{(1)}`$, $`\stackrel{~}{L}^{(3)}`$, $`\stackrel{~}{L}^{(5)}`$ are unitarily adjoint.
For $`j=2,4,6`$ analogous formulae hold:
$$\stackrel{~}{L}^{(4)}=\sqrt{\frac{w_n}{w_{n+T_1}}}T_1\stackrel{~}{L}^{(6)}T_1^1\sqrt{\frac{w_n}{w_{n+T_1}}}=\sqrt{\frac{w_n}{w_{n+T_2}}}T_2\stackrel{~}{L}^{(2)}T_2^1\sqrt{\frac{w_n}{w_{n+T_2}}}.$$
Remark $`8`$ In the definition of Laplace transformations we can discard the requirement $`w_n>0`$, replacing it by the condition $`w_n0`$, and set
$$\stackrel{~}{L}^{(j)}=Q_j^+w_{jn}^1Q_j+1,\stackrel{~}{\psi }=Q_j^+\psi .$$
We obtain the same transformations up to equivalence (104), but $`P_j`$ and $`P_{j+3}`$ are not exactly mutually inverse. We note that in the discrete case, as opposed to the continuous one, the condition $`w0`$ does not mean that $`w`$ has constant sign.
Remark $`9`$ The relations (103) were not noted in .
Remark $`10`$ If the potential $`w_nw_0`$ is constant (for example, for the first factorization), then the Laplace transformation
$$P_1L\stackrel{~}{L}^{(1)}$$
acts on the eigenfunctions of all levels $`L\psi =\lambda \psi `$:
$$P_1\psi =Q_1^+\psi ,P_1L=Q_1^+Q_1+w_0=\stackrel{~}{L}^{(1)}.$$
In the framework of the equivalence class (104) of real self-adjoint operators we can always arrange that $`w_n=\text{const}`$ if $`w_n=\mathrm{exp}f_n0`$. However, these transformations realize a formal equivalence of the spectral theories only on the zero level $`L\psi =0`$, and therefore we can apply the Laplace transformation to all levels, generally speaking, only once (for special cases see below).
The proof of the lemma can be obtained by direct verification.
An investigation of the cyclic chains of difference operators has not been carried out. The classes of purely factorized operators of the form a) and b):
$$\text{a)}L=Q^+Q;\text{b)}\stackrel{~}{L}=QQ^+$$
are of interest, where
$$Q=1+ce^{l_1(n)}T_1+de^{l_2(n)}T_2,05$$
and
$$l_j(n)=l_{j1}n_1+l_{j2}n_2$$
are linear forms in $`n=(n_1,n_2)`$.
Following , we consider the ‘equation of black triangles’, which have the form $`n,n+T_1,n+T_2`$:
$$Q\psi =0,$$
and the ‘equation of white triangles’, which have the form $`n,nT_1,nT_2`$:
$$Q^+\psi =0.$$
###### Theorem 4
$`1)`$ The equation of black triangles $`Q\psi =0`$ certainly has an infinite-dimensional space of solutions $`\psi _2(^2)`$ if one of the following conditions is satisfied:
* $`l_{11},l_{22}>0`$, $`l_{11}l_{22}l_{12}^2>0;`$
* $`l_{11},l_{22}>0`$, $`l_{11}l_{22}l_{21}^2>0;`$
* $`l_{11}>0`$, $`l_{11}l_{22}l_{12}^2>l_{11}(l_{21}l_{12});`$
* $`l_{22}>0`$, $`l_{11}l_{22}l_{21}^2>l_{22}(l_{21}l_{12})`$.
$`2)`$ The equation of white triangles $`Q^+\psi =0`$ certainly has an infinite-dimensional space of solutions $`\psi _2(^2)`$ if one of the following conditions, which are obtained from the previous ones by the transformation $`l_{ij}l_{ij}`$, is satisfied:
* $`l_{11},l_{22}<0`$, $`l_{11}l_{22}l_{12}^2>0;`$
* $`l_{11},l_{22}<0`$, $`l_{11}l_{22}l_{21}^2>0;`$
* $`l_{11}<0`$, $`l_{11}l_{22}l_{12}^2>l_{11}(l_{12}l_{21});`$
* $`l_{22}<0`$, $`l_{11}l_{22}l_{21}^2>l_{22}(l_{12}l_{21})`$.
Proof We construct explicitly the solutions of our equations if the conditions presented above are satisfied. We examine first the case of black triangles. We are looking for a solution of the form
$$\psi _n=e^{K_2(n)}\chi _n,$$
where
$$K_2(n)=\alpha n_1^2+2\beta n_1n_2+\delta n_2^2.$$
For $`\chi _n`$ we obtain the equation
$$0=\chi _n+ce^\alpha e^{(l_{11}2\alpha )n_1+(l_{12}2\beta )n_2}\chi _{n+T_1}+de^\delta e^{(l_{21}2\beta )n_1+(l_{22}2\delta )n_2}\chi _{n+T_2}.06$$
We consider three cases:
* the coefficients (106) depend only on $`n_1`$;
* the coefficients (106) depend only on $`n_2`$;
* the coefficients (106) depend only on $`n_1+n_2`$.
This leads, respectively, to the conditions:
* $`l_{12}=2\beta `$, $`l_{22}=2\delta `$;
* $`l_{11}=2\alpha `$, $`l_{21}=2\beta `$;
* $`l_{11}2\alpha =l_{12}2\beta `$, $`l_{21}2\beta =l_{22}2\delta `$.
In the case a) we make the substitution
$$\chi _n=w^{n_2}\phi _{n_1}07$$
and obtain the equation for $`\phi _m`$:
$$0=\phi _m+ce^\alpha e^{(l_{11}2\alpha )(m+1)}\phi _{m+1}+wde^\delta e^{(l_{21}l_{12})m}\phi _m.$$
In addition we put $`l_{11}=2\alpha `$. After this we obtain
$$(1+wce^{l_{22}/2}e^{(l_{21}l_{12})m})\phi _m=ce^\alpha \phi _{m+1}.08$$
Two cases are possible.
Case 1. $`l_{21}>l_{12}`$. We choose the value of $`w_q`$ such that
$$1+w_qde^{\delta /2}e^{(l_{21}2\beta )q}=0,09$$
where $`2\delta =l_{22}`$, $`2\beta =l_{12}`$. ¿From this we obtain the ‘quantization condition’, which selects a discrete series of admissible values $`w=w_q`$, where $`q`$ runs over $``$. Under this condition $`\phi _m^{(q)}=0`$ for $`m>q`$.
Further, we consider the relation
$$\phi _m^{(q)}=(ce^\alpha )(1+w_qe^{k_2\delta }e^{(l_{21}2\beta )m})^1\phi _{m+1}$$
as $`m\mathrm{}`$. If $`l_{21}2\beta >0`$, then for $`m<0`$ and $`|m|\mathrm{}`$ we have the asymptotics
$$\phi _m^{(q)}(\text{const})^{|m|}.$$
Our solution is constructed in the form
$$\psi _n^{(q)}=e^{K_2(n)}w^{n_2}\phi _{n_1}^{(q)},10$$
where $`2K_2(n)=l_{11}n_1^2+2l_{12}n_1n_2+l_{22}n_2^2`$. This form is strictly positive if and only if
$$l_{11}>0,l_{22}>0,l_{11}l_{22}l_{12}^2>0.$$
The function $`\psi _n^{(q)}`$ certainly belongs to the space $`_2(^2)`$ and takes the value zero for $`n_1>q`$.
Case 2. $`l_{12}l_{21}`$. In this case a solution $`\phi _m`$ of equation (108) grows no faster than $`(\text{const})^{|m|}`$, $`m\pm \mathrm{}`$, for any $`w`$. Thus we have constructed the necessary solution for the case a) of black triangles.
In the case b) we interchange the position of $`n_1`$ and $`n_2`$. In the case c) we require that $`\phi `$ depends only on $`n_1+n_2`$. For example, in case c) we put
$$2K_2(n)=l_{11}n_1^2+2l_{12}n_1n_2+(l_{22}l_{21}+l_{12})n_2^2$$
and make the substitution
$$\chi _n=w^{n_1}\phi _{n_1+n_2}.$$
For the function $`\phi _n`$ we obtain the following equation:
$$\phi _n+(cwe^{l_{11}/2}+de^{(l_{21}l_{12}l_{22})/2}e^{(l_{21}l_{12})(n_1+n_2)})\phi _{n+1}=0.$$
In the cases a)–c) for white triangles we proceed completely analogously. $`\mathrm{}`$
Other cases, when the solutions lie in $`_2(^2)`$, have not been found so far.
The form $`K_2(n)`$ can be written as follows.
The case of black triangles:
* $`2K_2=\left(\begin{array}{cc}l_{11}& l_{12}\\ l_{12}& l_{22}\end{array}\right)`$;
* $`2K_2=\left(\begin{array}{cc}l_{11}& l_{21}\\ l_{21}& l_{22}\end{array}\right)`$;
* $`2K_2=\left(\begin{array}{cc}l_{11}& l_{12}\\ l_{12}& l_{22}l_{21}+l_{12}\end{array}\right)`$;
* $`2K_2=\left(\begin{array}{cc}l_{11}l_{21}+l_{12}& l_{21}\\ l_{21}& l_{22}\end{array}\right)`$.
The case of white triangles:
* $`2K_2=\left(\begin{array}{cc}l_{11}& l_{12}\\ l_{12}& l_{22}\end{array}\right)`$;
* $`2K_2=\left(\begin{array}{cc}l_{11}& l_{21}\\ l_{21}& l_{22}\end{array}\right)`$;
* $`2K_2=\left(\begin{array}{cc}l_{11}& l_{12}\\ l_{12}& l_{22}+l_{21}l_{12}\end{array}\right)`$;
* $`2K_2=\left(\begin{array}{cc}l_{11}+l_{21}l_{12}& l_{21}\\ l_{21}& l_{22}\end{array}\right)`$.
Remark $`11`$ We can impose analogous conditions on the coefficients $`l_{ij}`$ and their relation with $`\alpha ,\beta ,\delta `$, demanding that the coefficients of equation (106) for $`\chi _n`$ be dependent only on a combination of the form $`(\kappa n_1+\tau n_2)`$. If $`(\kappa n_1+\tau n_2)`$ and $`(un_1+vn_2)`$ are the basis of the lattice, then the condition of dependency on only one variable has the form:
$$\begin{array}{cc}\hfill (l_{11}2\alpha )\tau & =(l_{12}2\beta )\kappa ,\hfill \\ \hfill (l_{21}2\beta )\tau & =(l_{22}2\delta )\kappa .\hfill \end{array}11$$
Now we can seek a solution in the form
$$\chi _n=w^{un_1+vn_2}\phi _{\kappa n_1+\tau n_2}.12$$
Further, for the functions $`\phi _s`$, $`s`$, we obtain a difference equation of the form
$$0=\phi _s+A(s)\phi _{s+\kappa }+B(s)\phi _{s+\tau }.13$$
Only in the three cases mentioned above can this equation be easily solved:
$$\text{a)}\kappa =1,\tau =0;\text{b)}\kappa =0,\tau =1;\text{c)}\kappa =\tau =1.$$
In the other cases we obtain difference equations of order 2 and higher. We do not know how we could find solutions of them that lead to functions $`\psi _2(^2)`$.
¿From the theorem we have the following corollary.
###### Corollary 1
Under the conditions stated in the theorem the positive operators $`L=QQ^+`$ or $`\stackrel{~}{L}=Q^+Q`$ have spectrum in $`_2(^2)`$ such that the point $`\lambda =0`$ is an infinitely degenerate point of the discrete spectrum. The eigenfunctions of this point satisfy the equation of black or white triangles
$$1)Q\psi =0\text{or}2)Q^+\psi =0.$$
The eigenfunctions of the ground state $`\lambda =0`$ can be determined by explicit formulae, which follow from the construction given above.
This assertion is a difference analogue of results in , on the ground states of the Pauli operator in a continuous purely magnetic case.
Problems 1. Prove that the constructed spaces of solutions of the equations $`Q\psi =0`$ and $`Q^+\psi =0`$ are complete in $`_2(^2)`$ among all solutions, and find appropriate orthonormal bases. Explain what kind of cases of solubility in $`_2(^2)`$ of the equations of black or white triangles exist, apart from the solutions found above. Even in the class of operators whose coefficients are exponents of linear forms $`l_j(n)`$, this problem has not been solved.
2. Prove that if the conditions of the theorem are satisfied, then the spectrum of the operators $`L=QQ^+`$ and $`\stackrel{~}{L}=Q^+Q`$ has a non-trivial gap $`\mathrm{\Delta }`$ between the ground state $`\lambda _0=0`$ and the next level $`\lambda _1\mathrm{\Delta }>0`$.
3. Find the class of operators $`Q`$ such that the spectrum of the operators $`L=QQ^+`$ and $`\stackrel{~}{L}=Q^+Q`$ begins at the point $`\lambda =0`$ and is continuous, that is, the point $`\lambda =0`$ is the bottom of the continuous spectrum ($`\mathrm{\Delta }=0`$).
The last problem is of special interest. Below we consider algebraic conditions of the type of commutation of the operators $`Q`$$`Q^+`$, which supposedly lead to the situation discussed in Problem 3.
To begin with we consider the difference (purely real) $`q`$-analogues of the continuous Schrödinger–Landau operator in a homogeneous (constant) magnetic field.
Suppose that the operators $`Q`$, $`Q^+`$ have the form (105):
$$Q_{c,d}=1+ce^{l_1(n)}T_1+de^{l_2(n)}T_2,$$
and depend on the parameters
$$c=e^{k_1},d=e^{k_2},l_j=l_{j1}n_1+l_{j2}n_2.$$
###### Lemma 16
6 Operators $`Q`$, $`Q^+`$ of the form $`(105)`$ satisfy the relation
$$Q_{c,d}Q_{c,d}^+1=q(Q_{c^{},d^{}}^+Q_{c^{},d^{}}1)14$$
if the matrix $`l_{ij}`$ and the parameters $`c`$, $`d`$, $`c^{}`$, $`d^{}`$ are related by the following equalities:
$$c=u^2c^{},d=u^2d^{},q=u^2,u=e^{l_{11}},$$
$$2l_{11}=2l_{22}=l_{12}+l_{21}.15$$
Remark $`12`$ In the notation
$$e^{l_{11}}=u,e^{l_{12}}=v$$
the operator $`Q`$ is written in the form
$$Q_{c,d}=1+cu^{n_1}v^{n_2}T_1+d(u^2/v)^{n_1}u^{n_2}T_2.16$$
Dependence on the parameters $`(u,v)`$ in (114) is omitted. Such notation was used in , where these operators were introduced for the first time and their eigenfunctions were found.
The proof of the lemma is by direct verification.
Comparing the results of Lemma 16 and Theorem 4, we obtain the following result ().
###### Theorem 5
We consider the operators $`L=QQ^+`$ and $`\stackrel{~}{L}=Q^+Q`$ under the conditions $`(115)`$. The spectrum of these operators in $`_2()`$ for $`0\lambda <1`$ is purely discrete, infinitely degenerate, and can only lie at the following points:
$$\text{a)}\lambda _j=1u^{2j},j0,u<1;17$$
$$\text{b)}\lambda _j=1u^{2j},j0,u>1.18$$
In the following cases the spectrum of the operator $`L`$ occupies all points $`(117)`$, and the spectrum of the operator $`\stackrel{~}{L}`$ occupies all points $`(117)`$ except $`\lambda _0`$:
$$\text{a}\text{}\text{)}u^3>v^1>u^1>1;\text{a}\text{′′}\text{)}u^1>\mathrm{max}(v,v^1)1.19$$
Analogously, the spectrum of the operator $`\stackrel{~}{L}`$ occupies all points $`(118)`$, and $`L`$ occupies all points except $`\lambda _0`$, if one of the following conditions is satisfied:
$$\text{b}\text{}\text{)}u^3>v>u>1;\text{b}\text{′′}\text{)}u>\mathrm{max}(v,v^1)1.20$$
All eigenfunctions discussed in the theorem are obtained from solutions of the equations $`Q\psi =0`$ or $`Q^+\psi =0`$ for proper values of the constants $`c`$$`d`$$`u`$$`v`$ by the use of ‘creation operators’:
$`1)`$ $`Q_{c,d}\stackrel{~}{\psi }_0=0`$, $`\stackrel{~}{L}_{c,d}\stackrel{~}{\psi }_0=Q_{c_0,d_0}^+Q_{c_0,d_0}\stackrel{~}{\psi }_0=0`$, $`\stackrel{~}{\psi }=Q_{c_0,d_0}^+\mathrm{}Q_{c_{k1},d_{k1}}^+\stackrel{~}{\psi }_0`$, $`\stackrel{~}{L}_{c_0,d_0}\psi =(1u^{\pm 2k})\psi ;`$
$`2)`$ $`Q^+\psi _0=0`$, $`L_{c,d}\psi _0=Q_{c,d}Q_{c,d}^+\psi _0=0`$, $`\psi =Q_{c_0,d_0}\mathrm{}Q_{c_{k1},d_{k1}}\psi _0`$, $`L_{c_0,d_0}\psi =(1u^{\pm 2k})\psi `$,
$$c_j=u^2c_{j1},d_j=u^2d_{j1}.$$
Remark $`13`$ Of the two cases mentioned in the proof of Theorem 4, in the given situation only case 1 is realized, in which the eigenfunctions $`\psi ^{(q)}`$ that we found vanish on the whole subplane.
The situation described in Theorem 5 is completely analogous to , that is, to Theorem 1 in §3. The difference is that for $`n=2`$ the spaces of solutions are infinite-dimensional (the completeness of the solutions found on the corresponding levels has not been established so far).
We note that conditions (119), (120) are obtained directly from the condition of Theorem 4, but the case c) here is not realized under conditions (115).
Problem Prove that the spectrum of the operators $`L=Q_{c,d}Q_{c,d}^+`$ and $`\stackrel{~}{L}=Q_{c,d}^+Q_{c,d}`$ in the space $`_2(^2)`$ for $`\lambda 1`$ is continuous and runs over the whole semi-axis $`\lambda 1`$. As we assume, this is a simple Lebesgue spectrum.
We point out that our operators $`Q`$ depend on the constants $`c`$$`d`$$`u`$$`v`$ (if the dependence on $`u`$$`v`$ is not written explicitly, then in the given formula we assume that $`u`$$`v`$ have not changed). The following formulae are true:
$$\begin{array}{cccccc}\hfill T_1Q_{c,d}T_1^1& =Q_{c^{},d^{}},\hfill & \hfill c^{}& =cd^{l_{11}},\hfill & \hfill d^{}& =de^{l_{12}},\hfill \\ \hfill T_2Q_{c,d}^+T_2^1& =Q_{c^{\prime \prime },d^{\prime \prime }},\hfill & \hfill c^{\prime \prime }& =cd^{l_{12}},\hfill & \hfill d^{\prime \prime }& =de^{l_{22}},\hfill \end{array}21$$
where
$`Q_{c,d}`$ $`=1+ce^{l_1(n)}T_1+de^{l_2(n)}T_2,`$
$`Q_{c,d}^+`$ $`=q+ce^{l_1(nT_1)}T_1^1+de^{l_2(nT_2)}T_2^1.`$
The same is also true for the operators $`L=QQ^+`$, $`\stackrel{~}{L}=Q^+Q`$:
$`T_1L_{c,d}T_1^1`$ $`=L_{c^{},d^{}},`$
$`T_2L_{c,d}T_2^1`$ $`=L_{c^{\prime \prime },d^{\prime \prime }}.`$
Definition 11 A characteristic vector-section is a function $`\psi _{c,d}(n)`$ that is analytically dependent on all the parameters $`c`$$`d`$$`u`$$`v`$ and such that
$$L\psi =\lambda \psi \text{or}\stackrel{~}{L}\psi =\lambda \psi ,$$
where $`\lambda `$ does not depend on the parameters. A Bloch vector-section $`\psi _{c,d}(n,p)`$ is a function $`\psi `$ such that
$$L\psi =\lambda \psi ,T_1\psi _{c,d}=e^{ip_1}\psi _{c^{},d^{}},T_2\psi _{c,d}=e^{ip_2}\psi _{c^{\prime \prime },d^{\prime \prime }}.$$
By (121) the functions $`T_j\psi _{c,d}`$ satisfy these equations, but with shifted parameters $`(c^{},d^{})`$ or $`(c^{\prime \prime },d^{\prime \prime })`$. This definition makes sense also in the one-dimensional case (54), where we had
$`Q_c=1+ce^{l(n)}T,l(n)=ln,`$
$`T=T_1,n=n_1,T_1Q_cT_1^1=Q_{ca},a=e^l.`$
Let us construct the Bloch vector-section for $`\lambda =0`$ when $`a>1`$. We choose $`c=c_0>0`$. Choosing an arbitrary $`\psi _{0,c_0}`$, we put
$$\psi _{c_0a^m}(n,p)=T_1^m\left(\psi _{0,c_0}(n)\right)e^{imp}.22$$
Limitations on the construction appear as $`a1`$. There are no solutions at all in $`_2()`$ for $`a<1`$ if we are talking about the equation $`Q_c\psi =0`$ corresponding to $`\lambda =0`$. Therefore, $`a1`$, $`c0`$.
As $`a1+0`$ the operator tends to an operator with constant coefficients, where there are singularities of the Bloch vector-section.
Apparently, as $`|c|1,a1`$ we can talk about the transition of the Bloch vector-section corresponding to $`\lambda 1`$ into an ordinary Bloch solution for the equation with constant coefficients (that is, a solution of pure exponent type), since for $`c0`$ the spectrum of the operator for $`a=1`$ has the form $`(1|c|)^2\lambda (1+|c|)^2`$.
In addition, for $`a=1`$ the spectrum is bounded in $`_2`$, $`|\lambda |(1+|c|)^2`$, and for $`a>1`$ the operator is of course unbounded. Hence, we necessarily have a singularity here. For $`\lambda 1`$ the situation is not clear.
Problem For the operators $`L=QQ^+`$ or $`\stackrel{~}{L}=Q^+Q`$, whose coefficients are exponents of linear functions, construct a complete basis of eigenfunctions in the form of ‘Bloch sections’ that depend analytically on all the parameters $`c`$$`d`$$`l_{ij}`$, which has minimally possible singularities in the submanifolds, where we have operators with constant coefficients in one of the variables.
We return now to the case when the point $`\lambda =0`$ is supposedly the lower bound of the continuous spectrum for operators of the form
$$L=QQ^+,23$$
where the $`Q`$ have the form (105).
Hypothesis If the operators $`Q`$, $`Q^+`$ commute, $`QQ^+=Q^+Q`$, and they have a general solution $`Q\psi =0`$, $`Q^+\psi =0`$ that is bounded on the lattice $`^2`$ (or is growing sufficiently slowly?), then the point $`\lambda =0`$ is the lower bound of the continuous spectrum in $`_2(^2)`$ for the operator $`L=\stackrel{~}{L}`$. We state below a natural generalization of this hypothesis.
Example 10 Let the operators $`Q_{c,d}`$, $`Q_{c,d}^+`$ be such that $`l_{11}=l_{22}=0`$ or $`u=1`$. Then we have
$$Q_{c,d}Q_{c,d}^+=Q_{c,d}^+Q_{c,d},24$$
as follows from (114). The operators have the form
$$\begin{array}{cc}\hfill Q_{c,d}& =1+cv^{n_2}T_1+dv^{n_1}T_2,\hfill \\ \hfill Q_{c,d}^+& =1+cv^{n_2}T_1^1+dv^{n_1}T_2^1.\hfill \end{array}25$$
We pose the following question. Let two operators be given in the form
$$Q_1=1+a_nT_1+b_nT_2,Q_2=1+c_nT_1^1+d_nT_2^126$$
with non-vanishing coefficients. We consider the system of equations of black and white triangles simultaneously:
$$\{\begin{array}{cc}Q_1\psi =0,\hfill & \\ Q_2\psi =0.\hfill & \end{array}27$$
In this case is the system (127) completely locally consistent?
Richer in content is the following formulation. Let the ‘initial condition’ for (127) be given in the form of two arbitrary values of $`\psi `$ at the ends of any edge of the lattice (for example, $`\psi _{nT_1}`$ and $`\psi _n`$). Using equations (127), we can solve the system and find the value at any other point $`n^{}^2`$, moving along the paths of the triangles, black and white: knowing the values at the ends of any edge, we find the value at the third vertex of any black or white triangle that has this edge as its border. Considering a bundle of six triangles with a common vertex $`n`$, we can pass from one cycle and return again to the edge $`[nT_1,n]`$, and the linear ‘transformation of curvature’ will be given by an upper triangular $`(2\times 2)`$-matrix:
$`\psi _{nT_1}^{(new)}=A_n\psi _{nT_1}^{(old)}+B_n\psi _n^{(old)},`$
$`\psi _n^{(new)}=\psi _n^{(old)}.`$
The matrix coefficients are the ‘curvatures’ $`A_n`$, $`B_n`$. They are expressed in terms of the coefficients of the system (127):
$$\begin{array}{cc}\hfill A_n& =\frac{b_{nT_1}d_{n+T_1}}{b_nc_{n+T_2}d_na_{nT_2}},\hfill \\ \hfill B_n& =\frac{b_{nT_1}}{c_{n+T_2}}\left(\frac{1}{b_n}\left(\frac{d_{n+T_1}}{a_{nT_2}}\left(\frac{1}{d_n}b_{nT_2}\right)c_{n+T_1}+1\right)d_{n+T_2}\right)a_{nT_1}.\hfill \end{array}28$$
###### Lemma 17
7 The condition $`A_n=1`$, $`B_n=0`$ for all $`n^2`$ is necessary and sufficient for the possibility of a unique solution of the system $`(127)`$ on the plane $`^2`$ under arbitrary initial conditions imposed at the ends of any fixed edge. For this it is also necessary and sufficient to satisfy the following algebraic relation for the operators $`Q_1`$$`Q_2`$:
$$\left((Q_11)(Q_21)1\right)=f_n\left((Q_21)(Q_11)1\right),f_n0.29$$
We omit the proof of the lemma.
###### Corollary 2
If the system $`(127)`$ is completely consistent, then its solution is determined by an arbitrary solution of a one-dimensional difference equation of the second order along any path of edges without self-crossing, in particular, along the path $`n_1=\text{const}`$ (we obtain an equation for the variable $`n_2)`$ or $`n_2=\text{const}`$ (we obtain an equation for the variable $`n_1)`$. If the curvature is trivial, then all these equations are equivalent.
Example 11 For the commuting operators $`Q_1=Q_2^+=Q_{c,d}`$ from (125) the condition for consistency is satisfied. For the variable $`n_1`$ ($`n_2=\text{const}`$) we get the equation
$$cv^{n_2}(\psi _{nT_1}+\psi _{n+T_1})+(1+c^2v^{2n_2}dv^{2n_1})\psi _n=0.30$$
To end this section we give some information on factorizations and Laplace transformations for general non-self-adjoint operators $`L`$ of the second order on an equilateral triangular lattice, and also on complex Hermite operators. The following assertions hold.
###### Theorem 6
To represent the operator
$$L=a_n+b_nT_1+c_nT_2+d_{nT_1}T_1^1T_2+e_{nT_1}T_1^1+f_{nT_2}T_2^1+g_{nT_2}T_1T_2^131$$
in the form
$$\text{a)}L=Q_1Q_2+w_n\text{or}\text{b)}L=Q_2Q_1+w_n,$$
where $`Q_1`$ and $`Q_2`$ have the form
$$Q_1=x_n+y_nT_1+z_nT_2,Q_2=p_n+q_{nT_1}T_1^1+r_{nT_2}T_2^1,$$
it is necessary and sufficient to satisfy the conditions
$$\begin{array}{cc}\hfill \text{a)}& b_{n+T_2}d_nf_{n+T_1}=g_ne_{n+T_2}c_{n+T_1},\hfill \\ \hfill \text{b)}& f_nd_nb_n=c_ne_ng_n.\hfill \end{array}32$$
If both conditions are satisfied, then the Laplace transformations
$$\text{a)}L=Q_1Q_2+w\stackrel{~}{L}=Q_2w^1Q_1+1,33$$
$$\text{b)}L=Q_2Q_1+v\stackrel{~}{L}=Q_1w^1Q_2+134$$
can be iterated, moreover infinitely many times (if $`w,v0)`$.
Remark $`14`$ The factorization conditions (132) are invariant with respect to the rotation of the lattice by the angle $`2\pi /3`$. By analogy with Definition 10 we can introduce six types of Laplace transformations. Then the 1st, 3rd and 5th transformations are unitarily adjoint and inverse to the 4th, 6th and 2nd transformations respectively. The same applies to the following assertion.
###### Theorem 7
To factorize an operator $`L`$ of the form
$$L=a_n+b_nT_1+\overline{c}_nT_2+d_{nT_1}T_2T_1^1+\overline{b}_{nT_1}T_1^1+c_{nT_2}T_2^1+\overline{d}_{nT_2}T_1T_2^135$$
in the form
$$\text{a)}L=QQ^++w_n\text{or}\text{b)}L=Q^+Q+w_n,$$
it is necessary and sufficient to satisfy the following conditions:
$$\begin{array}{cc}\hfill \text{a)}& b_{nT_1}c_{nT_2}d_{nT_1T_2},\hfill \\ \hfill \text{b)}& d_nc_nb_n.\hfill \end{array}36$$
In particular, if both conditions are satisfied, then the Laplace transformations can be iterated infinitely many times.
The proof of these theorems is straightforward.
For complex Hermite operators $`L`$ it is natural to define a class of ‘phase’ gauge transformations
$$Le^{if_n}Le^{if_n},\psi _ne^{if_n}\psi _n$$
such that $`f_n`$. These transformations keep the operator formally Hermitian with respect to the previous standard scalar product in $`_2(^2)`$
$$\phi ,\psi =\underset{n^2}{}\phi _n\overline{\psi }_n.37$$
The operator $`L`$ naturally has ‘real’ and ‘phase’ projections
$`L_{}`$ $`=a_n+\beta _nT_1+\gamma _nT_2+\delta _{nT_1}T_2T_1^1+\beta _{nT_1}T_1^1+\gamma _{nT_2}T_2^1+\delta _{nT_2}T_1T_2^1,`$
$`L_\mathrm{\Phi }`$ $`=a_n+B_nT_1+\overline{C}_nT_2+D_{nT_1}T_2T_1^1+\overline{B}_{nT_1}T_1^1+C_{nT_2}T_2^1+\overline{D}_{nT_2}T_1T_2^1,`$
$`b_n=\beta _nB_n`$, $`c_n=\gamma _nC_n`$, $`d_n=\delta _nD_n`$, where $`\beta _n`$, $`\gamma _n`$, $`\delta _n`$, $`a_n`$$`w_n`$ are real numbers and $`B_n`$$`C_n`$$`D_n`$ are complex with modulus equal to 1.
Definition 12 A physical magnetic field is the phase part of the product of the coefficients of an operator (of connection) along the border of any black or white triangle
$$e^{i\mathrm{\Phi }_n^{(1)}}=B_{nT_1}C_{nT_2}D_{nT_1T_2},lack$$
$$e^{i\mathrm{\Phi }_n^{(2)}}=D_nC_nB_n.hite$$
The following lemma is obvious.
###### Lemma 18
8 The operator $`L`$ is reduced by the phase gauge transformation $`e^{if_n}Le^{if_n}`$ to a purely real form if and only if the magnetic field (that is, all ‘magnetic fluxes’ $`\mathrm{\Phi }_n^{(j)})`$ is trivial:
$$\mathrm{\Phi }_n^{(j)}=0,n^2,j=1,2.$$
###### Corollary 3
A complex Hermite operator $`L`$ admits an unbounded number of Laplace transformations of all six forms if and only if it is reduced by a phase gauge transformation to a purely real operator.
Thus, the $`q`$-analogues of the Schrödinger–Landau operator on the lattice have no connection with the magnetic field. Difference operators in a physical magnetic field do not factorize, generally speaking. A proper analogue of an operator in a homogeneous magnetic field can (having chosen a specific gauge transformation, that is, a ‘vector-potential’ composed of coefficients $`b_n,c_n,d_n`$ such that $`|b_n|=|c_n|=|d_n|=1`$) be written in the form
$$\begin{array}{c}\hfill L=6e^{i\mathrm{\Phi }n_2}T_1e^{i\mathrm{\Phi }n_1}T_2e^{i\mathrm{\Phi }(n_1+n_2)}T_2T_1^1\\ \hfill e^{i\mathrm{\Phi }n_2}T_1^1e^{i\mathrm{\Phi }n_1}T_2^1e^{i\mathrm{\Phi }(n_1+n_2)}T_1T_2^1.\end{array}38$$
## §7. Factorizations and Laplace transformations on many-dimensional lattices of regular tetahedra in $`^𝐍`$
We consider a lattice in $`^N`$ such that the ends of the basis vectors $`T_1,\mathrm{},T_N`$ of the lattice together with the point 0 form a regular $`N`$-dimensional simplex (a tetrahedron for $`N=3`$). We call this simplex, and also all others obtained from it through translations by a vector of the lattice, black tetrahedra. White tetrahedra ($`N`$-dimensional simplexes) are the tetrahedron $`0,T_1,\mathrm{},T_N`$ and all others obtained from it through integer-valued translations.
Two black (white) tetrahedra can have no more than one vertex in common. A black and a white tetrahedron can have the longest edge in common. This is true for any dimension $`2`$.
We consider a real self-adjoint operator of the form
$$L=a_n+\underset{k=1}{\overset{N}{}}(b_{k,n}T_k+b_{k,nT_k}T_k^1)+\underset{kj}{}(c_{kj,nT_j}T_kT_j^1+c_{kj,nT_k}T_jT_k^1).$$
We assume that the coefficients $`b_{k,n},c_{kj,n}`$ are a real connection, and they are all non-zero.
###### Theorem 8
The operator $`L`$ admits a representation in the form
$$L=QQ^++w_n,$$
where
$$Q=x_n+\underset{k=1}{\overset{N}{}}y_{k,n}T_k,$$
if and only if the following condition for the coefficients $`b_{k,n}`$$`c_{kj,n}`$ is satisfied:
$$\frac{b_{k,nT_k}b_{j,nT_j}}{c_{kj,nT_kT_j}}=x_n\text{ does not depend on }k\text{}j.39$$
Analogously, the condition for factorization $`L=Q^+Q+w_n`$ is the independence of the expression $`b_{k,n}b_{j,n}/c_{kj,n}`$ on $`k`$$`j`$.
Remark $`15`$ For $`N=2`$ we do not have any conditions. For $`N=3`$ we have two conditions for any white tetrahedron: the product of the coefficients of connection that sit on pairs of opposite (skew) edges of the tetrahedron is the same. In all, the tetrahedron has six edges, which make up three skew pairs, hence we have two conditions. If we want to have all forms of factorizations, then this condition should also be satisfied for black tetrahedra.
The proof of the theorem, as before, is by direct verification.
We now consider factored operators, for which the coefficients of the corresponding operators $`Q`$$`Q^+`$ have the form of exponents of linear functions, as for $`N=2`$:
$$Q=1+ce^{l_1(n)}T_1+de^{l_2(n)}T_2+fe^{l_3(n)}T_3,40$$
$$l_j(n)=\underset{i=1}{\overset{N}{}}l_{ji}n_i.$$
For such operators the following results are true.
###### Theorem 9
The equations
$$\text{1)}Q\psi =0\text{ }\text{(}\text{the equation of black triangles}\text{)}41$$
or
$$\text{2)}Q^+\psi =0\text{ }\text{(}\text{the equation of white triangles}\text{)}42$$
certainly have infinite-dimensional spaces of solutions $`\psi _2(^N)`$ if the following conditions are satisfied:
$`1)`$
$$\text{1a)}\left(\begin{array}{ccc}l_{11}& l_{12}& l_{13}\\ l_{12}& l_{22}& l_{23}\\ l_{13}& l_{23}& l_{33}\end{array}\right)>0,l_{32}=l_{23},43$$
$$\text{1b)}\left(\begin{array}{ccc}l_{11}& l_{12}& l_{13}\\ l_{12}& l_{22}l_{21}+l_{12}& l_{23}\\ l_{13}& l_{23}& l_{33}\end{array}\right)>0,l_{23}+l_{31}=l_{32}+l_{13},44$$
$$\text{1c)}\left(\begin{array}{ccc}l_{11}& l_{12}& l_{13}\\ l_{12}& l_{22}l_{21}+l_{12}& l_{23}l_{21}+l_{12}\\ l_{13}& l_{23}l_{21}+l_{12}& l_{33}l_{31}+l_{13}\end{array}\right)>0,l_{23}l_{21}+l_{12}=l_{32}l_{31}+l_{13};45$$
$`2)`$ the same with $`l_{ij}l_{ij};`$
and also the conditions obtained from those stated above by an arbitrary permutation of the indices $`1,2,3`$.
###### Corollary 4
If the conditions of the theorem are satisfied, the operators $`L=QQ^+`$ and $`\stackrel{~}{L}=Q^+Q`$ have $`\lambda =0`$ as the point of the discrete spectrum that is infinitely degenerate.
Problem Prove that the eigenfunctions constructed according to the scheme of the proof of Theorem 4 give a complete basis for $`\lambda =0`$. Prove that the remaining spectrum is separated from $`\lambda =0`$ by a finite gap $`\mathrm{\Delta }>0`$ in the space $`_2(^3)`$.
###### Theorem 10
0 If the relations
$$l_{ij}+l_{ji}=h46$$
are satisfied, then the operators $`Q`$, $`Q^+`$ satisfy the relations
$$Q_{c,d,f}Q_{c,d,f}^+1=q(Q_{c^{},d^{},f^{}}^+Q_{c^{},d^{},f^{}}1),47$$
where
$$q=e^h,c^{}=e^hc,d^{}=e^hd,f^{}=e^hf.$$
This theorem generalizes Lemma 16, formulated for $`N=2`$.
It is essential to note that the conditions (146) contradict the condition of Theorem 9, and we cannot explicitly find solutions of the equations
$$Q_{c,d,f}\psi =0,48$$
$$Q_{c,d,f}^+\psi =0,49$$
that belong to $`_2(^2)`$ and are necessary to construct the spectrum of the operators $`L`$$`\stackrel{~}{L}`$.
Remark $`16`$ As before, we can seek solutions of (141) in the form
$$\psi _n=e^{K_2(n)}\chi _n,$$
where $`K_2(n)`$ is chosen in such a way that the coefficients of the equation for $`\chi _n`$ do not contain the variable $`n_1`$. However, we have not succeeded in reducing the resulting equation to one variable. After the substitution
$$\chi _n=w^{n_1}\phi _{n_2,n_3}$$
we arrive at the two-dimensional equation of triangles
$$A_n\phi _n+B_n\phi _{n+T_2}+C_n\phi _{n+T_3}=0.50$$
Unfortunately, the coefficients of this equation are very complicated, and so far we do not know how to find solutions of it explicitly.
Obviously, one equation (141) can always be solved in the half-space of a definite direction if the initial conditions are imposed arbitrarily on vertices in any plane that has one of the four forms:
$$n_1=\text{const},\text{or}n_2=\text{const},\text{or}n_3=\text{const},\text{or}n_1+n_2+n_3=\text{const}.51$$
Here the equation of a tetrahedron (as of a triangle for $`N=2`$) is treated as the evolution equation. It is not reversible, as it is for $`N=2`$, since a solution in the inverse direction of ‘time’ is no longer local; its solubility depends on the initial condition belonging to a special functional class.
Now we consider the condition for consistency of the pair of equations (black and white)
$$\begin{array}{cc}\hfill Q_1\psi & =0,Q_1=1+x_nT_1+y_nT_2+z_nT_3,\hfill \\ \hfill Q_2\psi & =0,Q_2=1+p_nT_1^1+q_nT_2^1+r_nT_3^1.\hfill \end{array}52$$
###### Theorem 11
1 The system $`(152)`$ is completely consistent, that is, its solution is determined by an arbitrary solution of some difference equation of the second order in one of the planes $`(151)`$, if the relation
$$(1Q_1)(1Q_2)1=f_n((1Q_2)(1Q_1)1)53$$
is satisfied, where $`f_n`$ is a non-zero function (in particular, for $`f_n1`$ this equality is transformed into the commutation condition $`Q_1Q_2=Q_2Q_1)`$. The corresponding equation of the second order, for example for the plane $`n_1+n_2+n_3=\text{const}`$, has the form
$$(1Q_1)(1Q_2)\psi =\psi .$$
Proof Since equations (152) have the form
$$(1Q_1)\psi =\psi ,(1Q_2)\psi =\psi ,$$
we see that the conditions
$$(1Q_1)(1Q_2)\psi =\psi $$
and
$$(1Q_2)(1Q_1)\psi =\psi ,$$
which are equations of the second order in the plane $`n_1+n_2+n_3=\text{const}`$, should be equivalent. Hence it follows that there is a non-zero function $`f_n`$ such that
$$(1Q_1)(1Q_2)1=f_n\left((1Q_2)(1Q_1)1\right).$$
Thus, we have proved the theorem for the plane $`n_1+n_2+n_3=\text{const}`$.
Writing our conditions locally, in the star of any vertex, we see that it is invariant with respect to rotations that transform the lattice into itself. All four directions of the planes mentioned above are equivalent. We have proved Theorem 11. $`\mathrm{}`$
Example 12 Let us consider the operators (147), where $`h=0`$. In this case the operators $`Q`$$`Q^+`$ commute. The corresponding equation in the hyperplane $`n_3=\text{const}`$ has the form
$$\begin{array}{c}\hfill 0=(A+C^2e^{2\alpha n_2}+D^2e^{2\alpha n_1})\psi _n+Ce^{\alpha n_2}(\psi _{n+T_1}+\psi _{nT_1})\\ \hfill +De^{\alpha n_1}(\psi _{n+T_2}+\psi _{nT_2})+CDe^{\alpha (n_2n_1)}(e^\alpha \psi _{n+T_2T_1}+e^\alpha \psi _{n+T_1T_2}).\end{array}54$$
Now we consider a ‘vector factorization’ for $`N=3`$, where the operators $`Q^+`$ represent vectors $`(Q^{+\alpha })`$, $`\alpha =1,2`$. Hence, the factorization of the operator is given in the form
$$\begin{array}{cc}\hfill \text{a)}& \text{WW}L=\underset{\alpha =1}{\overset{2}{}}Q^\alpha Q^{+\alpha }+w_n,\hfill \\ \hfill \text{b)}& \text{BB}L=\underset{\alpha =1}{\overset{2}{}}Q^{+\alpha }Q^\alpha +w_n,\hfill \\ \hfill \text{c)}& \text{BW}L=Q^1Q^{+1}+Q^{+2}Q^2+w_n,\hfill \end{array}55$$
where
$$Q^\alpha =x_n^\alpha +y_n^\alpha T_1+z_n^\alpha T_2+t_n^\alpha T_3.$$
###### Lemma 19
9 Vector factorization of a self-adjoint operator $`L`$ is always possible (although it is not unique). Moreover, we can regard the potential as a constant $`w_n=w_0`$.
The vector factorizations of all three forms generate Laplace transformations which, however, cannot be iterated.
The search for eigenfunctions of the ground state can sometimes be carried out in one of them:
$$\begin{array}{cc}\hfill \text{a)}& Q^{+1}\psi =0,Q^{+2}\psi =0\text{ for (WW)},\hfill \\ \hfill \text{b)}& Q^1\psi =0,Q^2\psi =0\text{ for (BB)},\hfill \\ \hfill \text{c)}& Q^{+1}\psi =0,Q^2\psi =0\text{ for (BW)}.\hfill \end{array}56$$
If the constant $`w_0`$ for the factorization is chosen correctly, then the lowest level is obtained from equations (156), a), b), or c).
It is difficult to find a criterion for the existence of one solution for the systems (156). However, as before, we state here a criterion for complete local consistency of these systems.
###### Theorem 12
2 The system (156) a) (or b) has solutions that are uniquely determined by arbitrary initial conditions imposed on the vertices of any straight line with direction vector $`T_1`$$`T_2`$$`T_3`$ or $`T_iT_j^1`$, $`ij`$, inside the dihedral angle in which lie the white (respectively, black) tetrahedra adjoining this straight line along an edge, if and only if the following condition is satisfied: for any four white (respectively, black) tetahedra that are located in angles of a tetrahedron twice the size, the eight equations corresponding to them are linearly dependent.
Proof For definiteness we consider the system (156) b). For the values of $`\psi `$ at the vertices of every black tetrahedron we have two equations. This means that if we know the values of $`\psi `$ at two vertices of a black tetrahedron, we can find the values at the other two. Hence we can easily see that if $`\psi `$ is given on all the vertices of some straight line, then our system dictates the extension of $`\psi `$ into the interior of the stated sector. Thus, we only need to clarify when this extension is possible for any initial conditions.
We consider a tetrahedron $`T_N`$, similar to the black one, with edge of length $`N`$. Let $`V(N)`$ be the number of black tetrahedra that lie in $`T_N`$, $`V(N)=N(N+1)(N+2)/6`$. For the values of $`\psi `$ at the $`V(N+1)`$ vertices that lie inside and on the border of $`T_N`$ we have $`2V(N)`$ equations, of which only $`V(N+1)N1`$ should be linearly independent in order to satisfy the condition for complete consistency. But $`2V(N)V(N+1)+N+1=V(N1)`$ is the number of double-size tetrahedra $`T_2T_N`$. $`\mathrm{}`$
The condition for consistency of the system (156) c) was considered in Theorem 11.
## §8. Factorizations of operators and Laplace transformations on two-dimensional surfaces
Let us consider a two-dimensional manifold without boundary, triangulated in such a way that two-dimensional simplexes can be painted in two colours (black and white) so that two triangles that are adjacent along an edge have different colours. In this case an even number of triangles should meet at each vertex.
The metric on the surface is chosen in such a way that all triangles are equivalent to an equilateral triangle in the Euclidean plane. Hence it follows immediately that at any vertex the total curvature has the form $`(2\pi N_{tr}\pi /3)`$, where $`N_{tr}`$ is the number of triangles meeting at a given vertex. If $`N_{tr}=6`$, then the curvature is equal to zero. If $`N_{tr}=4`$, then the curvature is positive. If $`N_{tr}>6`$, then the curvature is negative.
Definition 13 A vertex scalar Schrödinger operator is an operator $`L`$ that acts on a function from a vertex according to the formula
$$(L\psi )_P=\underset{P^{}}{}b_{P:P^{}}\psi _P^{}+a_P\psi _P,57$$
where summation goes over the vertices $`P^{}`$ that are nearest to $`P`$ but do not coincide with $`P`$. The condition that the operator $`L`$ is self-adjoint has the form
$$b_{P^{}:P}=\overline{b}_{P:P^{}},a_P.58$$
Definition 14 1) A triangular (black) Schrödinger operator of type I is an operator $`L`$ that acts on functions $`\psi _T`$ of the black triangles $`T`$ in such a way that
$$(L\psi )_T=\underset{T^{}}{}b_{T:T^{}}\psi _T^{}+a_T\psi _T,59$$
where $`T^{}`$ are all the black triangles that have a common vertex with $`T`$.
2) A triangular (black) Schrödinger operator of type II is an operator $`L`$ that acts on $`\psi _T`$ according to formula (159), but now $`T^{}`$ runs over all black triangles that have with $`T`$ a common neighbouring (that is, adjacent along an edge) white triangle.
Analogously we define white triangular operators of types I and II. In both cases the self-adjointness condition with respect to the standard scalar product
$$\phi ,\psi =\underset{T}{}\phi _T\overline{\psi }_T60$$
has the form
$$b_{T^{}:T}=\overline{b}_{T:T^{}},a_T.$$
###### Theorem 13
3 Any real self-adjoint vertex operator $`L`$ admits a unique factorization of the form
$$L=QQ^++w_P,$$
where
$$(Q^+\psi )_T=\underset{P}{}y_{T:P}\psi _P+x_T\psi _T,61$$
and the sum is taken over all vertices $`P`$ of the black triangle $`T`$. For any black triangle $`T`$ and two of its vertices $`P_1`$$`P_2`$ we have the equality (compare with $`(101))`$:
$$y_{T:P_1}y_{T:P_2}=b_{P_1:P_2}.62$$
We have also an analogous assertion for white triangles.
###### Theorem 14
4 A real triangular black self-adjoint operator $`\stackrel{~}{L}`$ of type I admits a factorization of the form
$$\stackrel{~}{L}=Q^+Q+v_T,63$$
where the operator $`Q^+`$ has the form $`(161)`$, if and only if for every vertex $`P`$ the matrix $`B_P=(b_{T:T^{}})`$, where the black triangles $`T`$$`T^{}`$ belong to the star of the vertex $`P`$, has the form
$$B_P=Diag+\mathrm{\Lambda }_P,rk\mathrm{\Lambda }_P=1.64$$
For multiplicities $`N_{tr}=4,6`$ this condition is always satisfied.
The operator $`Q`$ in $`(163)`$ is defined up to a transformation that does not change any product of the form
$$y_{T:P}y_{T^{}:P}(TT^{}).$$
In particular, for $`N_{tr}6`$ the coefficients $`y_{T:P}`$ are defined uniquely, and for $`N_{tr}=4`$ up to the transformation $`(TT^{})`$
$$y_{T:P}\mu _Py_{T:P},y_{T^{}:P}\mu _P^1y_{T^{}:P},$$
where $`0\mu _P`$ is any non-zero function from vertices with $`N_{tr}(P)=4`$.
###### Theorem 15
5 A triangular (black) Schrödinger operator of type II always admits a factorization of the form
$$L=QQ^++u_T,65$$
where
$$(Q^+\psi )_{T_1}=\underset{T}{}y_{T_1:T}\psi _T+x_{T_1}\psi _{T_1},$$
$`T_1`$ is a white triangle, and the sum is taken over all black triangles $`T`$ adjacent to $`T_1`$. In fact, the coefficients $`y_{T_1:T}`$ correspond to the edges of the triangulation.
We have the equalities
$$y_{T:T_1}y_{T^{}:T_1}=b_{T:T^{}}66$$
for any three triangles $`T`$, $`T_1`$$`T^{}`$ sequentially adjacent to each other (black, white, black) from the star of a vertex with $`N_{tr}6;`$
$$y_{T:T_1}y_{T^{}:T_1}+y_{T:T_1^{}}y_{T^{}:T_1^{}}=b_{T:T^{}}67$$
for black triangles $`T`$, $`T^{}`$ and white triangles $`T_1`$, $`T_1^{}`$ from the star of a vertex with $`N_{tr}=4`$.
The factorization is unique in the neighbourhood of all vertices $`P`$ such that $`N_{tr}(P)6`$. In the neighbourhood of vertices $`P`$ with $`N_{tr}(P)=4`$ the factorization is not unique: every equation of the form $`(167)`$ can be replaced by two:
$$y_{T:T_1}y_{T^{}:T_1}=b_{T:T^{}}^{(1)},y_{T:T_1^{}}y_{T^{}:T_1^{}}=b_{T:T^{}}^{(2)},68$$
where $`b_{T:T^{}}=b_{T:T^{}}^{(1)}+b_{T:T^{}}^{(2)}`$ is an arbitrary decomposition of $`b_{T:T^{}}`$ into a sum of two non-zero terms. The resulting system is uniquely soluble.
In this theorem the black and white colours can be interchanged.
Remark $`17`$ If for all vertices of the triangulation we have $`N_{tr}=4,6`$, then the classes of operators of types I and II coincide.
###### Corollary 5
$`1)`$ If for all vertices we have $`N_{tr}6`$ (that is, the curvature is non-negative), then factorizations of all types are defined. Correspondingly, all the Laplace transformations related to them are also defined. For example, for a vertex operator the Laplace transformation has the form
$$L=QQ^++w\stackrel{~}{L}=Q^+w^1Q+1.69$$
$`2)`$ Factorizations of triangular Schrödinger operators $`L`$ of type II are always defined in terms of triangles of the opposite colour. The degree of non-uniqueness of these factorizations and the corresponding Laplace transformations is determined by the number of vertices $`P`$ of positive curvature, $`N_{tr}(P)=4`$.
Remark $`18`$ For an equilateral triangular lattice in the plane the spaces of functions on vertices, on black and on white triangles, can be identified, and then these factorizations coincide (up to equivalence (104)) with the factorizations of operators on a regular lattice considered in §6.
Conclusion. Formulation of the problem on cyclic Laplace chains arises in two cases.
Case 1: for $`N_{tr}6`$ (Corollary 5).
Case 2: for vertices with $`N_{tr}=4`$ under the conditions of Corollary 5.
In case 2 we can limit ourself to Laplace transformations for operators of type II, which transform functions on black triangles into functions on white triangles and vice versa. This is similar to one-dimensional Darboux transformations.
## §9. Simplicial connections. Generalizations
We consider a simplicial complex $`K`$. Everywhere we denote a simplex of dimension $`l`$ by $`\sigma ^l`$.
Definition 15 A simplicial connection of type $`(q,j,k)`$, $`0j<k`$, is an equation
$$\underset{\sigma ^q\sigma ^{q+k}}{}c_{\sigma ^{q+k}:\sigma ^q}\psi _{\sigma ^q}=0,70$$
determined by a vector-function which ascribes to every pair of simplexes $`\sigma ^q,\sigma ^{q+k}K`$ such that $`\sigma ^q\sigma ^{q+k}`$ a collection
$$c_{\sigma ^{q+k}:\sigma ^q}=(c_{\sigma ^{q+k}:\sigma ^q}^\alpha )^m,\alpha =1,\mathrm{},m,71$$
where
$$m=C_{q+k+1}^{q+1}C_{q+j+1}^{q+1}.$$
We require that the function $`c_{\sigma ^{q+k}:\sigma ^q}`$ satisfies the conditions for non-degeneracy and localization (see below).
Such a function defines an operator $`Q^+`$ which transforms the space of numerical functions of simplexes $`\sigma ^q`$ into the space of $`m`$-vector-functions of simplexes $`\sigma ^{q+k}`$:
$$(Q^+\psi )_{\sigma ^{q+k}}=\underset{\sigma ^q}{}c_{\sigma ^{q+k}:\sigma ^q}\psi _{\sigma ^q}^m.72$$
Simplicial connection depends only on the zero-space of this operator:
$$Q^{+\alpha }\psi =0,\alpha =1,\mathrm{},m.73$$
In fact, this relation is written separately in each simplex $`\sigma ^{q+k}`$.
Requirement of non-degeneracy. Equation (197) should be such that for any subsimplex $`\sigma ^{q+k}K`$ an arbitrarily given collection of values $`\psi _{\sigma ^q}`$ in simplexes $`\sigma ^q`$, forming a $`q`$-dimensional skeleton of any simplex $`\sigma ^{q+j}\sigma ^{q+k}`$, should uniquely and consistently determine values of $`\psi `$ on all remaining $`q`$-dimensional subsimplexes in $`\sigma ^{q+k}K`$. Thus, we can arbitrarily define values of $`\psi `$ in the $`q`$-dimensional skeleton of simplexes $`\sigma ^{q+j}\sigma ^{q+k}`$.
###### Lemma 20
0 Suppose we are given a path $`\gamma `$ composed of $`(q+k)`$-dimensional simplexes $`\sigma _1^{q+k},\sigma _2^{q+k},`$, where $`\sigma _s^{q+k}`$ and $`\sigma _{s+1}^{q+k}`$ intersect exactly along a face of dimension $`q+j`$ for all $`s`$. Then the simplicial connection consistently defines a solution of $`(170)`$ along the path $`\gamma `$, starting from arbitrary initial data defined in a $`q`$-dimensional skeleton of an arbitrary $`(q+j)`$-dimensional face of any of the simplexes $`\sigma _s^{q+k}`$ that make up this path.
We obtain the proof of the lemma in a simple way from the definitions: a solution is constructed by transition from the simplex $`\sigma _s^{q+k}`$ to the simplex $`\sigma _{s+1}^{q+k}`$, $`s=1,2,\mathrm{}`$ .
The analogue of curvature arises naturally for ‘closed’ paths $`\gamma `$ in which $`\sigma _1^{q+k}=\sigma _N^{q+k}`$. In this case there arises a ‘holonomy transformation’ of the simplicial connection (170) along the path $`\gamma `$: solving (170) along the path according to the lemma, starting from some face $`\sigma _0^{q+j}\sigma _1^{q+k}`$, we finally find the value of $`\psi `$ on all $`q`$-dimensional faces of the simplex $`\sigma _N^{q+k}`$, including $`\sigma ^q\sigma _0^{q+j}`$. These values may not coincide with the initial values. There arises the linear transformation
$$R_\gamma ^M^M,$$
where $`M=C_{q+j+1}^{q+1}`$ is the number of $`q`$-dimensional faces in the simplex $`\sigma _0^{q+j}\sigma _1^{q+k}`$ on which the values of the function $`\psi _{\sigma ^q}`$ were given arbitrarily.
Starting from the requirement of non-degeneracy we can reduce equation (170) in any simplex $`\sigma ^{q+k}`$ to such a form that inverse operators are defined for any pair of simplexes $`\sigma _1^{q+j},\sigma _2^{q+j}\sigma ^{q+k}`$:
$$L_{12}_1^M_2^M,$$
where the space $`_\epsilon ^M`$ consists of all possible values of $`\psi `$ on simplexes $`\sigma ^q\sigma _\epsilon ^{q+j}`$, $`\epsilon =1,2`$, and the bases in them are $`\delta `$-functions of the simplexes $`\sigma ^q`$.
Requirement of localization. All operators $`L_{12}`$ are uniquely defined only by the minimal simplex containing $`\sigma _1^{q+j}`$ and $`\sigma _2^{q+j}`$:
$$(\sigma _1^{q+j}\sigma _2^{q+j})\sigma ^{q+j+s}\sigma ^{q+k}.$$
This means that if two simplexes $`\sigma _1^{q+k}`$ and $`\sigma _2^{q+k}`$ intersect along the simplex $`\sigma ^{q+j+s}`$, where $`s>0`$, then for them all the operators $`L_{12}`$ inside the simplex $`\sigma ^{q+j+s}`$ coincide.
The requirment of localization is automatically satisfied in two cases.
Case 1. $`k=j+1`$. Here always $`\sigma ^{q+j+1}=\sigma ^{q+k}`$.
Case 2. In a simplicial complex $`K`$ the simplexes $`\sigma _1^{q+k}`$ and $`\sigma _2^{q+k}`$ coincide if they intersect in a face of dimension greater than $`q+j`$.
In the examples already considered in this work (above) we have always had $`q=0`$, that is, functions $`\psi `$ were defined on vertices. Localization corresponded to cases 1 or 2.
A trivial example Let $`k=1,q=0,j=0`$. We have the usual Abelian connection, sitting on edges of the complex $`K`$. In fact, from (170)
$$c_{\sigma ^1:\sigma _1^0}\psi _{\sigma _1^0}+c_{\sigma ^1:\sigma _2^0}\psi _{\sigma _2^0}=0$$
we arrive in this case at the operator of multiplication
$$L_{12}=\frac{c_{\sigma ^1:\sigma _1^0}}{c_{\sigma ^1:\sigma _2^0}}\psi _{\sigma _1^0}\psi _{\sigma _2^0}=L_{12}(\psi _{\sigma _1^0}).$$
Example 13 a) Let $`q=0,k=2,j=1`$ and let $`K`$ be a triangulation of a two-dimensional surface. Here we have one equation. This situation was studied above (see §6).
For $`K=^2`$ with an equilateral triangular lattice we paid particular attention to the case of ‘zero local curvature’ where a solution of the equation $`Q^+\psi =0`$ reduces to a difference equation of the second order on the straight line (see Example 11). According to a hypothesis of the authors, this situation arises in some interesting cases when the point $`\lambda =0`$ is supposedly the lower bound of the continuous spectrum of the Schrödinger operator (123). On surfaces with non-trivial topology the case of zero local curvature leads to the global monodromy defined on the group $`\pi _1`$.
b) Let $`q=0,j=1,k=2`$ and let $`K`$ be the ‘black part’ (or ‘white part’) of the black and white triangulation of a surface. Here we also have one equation $`Q^+\psi =0`$, but we do not have local curvature, since black triangles are adjacent to each other at vertices (there are no common edges). Therefore there are no non-trivial paths.
c) Let $`q=0,j=0,k=2`$, and let $`K`$ be the same as in b). Here we have curvature.
Example 14 Let $`q=0`$, $`k=3`$, and let $`K`$ be the lattice of regular tetrahedra in $`^3`$. We have $`K=K_1K_2`$, where $`K_1`$ is the ‘black’ part, $`K_2`$ is the ‘white’ part, and $`K_1K_2`$ is a one-dimensional skeleton. Adjacency of two tetrahedra in $`K`$ is possible along no more than an edge, and adjacency of tetrahedra inside $`K_1`$ (or $`K_2`$) is only possible at a vertex. We have already considered the following cases a), b), see §7:
a) $`j=2`$, complex $`K_1`$ (or $`K_2`$). There is no local curvature here.
b) $`j=2`$, complex $`K`$. There is also no local curvature here, since there are no paths $`\gamma `$. Nevertheless, there is a non-trivial condition for consistency (153) which ensures an extensive space of solutions of the equation $`Q^+\psi =0`$. This particular analogue of curvature has not been studied in a general form.
c) $`j=1`$, complex $`K`$. Here the concept of local curvature arises, since we have many paths $`\gamma `$. The system (170) has a solution in the whole of $`^3`$, which is uniquely determined by an arbitrary pair of values of $`\psi `$ at the vertices of any edge if and only if the following condition is satisfied: eight equations, corresponding to each set of four tetrahedra, a pair of black and a pair of white, from the star of one vertex, such that both black tetrahedra adjoin each of the white ones along an edge, are linearly dependent. This assertion means that our system in the star of every vertex has a two-dimensional local space of solutions. Geometrically, ‘curvatures’ that obstruct this solution correspond on the border of the star—a sphere $`S^2`$ divided into triangles and squares—just to squares. In all we have six conditions at each vertex (the number of squares), but one of them is dependent.
d) Let $`q=0,j=0`$, and let $`K`$ be $`K_1`$. There are paths here composed of tetrahedra, linked along vertices. The space $`^M`$ for $`j=0`$ is one-dimensional, and we arrive only at Abelian connections. This is true for $`j=0`$ in any simplicial complexes for any $`q`$.
###### Lemma 21
1 A simplicial connection of type $`(q,j,k)`$ that satisfies the conditions for non-degeneracy and localization defines a multiplicative curvature transformation
$$R_\gamma ^M^M,R_{\gamma _1\gamma _2}=R_{\gamma _1}R_{\gamma _2},R_{\gamma ^1}=R_\gamma ^174$$
for any path $`\gamma `$,
$$\gamma =\sigma _1^{q+k}\sigma _2^{q+k}\mathrm{}\sigma _N^{q+k},$$
where $`\sigma _N^{q+k}=\sigma _1^{q+k}`$ and the intersection $`\sigma _s^{q+k}\sigma _{s+1}^{q+k}`$ is a face of dimension greater than or equal to $`q+j`$ for all $`s=1,\mathrm{},N`$.
The proof of the lemma is easily obtained from the previous one.
We now consider real discrete self-adjoint operators
$$(L\psi )_{\sigma _2^q}=\underset{\sigma _1^q}{}b_{\sigma _1^q:\sigma _2^q}\psi _{\sigma _2^q},75$$
acting on functions of $`q`$-simplexes of the complex $`K`$. We consider only ‘operators of the second order’, where the coefficients $`b_{\sigma _1^q:\sigma _2^q}`$ are different from zero only for ‘nearest neighbours’.
Definition 16 The simplexes $`\sigma _1^q`$, $`\sigma _2^q`$ are $`k_+`$-nearest ($`k_{}`$-nearest) if this pair is contained in some simplex $`\sigma ^{q+k_+}`$ (respectively, if the intersection $`\sigma _1^q\sigma _2^q`$ is a simplex $`\sigma ^{qk_{}}`$).
We can represent the operator $`L`$ as the sum of operators of two types $`L=L_++L_{}`$. We consider these types separately;
$$(L_+\psi )_{\sigma _2^q}=\underset{\sigma _1^q,\sigma _2^q\sigma ^{q+k_+}}{}b_{\sigma _2^q:\sigma ^{q+k_+}:\sigma _1^q}\psi _{\sigma _1^q},$$
$$(L_{}\psi )_{\sigma _4^q}=\underset{\sigma _3^q\sigma _4^q\sigma ^{qk_{}}}{}b_{\sigma _4^q:\sigma ^{qk_{}}:\sigma _3^q}\psi _{\sigma _3^q},$$
where $`\sigma _1^q,\sigma _2^q`$ are $`k_+`$-nearest and $`\sigma _3^q,\sigma _4^q`$ are $`k_{}`$-nearest. Let $`k_+=k_{}=k`$. We consider an operator $`Q^+`$ of the same type as in the definition of a simplicial connection, and the operator $`Q_1`$:
$$(Q^+\psi )_{\sigma ^{q+k}}=\underset{\sigma ^q}{}c_{\sigma ^{q+k}:\sigma ^q}\psi _{\sigma ^q},(Q_1\psi )_{\sigma ^{qk}}=\underset{\sigma ^q}{}c_{\sigma ^q:\sigma ^{qk}}\psi _{\sigma ^q}.$$
Here the images are vector-functions, and the $`\psi `$ are scalars.
Definition 17 1) A factorization of the first type is a representation of $`L_+`$ in the form
$$L_+=QQ^++w,76$$
where $`Q^+`$ is the operator adjoint to $`Q`$ and $`w=w_{\sigma ^q}`$ is the operator of multiplication by a numerical function.
A special factorization of the first type is a representation (176), where $`w=\text{const}`$.
2) A factorization of the second type is a representation of the operator $`L_{}`$ in the form
$$L_{}=Q_1^+Q_1+v,77$$
where $`v`$ is the operator of multiplication by a function of $`\sigma ^q`$ and the operator $`Q_1^+`$ is adjoint to $`Q_1`$.
A special factorization of the second type is the case when $`v=\text{const}`$.
3) A Laplace transformation of the operators $`L_\pm `$ is defined by the formulae
$$\begin{array}{cc}\hfill L_+& \stackrel{~}{L}_+=Q^+w^1Q+1,\hfill \\ \hfill L_{}& \stackrel{~}{L}_{}=Q_1v^1Q_1^++1,\hfill \end{array}78$$
and their zero eigenvectors are transformed by well-known formulae into the eigenvectors of $`\stackrel{~}{L}_\pm `$:
$`\psi ^+\stackrel{~}{\psi }^+=Q^+\psi ^+,\stackrel{~}{L}_+\stackrel{~}{\psi }^+=0,L_+\psi ^+=0,`$
$`\psi ^{}\stackrel{~}{\psi }^{}=Q_1\psi ^{},\stackrel{~}{L}_{}\stackrel{~}{\psi }^{}=0,L_{}\psi ^{}=0.`$
Obviously, the operator $`\stackrel{~}{L}_+`$ is written analogously to $`L_{}`$, where $`\stackrel{~}{Q}^+=v^{1/2}Q^+`$. These formulae assume that $`w0`$, $`v0`$, although this does not play a serious role here.
Detailed studies have been devoted above to the investigation of factorizations and Laplace transformations in particular cases. Here we consider only the simplest example, that is, a standard discrete real Schrödinger operator.
Example 15 Let $`q=0`$, $`k=1`$, $`j=0`$. We have an operator defined on functions of vertices
$$L=L_+,(L\psi )_{\sigma _2^0}=\underset{\sigma _1^0\sigma _2^0=\sigma ^1}{}b_{\sigma _2^0:\sigma ^1:\sigma _1^0}\psi _{\sigma _1^0}+b_{\sigma _1^0}\delta (\sigma _2^0,\sigma _1^0)\psi _{\sigma _1^0}.$$
(The last term is present only for $`\sigma _2^0=\sigma _1^0`$; this is shown by the $`\delta `$-function.)
We can always factorize such an operator in the form (we always suppose that $`b_{\sigma _2^0:\sigma ^1:\sigma _1^0}0`$)
$$L=QQ^++w,79$$
where $`w`$ is some function. This factorization is not unique.
Question Is there always a ‘special factorization’, where $`w=\text{const}`$? It exists for the complex $`K=^1`$, see §3, but in the general case the answer is not clear (we assume that $`b_{\sigma ^0}0`$).
Remark $`19`$ Earlier (see §8) in connection with the Laplace transformation on surfaces with black-white triangulation we considered yet another form of the factorization: let the complex $`K`$ and two subcomplexes $`K_1`$$`K_2`$ be given, where $`K_1K_2`$ is a $`(q1)`$-dimensional skeleton of the complex $`K`$, and the operator $`L`$ acts on functions of $`\sigma ^qK_1`$ (white $`q`$-simplexes):
$$(L\psi )_{\sigma _2^q}=\underset{\sigma _1^q}{}b_{\sigma _1^q:\sigma _2^q}\psi _{\sigma _1^q},\sigma _1^q,\sigma _2^qK_1.$$
The notion of the ‘nearest $`k`$-neighbourhood’ is defined for simplexes from $`K_1`$ by means of faces of dimension $`qk`$, which are common for them with an arbitrary ‘black’ $`q`$-simplex from $`K_2`$. The factorization is sought in the form (179), where $`w`$ is the operator of multiplication by a function, and $`Q^+`$ is an operator of the form
$$(Q^+\psi )_{\overline{\sigma }^q}=c_{\overline{\sigma }^q:\sigma ^{qk}:\sigma ^q}\psi _{\sigma ^q},80$$
where $`\overline{\sigma }^qK_2`$, $`\sigma ^qK_1`$, $`\sigma ^{qk}\overline{\sigma }^q\sigma ^q`$. Analogously, the operators $`\overline{L}`$ act on ‘black’ $`q`$-simplexes $`\overline{\sigma }^qK_2`$ and the factorization is sought in the form
$$\overline{L}=Q^+Q+v.$$
Previous formulae define Laplace transformations and also special Laplace transformations if the potentials $`v,w`$ are constant.
Example 16 Let $`K`$ be a $`q`$-dimensional triangulated manifold with black-white coloured $`q`$-dimensional simplexes, $`K=K_1K_2`$, and let $`K_1K_2`$ be a $`(q1)`$-dimensional skeleton. For $`k=1`$ we have $`\sigma ^{q1}`$ as a common face of exactly two simplexes $`\overline{\sigma }^q,\sigma ^q`$, that is, black and white. By analogy with the case $`q=2`$ (see §8), the condition for factorization depends on the $`N(\sigma ^{q2})`$-multiplicity of $`(q2)`$-dimensional simplexes $`\sigma ^{q2}K`$, which is equal to the number of adjacent $`(q1)`$-dimensional $`\sigma ^{q1}\sigma ^{q2}`$. This number is even, $`N4`$. For $`\sigma _j^qK_1`$, $`\overline{\sigma }_j^qK_2`$ we have:
* $`b_{\sigma _1^q:\sigma _2^q}=c_{\sigma _1^q:\sigma _1^{q1}:\overline{\sigma }^q}c_{\sigma _2^q:\sigma _2^{q1}:\overline{\sigma }^q}`$, $`\sigma ^{q2}=\sigma _1^{q1}\sigma _2^{q1}`$, $`N(\sigma ^{q2})6`$;
* $`b_{\sigma _1^q:\sigma _2^q}=c_{\sigma _1^q:\sigma _1^{q1}:\overline{\sigma }_1^q}c_{\sigma _2^q:\sigma _2^{q1}:\overline{\sigma }_1^q}+c_{\sigma _1^q:\sigma _3^{q1}:\overline{\sigma }_2^q}c_{\sigma _2^q:\sigma _4^{q1}:\overline{\sigma }_2^q}`$, $`\sigma ^{q2}=\sigma _1^{q1}\sigma _2^{q1}\sigma _3^{q1}\sigma _4^{q1}`$, $`N(\sigma ^{q2})=4`$.
Let us consider a): the number of equations is equal to the number of $`(q2)`$-dimensional faces of the simplex $`\overline{\sigma }^q`$, that is, $`l=q(q+1)/2`$. The number of unknowns is $`m(q+1)`$, where $`m`$ is the dimension of the vector $`c_{\sigma ^a:\sigma ^{q1}:\overline{\sigma }^q}`$. Thus we have the condition $`m(q+1)q(q+1)/2`$ or $`mq/2`$. The scalar factorization ($`m=1`$) needs to satisfy additional conditions for $`q3`$.
Example 17 We again consider $`K`$, a $`(q+2)`$-dimensional manifold with ‘black-white’ coloured $`(q+2)`$-simplexes, $`K=K_1K_2`$, and $`K_1K_2`$, a $`(q+1)`$-skeleton.Suppose we are given an operator of some other type, acting on functions of $`q`$-simplexes from $`K`$
$$(L\psi )_{\sigma _2^q}=\underset{\sigma _1^q}{}b_{\sigma _2^q:\sigma _1^q}\psi _{\sigma _1^q},$$
and we seek a ‘white’ factorization (179), where $`w`$ is the operator of multiplication by a function, and
$$(Q^+\psi )_{\sigma ^{q+2}}=\underset{\sigma ^q}{}c_{\sigma ^{q+2}:\sigma ^q}\psi _{\sigma ^q},\sigma ^{q+2}K.$$
We arrive at the relation
$$b_{\sigma _1^q:\sigma _2^q}=c_{\sigma ^{q+2}:\sigma _1^q}c_{\sigma ^{q+2}:\sigma _2^q}.$$
The number of unknowns is equal to $`m(q+2)(q+3)/2=lm`$ (if $`c_{\sigma ^{q+2}:\sigma _1^q}`$ is an $`m`$-vector) and the number of equations is equal to $`l(l1)/2`$. The factorization condition thus has the form
$$lml(l1)/2,81$$
$`(q+2)(q+3)/2=l`$, $`m(l1)/2`$.
For $`q=1,k=2`$ we see that $`m=3`$ or $`5`$ is acceptable. In these cases factorization is always possible in the form (179). The possibility of the special factorization $`w=\text{const}`$ is of interest: one would like to clarify this question. If the factorization is special, then the question of zero modes of the operator $`L=QQ^+`$ is of interest, that is, the question of the solubility in the space $`\psi _2`$ of the equation $`Q^+\psi =0`$. For $`m=3`$ and $`m=5`$ equations of this type determine the connections (if the conditions for non-degeneracy and localization are satisfied). Paths along which ‘parallel transport’ is realized and curvature is defined consist of 3-simplexes ($`q=1`$) adjoining each other along edges ($`m=5`$) or along faces ($`m=3`$). The requirement of localization is automatically satisfied for the case $`m=5`$, $`K=K_1`$ (white tetrahedra).
If $`m=3`$, then purely white paths do not exist. The curvature is defined by the pair of equations: $`Q_1^+\psi =0`$ (white part), $`Q_2^+\psi =0`$ (black part), that is, $`K=K_1K_2`$.
Now suppose we are given an arbitrary simplicial connection on a complex $`K`$ of type $`(q,j,k)`$.
Definition 18 The local curvature of the vertex $`\sigma ^0K`$ is the curvature of this connection on special paths of the subcomplex $`K_P`$, which is the simplicial star of the vertex $`\sigma ^0`$: we take all simplexes $`\sigma _c^q`$ with vertex $`\sigma ^0`$, $`\sigma ^qK_P`$, paths of the form
$$\gamma =(\sigma _1^{q+k},\sigma _2^{q+k},\mathrm{},\sigma _N^{q+k}=\sigma _1^{q+k}),82$$
such that $`\sigma _j^{q+k}=\sigma ^0\sigma _j^{q1+k}`$.
A transformation $`R_\gamma `$ determined only by paths $`\gamma K_P`$ (182) is called the local curvature of the simplicial connection at the point $`\sigma ^0K`$.
Hypothesis For manifolds $`K`$ the curvature is trivial if the local curvatures of all the vertices are trivial (we note that in a number of cases above $`K`$ was not a manifold). |
warning/0003/nlin0003030.html | ar5iv | text | # The effect of synchronized area on SOC behavior in a kind of Neural Network Model 1footnote 11footnote 1The project supported by P.R.China National Basic Research Project ”Nonlinear Science”
## I Introduction
A few years ago, Bak, Tang and Wiesenfeld introduced the concept of the ”Self-Organized Criticality” (SOC) in sand pile model 1 . From then on, this concept has been widely studied in some extended dissipative dynamical systems, such as earthquake 2 , biology evolution 3 , and so on. It is shown that all these systems can naturally evolve into a ”critical state” with no intrinsic spatial and temporal scales through a self-organized process without the need to fine-tune parameters of the system. This critical state is characterized by a power-law distribution of avalanche sizes, where the size is the total number of toppling events or unstable units.
Now,the research on SOC has come into a new level. One has studied several factors’ influence on SOC behavior, such as network size, periodic or nonperiodic boundary conditions,local dynamics variable is conservative or not, and so on. Many investigators believe that it is intrinsic-stability(order) and variability(disorder)’s common action make the system evolves into a ”frozen disorder” (SOC) state 4 ; 5 ; 6 . It is the combination feature of stability and variability, and it’s complex spatial-temporal dynamic behavior, make the system in SOC state have maximum complexity and latent computing potency.
The brain is a complex system and its information process has the properties of stability and variability – on one hand, there are relative stable information stored mechanism, and stored area in brain(such as feature area in cortex) ; on the other hand, the brain is influenced by the environment and one should continuous update knowledge and concepts. The similarity between the SOC systems and the brain has lead us to study Artifical Neural Network(ANN) and SOC together. There is some SOC behavior shown in the neuron network model introduced by our group 7 .
The brain is also a complex system with highly complexity, highly order and special structure. The structure must have the effect on the brain’s dynamics behavior. Our neuron network can also produce some special structure, so we believe that the SOC behavior shown in our model must have it’s own special behavior and rule.
In this paper, the feature area(produced by self-organized process)’s definite effect on SOC behavior has been studied.
## II Model
Here we propose a two-dimensional neural network model of square lattice. This model is a kind of serial self-organized neural network model, based on the LISSOM model 8 . When a $`h`$\- dimensional vector $`\zeta `$ is inputted, the state $`\eta _{ij}(t)`$ of the neuron $`(i,j)`$ at time $`t`$ is changed according to the formula:
$`\eta _{ij}`$ $`=`$ $`\sigma \{{\displaystyle \underset{h}{}}\mu _{ij,h}\zeta _h+\gamma _e{\displaystyle \underset{kl}{}}E_{ij,kl}\eta _{kl}(t1)`$ (1)
$`\gamma _i{\displaystyle \underset{k^{}l^{}}{}}I_{ij,k^{}l^{}}\eta _{k^{}l^{}}(t1)\}`$
$`=`$ $`\sigma \{f_{ij}(t)\}`$
where $`\sigma (x)`$ is an active function, and we design it as sign function, i.e., if $`x0`$, then $`\sigma (x)=1`$ , otherwise $`\sigma (x)=1`$ . $`\mu _{ij,h}`$ is an afferent input weight vector; $`E_{ij,kl}`$ is the excitatory lateral connection weight on the connection from the neuron $`(k,l)`$ to the $`(i,j)`$ neuron; $`I(ij,kl)`$ is the inhibitory connection weight. $`f_{ij}(t)`$ is the local field of the neuron $`(i,j)`$ at time $`t`$. $`\gamma _e`$ and $`\gamma _i`$ are constant factors. The adjustment of those three connection weights is as following according to the dynamic Hebb rule:
$`\mu _{ij,h}(t+1)={\displaystyle \frac{\mu _{ij,h}(t)+\alpha \eta _{ij}\xi _h}{\{_h\left[\mu _{ij,h}(t)+\alpha \eta _{ij}\xi _h\right]^2\}^{1/2}}}\text{,}`$
$`E_{ij,kl}(t+1)={\displaystyle \frac{E_{ij,h}(t)+\alpha _E\eta _{ij}\eta _{kl}}{_{ij}\left[E_{ij,kl}(t)+\alpha _E\eta _{ij}\eta _{kl}\right]}}\text{,}`$
$`I_{ij,k^{}l^{}}(t+1)={\displaystyle \frac{I_{ij,k^{}l^{}}(t)+\alpha _I\eta _{ij}\eta _{k^{}l^{}}}{_{k^{}l^{}}\left[I_{ij,k^{}l^{}}(t)+\alpha _I\eta _{ij}\eta _{k^{}l^{}}\right]}}\text{.}`$ (2)
where $`\alpha ,\alpha _E,\alpha _I`$ are the learning rates.
Note the change of the neuron states is quick and the adjustment of the connection weights is slow. Usually, after over 10 iterations of the neuron states when any pattern is inputted (at this time, the network state becomes an attractor in the state space, often a fixed point.), all connection weights are updated once. Thus, after learning a while, the lateral connection weight self-evolves into the”Mexican hat” profile,the afferent input weight self-organizes into a topological map of the input space, the neuron network can produce some special feature areas, and the state of the neural network is evolved from disordered case to stable and topological case in state space.
Then we introduce the following interactive process between the neurons, similar with the pulse coupled interaction:
1) When the neuron $`(i,j)`$ is stable, i.e., $`\eta _{ij}(t)=\sigma \{f_{ij}(t1)\}`$ , it doesn’t influence the others;
2) If the neuron $`(i,j)`$ is unstable, i.e., $`\eta _{ij}\sigma \{f_{ij}(t1)\}`$ or $`f_{ij}(t1)=0`$ , then the nearest neighbors $`(i^{},j^{})`$ around this unstable neuron will receive a pulse respectively and their local fields will be changed. At the same time, the neuron $`(i,j)`$ becomes stable again, depending on the formula (1);
3) When all neurons of the neural network are stable, we choose the minimum $`g`$ among the absolute values of all local fields $`f_{ij}`$ and drive the local field of every neuron, i.e.,
$$f_{ij}f_{ij}c\eta _{ij}g$$
(3)
where $`c`$ is a constant.
Now, we present the computer simulation procedure of this model in detail:
1) Variable initialization. In the 2-dimensional $`n\times n`$ neural network model, let the initiatory state and local field equal $`0`$; random initialize each connection weight among $`[1,1]`$ ; and produce $`M`$ random input patterns, $`\zeta _j^i[1,1]`$, $`(i=1,2,\mathrm{}M;j=1,2,\mathrm{}h)`$.
2) Learning process. According to formula (1), we input the pattern circularly and iterate the neuron state and local field. After period of time, the space state of the network reaches stability and we consider the $`M`$ input patterns have been stored.
3) Associative memory. Input a new pattern, and then search the unstable neuron $`(i,j)`$ as defined above in whole neural network. Due to being unstable, the neuron $`(i,j)`$ discharges a pulse to the each nearest neighbor $`(i^{},j^{})`$ and thus causes the local fields of them to change as following:
$$f_{i^{}j^{}}f_{i^{}j^{}}\frac{\gamma }{2}\eta _{i^{}j^{}}(1+|f_{ij}|)$$
(4)
where $`\gamma `$ represents the pulse intensity, symbol $`||`$ denotes absolute value.
Simultaneously, according to formula (1), the neuron $`(i,j)`$ becomes stable as: $`\eta _{ij}\sigma (f_{ij})`$ , $`f_{ij}\eta _{ij}`$ ; where $`\sigma (x)`$ is sign function.
Repeat this procedure until all neurons of the model are stable. Define one avalanche as all unstable neurons in this process. Then begin drive process by formula (3) and new avalanche.
## III Simulation results
Recently, Bak and Sneppen have investigated the power law distribution $`P(X)`$ of the distances $`X`$ between subsequent unstable sites in lattice of BS biology evolution model 3 , and J.De.Boer et al. have found some interested result with it 9 .So in this paper we studied not only the distribution $`P(S)`$ of the avalanche sizes $`S`$ but also the distribution $`P(X)`$ of the distances $`X`$ between the subsequent unstable sites . We find the distribution will deviate from the power law in some conditions.
### III.1 The effect of synchronized area on SOC behavior
The size of our lattice is $`40\times 40`$.We find that in associative memory process, the distribution of the avalanche sizes has power-law behavior, $`P(S)S^\tau `$ , $`\tau 0.90`$ . It is shown in Fig.1.a. The distribution $`P(X)`$ of the distances $`X`$ between the subsequent unstable sites has power-law behavior too, $`P(X)X^\beta `$ , $`\beta 2.23`$ , it is shown in Fig.1.b.
The SOC behavior changed with the scope of lateral connection has been studied. We increase the excitatory lateral connection radius $`d_e`$ and the inhibitory connection radius $`d_i=3d_e`$. It can be seen in Fig 1.a that avalanche size and occurring probability of large scale avalanche decrease with the slope $`d_e`$ increasing, and the distribution of $`P(X)`$ deviates from power law more and more in large $`X`$, the probability of large distance $`X`$ between the subsequent unstable sites also increases, we consider it is a deviation from SOC behavior, it is shown in Fig 1.b.
By investigating , we think our system is in a ”partly-synchronized” state, hence the dynamics behavior mentioned above can be seen.
A.Corral et al. propose SOC state and synchronization state might be considered as two uttermost state of system (just like two sides of the same coin) 4 . The inhomogeneity introduced by boundary or initialization conditions can propagate into interior of network, hence makes the system evolve into SOC state 10 ; 11 . When inhomogeneity is not large enough, the system finds a compromise between synchronization and SOC 6 . It can be considered as a partly-synchronized state.
If Our Model is only a pure OFC model without learning process, the system will present a macroscopic SOC behavior among almost all the lattices, but it is also a neuron network model. As a kind of self-organized feature map model, after learning a while, it’s neurons will develop a unique lateral interaction ”Mexican hat” profile that represents its long-term associations with each other. the afferent input weights will self-organize into a topological map of the input space 8 , it can form some special topological feature regions. We consider that in these regions, as the connection weights become more topographically ordered, neuron’s synchronization effect between each other will be reinforced. When order in these regions is applied to the model, the system has a tendency from SOC state to synchronized state. At last, the system finds a compromise between synchronization and SOC, it could be seen as ”partly-synchronized” state. With the process, the distribution of avalanche varies from the continuous distribution to a discrete one, the possibility of large scale avalanche propagating into the interior of the synchronization regions will reduce, and occurring probability of large scale avalanche will decrease too, the one-off isolated avalanche(only one unstable site in an avalanche)in these regions will increase greatly, it makes the distances $`X`$ between the subsequent unstable sites have a stochastic spatial even distribution in the area. This distribution has more effect on probability of large distance between unstable sites than probability of small distance. (Because probability of small distance is larger than one of large distance.)
So at this moment, there are some areas in synchronized state and another regions in SOC state. We can consider that with the increasing of $`d_e`$, the feature region(synchronized region) produced by self-organized process becomes wider, which results in the whole system dynamics deviating largely from SOC state, it can be seen in Fig.1.
To verify the idea mentioned above, we draw the avalanche’s distribution map of the whole system. We draw one avalanche’s distribution snapshot every 1000 avalanches, then overlap all the snapshots in one picture. The result can be seen in Fig2. It can be clearly seen that the blank region (seldom avalanche area) expands with the increasing of $`d_e`$. It means that the synchronized region introduced by self-organized expands and large scale avalanches reduce more and more. Even though, there are still some isolated unstable neurons in the area, it indirectly verifies our deduction of $`p(X)`$ distribution mentioned above.
We investigate the relation between average avalanche size $`S`$ and radius $`d_e`$. From Fig.3 , we can see that with the decreasing of $`d_e`$ , $`S`$ will increases, and with $`d_e`$ approaching 0, the slope of the curve becomes quite large. This result is approximate to the phenomena with the increasing of pulse discharging intensity $`\gamma `$ 7 , and is consistent with the result in Ref. 12 . It implies that the network approaches to SOC state when $`d_e`$ approaching 0 or $`\gamma `$ approaching 0.5.
When the ratio $`T`$ of radius $`d_i`$ with radius $`d_e`$ is changed , the probability distribution $`P(S)`$ and $`P(X)`$ and avalanche distribution are changed too, the tendency is similar to Fig.1, Fig.2. With the increment of $`T`$, the partly-synchronized behavior of system becomes distinctness. The phenomena can also be explained by the expanding of synchronized region , but it leads us to think that the inhibitory lateral connection and excitatory lateral connection have what different effect on synchronized process? It is look like that the inhibitory lateral connection has more important effect, but it is not very clear, there are still a lot of work to do.
### III.2 The approximate behavior in a kind of quasi-OFC earthquake model
OFC earthquake model is a kind of SOC model which has been widely studied in recent years 2 .If we use transformation in our model : $`1\eta _{ij}f_{ij}F_{ij}`$ , then formula (4) become $`F_{i^{}j^{}}F_{i^{}j^{}}+\frac{\gamma }{2}F_{ij}`$ . It is the avalanche mechanism of the OFC earthquake model , in fact these two models belong to the same class.Therefore we add some synchronized regions in OFC model and examine that if the system has the similarity partly-synchronized behavior as in our neuron network model. Hence we introduced a cellular automaton model based on OFC model, the steps are as follows:
1) We define a $`N\times N`$ square lattice ,and a $`L\times L`$ square area in the lattice ( $`L=KN<N`$) .
2) Initialize all sites to a random value $`F_{ij}`$ between 0 and 1.
3) If any $`F_{ij}F_{th}=1`$,then redistribute the force on $`F_{ij}`$ to its neighbors according to the rule:
$`F_{i^{}j^{}}F_{i^{}j^{}}+{\displaystyle \frac{\gamma }{2}}F_{ij}\{\begin{array}{cc}\gamma =\alpha <0.5\hfill & ,whensite(i,j)L\times Lregion,\hfill \\ \gamma =0.5\hfill & ,otherwise.\hfill \end{array}`$ (7)
$`F_{ij}0`$
4) Repeat step 3 until no $`F_{ij}F_{th}`$,we define the avalanche is fully evolved.
5) Locate the site with the largest strain ,$`F_{max}`$ ,then drive all sites
$$F_{ij}F_{ij}+(F_{th}F_{max})$$
(8)
and return to step 3. Here we use open boundary conditions ,and serial working mode.It is same as our neuron network model but different from traditional OFC model.
We still focus on the spatial distribution of $`S`$ and $`X`$. Let $`\alpha =0.1`$, then change $`K`$, the result is shown in Fig 4. It can be seen that the probability distribution $`P(X)`$ increases and deviates from power-law more with $`K`$ increasing at larger $`X`$ in Fig 4.a. The tendency is similar to that in our previous neuron network model. In Fig 4.b , we can see the occurring probability of large scale avalanche reduces with $`K`$ increasing too, but the phenomena is not distinctness as that in previous model. The reason may be that the previous model has more complex structure than this model. It produces many pieces of synchronized region, but here, we only add a piece of synchronized region in model.
A. Corral, Grassberger et al. have investigated the influence of pulse discharging intensity $`\gamma `$ in OFC model. They propose the model present a macroscopic synchronization among all the elements of the lattice when $`\gamma `$ is small 4 ; 13 . Therefore in this model , with expanding of $`L\times L`$ area($`\gamma =\alpha =0.1`$), the synchronized area in the network expands, the synchronization behavior becomes distinctness, and the whole system evolves into a partly-synchronized state. The dynamics behavior of the previous model is so consistent with this model, it suggested that maybe they have the similar dynamics mechanism . Our results tend to agree with this idea.
For studying the partly-synchronized phenomena further, the relation between distribution $`P(x)`$ and $`\alpha `$ has been investigated. Let $`K=0.8`$ , the partly-synchronized behavior become distinctness with $`\alpha `$ decreasing from 0.5 . But when $`\alpha `$ less than a number (about 0.4), there is no more distinctness partly-synchronized behavior introduced by $`\alpha `$ changing (seen in Fig 5) . It can show that the $`L\times L`$ area have evolved into synchronized state. We also draw the whole network’s avalanches distribution map(Fig .6). It can be clearly seen that the isolated avalanches’ number in $`L\times L`$ area increases with $`\alpha `$ decreasing, and the large scale avalanche can’t propagate into the interior of $`L\times L`$ area. These result are consistent with the work of C.Tang , Grassberger et al. 10 ; 13 .
## IV conclusion
In this paper, we analyze the dynamics of the proposed neural network model and find the distribution of the avalanche sizes and the distances $`X`$ between subsequent unstable sites show the power-law behavior. More important, we find the function area and weight distribution produced by self-organized process in our Neural Network model will let the $`P(S)`$ and $`P(X)`$ distribution deviate from power law behavior , and the system evolves into a partly-synchronized state at this time. To verify the explanation , we study a quasi-OFC earthquake model containing synchronized region , and find it will deviate from power-law , evolve into a partly-synchronized state in some conditions too.
Our self-organized feature map Neuron Network model is just a very simple simulation of brain. The real brain has very complex structure and more specific feature regions in Cortex. So brain may express a quasi-SOC(partly-synchronized) behavior more than a pure SOC behavior. Therefore the stored pattern in brain might be designed as a quasi-SOC attractor, and the associate memory process might be designed as the process of the input pattern evolving into the attractor.
Now the neuron synchronization in brain has been observed in many experiments 14 . Rodriguez et al. have investigated the long distance synchronization of human brain activity 15 . We think it would be interesting to further investigate the relationship between synchronization and cognitive , associate process. |
warning/0003/hep-ph0003270.html | ar5iv | text | # TURKU-FL-P34-00 Q-ball collisions in the MSSM: gravity-mediated supersymmetry breaking
## 1 Introduction
Stable non-topological solitons , Q-balls , can be present in several field theory models. In particular the supersymmetric extensions of the Standard Model may contain them. A Q-ball is a coherent state of a complex scalar field that carries a conserved charge, typically a $`U(1)`$-charge. In the sector of fixed charge a Q-ball is a ground state so that the conservation of charge assures that the Q-ball is stable. In the Minimal Supersymmetric Standard Model (MSSM) Q-balls carrying lepton or baryon number are present due to the existence of flat directions in the scalar sector of the theory .
The cosmological significance of Q-balls can present itself in many forms. Stable (or long living) Q-balls are natural candidates for dark matter and the decay of Q-balls can explain the baryon to dark matter ratio of the universe . Q-balls can also protect the baryon asymmetry from sphalerons at the electroweak phase transition and decaying Q-balls may be responsible for the baryon asymmetry of the universe . Furthermore, Q-balls can play an important role in considering the stability of neutron stars .
The mechanism by which supersymmetry (SUSY) is broken in the theory is significant for the charges and stability of Q-balls in the theory. If SUSY is broken by a gauge-mediated mechanism, the baryon number carrying B-balls can have very large charges due to to the flatness of the potential. Assuming that the charge is large enough, they can then be stable against decay into nucleons . If, however, SUSY is broken by a gravitationally coupled hidden sector, the potential is not completely flat but Q-balls may still exist due to radiative corrections . In this case the Q-balls can decay (evaporate ) into baryons or supersymmetric particles.
The formation of Q-balls from an Affleck-Dine (AD) condensate in the early universe has been studied recently with numerical simulations . In these simulations both the gauge- and gravity-mediated SUSY breaking scenarios have been considered. In both cases it was found that Q-balls do form from the AD condensate. In the gravity-mediated case it was especially noted that the formed Q-balls have non-zero velocities and can hence collide with each other . Q-ball collisions were also simulated on a one dimensional lattice and it was found that Q-balls typically merge, exchange charge or pass through each other . Since the charge can change due to collisions, they may play an important role in the determination of the Q-ball charge distribution after their formation. On the other hand the charge distribution is important in evaluating the significance of Q-balls for the evolution of the universe. It is hence worthwhile to study Q-ball collisions in more detail. Q-ball collisions have also been studied previously in various potentials in - but not to our knowledge in either the gauge- or gravity-mediated scenarios.
In this paper we have studied numerically collisions of Q-balls in the gravity-mediated scenario on a two dimensional lattice. The gauge mediated scenario will be analyzed in a forthcoming paper.
## 2 Q-ball solutions
Consider a field theory with a U(1) symmetric scalar potential, $`U(\varphi )`$, with a global minimum at $`\varphi =0`$. The complex scalar field $`\varphi `$ carries a unit quantum number with respect to the $`U(1)`$-symmetry. The charge and energy of a given field configuration $`\varphi `$ are given by
$$Q=\frac{1}{i}(\varphi ^{}_t\varphi \varphi _t\varphi ^{})d^Dx$$
(1)
and
$$E=[|\dot{\varphi }|^2+|\varphi |^2+U(\varphi ^{}\varphi )]d^Dx.$$
(2)
The single Q-ball solution corresponds to the minimum energy configuration at a fixed charge. If it is energetically more favourable to store charge in a Q-ball compared to free particles, the Q-ball will be stable against radiative decays into $`\varphi `$-scalars. Hence, for a stable Q-ball, condition
$$E<mQ,$$
(3)
where $`m`$ is the mass of the $`\varphi `$-scalar, must hold.
Minimizing the energy is straightforward and is easily done by using Lagrange multipliers. The Q-ball solution can be shown to be of the form
$$\varphi (x,t)=e^{i\omega t}\varphi (r),$$
(4)
where $`\varphi (x)`$ is now time independent and real, $`\omega `$ is the Q-ball frequency, $`|\omega |[0,m]`$ and $`\varphi `$ is spherically symmetric. The charge of a spherically symmetric Q-ball in D-dimensions reads
$$Q=2\omega \varphi (r)^2d^Dr.$$
(5)
The equation of motion at a fixed $`\omega `$ is
$$\frac{d^2\varphi }{dr^2}+\frac{D1}{r}\frac{d\varphi }{dr}=\varphi \frac{U(\varphi ^2)}{\varphi ^2}\omega ^2\varphi .$$
(6)
To obtain the Q-ball profiles we must solve (6) with boundary conditions $`\varphi ^{}(0)=0,\varphi (\mathrm{})=0`$.
In the present paper we consider a potential of the form
$$U(\varphi )=m^2\varphi ^2(1K\mathrm{log}(\frac{\varphi ^2}{M^2}))+\lambda \varphi ^{10},$$
(7)
where the parameter values are chosen to be $`m=100`$ GeV, $`K=0.1`$ and $`\lambda =M_{\mathrm{Pl}}^6`$, where $`M_{\mathrm{Pl}}`$ is the reduced Planck mass. This choice of potential corresponds to a D-flat direction in the full scalar potential in the MSSM where supersymmetry has been broken by a gravitationally coupled hidden sector . The large mass scale M is chosen such that the minimum is degenerate
$$M=(\frac{1}{4}Km^2\lambda ^1\mathrm{exp}(1\frac{4}{K}))^{\frac{1}{8}}10^{11}\mathrm{GeV}.$$
(8)
We have calculated the charge and energy of Q-balls for different values of $`\omega `$. Energy vs. charge curves are shown in Figure 1(a). The axis scales are chosen differently for two and three dimensions; for two dimensions, $`Q_0=8000(M/\mathrm{GeV})^2`$, $`E_0=8\times 10^5M^2\mathrm{GeV}^1`$ and for three dimensions, $`Q_0=600(M/\mathrm{GeV})^2`$, $`E_0=6\times 10^4M^2\mathrm{GeV}^1`$. The dashed line is the stability line, $`E=mQ`$, that indicates that the Q-balls considered here are stable with respect to scalar decays.
From the figure it can be seen that the energy vs. charge curves are of a similar shape in two and three dimensions. Q-ball profiles are plotted in Figure 1(b) for different values of $`\omega `$ in two and three dimensions. The profiles appear very similar in these two cases. These figures suggest that the collision processes calculated in two spatial dimensions are likely to be similar to collisions in three spatial dimensions. Here it is worth noting that, as from the equation of motion (6) can be seen, a non-zero dissipation term is present for dimensions larger than one. This suggests that collisions in one dimension may differ quite significantly from collision processes in higher dimensions. This also seemed to be the case in the one dimensional simulations we have done.
## 3 Collisions
We have studied collisions of Q-balls with equal charges in the potential (7) . The studied values of $`\omega `$:s were $`\omega /m=0.50,0.60,0.75,0.90,0.99`$. These values correspond to charges $`1450,645,110,12.6,2.24`$ in units of $`(M/\mathrm{GeV})^2`$ in the case with two spatial dimensions so that in terms of $`\varphi `$-scalars the considered range of charges is $`10^{24}10^{26}`$. For the same range of $`\omega `$:s, the charges in the three dimensional space are in the range $`10^{23}10^{26}`$.
The relative phase of the Q-balls is also accounted for. By the relative phase, we mean the difference in individual phases at the point where the distance between Q-balls is at a minimum assuming there is no interaction between them. Position is defined as the point of the maximum value of the amplitude of $`\varphi `$. The relative phase is allowed to have values in the range $`0\mathrm{\Delta }\varphi 2\pi `$. The impact parameter is also varied to study the cross-sections. The Q-balls studied here are of the thick-wall type so that there is no natural definition of the Q-ball size. Therefore we have defined the size of a Q-ball by a Gaussian fit; we fit a Gaussian $`\varphi =Ae^{Br^2}`$ to the profiles and define the radius of the ball as $`R=B^{\frac{1}{2}}`$. The cross-sections quoted here are three dimensional cross-sections with the interaction radius taken from the two-dimensional simulations.
Collisions were simulated on a 2+1 -dimensional lattice. The lattice size typically used was $`200^2`$ with continuous boundary conditions. A 9-point Laplacian operator and a step size of $`5\times 10^3`$ was used in all calculations. Collisions were studied for different initial velocities of Q-balls, $`v=10^3`$ and $`v=10^2`$.
### 3.1 Numerical Results
Collisions can roughly be divided into three types; fusion, charge exchange and elastic scattering. Fusion is defined as a process where most of the initial charge is in a single Q-ball after the collision and the rest of the charge is lost either as radiation or as small Q-balls. By charge exchange we mean a process where Q-balls exchange some of their charge while the total amount of charge carried by the two balls is essentially conserved. An elastic scattering is defined to be a process where less than $`1\%`$ of the total charge is exchanged. After the collision the ratio of the charge in the largest Q-ball to the total initial charge as a function of the relative phase has been plotted in Figures 2 and 3.
From the figures the two different types of processes can be distinguished. In a fusion process typically $`1020\%`$ of the initial charge is lost as radiation and small Q-balls and the rest of the charge is in a single Q-ball. On the other hand, if charge is exchanged the larger Q-ball carries usually less than $`70\%`$ of the total charge. As from Figs. 2 and 3 can be seen, fusion occurs generally only when the relative phase is small and is more likely to occur with smaller $`\omega `$. The amount of exchanged charge decreases substantially with increasing relative phase. Increased velocity does not seem to have a large effect for the range of $`\omega `$ where fusion occurs. However, more charge is exchanged between balls of equal size when the initial velocity is larger. The relative changes in size are weakly dependent on $`\omega `$ (in both cases the standard deviation is less than $`1\%`$) and hence on the size of the Q-balls. Averaging over the relative phase (assuming a random distribution for the $`\mathrm{\Delta }\varphi `$:s) and $`\omega `$:s, the relative change in the size of a Q-ball is $`10\%`$ for $`v=10^3`$ and $`14\%`$ for $`v=10^2`$.
We are now ready to calculate the fusion cross-section, $`\sigma _\mathrm{F}`$, and the cross-section for the charge exchange, $`\sigma _\mathrm{Q}`$, as a function of $`\omega `$. These are plotted in Figure 4 with the geometrical cross-section, $`\sigma _\mathrm{G}`$.
Clearly the fusion cross-section is strongly dependent on $`\omega `$; larger Q-balls fuse more easily than smaller ones. This effect is clearly not explained by the different geometrical sizes of the Q-balls as can be noted from the geometrical cross-section. The fusion cross-section decreases with increasing $`\omega `$ because the Q-balls with higher $`\omega `$ have more regions with differing relative phases. The field dynamics cannot then even out the relative phase differences quickly enough to keep the colliding balls together. From the simulations it can be seen that balls with larger $`\omega `$ are less likely to fuse than balls with the same phase difference but smaller $`\omega `$. The effect of increasing the initial velocity on the cross-sections can also be noted from Fig. 4. The fusion cross-section is slightly decreased as velocity increases while the charge exchange cross-section increases. The increase in $`\sigma _Q`$ is due to the fact that now the Q-balls have more kinetic energy to overcome the repulsion resulting from the relative phase difference.
The total cross-section including all the cases i.e. when the balls fuse, exchange charge or scatter elastically, is also dependent on $`\omega `$, but only weakly. Averaging over the $`\omega `$:s, $`\sigma _{\text{tot}}=0.27\pm 0.01\mathrm{GeV}^2`$ ($`v=10^3`$) and $`\sigma _{\text{tot}}=0.19\pm 0.01\mathrm{GeV}^2`$ ($`v=10^2`$).
We have also studied Q-ball collisions with larger initial velocities for a more limited set of parameter values. When the initial velocity is increased to $`v=10^1`$, the fusion cross-section is reduced significantly from its value when $`v=10^3`$. Furthermore, at such high velocities we also see processes where the Q-balls pass through each other essentially without exchanging any charge. This is a similar process that was reported to occur in one dimension in and which we have also observed in our one dimensional simulations.
Charge exchange also affects the final velocities of the Q-balls. As charge is exchanged the speed of the Q-balls typically increase. The final velocity of the smaller ball can be quite large; we have often noted final velocities ten times larger than the initial velocity.
## 4 Conclusions
In this paper we have studied Q-ball collisions in the MSSM with supersymmetry broken by a gravitational hidden sector. For the studied range of charges the total cross-section was found to be approximately constant. The cross-section for fusion, $`\sigma _\mathrm{F}`$, appeared to be smaller than the geometrical cross-section, $`\sigma _\mathrm{G}`$, whereas the cross-section for charge exchange, $`\sigma _\mathrm{Q}`$, was larger than $`\sigma _\mathrm{G}`$. In a collision it is hence more probable that a charge exchanging process occurs rather than a fusion process. This probability increases with increasing $`\omega `$ (or with decreasing charge). Averaging over the fusion and charge exchanging processes the average charge increase of the largest Q-ball emerging from a collision was found to be approximately constant. For the considered range of charges and velocities it was $`10\%(v=10^3)`$ and $`14\%(v=10^2)`$.
In a cosmological context, Q-ball collisions may have a significant effect on the charge distribution of Q-balls. Clearly for collisions to be important the number density of Q-balls must be high enough and the balls must have large enough velocities for the rate of interaction to be significant. In the early universe this obviously means that the interaction rate must be larger than the Hubble rate. If collisions typically do occur the resulting charge distribution can then be altered by the fusion and charge exchange processes. Based on the results presented in this paper, the relative phase, size and the initial velocity of the balls then play important roles in studying the evolution of the Q-ball charge distribution.
If the balls that are formed from the AD condensate are in the same phase, fusion processes will dominate and the average size of a Q-ball grows substantially in a collision. Since most of the charge is left in the remaining ball and the rest is in the form of several small, quickly evaporating , Q-balls and radiation, the number density of Q-balls reduces rapidly. Collisions can therefore freeze the Q-ball distribution quickly in the early phases of the universe. If, on the other hand, the phases are randomly distributed the probability for fusion is greatly reduced and the distribution will not change as significantly as in the previous case. Collision processes can then also continue for a longer period of time.
The typical size of Q-balls is obviously an important factor. The total scattering cross-section depends quite weakly on the size of the Q-balls but the fusion and charge exchange cross-sections do have a strong $`\omega `$-dependence.
The initial velocity of the Q-balls is also significant in the evolution of the Q-ball distribution. A larger velocity means that the interaction rate is increased but on the other hand if the initial velocity is too large the cross-sections decrease due to a decreased interaction time. Collisions can also significantly change the velocities of the Q-balls so that an initially uniform velocity distribution can be spread out by the collision processes.
The effect of collisions can be important in deciding the exact role and significance of Q-balls in the evolution of the universe. In determining their importance on cosmology, more information is needed about the Q-ball distribution after their formation and also about the effects of collisions on the initial distribution. To quantify the effects of the different collision processes described in this paper, a more detailed analysis is needed which gives motivation for future work.
Acknowledgements. We thank K. Enqvist for discussions and the Center for Scientific Computing for computation resources. This work has been supported by the Academy of Finland. |
warning/0003/cond-mat0003500.html | ar5iv | text | # Enhanced Stability of Layered Phases in Parallel Hard-Spherocylinders Due to Addition of Hard-Spheres
## I Introduction
In hard particle fluids all allowed configurations have the same energy and therefore it is the number of states, or equivalently the entropy of a system that determines the equilibrium phase. Examples of well known phase transitions where the formation of ordered structures are driven solely by an increase in entropy are the liquid to crystal transition in hard spheres , the isotropic to nematic and the nematic to smectic transition in hard-rods . Because of their high degree of monodispersity, and because of the dominant role of steric repulsion in the pair-potential, colloidal suspensions of polystyrene latex and rod-like viruses have often been used as experimental model systems for the study of entropy induced ordering in hard-sphere and hard-rod systems, respectively .
A natural extension of the above work is to the phase behavior of mixtures, with a number of recent experimental and theoretical studies focusing on the phase behavior of binary mixtures of hard-spheres . We have recently begun work on less studied systems that closely approximate hard-rod/hard-sphere and hard-rod/polymer mixtures . As a model for hard-rods we used either $`\mathrm{𝑓𝑑}`$ or TMV virus, as hard-spheres we used polystyrene latex, and as polymers we used poly(ethylene-oxide) with varying molecular weights . The part of the phase diagram explored consisted of pure rods in either the isotropic, nematic, or smectic phase to which a small volume fraction of spheres or polymers was added. Remarkably, besides the expected uniform mixtures and bulk demixing, we also observed a variety of microphases for a wide range of sphere sizes and concentrations . In microphase separation the system starts separating into liquid-like regions that are rich in either spheres or rods. However, unlike bulk demixing where rod and sphere rich regions grow until reaching macroscopic dimensions, in microphase demixing these liquid-like regions increase only to a critical size after which they order into well defined three dimensional equilibrium structures. One of the micro-separated phases observed, named the lamellar microphase, consists of alternating two-dimensional liquid-like layers of rods and spheres and is the subject of theoretical analysis in this paper.
In this paper we use the second virial approximation first studied by Koda et. al. to examine the influence of molecular parameters such as shape and size, on the phase behavior of rod/sphere mixtures. As the second-virial theory is approximate in nature, we validate the theoretical predictions by comparing them with either computer simulations or experimental results. The remainder of this paper is organized as follows: In section II we formulate the second virial approximation for the rod/sphere mixture. The general features of the phase diagram are discussed and a physical picture of the factors responsible for the enhanced stability of the layered phase due to the presence of spheres is presented. In Section III the influence of varying the spherocylinder length on the phase behavior of spherocylinder/sphere mixtures is studied using computer simulations and the results are compared to theoretical predictions. Section IV examines how changes of the sphere diameter influence the phase behavior of spherocylinder/sphere mixtures. Finally in Section V we present our conclusions.
## II General features of a phase diagram of a spherocylinder-sphere mixture
Although the equilibrium phases of all hard particle fluids are determined by maximizing the entropy, ordering transitions are still possible because the expression for the total entropy, or equivalently free energy, splits into two parts. The ideal contribution to the entropy is of the form $`\rho \mathrm{ln}\rho `$, where $`\rho `$ is the density distribution function. This contribution to the entropy attains a maximum for a uniform density distribution and therefore always suppresses transitions from uniform to modulated phases. In contrast, excluded volume entropy sometimes increases with increasing order and therefore drives the system towards a modulated phase. In this paper we use a highly simplified second virial approximation to calculate the excluded volume entropy.
The equilibrium phase in a spherocylinder/sphere mixture is determined by four parameters: length over diameter of a spherocylinder $`(L/D_{sc})`$, diameter of spherocylinder over diameter of sphere ($`D_{sc}/D_{sp}`$), total volume fraction of spheres and spherocylinders ($`\eta `$) and partial volume fraction of spheres ($`\rho _{sp}`$). To help us in interpretation of our results we first define the slope
$$\tau =\underset{\rho _{sp}0}{lim}\frac{\eta (\rho _{sp})\eta (0)}{\rho _{sp}}$$
(1)
where $`\eta (\rho _{sp})`$ is the total volume fraction of the rod-sphere mixture at the layering transition after the introduction of spheres at partial volume fraction $`\rho _{sp}`$. A positive value of $`\tau `$ implies that adding a second component stabilizes the nematic phase by displacing the smectic transition to higher densities. For the case when both components are spherocylinders of different lengths but with the same diameter, slope $`\tau `$ is positive if the ratio of lengths is less then approximately 7 . In the same manner, negative values of $`\tau `$ imply that the second component stabilizes the smectic phase. There are predictions of a negative value of $`\tau `$ in a bidisperse rod mixture when the ratio of rod lengths is large enough , or when added rods have a larger diameter . In this section we focus on the phase behavior of the spherocyinder-sphere mixture for the specific microscopic parameters $`L/D_{sc}=20`$ and $`D_{sc}/D_{sp}=1`$. We present a physical picture of excluded volume effects that are responsible for the enhanced stability of the lamellar phase. In the next two sections we extend our study on how changes in the molecular parameters $`L/D_{sc}`$ and $`D_{sc}/D_{sp}`$ modify the phase behavior and in particular, their influence on the magnitude and sign of the slope $`\tau `$.
### A Second virial approximation
The second virial approximation for a mixture of perfectly aligned spherocylinders and spheres of equal diameter was proposed by Koda, Numajiri and Ikeda and is generalized for arbitrary $`L/D_{sc}`$ and $`D_{sc}/D_{sp}`$ in the appendix. It was previously shown that the second virial approximation described qualitatively the formation and various features of the smectic phase of hard rods . Here we study how the addition of spheres perturbs the formation of the smectic phase. Since the sphere volume fraction is very low we expect that the second virial approximation is still qualitatively correct for these mixtures. We consider a sinusoidal perturbation from the uniform density for both spherocylinders and spheres. From equations (7) and (16) in the appendix we obtain the free energy difference between the uniformly mixed and layered state in a spherocylinder/sphere mixture:
$$\delta F=a_1^2\left(S_{11}2\frac{a_1}{a_2}S_{12}+\left(\frac{a_1}{a_2}\right)^2S_{22}\right)=0$$
(2)
.
The phase diagram obtained within this approximation for microscopic parameters $`L/D_{sc}=20`$ and $`D_{sc}/D_{sp}=1`$ is shown in Fig. 1. From the phase diagram we see that the first prediction of the model is that spheres, upon addition to a smectic phase, will preferentially occupy space between smectic layers and therefore create a stable micro-separated lamellar phase. The second prediction is that the total volume fraction at which the system undergoes a transition from a uniform miscible state to a layered lamellar state is lowered by increasing the partial volume fraction of spheres. This implies that the slope $`\tau `$ is negative for this particular spherocylinder/sphere mixture and we conclude that in this case spheres enhance the layering transition.
We can assign a simple physical origin to every term given in Eq. (2) above and Eq.(12) of the Appendix. The parts of the spherocylinder-spherocylinder interaction term $`S_{22}`$ and sphere-sphere term $`S_{11}`$ that scale as $`\eta `$ are due to the ideal (id) contribution to the free energy, also known as the entropy of mixing and are denoted as $`S_{22}^{id}`$ and $`S_{11}^{id}`$, respectively. The terms having a $`\eta ^2`$ dependence in $`S_{22}`$, $`S_{12}`$, $`S_{11}`$ are due to the spherocylinder-spherocylinder, spherocylinder-sphere and sphere-sphere excluded volume (ex) interaction, respectively and are denoted as $`S_{22}^{ex}`$, $`S_{12}^{ex}`$ and $`S_{11}^{ex}`$. Since the instability is defined as $`\delta F(\eta _c,k_c)=0`$, at a critical density $`\eta _c`$ and at a critical wavevector $`k_c`$ all individual contributions to the free energy difference in Eq. (2) must add up to zero. In Fig. 2 we show the value of all terms with distinct physical origins at the instability density $`\eta _c`$ and wavevector $`k_c`$ as a function of partial volume fraction of spheres. Since from our analysis we cannot determine the absolute amplitude of $`a_1`$ we only plot the ratios of all free energy components to the absolute value of the spherocylinder-spherocylinder excluded volume $`S_{22}^{ex}`$. If we set the partial volume fraction of spheres to zero $`(\rho _{sp}=0)`$ in Eq.(2) we obtain an equation whose solution indicates the nematic-smectic stability limit in a pure suspension of aligned spherocylinders . For these conditions the only two nonzero components of free energy are $`S_{22}^{ex}`$, which is negative and therefore drives the transition and $`S_{22}^{id}`$, which is positive and therefore suppresses the transition. As we start increasing the partial sphere volume fraction $`\rho _{sp}`$, the spherocylinder-sphere free volume term $`S_{12}^{ex}`$ rapidly assumes large negative values as evidenced by the rapidly decreasing ratio of $`S_{12}^{ex}/S_{22}^{id}`$. This implies that layering the mixture significantly decreases the excluded volume that is due to the spherocylinder-sphere interaction.
We can use the information gained from the second virial approximation to obtain a clear physical picture of excluded volume effects in spherocylinder/sphere mixtures and explain the enhanced stability of the lamellar phase. Taking any single spherocylinder in a uniform spherocylinder/sphere mixture and replacing it by two spheres will leave the value of excluded volume virtually unchanged. The reason for this lies in the fact that the volume excluded to the spherocylinder due to the presence of a sphere with equal diameter, under the constraint of uniform packing, is a spherocylinder with diameter $`2D_{sc}`$ and length $`(L+2D_{sc})`$ where $`L`$ and $`D_{sc}`$ are defined in Fig. 3. However, the excluded volume between any two spherocylinders with large $`L/D_{sc}`$ is only about twice this value as illustrated in Fig. 3. Although replacing spherocylinders by spheres in such a manner leaves the excluded volume almost unchanged, it does significantly decrease the total volume fraction of the mixture since the volume of two spheres is much smaller then the volume of a spherocylinder with large $`L/D_{sc}`$. Therefore in the spherocylinder/sphere mixture we encounter excluded volume problems similar to those found in a pure spherocylinder solution, but at a lower total volume fraction. As in pure spherocylinders, the system reduces the excluded volume by undergoing a transition to a layered phase. The excluded volume is reduced in the lamellar state because a periodic density distribution forces spheres and spherocylinder into alternate layers thus decreasing the probability of the very unfavorable sphere-spherocylinder contacts as illustrated in Fig. 4. This explains the large decrease in the value of the $`S_{12}^{ex}`$ term at the lamellar transition that we observe in the second virial theory. This term is responsible for the enhanced stability of the lamellar phase in a sphere/spherocylinder mixture. In conclusion, it is the inability to efficiently pack a uniform mixture of spherocylinders and spheres, as reflected in the large spherocylinder/sphere excluded volume term, that destabilizes the nematic phase and enhances the formation of a layered phase.
An alternate way to think about the formation of a layered phase is to focus on the effects of spherocylinder ends . The nematic phase in our simplified model is characterized by random distribution of spherocylinders along their axial and radial direction as illustrated in Fig. 5. This end effect is responsible for the formation of the smectic phase, which is characterized by a periodic density distribution. In similar fashion, introducing a sphere into the nematic phase will have the same effect on the surrounding spherocylinders as another spherocylinder end. Therefore adding spheres very effectively increases the density of “spherocylinder ends” and decreases the total volume fraction. To resolve the difficulties in efficient packing due to these extra “spherocylinder ends”, the mixture layers at a lower total volume fraction.
### B Monte Carlo Simulation
In the previous section we discussed two predictions of the second virial theory for a spherocylinder/sphere mixture with $`L/D_{sc}=20`$ and $`D_{sp}/D_{sc}=1`$; the existence of the lamellar phase and the enhanced stability of the lamellar phase when compared to a smectic phase of pure spherocylinders. Our results are in agreement with previous studies of Koda et. al. . However, the second virial approximation is highly approximate and there is reasonable concern about the influence of higher terms on the topology of the phase diagram. To support their conclusions Koda et. al. performed computer simulations, which indicated the existence of an lamellar phase . Still, the question of whether spheres simply fill the voids between layers in an already formed smectic phase, or actually induce layering at lower total volume fraction was not addressed. In this section, using Monte Carlo simulations we address the question of the influence of adding spheres on the phase behavior of spherocylinders by determining the slope $`\tau `$ in Eq. 1 in a mixture of spherocylinders and spheres with parameters $`L/D_{sc}=20`$ and $`D_{sp}/D_{sc}=1`$
A Monte Carlo simulation of a mixture of hard-spheres and perfectly aligned hard-spherocylinders was performed at constant pressure and number of particles . Most simulations contained 392 spherocylinders and a variable number of spheres. To check for finite size effects we also ran simulations with 784 spherocylinders, but saw no significant difference in the results obtained. In one sweep, pressure was increased from a dilute homogeneous mixture up to a well ordered, dense smectic or lamellar phase. At each value of the pressure, the density of spheres and spherocylinders and their corresponding smectic order parameter were measured after the system was allowed to equilibrate. Identical results were obtained when the pressure was slowly decreased from a initially dense phase composed of alternating layers of spherocylinders and spheres to a dilute homogeneous mixture.
Besides lamellar transitions there is a possible demixing transition where spherocylinders and spheres phase separate into macroscopically distinct phases. However, once a layered phase is formed the exchange of spheres between layers drops to a negligible amount, leaving open the possibility that system would undergo a demixing transition, but is stuck in a lamellar phase, which is only a metastable state. To find out the location of the demixing transition it is necessary to measure the chemical potential of both spherocylinders and spheres in a spherocylinder/sphere mixture . This possibility was not examined in this work, primarily because we are only interested in how low concentrations of spheres perturb the formation of the layered phase. Therefore it is reasonable to expect that at a very low volume fraction of spheres, the lamellar transition is going to be more stable than the demixing transitions as predicted by the second virial theory.
A plot of the smectic order parameter for spherocylinders with $`L/D_{sc}=20`$ as a function of increasing total density for different partial volume fractions of spheres is shown in Fig. 6. As the system approaches a certain critical density we observe a rapid non-linear increase in the smectic order parameter that we interpret as a signature of the nematic to smectic phase transition. This critical density shifts to lower values of the total volume fraction as the partial volume fraction of spheres is increased. To reconstruct a phase diagram from the above data we define a phase as layered when its smectic order parameter reaches a value of 0.3 . For a pure spherocylinder suspension this value yields good agreement with previous studies of the volume fraction of the nematic-smectic phase transition . Since we are mostly interested in the qualitative behavior of a spherocylinder/sphere mixture this method should suffice our purposes. Using this phenomenological rule, the phase diagram for a mixture of spherocylinders and spheres ($`L/D_{sc}=20`$, $`D_{sc}/D_{sp}=1`$) is reconstructed and compared to the second virial theory in Fig. 1. An immediate conclusion drawn from Fig. 1 is that adding spheres to aligned spherocylinders enhances the stability of the lamellar phase, which is indicated by the negative value of slope $`\tau `$, in agreement with the prediction of the second virial approximation.
## III The effects of spherocylinder length on the phase diagram
Next we proceed to investigate the influence of varying the spherocylinder length on the magnitude of slope $`\tau `$. The predictions of the second virial theory for the nematic-lamellar instability are shown in Fig. 7a. The second virial theory clearly predicts increasing stability of the lamellar phase with increasing length of spherocylinder. To verify this prediction we repeated Monte Carlo simulations for spherocylinders with different $`L/D_{sc}`$ and used the same rule as before to identify the volume fraction of the nematic-lamellar transition. The simulation results for the location of the nematic to layered transition are shown in Fig. 7b. We can conclude that our simulations confirm predictions of the second virial model and that the length of the spherocylinder is an important parameter in forming the lamellar phase, with longer spherocylinders showing an increasing tendency to form a layered phase at a lower volume fraction of added spheres.
Using the physical picture of the excluded volume effects developed in the previous section provides a natural explanation for our simulation results in Fig. 7. With increasing spherocylinder length the excluded volume due to the spherocylinder-sphere interaction grows proportionally to the spherocylinder length and consequently the value of the $`S_{12}^{ex}`$ term increases in magnitude. As we have seen before, the larger the $`S_{12}^{ex}`$ term, the more likely it is for the system to form a layered phase.
It is interesting to consider the limit of spherocylinders with infinite aspect ratio. In the density regime of the nematic-smectic transition, this model can be mapped onto a system with skewed cylinders with an aspect ratio close to one. The nematic-smectic transition in this model has been studied numerically . If we consider the addition of spheres to this system, then the same affine transformation that maps the infinite spherocylinders onto squat, skewed spherocylinders, will map the spheres onto infinitely thin, parallel disks. As the disks are infinitely thin, they do not interact with each other but only with the cylinders. Inside the nematic phase, most volume is excluded for these disks. However, in the smectic phase, there is ample space for the disks between the layers. In fact, the stronger the layering, the larger the accessible volume. Hence in this limit, the addition of spheres will strongly stabilize the smectic phase.
## IV The effects of sphere diameter on the phase diagram
In this section we investigate the influence of sphere diameter on the value of slope $`\tau `$. Fig. 8 shows the prediction of the second virial theory for the dependence of slope $`\tau `$ on the ratio of spherocylinder to sphere diameter $`(D_{sc}/D_{sp})`$ for spherocylinders with different $`L/D_{sc}`$. In section A we examine the phase behavior of sphere/spherocylinder mixtures when the sphere diameter is smaller then spherocylinder diameter and in section B we examine the other case when the sphere diameter is larger then the spherocylinder diameter. In our model the presence of the spheres cannot alter the orientational distribution function of spherocylinders, which are always perfectly parallel to each other. It is reasonable to expect that this assumption holds for spheres smaller then the spherocylinder length, but as a sphere becomes larger then the spherocylinder length, long wavelength elastic effects start to dominate the behavior of the system and hard spherocylinders will tend to align parallel to the surface of the sphere . Therefore in Fig. 8 we plot the values of slope $`\tau `$ only for those values of $`D_{sc}/D_{sp}`$ for which our assumptions are at least qualitatively correct. As we increase the sphere size beyond this limit our model describes a highly artificial system of large spheres and parallel spherocylinders. In this regime we observe oscillations in the value of slope $`\tau `$ similar to what is observed in binary mixtures of parallel spherocylinders .
### A Sphere diameter smaller than spherocylinder diameter
In the regime where $`D_{sc}/D_{sp}>1`$ (for spherocylinders of any $`L/D_{sc}`$), decreasing the sphere size increases the stability of the lamellar phase as indicated by the increasing negative value of slope $`\tau `$ seen in the right hand side of Fig. 8. This prediction of the theory has a simple explanation in our picture of excluded volume in a sphere/spherocylinder mixture. If we halve the sphere radius $`D_{sp}`$, while keeping constant the volume fraction of spheres, we increase the number of spheres eight times. At the same time, the result of reducing the sphere size is to decrease the excluded volume of the spherocylinder-sphere interaction. However, the eightfold increase in the number of spherocylinder-sphere interactions more then compensates for the decrease in excluded volume between the sphere and spherocylinder and consequently the magnitude of $`S_{12}^{ex}`$ increases with decreasing sphere diameter. This leads to the increased stability of the layered phase with decreasing sphere size.
It becomes difficult to verify this prediction using computer simulations. As the sphere size decreases at constant total volume fraction $`\eta `$, the number of particles in a simulation rapidly reaches the order of thousands requiring simulation times that are prohibitively long. As the ratio of spherocylinder to sphere diameter $`(D_{sc}/D_{sp})`$ was varied within the accessible range between 0.5 to 2 we did not observe any changes in the value of slope $`\tau `$ that were larger than our measurement error. Larger and longer simulations are needed for a careful analysis of spherocylinder/sphere mixtures with extreme values of the ratio $`D_{sc}/D_{sp}`$.
### B Sphere diameter larger than spherocylinder diameter
For spherocylinders with small $`L/D_{sc}`$, Fig. 8 shows that the magnitude of slope $`\tau `$ uniformly decreases with increasing sphere size. Eventually the slope $`\tau `$ changes sign and becomes positive, implying that large spheres stabilize the nematic and not the smectic phase. The phase diagram under conditions where slope $`\tau `$ is positive is shown in Fig. 9. The wavevector associated with the layering transition, indicated with a solid line in Fig. 9, remains at an almost constant value. Another important point is that the amplitude ratio in Eq. (16) is positive. This means that the periodic density modulations of the spherocylinders and spheres are in phase, which implies that spheres no longer go into the gap between two spherocylinder layers, but rather fit into the spherocylinder layer. However, as the partial volume fraction of spheres $`(\rho _{sp})`$ is increased further we observe a discontinuous jump in the wavevector to zero value. This implies that there is a discontinuous change from a layering to a demixing transition. As the demixing transition is reached there is also a change in sign of the amplitude ratio, which becomes negative and the spherocylinders and spheres bulk separate. In contrast, the phase diagram for mixtures of small spheres and spherocylinders shown in Fig. 1 looks quite different. The amplitude ratio for this case is always negative implying formation of the lamellar phase. Another contrast is that in a mixture of small spheres and spherocylinders the wavevector associated with the layering transition decreases in a continuous fashion until it reaches zero value.
We now examine the behavior of individual terms in Eq. (2) for a mixture of large spheres and short spherocylinders shown in Fig. 9. Most notably, we find that at low volume fractions of spheres where the system undergoes the layering transition, the ratio $`S_{12}^{ex}/S_{22}^{ex}<<1`$. This implies that upon layering there is almost no reduction of the unfavorable sphere/sperocylinder interaction and that the spherocylinder/spherocylinder interaction alone drives the formation of the layered phase. In contrast, for small spheres this ratio was large and was responsible for enhanced stability of the lamellar phase as was shown in Fig. 2. At a higher volume fraction of large spheres where the mixture directly bulk phase separates we find that the ratio $`S_{12}^{ex}/S_{22}^{ex}>>1`$. This implies, as expected, that demixing very effectively reduces the unfavorable sphere/spherocylinder interactions. These results suggest a physical picture of the excluded volume effect. Unlike small spheres, large spheres can not fit into the gap between smectic layers and consequently there is no way to gain free volume by undergoing the layering transition. As an alternative, to gain free volume the system bulk phase separates at the lowest volume fraction of spheres possible.
While for short spherocylinders the magnitude of slope $`\tau `$ uniformly decreases with increasing sphere size, longer spherocylinders exhibit a qualitatively different behavior. For a mixture of spherocylinders with $`L/D_{sc}=100`$ and spheres with $`D_{sc}/D_{sp}=0.1`$ there is a pronounced increase in the stability of the lamellar phase as shown in Fig. 8. By increasing the length of spherocylinders to even larger values, the region of increased stability of the lamellar phase shifts to higher values of the sphere radius. Two conditions emerge, which when satisfied lead to enhanced stability of the lamellar phase. First, it is necessary for a sphere to fit between two smectic layers without disturbing them. This condition is satisfied when $`D_{sp}/L0.1`$. The second condition is that $`D_{sp}/D_{sc}>>1`$. It was argued before that under these condition large spheres are able to induce smectic correlations amongst neighboring spherocylinders , which in turn can enhance the formation of the lamellar phase.
Because of the large size asymmetry it was not feasible to carry out simulations for mixture of spherocylinders and spheres with $`L/D_{sc}100`$ and $`D_{sc}/D_{sp}0.1`$. However, these conditions are closely approximated by recent experiments on rod-like $`\mathrm{𝑓𝑑}`$ ($`L=1\mu `$m, $`L/D_{sc}`$ 100) and polystyrene spheres . Therefore, we compare theoretical results of slope $`\tau `$ for spherocylinders with $`L/D_{sc}=100`$ shown in Fig. 8 to these experimental results . When large spheres $`D_{sp}1\mu `$m, $`(D_{sc}/D_{sp}0.01)`$ are mixed with $`\mathrm{𝑓𝑑}`$ at any concentration for which the nematic phase is stable, we observe no formation of the layered phase. Instead, large spheres phase separate into dense aggregates elongated along the nematic director indicating that the value of slope $`\tau `$ is larger then zero. When the size of the sphere was decreased to $`D_{sp}=0.1\mu `$m, $`(D_{sc}/D_{sp}10)`$ we observed a transition to a layered state at a $`\mathrm{𝑓𝑑}`$ concentration of 20 mg/ml. The formation of a smectic phase in a pure fd suspension at the same ionic strength occurs at 65 mg/ml. The fact that adding spheres diminishes the rod density by a factor of three indicates a large negative value of slope $`\tau `$. As the sphere size was further decreased $`D_{sp}=0.022\mu `$m, ($`D_{sc}/D_{sp}=0.46`$) there was again indication of a lamellar phase, but this time at a much higher concentration of rods of about 50 mg/ml. Thus, although small spheres still stabilize the layering transition, implying a negative value of slope $`\tau `$, the magnitude of slope $`\tau `$ is much less for $`D_{sc}/D_{sp}0.46`$ then for $`D_{sc}/D_{sp}0.1`$. These qualitative trends of the non-monotonic behavior of slope $`\tau `$ with sphere size observed in experiments of fd-polystyrene mixtures are very similar to the theoretical prediction shown in Fig. 8 for spherocylinders with $`L/D_{sc}=100`$.
## V Conclusions
In this paper we have presented the predictions of the second virial theory for a mixture of parallel hard-spherocylinders and hard-spheres undergoing one dimensional microphase separation. We have been able to verify a number of these predictions using Monte Carlo simulations. We found that spheres induce layering, which implies a negative value of the slope $`\tau `$, which is the change in total volume fraction of the mixture at the point of nematic-smectic instability with respect to the partial volume fraction of added spheres (Eq. 1) . At the same time the magnitude of the slope $`\tau `$ increases with increasing spherocylinder length. In other words, spheres at the same partial volume fraction stabilize layering of longer spherocylinders more then shorter spherocylinders. Besides this, the theory predicts an unusual non-monotonic behavior in slope $`\tau `$ as a function of sphere to spherocylinder diameter. Although the physical origin of this effect is not clear, it is intriguing that similar qualitative trends are observed in experiments of mixtures of the spherocylinder-like $`\mathrm{𝑓𝑑}`$ and polystyrene spheres. However, in real experiments spherocylinders are free to rotate, are flexible, and have charge associated with them. Before quantitative comparisons with experiments are possible it will be necessary to perform simulations and formulate theories that take into account these effects mostly ignored in this highly idealized treatment.
## VI Acknowledgments
We acknowledge useful discussions with Richard Sear and Bulbul Chakraborty. We thank Tomonori Koda for critical reading of the manuscript. This research was supported by NSF DMR-9705336 and NSF INT-9113312. The work of the FOM Institute is supported by FOM (“Stichting Fundamenteel Onderzoek der Materie”) with financial aid from NWO (“Nederlandse Organisatie voor Wetenschappelijk Onderzoek”).
## VII Appendix
A general expression for the free energy of bidisperse mixture at the second virial level is
$`\beta F(\rho _1,\rho _2)={\displaystyle \underset{i=1,2}{}}{\displaystyle _V}d(𝐫)\rho _i(𝐫)\mathrm{ln}(\rho _i(𝐫))`$ (3)
$`{\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1,2}{}}{\displaystyle \underset{j=1,2}{}}{\displaystyle _V}𝑑𝐫_\mathrm{𝟏}{\displaystyle _V}𝑑𝐫_\mathrm{𝟐}\rho _i(𝐫_\mathrm{𝟏})\rho _j(𝐫_\mathrm{𝟐})f_{i,j}(𝐫_\mathrm{𝟏},𝐫_\mathrm{𝟐})`$ (4)
where the function $`f_{i,j}`$ is the overlap function between two spheres, sphere and spherocylinder or two spherocylinders . It attains the value of -1 if two particles overlap, otherwise it is equal to 0. The terms involving $`\rho \mathrm{ln}\rho `$ represent the entropy of mixing while the terms involving $`f_{i,j}`$ represent the free volume entropy. Since we are interested in one dimensional layering we look at the response of the system to following density pertubation
$`\delta \rho _1(z)=a_1\mathrm{cos}(k_zz)`$ (5)
$`\delta \rho _2(z)=a_2\mathrm{cos}(k_zz)`$ (6)
The free energy difference between the uniform and perturbed state is
$$\delta F=F(1+\delta \rho _1(z),1+\delta \rho _2(z))F(1,1)=\stackrel{~}{𝐚}\mathrm{𝐒𝐚}$$
(7)
where $`\stackrel{~}{𝐚}=(a_1,a_2)`$ and $`𝐒`$ is a two dimensional stability matrix. To find the limit of stability we have to solve the equation $`\text{det}(𝐒)=0`$. For latter convenience we define the following function
$`S({\displaystyle \frac{L}{D_{sc}}},\sigma ,k)`$ $`=`$ $`{\displaystyle \frac{3\mathrm{sin}(k\sigma (2+2\frac{L}{D_{sc}}))}{4k^3}}`$ (9)
$`{\displaystyle \frac{2k\sigma \mathrm{cos}(k\sigma (2+2\frac{L}{D_{sc}}))\mathrm{sin}(k2\sigma \frac{L}{D_{sc}})}{4k^3}}`$
.
The above expression depends only on geometrical factors and is related to the Fourier transform of the spherocylinder which is specified by the excluded volume between a sphere of diameter $`D_{sp}`$ and a spherocylinder of length $`L`$ and diameter $`D_{sc}`$. Wavevector $`k`$ is dimensionless because it is rescaled with the spherocylinder diameter ($`D_{sc}`$). The parameter $`\sigma `$ is defined as ratio of sphere diameter to spherocylinder diameter $`(\sigma =D_{sp}/D_{sc})`$. In the limit of $`L/D_{sc}0`$ the above expression reduces to a Fourier transform of a sphere with unit diameter. The stability matrix $`𝐒`$ for a mixture of spherocylinders and spheres has the following form
$`𝐒=(\begin{array}{cc}\frac{\eta \left(1\rho _{sp}\right)\left(1+4\left(1\rho _{sp}\right)\eta S(0,1,k)\right)}{4}& \\ \frac{2\rho _{sp}\left(1\rho _{sp}\right)\eta ^2S({\displaystyle \frac{L}{D_{sc}}},1+\sigma ,k)}{\sigma ^6\left({\displaystyle \frac{2}{3}}{\displaystyle \frac{L}{D_{sc}}}+1\right)^2}& \end{array}`$ (12)
$`\begin{array}{cc}\frac{2\rho _{sp}\left(1\rho _{sp}\right)\eta ^2S({\displaystyle \frac{L}{D_{sc}}},1+\sigma ,k)}{\sigma ^6\left({\displaystyle \frac{2}{3}}{\displaystyle \frac{L}{D_{sc}}}+1\right)^2}& \\ \frac{\eta \rho _{sp}\left({\displaystyle \frac{\sigma ^6({\displaystyle \frac{3}{2}}{\displaystyle \frac{L}{D_{sc}}}+1)}{4}}+\eta \rho _{sp}S(2{\displaystyle \frac{L}{D_{sc}}},2\sigma ,k)\right)}{\sigma ^6\left({\displaystyle \frac{2}{3}}{\displaystyle \frac{L}{D_{sc}}}+1\right)^2}& \end{array})`$ (15)
where $`\rho _{sp}`$ denotes partial volume fraction of spheres and varies between 0 and 1 while $`\eta `$ denotes total volume fraction. Note that the terms in matrix elements $`S_{11}`$ and $`S_{22}`$ proportional to $`\eta `$ are due to configurational entropy while terms proportional to $`\eta ^2`$ are due to free volume entropy. As $`k0`$ the condition $`\text{det}(𝐒)=0`$ reduces to the usual thermodynamic condition for the stability of the system against bulk phase separation.
To reconstruct the stability diagram from the determinant we slowly increase the total volume fraction $`\eta `$. At a certain value of total volume fraction $`(\eta _c)`$ the determinant of $`𝐒`$ will equal zero for a specific wavevector $`(k_c)`$. If the wavevector $`k_c`$ obtained has a finite value it implies that system is undergoing a layering transition. On the other hand, the condition $`\text{det}(𝐒)=0`$ when $`k_c=0`$ implies complete demixing. Once we obtain values of $`\eta _c`$ and $`k_c`$ we can find out the ratio of amplitudes from the following formula
$$\frac{a_1}{a_2}=\frac{S_{12}(\eta _c,k_c)}{S_{11}(\eta _c,k_c)}.$$
(16)
A positive value of the amplitude ratio implies that the spheres and spherocylinders are in the same layer (the periodic modulations are in phase), while a negative value implies that the spheres and spherocylinders intercalate (the periodic modulations are out of phase). |
warning/0003/astro-ph0003339.html | ar5iv | text | # CORRELATED MIXTURES OF ADIABATIC AND ISOCURVATURE COSMOLOGICAL PERTURBATIONS
## 1 Introduction
The CMB anisotropies can indirectly measure the cosmological parameters by looking at the evolution of the cosmological perturbations between the end of inflation and the recombination epoch. In order to do so, one must implicitly assume a simple form for the initial perturbations. Usually, one considers only adiabatic fluctuations, with a power law spectrum.
Adiabatic fluctuations arise in the context of the simplest inflationary scenario. However, as soon as one has a multiple inflation scenario, isocurvature fluctuations can be generated, the amplitude of which, as well as their correlation with the adiabatic part, depend on the parameters of the model. In many of the models already studied $`^\mathrm{?}`$, the isocurvature and the adiabatic parts of the fluctuations are uncorrelated, but it is possible to have a correlated mixture of such perturbations, as was already stressed by one of us $`^\mathrm{?}`$ in the study of a specific inflation model with two massive non interacting scalar fields.
In this communication, we study this issue in a phenomenological way, and look at the observable consequences of a correlated mixture of adiabatic and isocurvature cosmological perturbations. These perturbations happen to have a richer structure than the more usual uncorrelated adiabatic and isocurvature ones.
## 2 Definitions and notations
When considering a mixture of several fluids, one can define an entropy perturbation $`S_{A,B}`$ for any pair of components $`A`$ and $`B`$. It is non zero as soon as the different particle number density contrasts $`\delta n_X/n_X`$ ($`X=A,B`$), are not equal. It can also be written in terms of energy density contrasts $`\delta _X`$ :
$$S_{A,B}\frac{\delta n_A}{n_A}\frac{\delta n_B}{n_B}=\frac{\delta _A}{1+\omega _A}\frac{\delta _B}{1+\omega _B},$$
(1)
where $`\omega _Xp_X/\rho _X`$ is the equation of state parameter for the species $`X`$. (Note that this definition is gauge-invariant.) Adiabatic initial conditions are defined such that all the entropy perturbations are zero. In cosmology, the perturbations are actually considered as random fields, usually assumed to be Gaussian, and described by their power spectra. When one has to deal with several random fields, one must also impose the form of the cross-correlation between them. In what follows, we will consider only *totally correlated* adiabatic and isocurvature perturbations, where all the random fields are described in terms of a single random variable. We further assume that only one species deviates from adiabaticity. In this case, one simply has to define which species deviates from adiabaticity, say $`X`$ (i.e. $`S_{A,B}=0`$ when $`A,BX`$), and the relative initial amplitude between the entropy perturbation $`S_{X,Y}`$ and the Bardeen potential $`\mathrm{\Phi }`$ (where $`Y`$ is another species which does not deviate from adiabaticity) :
$$S_{X,Y}\lambda \mathrm{\Phi }.$$
(2)
We will then talk about “$`X`$ hybrid perturbation”.
## 3 An analytical estimate
Adiabatic scale invariant initial conditions make two predictions concerning the CMB anisotropies. First, they predict a flat (Sachs-Wolfe) plateau at low multipoles, which illustrates the fact that the gravitational potential is “frozen” as long as the modes have not yet entered into the Hubble radius. Second, one expects to find a serie of Doppler peaks at smaller angular scales, produced by acoustic oscillations in the photon-baryon plasma. The height of the first peak depends on almost all the cosmological parameters, but as soon as one considers adiabatic initial conditions and (very) conservative cosmological parameters, the peak is between $`2`$ and $`8`$ times higher than the Sachs-Wolfe plateau. This is a strong prediction of adiabatic models, and it is in good agreement with the current data. (In opposition, a pure isocurvature CDM model leads to a Sachs-Wolfe plateau higher than the first Doppler peak.) In the case of adiabatic and isocurvature mixtures, the CMB anisotropies and the matter power spectrum correspond, on large scales, to different combinations of the initial perturbations. Such a complementarity is all the more useful in our model that, contrarily to the adiabatic case, it does not generically predict the ratio of amplitude between the CMB and the matter power spectrum. Indeed, in the standard adiabatic case, it is a classic calculation to derive the temperature anisotropy as a function of the gravitational potential. In the long wavelength limit, the temperature anisotropies are one third of the gravitational potential, which has varied of a factor $`9/10`$ during the radiation-to-matter transition. In the case of CDM hybrid perturbations, this relation can be rewritten as :
$$\frac{\delta T}{T}|_{\mathrm{MD}}=\frac{3}{10}\left(1+\frac{4}{15}\mathrm{\Omega }_\nu ^{\mathrm{RD}}\frac{2}{5}\lambda \mathrm{\Omega }_c^{\mathrm{MD}}\right)\mathrm{\Phi }_{\mathrm{RD}},$$
(3)
where $`\mathrm{\Omega }_c`$ and $`\mathrm{\Omega }_\nu `$ are respectively the CDM and neutrino density parameters, and the indexes $`\mathrm{RD}`$ and $`\mathrm{MD}`$ mean that one considers the quantities during the radiation dominated and the matter dominated eras respectively. The relative amplitude of CMB anisotropies and matter power spectrum can therefore in principle be used to extract the isocurvature part of the initial conditions.
## 4 Numerical results
We have extensively studied the four hybrid perturbations in a recent paper $`^\mathrm{?}`$. The main result of our analysis is that the CMB anisotropy and the matter power spectra and rather strongly sensitive to the set of initial conditions we have considered. As a consequence, such models are already fairly well constrained : one cannot deviate strongly from adiabaticity. As an example, we have plotted on Fig. 1 the relative height of the first Doppler peak in the four hybrid models one can consider. Photon and CDM hybrid perturbations are the most constrained. As for the relative amplitude between the CMB anisotropies and the matter power spectrum, we have compared them on Fig. 2. It is clear that the amplitude of the first Doppler peak, which is already well measured by several ground and balloon experiments, is strongly sensitive to the parameter $`\lambda `$. This might help to discriminate between these models and the standard adiabatic model.
## References |
warning/0003/hep-ph0003030.html | ar5iv | text | # Renormalization Scheme and Higher Loop Stability in Hadronic 𝜏 Decay within Analytic Perturbation Theory
## I Introduction
The ratio of hadronic to leptonic widths for the inclusive decay of the $`\tau `$-lepton, $`R_\tau =\mathrm{\Gamma }(\tau ^{}hadrons\nu _\tau )/\mathrm{\Gamma }(\tau ^{}\mathrm{}\overline{\nu }_{\mathrm{}}\nu _\tau )`$, gives important information about the QCD running coupling at relatively small energy scales. The theoretical analysis of the hadronic decay of a heavy lepton was performed in before the experimental discovery of the $`\tau `$-lepton in 1975. Since then, the properties of the $`\tau `$ have been studied very intensively. Numerous publications are devoted to the QCD description of the inclusive decay of the $`\tau `$-lepton and determination of the QCD running coupling $`\alpha _s`$ at the $`\tau `$ mass scale. A detailed consideration of this subject has been given in . Recently, an updated QCD analysis has been performed by the ALEPH and OPAL collaborations, where applications of different theoretical approaches to the $`\tau `$-decay have been analyzed.
At present, the $`R_\tau `$-ratio is known experimentally to high accuracy, $`0.5\%`$. Nevertheless, the value of $`\alpha _s`$ extracted from the data has a rather large error, in which theoretical uncertainties are dominant. For example, the ALEPH Collaboration result is $`\alpha _s(M_\tau =1.777\text{GeV})=0.334\pm 0.007_{\mathrm{expt}}\pm 0.021_{\mathrm{theor}}`$ . It should be emphasized that nonperturbative terms, the values of which are not well known, do not dominate these uncertainties, because their contribution is rather small . The main difficulty is associated with the perturbative description.
The original theoretical expression for the width $`\mathrm{\Gamma }(\tau ^{}hadrons\nu _\tau )`$ involves integration over small values of timelike momentum . The perturbative description with the standard running coupling, which has unphysical singularities, becomes ill-defined in this region and some additional ansatz has to be applied to get a finite result for the hadronic width. To this end, one usually transforms to a contour representation for $`R_\tau `$ , which allows one to give meaning to the initial expression and, in principle, perform calculations in the framework of perturbative QCD. Assuming the validity of this transformation it is possible to present results in the form of a truncated power series with $`\alpha _s(M_\tau )`$ as the expansion parameter . There are also other approaches to evaluating the contour integral. The Le Diberder and Pich prescription allows one to improve the convergence properties of the approximate series and reduce the renormalization scheme (RS) dependence of theoretical predictions. The possibility of using different approaches in the perturbative description of $`\tau `$-decay leads to an uncertainty in the value of $`\alpha _s(M_\tau )`$ extracted from the same experimental data. Moreover, any perturbative description is based on this contour representation, i.e., on the possibility of converting the initial expression involving integration over timelike momenta into a contour integral in the complex momentum plane. To carry out this transition by using Cauchy’s theorem requires certain analytic properties of the hadronic correlator or of the corresponding Adler function. However, the required analytic properties are not automatically maintained in perturbative QCD resummed by the renormalization group. It is well known that at the one-loop level the so-called ghost pole occurs in the invariant charge. Higher-loop corrections do not solve this problem, but merely add some unphysical branch points. The occurrence of incorrect analytic properties in the conventional perturbative approximation makes it impossible to exploit Cauchy’s theorem in this manner and therefore prevents rewriting the initial expression for $`R_\tau `$ in the form of a contour integral in the complex momentum-plane.
In this paper we will use the analytic approach proposed in (see also for details). Being inspired by Käll$`\stackrel{´}{\mathrm{e}}`$n–Lehmann analyticity, which is based on general principles of quantum field theory, this method ensures that the running coupling possesses the correct analytic properties, leads to a self-consistent definition of the effective charge in the timelike region (which cannot be a symmetrical reflection of the spacelike one ), and provides equality between the initial $`R_\tau `$-expression and the corresponding contour representation . A distinguishing feature of the analytic approach is the existence of a universal infrared limiting value of the analytic running coupling at $`q^2=0`$ which is independent of both the QCD scale parameter $`\mathrm{\Lambda }`$ and the choice of renormalization scheme. This limiting value is defined by the general structure of the Lagrangian and turns out to be stable with respect to higher-loop corrections in contrast to the corresponding quantity in conventional perturbation theory (PT). The higher-loop stability of the analytic perturbation theory (APT) holds also for physical observables .
However, it is not sufficient to study the stability with respect to higher-loop corrections; one must also investigate the stability with respect to choice of renormalization scheme. This is also essential in order to estimate the uncertainty of the results obtained. The theoretical ambiguity which is connected with higher-loop corrections and with RS dependence becomes considerable at low energy scales (see, e.g., ). The APT method, as an invariant analytical version of perturbative QCD , improves the situation and gives very stable results over a wide range of renormalization schemes. This has been demonstrated for the $`e^+e^{}`$ annihilation ratio and for the Bjorken and Gross–Llewellyn Smith deep inelastic scattering sum rules.
The main aim of the paper is a study of the RS dependence which appears in the description of the inclusive $`\tau `$ decay within the APT approach. We will consider the $`R_\tau `$-ratio at the next-to-next-to-leading order (NNLO) and the next-to-leading order (NLO) and compare results obtained with those of standard perturbation theory.
## II QCD parametrization of $`R_\tau `$
The ratio of hadronic to leptonic $`\tau `$-decay widths can be written as
$$R_\tau =3S_{\mathrm{EW}}(|V_{ud}|^2+|V_{us}|^2)(1+\delta _{\mathrm{QCD}}),$$
(1)
where $`S_{\mathrm{EW}}=1.0194\pm 0.0040`$ is the electroweak factor, $`|V_{ud}|=0.9752\pm 0.0007`$ and $`|V_{us}|=0.2218\pm 0.0016`$ are the CKM matrix elements, and $`\delta _{\mathrm{QCD}}`$ is the QCD correction (see for details).
We first introduce some definitions: $`\mathrm{Im}\mathrm{\Pi }1+r`$ for the hadronic correlator $`\mathrm{\Pi }(q^2)`$ and $`D1+d`$ for the Adler function $`D(q^2)`$. Then for massless quarks one can write $`\delta _{\mathrm{QCD}}`$ as an integral over timelike momentum $`s`$:
$$\delta _{\mathrm{QCD}}=2_0^{M_\tau ^2}\frac{ds}{M_\tau ^2}\left(1\frac{s}{M_\tau ^2}\right)^2\left(1+2\frac{s}{M_\tau ^2}\right)r(s).$$
(2)
Within the conventional perturbative approximation of $`r(s)`$ this integral is ill-defined due to unphysical singularities of the running coupling lying in the range of integration. The most useful trick to rescue the situation is to appeal to analytic properties of the hadronic correlator $`\mathrm{\Pi }(q^2)`$. This opens up the possibility of exploiting Cauchy’s theorem by rewriting the integral in the form of a contour integral in the complex $`q^2`$-plane with the contour being a circle of radius $`M_\tau ^2`$:
$$\delta _{\mathrm{QCD}}=\frac{1}{2\pi i}_{|z|=M_\tau ^2}\frac{dz}{z}\left(1\frac{z}{M_\tau ^2}\right)^3\left(1+\frac{z}{M_\tau ^2}\right)d(z).$$
(3)
Starting from the contour representation (3) the PT description can be developed in the following two ways (see, e.g., ). One is Braaten’s approach in which the quantity (3) is represented in the form of truncated power series with the expansion parameter $`\alpha _s(M_\tau ^2)`$. The NNLO representation for $`\delta _{\mathrm{QCD}}`$ is written as follows
$$\delta _{\mathrm{QCD}}^{\mathrm{Br}}=a_\tau +r_1a_\tau ^2+r_2a_\tau ^3,$$
(4)
where $`a_\tau \alpha _s(M_\tau ^2)/\pi `$. The coefficients $`r_1`$ and $`r_2`$ in the $`\overline{\mathrm{MS}}`$ scheme with three active flavors are $`r_1=5.2023`$ and $`r_2=26.366`$ . The running coupling satisfies the renormalization group equation:
$$\mu ^2\frac{da}{d\mu ^2}=\frac{b}{2}a^2(1+c_1a+c_2a^2),$$
(5)
where $`b`$, $`c_1`$ and $`c_2`$ are the $`\beta `$-function coefficients. For three active flavors $`b=9/2`$, $`c_1=16/9`$ and $`c_2^{\overline{\mathrm{MS}}}=3863/864`$.
In the approach of Le Diberder and Pich (LP) , the PT expansion is applied to the $`d`$-function<sup>*</sup><sup>*</sup>*We use the definition $`q^2<0`$ in the Euclidean region. We have made a few changes in notation from that given in : now $`a=\alpha _s/\pi `$, and consequently $`d_1`$ and $`d_2`$ are what we called $`d_2`$ and $`d_3`$ previously.
$$d(q^2)=a(q^2)+d_1a^2(q^2)+d_2a^3(q^2),$$
(6)
where in the $`\overline{\mathrm{MS}}`$-scheme $`d_1^{\overline{\mathrm{MS}}}=1.6398`$ and $`d_2^{\overline{\mathrm{MS}}}=6.3710`$ for three active quarks. Substituting Eq. (6) into Eq. (3) leads to the following expansion, which is not a power series in $`a`$,
$$\delta _{\mathrm{QCD}}^{\mathrm{LP}}=A^{(1)}(a)+d_1A^{(2)}(a)+d_2A^{(3)}(a)$$
(7)
with
$$A^{(n)}(a)=\frac{1}{2\pi i}_{|z|=M_\tau ^2}\frac{dz}{z}\left(1\frac{z}{M_\tau ^2}\right)^3\left(1+\frac{z}{M_\tau ^2}\right)a^n(z).$$
(8)
As noted above, transition to the contour representation (3) requires certain analytic properties of the correlator, namely, that it must be an analytic function in the complex $`q^2`$-plane with a cut along the positive real axis. The correlator parametrized by the PT running coupling does not have this virtue . Moreover, the conventional renormalization group method determines the running coupling in the spacelike region, whereas the initial expression (2) for $`R_\tau `$ contains an integration over timelike momentum. Thus, we are in need of some method of continuing the running coupling from the spacelike to the timelike region that takes into account the proper analytic properties of the running coupling . Because of this failure of analyticity, Eqs. (2) and (3) are not equivalent in the framework of PT and if one remains within PT, nothing can be said about the errors introduced by this transition.
The analytic approach may eliminate these problems. To make our analysis more transparent and to demonstrate clearly the differences between the consequences of the PT and APT methods, we restrict our consideration here to massless NNLO. The NNLO analysis can be performed in a more rigorous way without model assumptions that allows us to avoid minor details and exhibit the principal features of the APT approach. Thus, other effects, such as nonperturbative terms, higher-loop corrections, and renormalon contributions lie outside of the purpose of this paper. Note, however, that the NNLO approximation is adequate to describe the actual physical situation because numerically the corresponding terms give the principal contribution to the $`R_\tau `$-ratio.
The function $`d(q^2)`$, which is analytic in the cut $`q^2`$-plane, can be expressed in terms of the effective spectral function $`\rho (\sigma )`$, the basic quantity in the APT method,
$$d(q^2)=\frac{1}{\pi }_0^{\mathrm{}}\frac{d\sigma }{\sigma q^2}\rho (\sigma ).$$
(9)
The connection between the QCD corrections to the $`D`$\- and $`R`$-functions can be written down in the form of the dispersion integral
$$d(q^2)=q^2_0^{\mathrm{}}\frac{ds}{(sq^2)^2}r(s),$$
(10)
which is inverted by the following formula
$$r(s)=\frac{1}{2\pi i}_{siϵ}^{s+iϵ}\frac{dz}{z}d(z).$$
(11)
Here, the contour lies in the region of analyticity of the $`D`$-function. In terms of $`\rho (\sigma )`$ the function $`r(s)`$ defined for timelike momenta can be expressed as follows :
$$r(s)=\frac{1}{\pi }_s^{\mathrm{}}\frac{d\sigma }{\sigma }\rho (\sigma ).$$
(12)
Eqs. (9) and (12) determine the QCD corrections $`d(q^2)`$, which is defined in the Euclidean (spacelike) region of momenta, and $`r(s)`$ defined for the Minkowskian (timelike) argument, in terms of the spectral function $`\rho (\sigma )`$. For $`\delta _{\mathrm{QCD}}`$, using Eq. (2) or equivalently Eq. (3), in terms of $`\rho (\sigma )`$, we find To distinguish APT and PT cases, we will use subscripts “an” and “pt”.
$$\delta _{\mathrm{an}}=\frac{1}{\pi }_0^{\mathrm{}}\frac{d\sigma }{\sigma }\rho (\sigma )\frac{1}{\pi }_0^{M_\tau ^2}\frac{d\sigma }{\sigma }\left(1\frac{\sigma }{M_\tau ^2}\right)^3\left(1+\frac{\sigma }{M_\tau ^2}\right)\rho (\sigma ).$$
(13)
In the APT approach, the spectral function is defined as the imaginary part of the perturbative approximation to $`d_{\mathrm{pt}}(q^2)`$ on the physical cut:
$$\rho (\sigma )=\varrho _0(\sigma )+d_1\varrho _1(\sigma )+d_2\varrho _2(\sigma ),$$
(14)
where
$$\varrho _n(\sigma )=\mathrm{Im}[a^{n+1}(\sigma +iϵ)].$$
(15)
Substituting Eq. (14) into Eq. (13), we can rewrite $`\delta _{\mathrm{an}}`$ in the form of the APT expansion
$$\delta _{\mathrm{an}}=\delta ^{(0)}+d_1\delta ^{(1)}+d_2\delta ^{(2)}.$$
(16)
Note that the APT representations of the $`d`$-function and the QCD correction $`\delta _{\mathrm{QCD}}`$ are not in the form of power series.
The function $`\varrho _0(\sigma )`$ in Eq. (14) defines the analytic spacelike running coupling
$$a_{\mathrm{an}}(q^2)=\frac{1}{\pi }_0^{\mathrm{}}\frac{d\sigma }{\sigma q^2}\varrho _0(\sigma ).$$
(17)
In the one-loop approximation it leads to
$$a_{\mathrm{an}}(q^2)=a_{\mathrm{pt}}(q^2)+\frac{2}{b}\frac{\mathrm{\Lambda }^2}{\mathrm{\Lambda }^2+q^2}.$$
(18)
Unlike the one-loop PT running coupling, $`a_{\mathrm{pt}}(q^2)=2/b\mathrm{ln}(q^2/\mathrm{\Lambda }^2)`$, the analytic running coupling (18) has no unphysical ghost pole and, therefore, possesses the correct analytic properties, arising from Källén-Lehmann analyticity that reflects the general principles of the theory. The nonperturbative (non-logarithmic) term, which appears in the analytic running coupling, does not change the ultraviolet limit of the theory and thus the APT and the PT approaches coincide with each other in the asymptotic region of high energies.
Thus, the APT approach provides a self-consistent description of the hadronic $`\tau `$ decay. This description can be equivalently phrased either on the basis of the original expression (2), which involves the Minkowskian quantity $`r(s)`$, or on the contour representation (3), which involves the Euclidean quantity $`d(q^2)`$.
An important feature of the APT approach is the fact that $`d_{\mathrm{an}}(q^2)`$ and $`a_{\mathrm{an}}(q^2)`$ have a universal limit at the point $`q^2=0`$. This limiting value, generally, is independent of both the scale parameter $`\mathrm{\Lambda }`$ and the order of the loop expansion being considered. Because $`d_{\mathrm{an}}(0)`$ and $`a_{\mathrm{an}}(0)`$ are equal to the reciprocal of the first coefficient of the QCD $`\beta `$-function, they are also RS invariant (we consider only gauge- and mass-independent RSs). The existence of this fixed point plays a decisive role in the improved convergence properties relative to PT and in the very weak RS dependence of our results.
To find the analytic function $`d(q^2)`$ involved in Eq. (3), we solve the transcendental equation for the running coupling
$$\frac{b}{2}\mathrm{ln}\left(\frac{q^2}{\mathrm{\Lambda }_{\overline{\mathrm{MS}}}^2}\right)i\pi \frac{b}{2}=d_1^{\overline{\mathrm{MS}}}d_1+\frac{1}{a}+c_1\mathrm{ln}\left(\frac{b}{2c_1}\right)+F^{(l)}(a),$$
(19)
where at the NLO
$$F^{(2)}(a)=c_1\mathrm{ln}\left(\frac{c_1a}{1+c_1a}\right),$$
(20)
and at the NNLO
$$F^{(3)}(a)=F^{(2)}(a)+c_2_0^a\frac{dx}{(1+c_1x)(1+c_1x+c_2x^2)},$$
(21)
on the physical cut lying along the positive real axis in the complex $`q^2`$-plane and then use Eqs. (14), (15) and (9). Eq. (19) holds in any MS-like renormalization scheme and allows us to normalize the results obtained by using the scale parameter $`\mathrm{\Lambda }_{\overline{\mathrm{MS}}}`$. Having found $`\mathrm{\Lambda }_{\overline{\mathrm{MS}}}`$, we can study how $`\delta _{\mathrm{an}}`$ varies with a change of renormalization scheme. To do that one has to select parameters which determine the RS. The function $`d(q^2)`$ in Eq. (3) is parametrized by a set of RS-dependent parameters. There are RS invariant combinations which constrain these parameters . At the NNLO there are two RS-invariant quantities; the first of them expresses an energy dependence, the second is just a number
$$\omega _2=c_2+d_2c_1d_1d_{1}^{}{}_{}{}^{2},$$
(22)
which in our case equals 5.2378. Here, $`c_1`$ is RS invariant and we can choose $`d_1`$ and $`c_2`$ as independent variables, which define some RS.
There are no fundamental principles upon which one can choose one or another preferable RS. Nevertheless, a natural way of studying the RS dependence is to supplement results in a certain scheme with an estimate of the variability of the predictions over a range of a priori acceptable schemes specified by some criterion. In it was proposed to consider the class of ‘natural’ RSs, which obey the condition
$$|c_2|+|d_2|+c_1|d_1|+d_{1}^{}{}_{}{}^{2}C|\omega _2|.$$
(23)
This inequality is called the “cancellation index criterion” which means that the degree of cancellation in the second RS invariant (22) should not be too large. To define a boundary of ‘acceptable’ schemes which is defined by the value of the cancellation index $`C`$, we will require no more cancellation than that which occurs in the scheme obeying the principle of minimal sensitivity (PMS) , which leads to $`C2`$.
## III APT: convergence properties and RS dependence
For various physical quantities, the APT approach allows one to construct a series that has improved convergence properties as compared to a perturbative expansion. To demonstrate this fact for the hadronic $`\tau `$-decay, we compare the convergence properties of the PT expansions (4) and (7) on the one hand, and the APT approach given by Eq. (16) on the other hand. For our calculation we take as input the TAU’98 conference value: $`R_\tau =3.642\pm 0.019`$ , which is consistent with the PDG’98 fit $`R_\tau =3.642\pm 0.024`$ . In Table I we present NNLO results obtained by the methods mentioned above for the central experimental value in the $`\overline{\mathrm{MS}}`$ scheme. The relative contributions of higher-order terms depends on the method which is applied. The convergence properties of the APT expansion seem to be much improved compared to those of the PT expansions.
The values of the scale parameter $`\mathrm{\Lambda }_{\overline{\mathrm{MS}}}`$ and the coupling $`\alpha _s(M_\tau ^2)`$ obtained from above PT and APT expansions are noticeably different from each other. The corresponding numerical estimations are given in Table II, in which, in order to clarify the situation concerning higher-loop stability of different expansions, we also present the NLO result. This table demonstrates that the theoretical ambiguity, which associated with different versions of the perturbative description, leads to a rather large uncertainty, $`\alpha _{\mathrm{PT}(\mathrm{Br})}^{\mathrm{NNLO}}\alpha _{\mathrm{PT}(\mathrm{LP})}^{\mathrm{NNLO}}=0.012`$. At the same time the experimental error is $`\mathrm{\Delta }\alpha _{\mathrm{expt}}=0.007`$$`0.009`$ . The distinction between NLO and NLLO running coupling values is $`12\%`$ for PT (Br) and $`5\%`$ for PT (LP) approaches, while for the APT approach it is less than $`0.5\%`$.
The non-logarithmic terms, which ensure the correct analytic properties and allow a self-consistent description of $`\tau `$ decay, turn out to be very important for the numerical analysis and influence in an essential way the value of $`\mathrm{\Lambda }`$ parameter extracted from the data. Indeed, at the one-loop level one can write a simple relation: $`\delta _{\mathrm{an}}(\mathrm{\Lambda })\delta _{\mathrm{pt}}^{\mathrm{LP}}(\mathrm{\Lambda })(2/b)\mathrm{\Lambda }^2/M_\tau ^2`$. The second term, which is ‘invisible’ in the perturbative expansion, turns out to be numerically important (see the detailed discussion in ). Note also that there is a difference between the shapes of the analytic and perturbative running couplings, for example, $`\alpha _{\mathrm{an}}(\mathrm{\Lambda }=907\text{MeV})=0.403`$, while at the same scale, the value of the perturbative coupling much larger, $`\alpha _{\mathrm{pt}}(\mathrm{\Lambda }=907\text{MeV})=0.796`$. Here the question may arise, how is the large APT value of $`\mathrm{\Lambda }`$ consistent with high energy experimental data? We have estimated the ratio of hadronic to leptonic Z-decay widths, $`R_\mathrm{Z}`$, using the above value of $`\mathrm{\Lambda }_{\mathrm{an}}`$ and the matching procedure proposed in . We obtained the value $`R_\mathrm{Z}=20.82`$, which lies within the range of experimental errors; for example, the PDG’98 average is $`\mathrm{R}_\mathrm{Z}=20.77\pm 0.07`$ . This fact can be understood if one takes into account that there are differences between the shapes of the analytic and perturbative running couplings and also in the terms of the corresponding series.
We found that the value of $`\delta _{\mathrm{an}}`$ depends so slightly on $`\mathrm{\Lambda }_{\overline{\mathrm{MS}}}`$ that a $`0.9\%`$ error in $`R_\tau `$ gives $`18\%`$ error in the value of $`\mathrm{\Lambda }_{\overline{\mathrm{MS}}}`$. (This is the reason why the errors in the values of $`\mathrm{\Lambda }_{\overline{\mathrm{MS}}}`$ and $`\alpha (M_\tau ^2)`$ given by APT are larger than those in PT.) We illustrate this feature in Table III. According to the table, when we change $`M_\tau ^2/\mathrm{\Lambda }^2`$ from 2.0 to 6.5 (corresponding to a variation of $`\mathrm{\Lambda }`$ from 1.256 GeV to 0.697 GeV), $`\delta _{\mathrm{an}}`$ is only altered by about $`20\%`$. The sensitivity to $`\mathrm{\Lambda }_{\overline{\mathrm{MS}}}`$ increases as $`M_\tau ^2/\mathrm{\Lambda }^2`$ gets smaller.
Consider now the RS dependence of the APT result and compare it with the perturbative LP approach,The LP approach is often called the contour-improved fixed-order PT (CIPT) . which of the two PT schemes is more preferable from the viewpoint of sensitivity to RS dependence. In the framework of the PT, the RS dependence of $`\delta _{\mathrm{QCD}}`$ has been discussed in detail in .
In the $`\overline{\mathrm{MS}}`$ scheme we adopt $`\delta _{\mathrm{an},\mathrm{pt}}=0.1906`$ and consider some RS belonging to the domain described above \[see Eq. (23)\]. Take two schemes, $`A`$ and $`B`$, located at the two lower corners of the boundary of the domain (see Fig. 1), i.e., they have the same cancellation index as does the PMS scheme, with $`A=(1.6183,0)`$ and $`B=(0.9575,0)`$, where the first coordinate is $`d_1`$ and the second is $`c_2`$. Then for the PT case in NNLO we get $`\delta _{\mathrm{pt}}(A)=0.2025`$ and $`\delta _{\mathrm{pt}}(B)=0.1911`$. Therefore, even for this sufficiently narrow class of RS the perturbative approach gave a $`6\%`$ deviation in $`\delta _{\mathrm{QCD}}`$ that corresponds to a RS uncertainty for the running coupling value in the $`\overline{\mathrm{MS}}`$-scheme of $`\mathrm{\Delta }\alpha _{\mathrm{pt}}^{\mathrm{RS}}=0.0153`$. The difference between APT results is much smaller: $`\delta _{\mathrm{an}}(A)=0.1890`$ and $`\delta _{\mathrm{an}}(B)=0.1905`$, and we have only $`0.8\%`$ deviation, which corresponds in the $`\overline{\mathrm{MS}}`$-scheme to a RS uncertainty of $`\mathrm{\Delta }\alpha _{\mathrm{an}}^{\mathrm{RS}}=0.0035`$. The similar RS stability holds also at the two-loop level: one has a $`5\%`$ deviation in the PT case and only a $`0.4\%`$ for the APT. We display our NNLO results in the form of a contour plot, in Fig. 1.
It is worthwhile to analyze some schemes lying outside the domain considered above with the relatively small value of the cancellation index $`C2`$. Among them there is, for instance, the commonly used $`\overline{\mathrm{MS}}`$ scheme which does not belong this domain. In it was shown that the so-called $`V`$ scheme lies very far from the domain described above and gives so a large value of $`\delta _{\mathrm{pt}}`$ that it cannot be used at this low energy. For the $`V`$ scheme we have $`d_1=0.109`$ and $`c_2=26.200`$. The three-loop perturbative result is $`\delta _{\mathrm{pt}}(V)=0.3060`$ that corresponds to about a $`61\%`$ deviation from the $`\overline{\mathrm{MS}}`$ scheme. On the other hand, if we turn to APT we have $`\delta _{\mathrm{an}}=0.1902`$, i.e., only about a $`0.2\%`$ deviation from the $`\overline{\mathrm{MS}}`$ scheme. So the $`V`$ scheme is still useful at this energy in APT.
In PT at high energies the weak RS dependence is a consequence of the small value of the coupling constant. At lower energies the uncertainty increases. In APT, at high energies, the situation is the same. However, at low energies the theory has a universal RS-invariant infrared limiting value $`d_{\mathrm{an}}(0)`$, which restricts the RS ambiguity over a very wide range of momentum. Another way to illustrate the remarkable stability of APT is to calculate the spectral functions $`\varrho _n(\sigma )`$ given by Eq. (15); one sees that $`\varrho _1(\sigma )`$ is much smaller than $`\varrho _0(\sigma )`$ over the whole spectral region. The same statement is true for the relationship between $`\varrho _1(\sigma )`$ and $`\varrho _2(\sigma )`$. This monotonically decreasing behavior reduces the RS dependence strongly, since the perturbative coefficients $`d_1`$ and $`d_2`$ in expression (14) for $`\rho (\sigma )`$ are multiplied by these functions. For the $`\overline{\mathrm{MS}}`$ scheme, this situation is demonstrated in Fig. 2.
## IV Conclusion
We have considered inclusive $`\tau `$-decay in three-loop order within analytic perturbation theory concentrating on the analysis of theoretical uncertainties coming from the perturbative short distance part of the QCD correction to the $`R_\tau `$-ratio, which defines the principal contribution to this physical quantity. For the low energy $`\tau `$-mass scale, the main source of theoretical uncertainties results from the inevitable truncation of the perturbative series, which leads to the essential RS dependence and higher loop sensitivity of the theoretical predictions. In order to resolve this problem within the conventional perturbative approach it is possible to try, in principle, to compute higher loop contributions. However, even if this were to be done, one has to keep in mind that from the rigorous point of view it will hardly be sufficient because the series is asymptotic, and, in any finite order, the analytic properties of the hadronic correlator, which arise from general principles of the theory, are violated. Thus, to resolve this problem one has to use a modification of the perturbative expansion at low energy scales.
Here, we have applied the analytic approach which is not inconsistent with the general principles of quantum field theory and which opens up the possibility of reducing the theoretical uncertainties associated with short distance contributions mentioned above. Let us summarize the important features of this method: (i) the method maintains the correct analytic properties and leads to a self-consistent definition of the procedure of analytic continuation from the spacelike to the timelike region; (ii) the APT approach has much improved convergence properties and turns out to be stable with respect to higher-loop corrections; (iii) the RS dependence of the results obtained is reduced drastically. For example, the $`V`$ scheme, which gives a very large discrepancy in standard perturbation theory, can be used in analytic perturbation theory without any difficulty and the APT predictions are practically RS independent over a wide region of RS parameters.
The nonperturbative power corrections coming from the operator product expansion (in this connection see a discussion in ), renormalon and other effects are beyond the scope of our present consideration. Note, however, that the process of enforcing analyticity modifies the perturbative contributions by incorporating some nonperturbative terms. The form of the APT running coupling and the non-power structure of the APT expansion are essentially different from the PT ones. Numerically, this difference becomes very important in the region less than a few GeV and in order to get the same physical quantity the contribution of power corrections should also be changed.
The value of $`\mathrm{\Lambda }_{\mathrm{APT}}`$ is very sensitive to the experimental value of $`R_\tau `$. For example, as has been demonstrated in the use of the value of $`R_\tau `$ obtained by the CLEO collaboration gives a value of the scale parameter some 30% smaller than that found here. Note also that the renormalon contribution influences the value of $`\mathrm{\Lambda }`$ extracted from the $`\tau `$ data (see for a review). Within the usual approach, renormalons reduce the value of $`\alpha _s(M_\tau ^2)`$ by about $`15\%`$. A similar situation holds also in APT and for the nonperturbative $`a`$-expansion approach , which allows one, as in APT, to maintain the required analyticity . These two analytic approaches often lead to rather similar consequences. For example, they allow one to get a good description of experimental data corresponding to the Euclidean and Minkowskian characteristics of the process of $`e^+e^{}`$ annihilation into hadrons down to the lowest energy scale .
Pure massless APT analysis, which has been performed here, leads to an unusually large value of the QCD scale parameter $`\mathrm{\Lambda }`$ as compared to the conventional PT value. This is connected with the presence of nonperturbative contributions that appear in the APT method which have a negative relative sign. The effects mentioned above can change the value of the scale parameter extracted from the $`\tau `$ data. However, this fact is not relevant for the essential conclusion which we have claimed in this paper, that the APT method provides predictions which are stable with respect to the choice of renormalization scheme and to the inclusion of higher loop corrections. Thus, the analytic approach discussed here is not in conflict with the general principles of the theory and allows one to reduce the uncertainties of theoretical predictions drastically.
## Acknowledgements
The authors would like to thank D.V. Shirkov for interest in this work. Partial support of the work by the US National Science Foundation, grant PHY-9600421, and by the US Department of Energy, grant DE-FG-03-98ER41066, and by the RFBR, grants 99-01-00091, 99-02-17727 is gratefully acknowledged. The work of ILS and OPS is also supported in part by the University of Oklahoma, through its College of Arts and Science, the Vice President for Research, and the Department of Physics and Astronomy. |
warning/0003/cond-mat0003121.html | ar5iv | text | # The Statistical Physics of Regular Low-Density Parity-Check Error-Correcting Codes
## I Introduction
Error-correcting codes are commonly used for reliable data transmission through noisy media, especially in the case of memoryless communication where corrupted messages cannot be repeatedly sent. These techniques play an important role in a wide range of applications from memory devices to deep space explorations, and are expected to become even more important due to the rapid development in mobile phones and satellite-based communication.
In a general scenario, the sender encodes an $`N`$ dimensional Boolean message vector $`𝝃`$, where $`\xi _i(0,1),i`$, to an $`M(>N)`$ dimensional Boolean codeword $`𝒛_0`$, which is then being transmitted through a noisy communication channel. Noise corruption during transmission can be modelled by the noise vector $`𝜻`$, where corrupted bits are marked by the value 1 and all other bits are zero, such that the received corrupted codeword takes the form $`𝒛=𝒛_0+𝜻\text{(mod 2)}`$. The received corrupted message is then decoded by the receiver for retrieving the original message $`𝝃`$.
The error-correcting ability comes at the expense of information redundancy. Shannon showed in his seminal work that error-free communication is theoretically possible if the code rate, representing the fraction of informative bits in the transmitted codeword, is below the channel capacity; in the case of unbiased messages transmitted through a Binary Symmetric Channel (BSC), which we will focus on here, $`R=N/M`$ satisfies
$$R<1+p\mathrm{log}_2p+(1p)\mathrm{log}_2(1p).$$
(1)
The expression on the right is termed Shannon’s bound. However, Shannon’s derivation is non-constructive and the quest for codes which saturate Eq.(1) has been one of the central topics of information theory ever since.
In this paper we examine the efficiency and limitations of Gallager-type error-correcting code, which attracted much interest recently among researchers in this field. This code was discovered almost forty years ago by Gallager but was abandoned shortly after its invention due to the computational limitations of the time. Since their recent rediscovery by MacKay and Neal, different variations of Gallager-type codes have been developed attempting to get as close as possible to saturating Shannon’s bound.
In this paper we will examine the typical properties of a family of codes based on one variation, the MN code , using the established methods of statistical physics, to provide a theoretical study based on the typical performance of codes rather on the worst case analysis.
This paper is organised as follows: In the next two sections, we introduce Gallager-type error-correcting codes in detail and link them to the statistical mechanics framework. We then examine the equilibrium properties of various members of this family of codes using the replica method (section IV) and compare the bit-error rate below criticality. In section V, we examine the relation between Belief-Propagation (BP) decoding and the Thouless-Anderson-Palmer (TAP) approach to diluted spin systems; we then use it for comparing the analytical results obtained via the replica method to those obtained from simulations using BP decoding. In section VI we show a computationally efficient construction for one of the more practical constructions. Finally, we present conclusions for the current work and suggest future research directions.
## II Gallager-type error-correcting codes
There are several variations in Gallager-type error-correcting codes. The one discussed in this paper is termed the MN code, recently introduced by MacKay and Neal. In these codes, a Boolean message $`𝝃`$ is encoded into a codeword $`𝒛_0`$ using two randomly constructed Boolean sparse matrices $`C_s`$ and $`C_n`$, which are characterised in the following manner.
The rectangular sparse matrix $`C_s`$ is of dimensionality $`M\times N`$, having randomly chosen $`K`$ non-zero unit elements per row and $`C`$ per column. The matrix $`C_n`$ is an $`M\times M`$ (mod 2)-invertible matrix having randomly chosen $`L`$ non-zero elements per row and column. These matrices are shared by the sender and the receiver.
Using these matrices, one can encode a message $`𝝃`$ into a codeword $`𝒛_0`$ in the following manner
$$𝒛_0=C_n^1C_s𝝃\text{(mod 2)},$$
(2)
which is then transmitted via a noisy channel. Note that all matrix and vector components are Boolean $`(0,1)`$, and all summations are carried out in this field. For simplicity, the noise process is modelled hereafter by a binary symmetric channel (BSC), where each bit is independently flipped with probability $`p`$. Extending the code presented here to other types of noise is straightforward.
During transmission, a noise vector $`𝜻`$ is added to $`𝒛_0`$ and a corrupted codeword $`𝒛=𝒛_0+𝜻`$ (mod 2) is received at the other end of the channel. Decoding is then carried out by taking the product of the matrix $`C_n`$ and the received codeword $`𝒛`$, which results in $`C_s𝝃+C_n𝜻=C_n𝒛𝑱`$. The equation
$$C_s𝑺+C_n𝝉=𝑱\text{(mod 2)},$$
(3)
is solved via the iterative methods of belief propagation (BP) to obtain the most probable Boolean vectors $`𝑺`$ and $`𝝉`$. BP methods in this context have recently been shown to be identical to a TAP based solution of a similar physical system.
## III A Statistical Mechanics Perspective
Sourlas was the first to point out that error-correcting codes of this type have a similarity to Ising spin systems in statistical physics; he demonstrated this using a simple version of the same nature. His work, that focused on extensively connected systems, was recently extended to finitely connected systems. We follow a similar approach in the current investigation; preliminary results have already been presented in.
To facilitate the statistical physics analysis, we first employ the binary representation $`(\pm 1)`$ of the dynamical variables $`𝑺`$ and $`𝝉`$ and of the check vector $`𝑱`$ rather than the Boolean one $`(0,1)`$. The $`\mu `$-th component of Eq.(3) is then rewritten as
$$\underset{i_s(\mu )}{}S_i\underset{j_n(\mu )}{}\tau _j=J_\mu ,$$
(4)
where $`_s(\mu )`$ and $`_n(\mu )`$ are the sets of all indices of non-zero elements in row $`\mu `$ of the sparse matrices $`C_s`$ and $`C_n`$, respectively. The check $`\mu `$ is given by message $`𝝃`$ and noise $`𝜻`$ as $`J_\mu =_{i_s(\mu )}\xi _i_{j_n(\mu )}\zeta _j`$; it should be emphasised that the message vector $`𝝃`$ and the noise vector $`𝜻`$ themselves are not known to the receiver.
An interesting link can now be formulated between the Bayesian framework of MN codes and Ising spin systems. Rewriting Kronecker’s delta for binary variables $`x`$ and $`y`$ as $`\delta [x;y]=(1+xy)/2=lim_\beta \mathrm{}\mathrm{exp}(\beta \delta [1;xy])`$, one may argue that, using it as a likelihood, equation (4) gives rise to the conditional probability of the check $`𝑱`$ for given $`𝑺`$, $`𝝉`$, $`C_s`$ and $`C_n`$
$`𝒫(𝑱|𝑺,𝝉,C_s,C_n)=\underset{\beta \mathrm{}}{lim}\mathrm{exp}\left(\beta {\displaystyle \underset{\mu =1}{\overset{M}{}}}\delta [1;J_\mu {\displaystyle \underset{i_s(\mu )}{}}S_i{\displaystyle \underset{j_n(\mu )}{}}\tau _j]\right).`$ (5)
Prior knowledge about possibly biased message and noise is represented by the prior distributions
$`𝒫_s(𝑺)={\displaystyle \frac{\mathrm{exp}\left(F_s\underset{i=1}{\overset{N}{}}S_i\right)}{\left(2\mathrm{cosh}F_s\right)^N}},𝒫_n(𝝉)={\displaystyle \frac{\mathrm{exp}\left(F_n\underset{j=1}{\overset{M}{}}\tau _j\right)}{\left(2\mathrm{cosh}F_n\right)^M}},`$ (6)
respectively. Non-zero field $`F_s`$ is introduced for biased message and $`F_n`$ is determined by flip rate $`p`$ of channel noise as $`F_n=(1/2)\mathrm{ln}\left((1p)/p\right)`$. Using equations (5) and (6), the posterior distribution of $`𝑺`$ and $`𝝉`$ for given check $`𝑱`$ and matrices $`C_s`$ and $`C_n`$ is of the form
$`𝒫(𝑺,𝝉|𝑱,C_s,C_n)`$ $`=`$ $`{\displaystyle \frac{𝒫(𝑱|𝑺,𝝉,C_s,C_n)𝒫_s(𝑺)𝒫_n(𝝉)}{𝒫\left(𝑱|C_s,C_n\right)}}`$ (7)
$`=`$ $`\underset{\beta \mathrm{}}{lim}{\displaystyle \frac{\mathrm{exp}\left(\beta (𝑺,𝝉|𝓙,𝒟)\right)}{𝒵(𝓙,𝒟)}},`$ (8)
where $`𝒫(𝑱|C_s,C_n)=_{\{𝑺,𝝉\}}𝒫(𝑱|𝑺,𝝉,C_s,C_n)𝒫_s(𝑺)𝒫_n(𝝉)`$,
$`(𝑺,𝝉|𝓙,𝒟)`$ $`=`$ $`{\displaystyle \underset{\mu =1}{\overset{M}{}}}\delta [1;J_\mu {\displaystyle \underset{i_s(\mu )}{}}S_i{\displaystyle \underset{j_n(\mu )}{}}\tau _j]{\displaystyle \frac{F_s}{\beta }}{\displaystyle \underset{i=1}{\overset{N}{}}}S_i{\displaystyle \frac{F_n}{\beta }}{\displaystyle \underset{j=1}{\overset{M}{}}}\tau _j`$ (9)
$`=`$ $`{\displaystyle \underset{<i_1,..,i_K;j_1,..,j_L>}{}}𝒟_{<i_1,..,i_K;j_1,..,j_L>}\delta [1;𝒥_{<i_1,..,i_K;j_1,..,j_L>}S_{i_1}\mathrm{}S_{i_K}\tau _{j_1}\mathrm{}\tau _{j_L}]`$ (10)
$`{\displaystyle \frac{F_s}{\beta }}{\displaystyle \underset{i=1}{\overset{N}{}}}S_i{\displaystyle \frac{F_n}{\beta }}{\displaystyle \underset{j=1}{\overset{M}{}}}\tau _j,`$ (11)
and
$`𝒵(𝓙,𝒟)`$ $`=`$ $`\underset{\beta \mathrm{}}{lim}{\displaystyle \underset{\{𝑺,𝝉\}}{}}\mathrm{exp}\left(\beta (𝑺,𝝉|𝓙,𝒟)\right)`$ (12)
$`=`$ $`{\displaystyle \underset{\{𝑺,𝝉\}}{}}{\displaystyle \underset{<i_1,..,i_K;j_1,..,j_L>}{}}\left[1+{\displaystyle \frac{1}{2}}𝒟_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}\left(𝒥_{<i_1,..,i_K;j_1,..,j_L>}S_{i_1}\mathrm{}S_{i_K}\tau _{j_1}\mathrm{}\tau _{j_L}1\right)\right]`$ (13)
$`\times `$ $`\mathrm{exp}\left(F_s{\displaystyle \underset{i=1}{\overset{N}{}}}S_i+F_n{\displaystyle \underset{j=1}{\overset{M}{}}}\tau _j\right).`$ (14)
The final form of posterior distribution (8) implies that the MN code is identical to an Ising spin system defined by the Hamiltonian (11) in the zero temperature limit $`T=\beta ^10`$. In equations (11) and (14), we introduced the sparse connectivity tensor $`𝒟_{<i_1,..,j_L>}`$ which takes the value 1 if the corresponding indices of both message and noise are chosen (i.e., if all corresponding indices of the matrices $`C_s`$ and $`C_n`$ are 1) and 0 otherwise, and coupling $`𝒥_{<i_1,..,i_K;j_1,..,j_L>}=\xi _{i_1}\xi _{i_2}\mathrm{}\xi _{i_K}\zeta _{j_1}\zeta _{j_2}\mathrm{}\zeta _{j_L}`$. These come to isolate the disorder in choosing the matrix connections, embedded in $`𝒟_{<i_1,..,j_L>}`$, and to simplify the notation.
The posterior distribution (8) can be used for decoding. One can show that expectation of the overlap between original message $`𝝃`$ and retrieved one $`\widehat{𝝃}`$
$$m=\frac{1}{N}\underset{i=1}{\overset{N}{}}\xi _i\widehat{\xi }_i,$$
(15)
is maximised by setting $`\widehat{𝝃}`$ to its Bayes-optimal estimator
$$\widehat{\xi }_i^B=\text{sign}(m_i^S),m_i^S=\underset{\{𝑺,𝝉\}}{}S_i𝒫(𝑺,𝝉|𝑱,C_s,C_n).$$
(16)
It is worth while noting that this optimal decoding is realized at zero temperature rather than at a finite temperature as in . The reason is that the true likelihood term (5) corresponds to the ground state of the first term of the Hamiltonian (11) due to the existence of more degrees of freedom, in the form of the dynamical variables $`𝝉`$, which do not appear in other systems. Introducing the additional variables $`𝝉`$, the degrees of freedom in the spin system increase from $`N`$ to $`N+M`$, while the number of constraints from the checks $`𝑱`$ remains $`M`$. This implies that in spite of the existence of quenched disorder caused by $`𝓙`$ and $`𝒟`$, the system is free from frustration even in the low temperature limit, which is useful for practical decoding using local search algorithms. The last two terms in Eq.(11) scale with $`\beta `$ remain finite even in the zero temperature limit $`\beta \mathrm{}`$ representing the true prior distributions, which dominates the statistical properties of the system, while the first term vanishes to satisfy the parity check condition (4).
## IV Equilibrium Properties: The Replica Method
As we use the methods of statistical mechanics, we concentrate on the case of long messages, in the of limit of $`N,M\mathrm{}`$ while keeping code rate $`R=N/M=K/C`$ finite. This limit is quite reasonable for this particular problem since Gallager-type codes are usually used in the transmission of long ($`10^410^5`$) messages, where finite size corrections are likely to be negligible.
Since the first part of the Hamiltonian (11) is invariant under the gauge transformation $`S_i\xi _iS_i`$, $`\tau _j\zeta _j\tau _j`$ and $`𝒥_{i_1,\mathrm{},j_L}1`$, it is useful to decouple the correlation between the vectors $`𝑺`$, $`𝝉`$ and $`𝝃`$, $`𝜻`$. Rewriting the Hamiltonian using this gauge, one obtains a similar expression to Eq.(11) apart from the second terms which become $`F_s/\beta _{i=1}\xi _iS_i`$ and $`F_n/\beta _{j=1}\zeta _j\tau _j`$.
Due to the existence of several types of quenched disorder in the system, it is natural to resort to replica method for investigating the typical properties in equilibrium. More specifically, we calculate expectation values of $`n`$-th power of partition function (14) with respect to the quenched variables $`𝝃`$, $`𝜻`$ and $`𝒟`$ and take the limit $`n0`$.
Carrying out the calculation in the zero temperature limit $`\beta \mathrm{}`$ gives rise to a set of order parameters
$`q_{\alpha ,\beta ,..,\gamma }={\displaystyle \frac{1}{N}}{\displaystyle \underset{i=1}{\overset{N}{}}}Z_iS_i^\alpha S_i^\beta ,..,S_i^\gamma _\beta \mathrm{},r_{\alpha ,\beta ,..,\gamma }={\displaystyle \frac{1}{M}}{\displaystyle \underset{i=1}{\overset{M}{}}}Y_j\tau _j^\alpha \tau _j^\beta ,..,\tau _j^\gamma _\beta \mathrm{}`$ (17)
where $`\alpha `$, $`\beta ,..`$ represent replica indices, and the variables $`Z_i`$ and $`Y_j`$ come from enforcing the restriction of $`C`$ and $`L`$ connections per index, respectively:
$$\delta \left(\underset{i_2,..,i_K}{}𝒟_{<i,i_2,..,j_L>}C\right)=_0^{2\pi }\frac{dZ}{2\pi }Z^{_{i_2,..,i_K}𝒟_{<i,i_2,..,j_L>}(C+1)},$$
(18)
and similarly for the restriction on the $`j`$ indices.
To proceed further, it is necessary to make an assumption about the order parameters symmetry. The assumption made here is that of replica symmetry in both the order parameters and the related conjugate variables
$`q_{\alpha ,\beta ..\gamma }`$ $`=`$ $`a_q{\displaystyle 𝑑x\pi (x)x^l},\widehat{q}_{\alpha ,\beta ..\gamma }=a_{\widehat{q}}{\displaystyle 𝑑\widehat{x}\widehat{\pi }(\widehat{x})\widehat{x}^l}`$ (19)
$`r_{\alpha ,\beta ..\gamma }`$ $`=`$ $`a_r{\displaystyle 𝑑y\rho (y)y^l},\widehat{r}_{\alpha ,\beta ..\gamma }=a_{\widehat{r}}{\displaystyle 𝑑\widehat{y}\widehat{\rho }(\widehat{y})\widehat{y}^l},`$ (20)
where $`l`$ is the number of replica indices, $`a_{}`$ are normalisation coefficients, and $`\pi (x),\widehat{\pi }(\widehat{x}),\rho (y)`$ and $`\widehat{\rho }(\widehat{y})`$ represent probability distributions. Unspecified integrals are over the range $`[1,+1]`$. This ansatz is supported by the facts that (i) the current system is free of frustration and (ii) there has never been observed replica symmetry breaking at Nishimori’s condition which corresponds to using correct priors $`F_s`$ and $`F_n`$ in our case. The results obtained hereafter also support this ansatz. Extremizing the partition function with respect to distributions $`\pi ()`$, $`\widehat{\pi }()`$, $`\rho ()`$ and $`\widehat{\rho }()`$, one then obtains the free energy per spin
$`f={\displaystyle \frac{1}{N}}\mathrm{ln}𝒵_{\xi ,\zeta ,𝒟}`$ (21)
$`=`$ $`\text{extr}_{\{\pi ,\widehat{\pi },\rho ,\rho \}}\{{\displaystyle \frac{C}{K}}\mathrm{ln}2+C{\displaystyle }dxd\widehat{x}\pi (x)\widehat{\pi }(\widehat{x})\mathrm{ln}(1+x\widehat{x})+{\displaystyle \frac{CL}{K}}{\displaystyle }dyd\widehat{y}\rho (y)\widehat{\rho }(\widehat{y})\mathrm{ln}(1+y\widehat{y})`$ (22)
$`{\displaystyle \frac{C}{K}}{\displaystyle \left[\underset{k=1}{\overset{K}{}}dx_k\pi (x_k)\right]\left[\underset{l=1}{\overset{L}{}}dy_l\rho (y_l)\right]\mathrm{ln}\left[1+\underset{k=1}{\overset{K}{}}x_k\underset{l=1}{\overset{L}{}}y_l\right]}`$ (23)
$`{\displaystyle \left[\underset{k=1}{\overset{C}{}}d\widehat{x}_k\widehat{\pi }(\widehat{x}_k)\right]\mathrm{ln}\left[e^{F_s\xi }\underset{k=1}{\overset{C}{}}(1+\widehat{x}_k)+e^{F_s\xi }\underset{k=1}{\overset{C}{}}(1\widehat{x}_k)\right]_\xi }`$ (24)
$`{\displaystyle \frac{C}{K}}{\displaystyle }\left[{\displaystyle \underset{l=1}{\overset{C}{}}}d\widehat{y}_l\widehat{\rho }(\widehat{y}_l)\right]\mathrm{ln}[e^{F_n\zeta }{\displaystyle \underset{l=1}{\overset{L}{}}}(1+\widehat{y}_l)+e^{F_n\zeta }{\displaystyle \underset{l=1}{\overset{L}{}}}(1\widehat{y}_l)]_\zeta \},`$ (25)
where angled brackets with subscript $`𝝃`$, $`𝜻`$ and $`𝒟`$ denote averages over the message and noise distributions respectively, and sparse connectivity tensor $`𝒟`$. Message averages take the form
$$\mathrm{}_\xi =\underset{\xi =\pm 1}{}\frac{1+\xi \mathrm{tanh}F_s}{2}\left(\mathrm{}\right)$$
(26)
and similarly for $`\mathrm{}_\zeta `$. Details of the derivation are given in Appendix A.
Taking the functional variation of $`f`$ with respect to the distributions $`\pi `$, $`\widehat{\pi }`$, $`\rho `$ and $`\widehat{\rho }`$, one obtains the following saddle point equations
$`\pi (x)`$ $`=`$ $`{\displaystyle \underset{l=1}{\overset{C1}{}}d\widehat{x}_l\widehat{\pi }(\widehat{x}_l)\delta \left(x\mathrm{tanh}\left(\xi F_s+\underset{l=1}{\overset{C1}{}}\mathrm{tanh}^1\widehat{x}_l\right)\right)_\xi },`$ (27)
$`\widehat{\pi }(\widehat{x})`$ $`=`$ $`{\displaystyle \underset{l=1}{\overset{K1}{}}dx_l\pi (x_l)\underset{l=1}{\overset{L}{}}dy_l\rho (y_l)\delta \left(\widehat{x}\underset{l=1}{\overset{K1}{}}x_l\underset{l=1}{\overset{L}{}}y_l\right)},`$ (28)
$`\rho (y)`$ $`=`$ $`{\displaystyle \underset{l=1}{\overset{L1}{}}d\widehat{y}_l\widehat{\rho }(\widehat{y}_l)\delta \left(y\mathrm{tanh}\left(\zeta F_n+\underset{l=1}{\overset{L1}{}}\mathrm{tanh}^1\widehat{y}_l\right)\right)_\zeta },`$ (29)
$`\widehat{\rho }(\widehat{y})`$ $`=`$ $`{\displaystyle \underset{l=1}{\overset{K}{}}dx_l\pi (x_l)\underset{l=1}{\overset{L1}{}}dy_l\rho (y_l)\delta \left(\widehat{y}\underset{l=1}{\overset{K}{}}x_l\underset{l=1}{\overset{L1}{}}y_l\right)}.`$ (30)
After solving these equations, the expectation of the overlap between the message $`𝝃`$ and the Bayesian optimal estimator (16), which serves as a performance measure, can be evaluated as
$`m={\displaystyle \frac{1}{N}}{\displaystyle \underset{i=1}{\overset{N}{}}}\xi _i\text{sign}S_i_\beta \mathrm{}_{𝝃,𝜻,𝒟}={\displaystyle 𝑑z\varphi (z)\text{sign}(z)},`$ (31)
where
$`\varphi (z)={\displaystyle \left[\underset{l=1}{\overset{C}{}}d\widehat{x}_l\widehat{\pi }(\widehat{x}_l)\right]\delta \left(z\mathrm{tanh}\left(F_s\xi +\underset{i=1}{\overset{C}{}}\mathrm{tanh}^1\widehat{x}_i\right)\right)_\xi }.`$ (32)
The derivation of Eqs.(31) and (32) is given in Appendix B.
Examining the physical properties of the solutions for various connectivity values exposes significant differences between the various cases. In particular, these solutions fall into three different categories: the cases of $`K=1`$ and general $`L`$ value, the case of $`K=L=2`$ and all other parameter values where either $`K3`$ or $`L3`$ (and $`K>1`$). We describe the results obtained for each one of these cases separately.
### A Analytical solution - the case of $`K3`$ or $`L3`$, $`K>1`$
Results for the cases of $`K3`$ or $`L3`$, $`K>1`$ can be obtained analytically and have a simple and transparent interpretation; we will therefore focus first on this simple case. For unbiased messages (with $`F_s=0`$), one can easily verify that the ferromagnetic phase, characterised by $`m=1`$, and the probability distributions
$`\pi (x)=\delta (x1),\widehat{\pi }(\widehat{x})=\delta (\widehat{x}1),\rho (y)=\delta (y1),\widehat{\rho }(\widehat{y})=\delta (\widehat{y}1);`$ (33)
and the paramagnetic state of $`m=0`$ with the probability distributions
$`\pi (x)`$ $`=`$ $`\delta (x),\widehat{\pi }(\widehat{x})=\delta (\widehat{x}),\widehat{\rho }(\widehat{y})=\delta (\widehat{y}),`$ (34)
$`\rho (y)`$ $`=`$ $`{\displaystyle \frac{1+\mathrm{tanh}F_n}{2}}\delta (y\mathrm{tanh}F_n)+{\displaystyle \frac{1\mathrm{tanh}F_n}{2}}\delta (y+\mathrm{tanh}F_n),`$ (35)
satisfy saddle point equations (30). Other solutions may be obtained numerically; here we have represented the distributions by $`10^310^4`$ bins and iterated Eqs.(30) $`100500`$ times with $`10^5`$ Monte Carlo sampling steps for each iteration. No solutions other than the above two have been discovered.
The thermodynamically-dominant state is found by evaluating the free energy of the two solutions using Eq.(25), which yields
$`f_{\text{ferro}}={\displaystyle \frac{C}{K}}F_n\mathrm{tanh}F_n={\displaystyle \frac{1}{R}}F_n\mathrm{tanh}F_n,`$ (36)
for the ferromagnetic solution and
$`f_{\text{para}}={\displaystyle \frac{C}{K}}\mathrm{ln}2\mathrm{ln}2{\displaystyle \frac{C}{K}}\mathrm{ln}2\mathrm{cosh}F_n={\displaystyle \frac{1}{R}}\mathrm{ln}2\mathrm{ln}2{\displaystyle \frac{1}{R}}\mathrm{ln}2\mathrm{cosh}F_n,`$ (37)
for the paramagnetic solution.
Figure 1(a) describes schematically the nature of the solutions for this case, in terms of the free energy and the magnetisation obtained, for various flip rate probabilities. The difference between the free energies of Eqs.(36) and (37)
$`f_{\text{ferro}}f_{\text{para}}={\displaystyle \frac{\mathrm{ln}2}{R}}\left[R1+H_2(p)\right],`$ (38)
vanishes in the boundary between the two phase
$`R_c=1H_2(p)=1+p\mathrm{log}_2(p)+(1p)\mathrm{log}_2(1p),`$ (39)
which coincides with Shannon’s channel capacity.
Equation (39) indicates that all constructions with either $`K3`$ or $`L3`$ (and $`K>1`$) can potentially realize error-free data transmission for $`R<R_c`$ in the limit where both message and codeword lengths $`N`$ and $`M`$ become infinite, thus saturating Shannon’s bound.
### B The case of $`K=L=2`$
All codes with either $`K=3`$ or $`L=3`$, $`K>1`$ potentially saturate Shannon’s bound and are characterised by a first order phase transition between the ferromagnetic and paramagnetic solutions. On the other hand, numerical investigation based on Monte Carlo methods indicates of significantly different physical characteristics for $`K=L=2`$ codes shown in Fig.1(b).
At the highest noise level, the paramagnetic solution (35) gives the unique extremum of the free energy until noise level reaches the first critical point $`p_1`$, at which the ferromagnetic solution (33) of higher free energy appears to be locally stable. As the noise level decreases, a second critical point $`p_2`$ appears, where the paramagnetic solution becomes unstable and a sub-optimal ferromagnetic solution and its mirror image emerge. Those solutions have lower free energy than the ferromagnetic solution until the noise level reaches the third critical point $`p_3`$. Below $`p_3`$, the ferromagnetic solution becomes the global minimum of the free energy, while the sub-optimal ferromagnetic solutions still remain locally stable. However, the sub-optimal solutions disappear at the spinodal point $`p_s`$ and the ferromagnetic solution (and its mirror image) becomes the unique stable solution of the saddle point Eqs.(30) as shown by the numerical investigation for all $`p<p_s`$.
The analysis implies that $`p_3`$, the critical noise level below which the ferromagnetic solution becomes thermodynamically dominant, is lower than $`p_c=H_2^1(1R)`$ which corresponds to Shannon’s bound. Namely, $`K=L=2`$ does not saturate Shannon’s bound in contrast to $`K3`$ codes even if optimally decoded. Nevertheless, it turns out that the free energy landscape, for noise levels $`0<p<p_s`$, offers significant advantages in the decoding dynamic comparing to that of other codes ($`K3`$ or $`L3`$, $`K>1`$).
### C General $`L`$ codes with $`K=1`$
The particular choice of $`K=1`$, independently of the value chosen for $`L`$, exhibits a different behaviour presented schematically in Fig.1(c); also in this case there are no simple analytical solutions and all solutions in this scenario, except for the ferromagnetic solution, have been obtained numerically. The first important difference to be noted is that the paramagnetic state (35) is no longer a solution of the saddle point equations (30) and is being replaced by a sub-optimal ferromagnetic state. Convergence to the perfect solution of $`m=1`$ can only be guaranteed for corruption rates smaller than that of the spinodal point, marking the maximal noise level for which only the ferromagnetic solution exists, $`p<p_s`$.
The $`K=1`$ codes do not saturate Shannon’s bound in general; however, we have found that at rates $`R<1/3`$ they outperform the $`K=L=2`$ code while offering slightly improved dynamical (decoding) properties. Studying the free energy in this case shows that as the corruption rate increases, sub-optimal ferromagnetic solutions (stable and unstable) emerge at the spinodal point $`p_s`$. When the noise increases further this sub-optimal state becomes the global minimum at $`p_1`$, dominating the system’s thermodynamics. The transition at $`p_1`$ must occur at noise levels lower or equal to the value predicted by Shannon’s bound. In Fig.2 we show free energy values computed for a given code rate and several values of $`L`$, marking Shannon’s bound by a dashed line; it is clear that the thermodynamical transition observed numerically (i.e. the point where the ferromagnetic free energy equals the sub-optimal ferromagnetic free energy) is bellow, but very close, to the channel capacity. It implies that these codes also do not quite saturate Shannon’s bound if optimally decoded but get quite close to it.
## V Decoding: Belief propagation/TAP approach
The Bayesian message estimate (16) potentially provides the optimal retrieval of the original messages. However, it is computationally difficult to follow the prescription exactly as it requires a sum over $`𝒪(2^N)`$ terms. Belief propagation (BP) can be used for obtaining an approximate estimate. It was recently shown that the BP algorithm can be derived, at least in the current context, from the TAP approach to diluted systems in statistical mechanics.
Both algorithms (BP/TAP) are iterative methods which effectively calculate the marginal posterior probabilities $`𝒫(S_i|𝑱,C_s,C_n)=_{\{\{S_{ki}\},𝝉\}}𝒫(𝑺,𝝉|𝑱,C_s,C_n)`$ and $`𝒫(\tau _j|𝑱,C_s,C_n)=_{\{𝑺,\{\tau _{kj}\}\}}𝒫(𝑺,𝝉|𝑱,C_s,C_n)`$ based on the following three assumptions:
1. The posterior distribution is factorizable with respect to dynamical variables $`S_{i=1,\mathrm{},N}`$ and $`\tau _{j=1,\mathrm{},M}`$.
2. The influence of check $`J_{\mu =1,\mathrm{},M}`$ on a specific site $`S_i`$ (or $`\tau _j`$) is also factorizable.
3. The contribution of a single variables $`S_{i=1,\mathrm{},N}`$, $`\tau _{j=1,\mathrm{},M}`$ and $`J_{\mu =1,\mathrm{},M}`$ to the macroscopic variables is small and can be isolated.
Parameterising pseudo-marginal posteriors and marginalized conditional probabilities as
$`𝒫(S_i|\{J_{\nu \mu }\},C_s,C_n)`$ $`=`$ $`{\displaystyle \frac{1+m_{\mu i}^SS_i}{2}},𝒫(\tau _j|\{J_{\nu \mu }\},C_s,C_n)={\displaystyle \frac{1+m_{\mu j}^n\tau _j}{2}},`$ (40)
$`𝒫(J_\mu |S_i,\{J_{\nu \mu }\},C_s,C_n)`$ $``$ $`{\displaystyle \frac{1+\widehat{m}_{\mu i}^SS_i}{2}},𝒫(J_\mu |\tau _j,\{J_{\nu \mu }\},C_s,C_n){\displaystyle \frac{1+\widehat{m}_{\mu j}^n\tau _j}{2}},`$ (41)
the above assumptions provide a set of self-consistent equations
$`m_{\mu l}^S=\mathrm{tanh}\left(F_s+{\displaystyle \underset{\nu _S(l)/\mu }{}}\mathrm{tanh}^1(\widehat{m}_{\nu l}^S)\right),m_{\mu l}^n=\mathrm{tanh}\left(F_n+{\displaystyle \underset{\nu _n(l)/\mu }{}}\mathrm{tanh}^1(\widehat{m}_{\nu l}^n)\right).`$ (42)
and
$`\widehat{m}_{\mu l}^S=J_\mu {\displaystyle \underset{k_S(\mu )/l}{}}m_{\mu k}^S{\displaystyle \underset{j_n(\mu )}{}}m_{\mu j}^n,\widehat{m}_{\mu l}^n=J_\mu {\displaystyle \underset{k_S(\mu )}{}}m_{\mu k}^S{\displaystyle \underset{j_n(\mu )/l}{}}m_{\mu j}^n.`$ (43)
Here, $`_s(l)`$ and $`_n(l)`$ indicate the set of all indices of non-zero components in the $`l`$-th column of the sparse matrices $`C_s`$ and $`C_n`$, respectively. Similarly, $`_s(\mu )`$ and $`_n(\mu )`$ denote the set of all indices of non-zero components in $`\mu `$-th row of the sparse matrices $`C_s`$ and $`C_n`$, respectively. The notation $`_s(\mu )/l`$ represents the set of all indices belonging to $`_s(\mu )`$ except the index $`l`$.
Equations (42) and (43) are solved iteratively using the appropriate initial conditions. After obtaining a solution to all $`m_{\mu l}`$ and $`\widehat{m}_{\mu l}`$, an approximated posterior mean can be calculated as
$`m_i^S=\mathrm{tanh}\left(F_s+{\displaystyle \underset{\mu _S(l)}{}}\mathrm{tanh}^1(\widehat{m}_{\mu i}^S)\right),`$ (44)
which provides an approximation to the Bayes-optimal estimator (16) in the form of $`\widehat{\xi }^B=\text{sign}(m_i^S)`$.
Notice that the rather vague meaning of the fields distributions introduced in the previous section becomes clear by introducing the new variables $`x=\xi _im_{\mu i}^S`$, $`\widehat{x}=\xi _i\widehat{m}_{\mu i}^S`$, $`y=\zeta _jm_{\mu j}^n`$ and $`\widehat{y}=\zeta _j\widehat{m}_{\mu j}^n`$. If one considers that these variables are independently drawn from the distributions $`\pi (x)`$, $`\widehat{\pi }(\widehat{x})`$, $`\rho (y)`$ and $`\widehat{\rho }(\widehat{y})`$, the replica symmetric saddle point equations (30) are recovered from the BP/TAP equations (42) and (43). This connection can be extended to the free energy as equations (42) and (43) extremize the TAP free energy
$`f_{\text{TAP}}(\{𝒎\},\{\widehat{𝒎}\})`$ $`=`$ $`{\displaystyle \frac{M}{N}}\mathrm{ln}2+{\displaystyle \frac{1}{N}}{\displaystyle \underset{\mu =1}{\overset{M}{}}}{\displaystyle \underset{i_S(\mu )}{}}\mathrm{ln}\left(1+m_{\mu i}^S\widehat{m}_{\mu i}^S\right)+{\displaystyle \frac{1}{N}}{\displaystyle \underset{\mu =1}{\overset{M}{}}}{\displaystyle \underset{j_n(\mu )}{}}\mathrm{ln}\left(1+m_{\mu j}^n\widehat{m}_{\mu j}^n\right)`$ (45)
$`{\displaystyle \frac{1}{N}}{\displaystyle \underset{\mu =1}{\overset{M}{}}}\mathrm{ln}\left(1+J_\mu {\displaystyle \underset{i_S(\mu )}{}}m_{\mu i}^S{\displaystyle \underset{j_n(\mu )}{}}m_{\mu j}^n\right)`$ (46)
$`{\displaystyle \frac{1}{N}}{\displaystyle \underset{i=1}{\overset{N}{}}}\mathrm{ln}\left[e^{F_s}{\displaystyle \underset{\mu _S(i)}{}}\left(1+\widehat{m}_{\mu i}^S\right)+e^{F_s}{\displaystyle \underset{\mu _S(i)}{}}\left(1\widehat{m}_{\mu i}^S\right)\right]`$ (47)
$`{\displaystyle \frac{1}{N}}{\displaystyle \underset{j=1}{\overset{M}{}}}\mathrm{ln}\left[e^{F_n}{\displaystyle \underset{\mu _n(j)}{}}\left(1+\widehat{m}_{\mu j}^n\right)+e^{F_n}{\displaystyle \underset{\mu _n(j)}{}}\left(1\widehat{m}_{\mu j}^n\right)\right].`$ (48)
This expression may be used for selecting the thermodynamically dominant state when Eqs.(42) and (43) have several solutions.
We have investigated the performance of the various codes using BP/TAP equations as the decoding algorithm. Solutions have been obtained by iterating the equations (42) and (43) $`100500`$ times under various initial conditions. Since the system is not frustrated, the dynamics converges within $`1030`$ updates in most cases except close to criticality. The numerical results mirror the behaviour predicted by the analytical solutions.
For either $`K3`$ or $`L3`$ ,$`K>1`$ codes, the ferromagnetic solution
$`m_{\mu i}^S=\xi _i,\widehat{m}_{\mu i}^S=\xi _i,m_{\mu j}^n=\zeta _j,\widehat{m}_{\mu j}^n=\zeta _j,`$ (49)
which provides perfect decoding ($`m=1`$) and the paramagnetic solution ($`m=0`$)
$`m_{\mu i}^S=0,\widehat{m}_{\mu i}^S=0,m_{\mu j}^n=\mathrm{tanh}F_n=12p,\widehat{m}_{\mu j}^n=0,`$ (50)
are obtained in various runs depending on the initial conditions (the message is assumed unbiased resulting in $`F_s=0`$). However, it is difficult to set the initial conditions within the basin of attraction of the ferromagnetic solution without prior knowledge about the transmitted message $`𝝃`$.
Biased coding is sometimes used for alleviating this difficulty . Using a redundant source of information (equivalent to the introduction of a non-zero field $`F_s`$ in the statistical physics description), one effectively increases the probability of the initial conditions being closer to the ferromagnetic solution. The main drawback of this method is that the information per transmitted bit is significantly reduced due to this redundancy. In order to investigate how the maximum performance is affected by transmitting biased messages, we have evaluated the critical information rate (i.e., code rate $`\times H_2(f_s=(1+\mathrm{tanh}F_s)/2)`$, the source redundancy), below which the ferromagnetic solution becomes thermodynamically dominant \[Fig.3(a)\]. The data were obtained by the BP/TAP method (diamonds) and numerical solutions of from replica framework (square); the dominant solution in the BP/TAP results, was selected by using the free energy (48). Numerical solutions have been obtained using $`10^310^4`$ bin models for each distribution and had been run for $`10^5`$ steps per noise level. The various results are highly consistent and practically saturate Shannon’s bound for the same noise level. However, it is important to point out that close to Shannon’s limit, prior knowledge on the original message is required for setting up appropriate initial conditions that ensure convergence to the ferromagnetic solution; such prior knowledge is not available in practice.
Although $`K,L3`$ codes seem to offer optimal performance when highly biased messages are transmitted, this seems to be of little relevance in most cases, characterised by the transmission of compressed unbiased messages or only slightly biased messages. In this sense, $`K=L=2`$ and $`K=1`$ codes can be considered more practical as the BP/TAP dynamics of these codes exhibit unique convergence to the ferromagnetic solution (or mirror image in the $`K=L=2`$ case) from any initial condition up to a certain noise level. This property results from the fact that the corresponding free energies have no local minima other than the ferromagnetic solution below $`p_s`$.
In figures 3(b) and (c) we show the value of $`p_s`$ for the cases of $`K=L=2`$ and $`K=1`$, $`L=2`$ respectively, evaluated by numerical solutions from the replica framework (diamonds) and using the BP/TAP method.
The case of $`K=L=2`$ shows consistent successful decoding for the code rates examined and up to noise levels slightly below, but close to, Shannon’s bound. It should be emphasised here that initial conditions are chosen almost randomly in the BP/TAP method, with a very slight bias of $`𝒪(10^{12})`$ in the initial magnetisation. This result suggests using $`K=L=2`$ codes (or similar), rather than $`K,L3`$ codes, although the latter may potentially have better equilibrium properties.
In Fig.3(c) we show that for code rates $`R<1/3`$, codes parametrised by $`K=1`$ and $`L=2`$ outperform $`K=L=2`$ codes with one additional advantage: Due to the absence of mirror symmetries these codes converge to the ferromagnetic state much faster, and there is no risk of convergence to the mirror solution. The difference in performance becomes even larger as the code rate decreases. Higher code rates will result in performance deterioration due to the low connectivity, eventually bringing the system below the percolation threshold.
In Fig.4 we examine the dependence of the noise level of the spinodal point $`p_s`$ on the value of $`L`$, and show that the choice of $`L=2`$ is optimal within this family. Codes with $`L=1`$ have very poor error-correction capabilities as their Hamiltonian (11) corresponds to the Mattis model, which is equivalent to a simple ferromagnet in a random field attaining magnetisation $`m=1`$ only in the noiseless case.
## VI Reducing Encoding Costs
The BP/TAP algorithm already offers an efficient decoding method, which requires $`𝒪(N)`$ operations; however, the current encoding scheme includes three costly processes: (a) The computational cost of constructing the generating matrix $`C_n^1C_s`$ requires $`𝒪(N^3)`$ operations for inverting the matrix $`C_n`$ and $`𝒪(N^2)`$ operations for the matrix multiplication. (b) The memory allocation for generating the matrix $`C_n^1C_s`$ scales as $`𝒪(N^2)`$ since this matrix is typically dense. (c) The encoding itself $`𝒛_0=C_n^1C_s𝝃`$ (mod $`2`$) requires $`𝒪(N^2)`$ operations.
These computational costs become significant when long messages $`N=10^410^5`$ are transmitted, which is typically the case for which Gallager-type codes are being used. This may require long encoding times and may delay the transmission.
These problems may be solved by utilising systematically constructed matrices instead of random ones, of some similarity to the constructions of . Here, we present a simple method to reduce the computational and memory costs to $`𝒪(N)`$ for $`K=L=2`$ and $`K=1`$, $`L=2`$ codes. Our proposal is mainly based on using a specific matrix for $`C_n`$,
$`\overline{C}_n=\left(\begin{array}{cccccccc}1& 0& 0& 0& \mathrm{}& 0& 0& \\ 1& 1& 0& 0& \mathrm{}& 0& 0& \\ 0& 1& 1& 0& \mathrm{}& 0& 0& \\ 0& 0& 1& 1& \mathrm{}& 0& 0& \\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \\ 0& 0& 0& 0& \mathrm{}& 1& 1& \end{array}\right),`$ (57)
instead of a randomly-constructed one. For $`C_s`$, we use a random matrix of $`K=2`$ (or $`K=1`$) non-zero elements per row as before.
The inverse (mod 2) of $`\overline{C}_n^1`$ becomes the lower triangular matrix
$`\overline{C}_n^1=\left(\begin{array}{cccccccc}1& 0& 0& 0& \mathrm{}& 0& 0& \\ 1& 1& 0& 0& \mathrm{}& 0& 0& \\ 1& 1& 1& 0& \mathrm{}& 0& 0& \\ 1& 1& 1& 1& \mathrm{}& 0& 0& \\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \\ 1& 1& 1& 1& \mathrm{}& 1& 1& \end{array}\right).`$ (64)
This suggests that encoding the message $`𝝃`$ into a codeword $`𝒛^0`$ would require only $`𝒪(N)`$ operations by carrying it out in two steps
$`t_\mu `$ $`=`$ $`(C_s𝝃)_\mu \text{(mod }2\text{)},\text{for }\mu =1,2,\mathrm{},M,`$ (65)
$`z_\mu ^0`$ $`=`$ $`(\overline{C}_n^1𝒕)_\mu =z_{\mu 1}^0+t_\mu \text{(mod }2\text{)},\text{for }\mu =2,\mathrm{},M,`$ (66)
with $`z_1^0=t_1`$. Both steps require $`𝒪(N)`$ operations due to the sparse nature of $`C_s`$. In addition, the required memory resources are also reduced to $`𝒪(N)`$ since only the sparse matrix $`C_s`$ should be stored.
The possible drawback of using the systematic matrix (57) is a deterioration in the error correction ability. We have examined numerically the performance of new construction to discover, to our surprise, that it is very similar to that of random matrix based codes as shown in Table I. Although our examination is only limited to BSC and i.i.d. messages, it seems to suggest that some deterministically constructed matrices may be implemented successfully in practice.
## VII Summary
In this paper, we have investigated the typical performance of the MN codes, a variation of Gallager-type error-correcting codes, by mapping them onto Ising spin models and making use of the established methods of statistical physics. We have discovered that for a certain choice of parameters, either $`K3`$ or $`L3`$, $`K>1`$ these codes potentially saturate the channel capacity, although this cannot be used efficiently in practice due to the decrease in the basin of attraction which typically diverts the decoding dynamics towards the undesired paramagnetic solution.
Codes with $`K=2`$ and $`L=2`$ show close to optimal performance while keeping a large basin of attraction, resulting in more practical codes. Constructions of the form $`K=1`$, $`L=2`$ outperform the $`K=L=2`$ codes for code rates $`R<1/3`$, having improved dynamical properties.
These results are complementary to those obtained so far by the information theory community and seem to indicate that worst-case analysis can be, in some situations, too pessimistic when compared to the typical performance results.
Beyond the theoretical aspects, we proposed an efficient method for reducing the computational costs and the required memory allocation by using a specific construction of the matrix $`C_n`$. These codes are highly attractive and provide lower computational costs for both encoding and decoding.
Various aspects that remain to be studied include a proper analysis of the finite size effects for rates below and above the channel capacity, which are of great practical relevance; and the use of statistical physics methods for optimising the matrix constructions.
## Acknowledgement
Support by the JSPS RFTF program (YK), The Royal Society and EPSRC grant GR/N00562 (DS) is acknowledged.
## A Replica Free Energy
The purpose of this appendix is to derive the averaged free energy per spin (25). Applying the gauge transformation
$`J_\mu `$ $``$ $`J_\mu {\displaystyle \underset{i_s(\mu )}{}}\xi _i{\displaystyle \underset{j_n(\mu )}{}}\zeta _j=1`$ (A1)
$`S_i`$ $``$ $`S_i\xi _i`$ (A2)
$`\tau _j`$ $``$ $`\tau _j\zeta _j,`$ (A3)
to eq. (14), one may rewrite the partition function in the form
$`𝒵(𝝃,𝜻,𝒟)={\displaystyle \underset{𝑺,𝝉}{}}\mathrm{exp}\left(F_s{\displaystyle \underset{i=1}{\overset{N}{}}}\xi _iS_i+F_n{\displaystyle \underset{j=1}{\overset{M}{}}}\zeta _j\tau _j\right)`$ (A4)
$`\times `$ $`{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}\left[1𝒟_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}+𝒟_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{\displaystyle \frac{1}{2}}\left(1+S_{i_1}\mathrm{}S_{i_K}\tau _{j_1}\mathrm{}\tau _{j_L}\right)\right].`$ (A5)
Using the replica method, one calculates the quenched average of the $`n`$-th power of the partition function given by
$`𝒵(𝝃,𝜻,𝒟)^n_{𝝃,𝜻,𝒟}={\displaystyle \underset{𝑺^1\mathrm{}𝑺^n}{}}{\displaystyle \underset{𝝉^1\mathrm{}𝝉^n}{}}\mathrm{exp}\left(F_s{\displaystyle \underset{i=1}{\overset{N}{}}}\xi _i{\displaystyle \underset{\alpha =1}{\overset{n}{}}}S_i^\alpha \right)_𝝃\mathrm{exp}\left(F_n{\displaystyle \underset{j=1}{\overset{M}{}}}\zeta _j{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\tau _i^\alpha \right)_𝜻`$ (A6)
$`\times `$ $`{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\left\{1+{\displaystyle \frac{1}{2}}𝒟_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}\left(S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha 1\right)\right\}_𝒟,`$ (A7)
where averages with respect to $`𝝃`$ can be easily performed
$`\mathrm{exp}\left(F_s{\displaystyle \underset{i=1}{\overset{N}{}}}\xi _i{\displaystyle \underset{\alpha =1}{\overset{n}{}}}S_i^\alpha \right)_𝝃`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}\left[\left({\displaystyle \frac{1+\mathrm{tanh}F_s}{2}}\right)e^{F_s_{\alpha =1}^nS_i^\alpha }+\left({\displaystyle \frac{1\mathrm{tanh}F_s}{2}}\right)e^{F_s_{\alpha =1}^nS_i^\alpha }\right]`$ (A8)
$`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}\mathrm{exp}\left(\xi F_s{\displaystyle \underset{a=1}{\overset{n}{}}}S_i^a\right)_\xi ,`$ (A9)
and similarly for $`\mathrm{}_𝜻`$. The main problem is in averages over the sparse tensor realisations $`𝒟`$, which have complicated constraints. Following the procedure introduced by Wong and Sherrington, it is being rewritten as
$`{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\left[1+{\displaystyle \frac{1}{2}}𝒟_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}\left(S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha 1\right)\right]_𝒟`$ (A10)
$`=`$ $`𝒩^1{\displaystyle \underset{𝒟}{}}{\displaystyle \underset{i=1}{\overset{N}{}}}\delta \left({\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}𝒟_{i,i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}C\right){\displaystyle \underset{j=1}{\overset{M}{}}}\delta \left({\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}𝒟_{i_1,\mathrm{},i_K;j,j_2,\mathrm{},j_L}L\right)`$ (A12)
$`\times {\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}{\displaystyle \underset{\alpha =1}{\overset{n}{}}}[1+{\displaystyle \frac{1}{2}}𝒟_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}(S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha 1)],`$
where $`\delta (\mathrm{})`$ represents Dirac’s $`\delta `$-function and
$`𝒩={\displaystyle \underset{𝒟}{}}{\displaystyle \underset{i=1}{\overset{N}{}}}\delta \left({\displaystyle \underset{i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}𝒟_{i,i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}C\right){\displaystyle \underset{j=1}{\overset{M}{}}}\delta \left({\displaystyle \underset{i_1,\mathrm{},i_K;j_2,\mathrm{},j_L}{}}𝒟_{i_1,\mathrm{},i_K;j,j_2,\mathrm{},j_L}L\right)`$ (A13)
represents the normalisation constant.
We first evaluate this normalisation constant using the integral representation of the $`\delta `$-function and Eq.(A13), to obtain
$`𝒩`$ $`=`$ $`{\displaystyle \underset{𝒟}{}}{\displaystyle \underset{i=1}{\overset{N}{}}}\delta \left({\displaystyle \underset{i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}𝒟_{i,i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}C\right){\displaystyle \underset{j=1}{\overset{M}{}}}\delta \left({\displaystyle \underset{i_1,\mathrm{},i_K;j_2,\mathrm{},j_L}{}}𝒟_{i_1,\mathrm{},i_K;j,j_2,\mathrm{},j_L}L\right)`$ (A14)
$`=`$ $`{\displaystyle \underset{𝒟}{}}{\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle _0^{2\pi }}{\displaystyle \frac{d\lambda _i}{2\pi }}\mathrm{exp}\left[i\lambda _i\left({\displaystyle \underset{i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}𝒟_{i,i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}C\right)\right]\right\}`$ (A16)
$`\times {\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle _0^{2\pi }}{\displaystyle \frac{d\lambda _j}{2\pi }}\mathrm{exp}\left[i\lambda _j({\displaystyle \underset{i_1,\mathrm{},i_K;j_2,\mathrm{},j_L}{}}𝒟_{i_1,\mathrm{},i_K;j,j_2,\mathrm{},j_L}L)\right]\right\}`$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle _0^{2\pi }}{\displaystyle \frac{d\lambda _i}{2\pi }}e^{iC\lambda _i}\right\}{\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle _0^{2\pi }}{\displaystyle \frac{d\nu _j}{2\pi }}e^{iL\nu _j}\right\}`$ (A17)
$`\times {\displaystyle \underset{𝒟}{}}{\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle \underset{i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}e^{i\lambda _i𝒟_{i,i_2,\mathrm{},i_K;j_1,\mathrm{},j_L}}\right\}{\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle \underset{i_1,\mathrm{},i_K;j_2,\mathrm{},j_L}{}}e^{i\nu _j𝒟_{i_1,\mathrm{},i_K;j,j_2,\mathrm{},j_L}}\right\}`$ (A18)
$`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle _0^{2\pi }}{\displaystyle \frac{d\lambda _i}{2\pi }}e^{iC\lambda _i}\right\}{\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle _0^{2\pi }}{\displaystyle \frac{d\nu _j}{2\pi }}e^{iL\nu _j}\right\}`$ (A19)
$`\times {\displaystyle \underset{𝒟}{}}{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}\left\{\left(e^{i\lambda _{i_1}}\mathrm{}e^{i\lambda _{i_K}}e^{i\nu _{j_1}}\mathrm{}e^{i\nu _{j_L}}\right)^{𝒟_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}}\right\}`$ (A20)
$`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle \frac{dZ_i}{2\pi i}Z_i^{(C+1)}}\right\}{\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle \frac{dY_j}{2\pi i}Y_j^{(L+1)}}\right\}{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}\left(1+Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}\right),`$ (A21)
where we made use of the transformations $`Z_i=e^{i\lambda _i},Y_j=e^{i\nu _j}`$, and carried out summations with respect to the realisation of $`𝒟`$. Expanding the product on the right hand side one obtains
$`{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}\left[1+\left(Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}\right)\right]`$ (A22)
$`=`$ $`\mathrm{exp}\left[{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}\mathrm{ln}\left\{1+\left(Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}\right)\right\}\right]`$ (A23)
$``$ $`\mathrm{exp}\left[{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}\left(Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}\right)\right]`$ (A24)
$``$ $`\mathrm{exp}\left[{\displaystyle \frac{1}{K!}}\left({\displaystyle \underset{i=1}{\overset{N}{}}}Z_i\right)^K{\displaystyle \frac{1}{L!}}\left({\displaystyle \underset{j=1}{\overset{M}{}}}Y_j\right)^L\right],`$ (A25)
in the thermodynamic limit. Using the identities
$`1={\displaystyle 𝑑q\delta \left(\underset{i=1}{\overset{N}{}}Z_iq\right)},1={\displaystyle 𝑑r\delta \left(\underset{j=1}{\overset{M}{}}Y_jr\right)}`$ (A26)
Eq. (A21) becomes
$`𝒩`$ $`=`$ $`{\displaystyle 𝑑q\delta \left(\underset{i=1}{\overset{N}{}}Z_iq\right)𝑑r\delta \left(\underset{j=1}{\overset{M}{}}Y_jr\right)}`$ (A27)
$`\times {\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle }{\displaystyle \frac{dZ_i}{2\pi i}}Z_i^{(C+1)}\right\}{\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle }{\displaystyle \frac{dY_j}{2\pi i}}Y_j^{(L+1)}\right\}\mathrm{exp}\left({\displaystyle \frac{q^K}{K!}}{\displaystyle \frac{r^L}{L!}}\right)`$ (A28)
$`=`$ $`{\displaystyle 𝑑q\frac{d\widehat{q}}{2\pi i}\mathrm{exp}\left[\widehat{q}\left(\underset{i=1}{\overset{N}{}}Z_iq\right)\right]𝑑r\frac{d\widehat{r}}{2\pi i}\mathrm{exp}\left[\widehat{r}\left(\underset{j=1}{\overset{M}{}}Y_jr\right)\right]}`$ (A29)
$`\times {\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle }{\displaystyle \frac{dZ_i}{2\pi i}}Z_i^{(C+1)}\right\}{\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle }{\displaystyle \frac{dY_j}{2\pi i}}Y_j^{(L+1)}\right\}\mathrm{exp}\left({\displaystyle \frac{q^K}{K!}}{\displaystyle \frac{r^L}{L!}}\right)`$ (A30)
$`=`$ $`{\displaystyle 𝑑q\frac{d\widehat{q}}{2\pi i}𝑑r\frac{d\widehat{r}}{2\pi i}\mathrm{exp}\left(\frac{q^K}{K!}\frac{r^L}{L!}q\widehat{q}r\widehat{r}\right)}`$ (A31)
$`\times {\displaystyle \underset{i=1}{\overset{N}{}}}\left[{\displaystyle }{\displaystyle \frac{dZ_i}{2\pi i}}Z_i^{(C+1)}\mathrm{exp}\left(\widehat{q}Z_i\right)\right]{\displaystyle \underset{j=1}{\overset{M}{}}}\left[{\displaystyle }{\displaystyle \frac{dY_j}{2\pi i}}Y_j^{(L+1)}\mathrm{exp}\left(\widehat{r}Y_j\right)\right].`$ (A32)
The contour integrals provide the following constants
$`{\displaystyle \underset{i=1}{\overset{N}{}}}\left[{\displaystyle \frac{dZ_i}{2\pi i}Z_i^{(C+1)}\mathrm{exp}\left(\widehat{q}Z_i\right)}\right]=\left({\displaystyle \frac{\widehat{q}^C}{C!}}\right)^N,{\displaystyle \underset{j=1}{\overset{M}{}}}\left[{\displaystyle \frac{dY_j}{2\pi i}Y_j^{(L+1)}\mathrm{exp}\left(\widehat{r}Y_j\right)}\right]=\left({\displaystyle \frac{\widehat{r}^L}{L!}}\right)^M,`$ (A33)
respectively. Applying the saddle point method to the remaining integrals, one obtains
$`𝒩=\text{extr}_{\{q,\widehat{q},r,\widehat{r}\}}\left\{\mathrm{exp}\left[{\displaystyle \frac{q^K}{K!}}{\displaystyle \frac{r^L}{L!}}q\widehat{q}r\widehat{r}+NC\mathrm{ln}\widehat{q}N\mathrm{ln}(C!)+ML\mathrm{ln}\widehat{r}M\mathrm{ln}(L!)\right]\right\},`$ (A34)
which yields the following saddle point equations with respect to $`q`$, $`r`$, $`\widehat{q}`$ and $`\widehat{r}`$
$`q`$ $`=`$ $`{\displaystyle \frac{NC}{\widehat{q}}},r={\displaystyle \frac{ML}{\widehat{r}}}`$ (A35)
$`\widehat{q}`$ $`=`$ $`{\displaystyle \frac{q^{K1}}{(K1)!}}{\displaystyle \frac{r^L}{L!}},\widehat{r}={\displaystyle \frac{r^{L1}}{(L1)!}}{\displaystyle \frac{q^K}{K!}},`$ (A36)
providing the normalisation constant
$`𝒩=\left({\displaystyle \frac{\widehat{q}^C}{C!}}\right)^N\left({\displaystyle \frac{\widehat{r}^L}{L!}}\right)^M\mathrm{exp}\left({\displaystyle \frac{q^K}{K!}}{\displaystyle \frac{r^L}{L!}}q\widehat{q}r\widehat{r}\right).`$ (A37)
Equation (A12) can be evaluated similarly. Following a similar calculation to that of Eq.(A21) provides
$`{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\left\{1+{\displaystyle \frac{1}{2}}𝒟_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}\left(S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha 1\right)\right\}_𝒟`$ (A38)
$`=`$ $`𝒩^1{\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle \frac{dZ_i}{2\pi i}Z_i^{(C+1)}}\right\}{\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle \frac{dY_j}{2\pi i}Y_j^{(L+1)}}\right\}`$ (A39)
$`\times {\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}[1+\left(Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}\right){\displaystyle \underset{\alpha =1}{\overset{n}{}}}{\displaystyle \frac{1}{2}}(1+S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha )].`$ (A40)
Using the expansion
$`{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\left(1+S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha \right)`$ (A41)
$`=`$ $`1+{\displaystyle \underset{\alpha =1}{\overset{n}{}}}S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha +{\displaystyle \underset{\alpha _1,\alpha _2}{}}(S_{i_1}^{\alpha _1}S_{i_1}^{\alpha _2})\mathrm{}(S_{i_K}^{\alpha _1}S_{i_K}^{\alpha _2})(\tau _{j_1}^{\alpha _1}\tau _{j_1}^{\alpha _2})\mathrm{}(\tau _{j_L}^{\alpha _1}\tau _{j_L}^{\alpha _2})`$ (A42)
$`+\mathrm{}+{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _n}{}}(S_{i_1}^{\alpha _1}\mathrm{}S_{i_1}^{\alpha _n})\mathrm{}(S_{i_K}^{\alpha _1}\mathrm{}S_{i_K}^{\alpha _n})(\tau _{j_1}^{\alpha _1}\mathrm{}\tau _{j_1}^{\alpha _n})\mathrm{}(\tau _{j_L}^{\alpha _1}\mathrm{}\tau _{j_L}^{\alpha _n})`$ (A43)
$`=`$ $`{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}(S_{i_1}^{\alpha _1}\mathrm{}S_{i_1}^{\alpha _m})\mathrm{}(S_{i_K}^{\alpha _1}\mathrm{}S_{i_K}^{\alpha _m})(\tau _{j_1}^{\alpha _1}\mathrm{}\tau _{j_1}^{\alpha _m})\mathrm{}(\tau _{j_L}^{\alpha _1}\mathrm{}\tau _{j_L}^{\alpha _m}),`$ (A44)
resulting in
$`{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}\left[1+\left(Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}\right){\displaystyle \underset{\alpha =1}{\overset{n}{}}}{\displaystyle \frac{1}{2}}\left(1+S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha \right)\right]`$ (A45)
$``$ $`e^{_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}_{\alpha =1}^n\frac{1}{2}\left(1+S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha \right)}`$ (A46)
$`=`$ $`e^{\frac{1}{2^n}_{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}_{m=0}^n_{\alpha _1,\mathrm{},\alpha _m}(S_{i_1}^{\alpha _1}\mathrm{}S_{i_1}^{\alpha _m})\mathrm{}(S_{i_K}^{\alpha _1}\mathrm{}S_{i_K}^{\alpha _m})(\tau _{j_1}^{\alpha _1}\mathrm{}\tau _{j_1}^{\alpha _m})\mathrm{}(\tau _{j_L}^{\alpha _1}\mathrm{}\tau _{j_L}^{\alpha _m})}`$ (A47)
$`=`$ $`e^{\frac{1}{2^n}\left\{_{m=0}^n_{\alpha _1,\mathrm{},\alpha _m}_{i_1,\mathrm{},i_K}\left(S_{i_1}^{\alpha _1}\mathrm{}S_{i_1}^{\alpha _m}Z_{i_1}\right)\mathrm{}\left(S_{i_K}^{\alpha _1}\mathrm{}S_{i_K}^{\alpha _m}Z_{i_K}\right)_{j_1,\mathrm{},j_L}\left(\tau _{j_1}^{\alpha _1}\mathrm{}\tau _{j_1}^{\alpha _m}Y_{j_1}\right)\mathrm{}\left(\tau _{j_L}^{\alpha _1}\mathrm{}\tau _{j_K}^{\alpha _m}Y_{j_L}\right)\right\}}`$ (A48)
$``$ $`e^{\frac{1}{2^n}\left\{_{m=0}^n_{\alpha _1,\mathrm{},\alpha _m}\frac{1}{K!}\left(_{i=1}^NS_i^{\alpha _1}\mathrm{}S_i^{\alpha _m}Z_i\right)^K\frac{1}{L!}\left(\tau _j^{\alpha _1}\mathrm{}\tau _j^{\alpha _m}Y_j\right)^L\right\}}.`$ (A49)
Using the identities
$`1`$ $`=`$ $`{\displaystyle 𝑑q_{\alpha _1,\mathrm{},\alpha _m}\delta \left(\underset{i=1}{\overset{N}{}}S_i^{\alpha _1}\mathrm{}S_i^{\alpha _m}Z_iq_{\alpha _1,\mathrm{},\alpha _m}\right)},`$ (A50)
$`1`$ $`=`$ $`{\displaystyle 𝑑r_{\alpha _1,\mathrm{},\alpha _m}\delta \left(\underset{j=1}{\overset{M}{}}\tau _j^{\alpha _1}\mathrm{}\tau _j^{\alpha _m}Y_jr_{\alpha _1,\mathrm{},\alpha _m}\right)}`$ (A51)
and going through the same steps as in Eqs. (A26 \- A34), we arrive at
$`{\displaystyle \underset{i_1,\mathrm{},i_K;j_1,\mathrm{},j_L}{}}\left[1+\left(Z_{i_1}\mathrm{}Z_{i_K}Y_{j_1}\mathrm{}Y_{j_L}\right){\displaystyle \underset{\alpha =1}{\overset{n}{}}}{\displaystyle \frac{1}{2}}\left(1+S_{i_1}^\alpha \mathrm{}S_{i_K}^\alpha \tau _{j_1}^\alpha \mathrm{}\tau _{j_L}^\alpha \right)\right]`$ (A52)
$`=`$ $`{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}{\displaystyle 𝑑q_{\alpha _1,\mathrm{},\alpha _m}\delta \left(\underset{i=1}{\overset{N}{}}S_i^{\alpha _1}\mathrm{}S_i^{\alpha _m}Z_iq_{\alpha 1,\mathrm{},\alpha _m}\right)}`$ (A53)
$`\times {\displaystyle }dr_{\alpha _1,\mathrm{},\alpha _m}\delta ({\displaystyle \underset{j=1}{\overset{M}{}}}\tau _j^{\alpha _1}\mathrm{}\tau _j^{\alpha _m}Y_jr_{\alpha 1,\mathrm{},\alpha _m})\mathrm{exp}\left({\displaystyle \frac{1}{2^n}}\left\{{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}{\displaystyle \frac{q_{\alpha _1,\mathrm{},\alpha _m}^K}{K!}}{\displaystyle \frac{r_{\alpha _1,\mathrm{},\alpha _m}^L}{L!}}\right\}\right)`$ (A54)
$``$ $`\text{extr}_{\{𝒒,\widehat{𝒒},𝒓,\widehat{𝒓}\}}\{\mathrm{exp}[{\displaystyle \frac{1}{2^n}}\left\{{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}{\displaystyle \frac{q_{\alpha _1,\mathrm{},\alpha _m}^K}{K!}}{\displaystyle \frac{r_{\alpha _1,\mathrm{},\alpha _m}^L}{L!}}\right\}`$ (A55)
$`{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}q_{\alpha _1,\mathrm{},\alpha _m}\widehat{q}_{\alpha _1,\mathrm{},\alpha _m}{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}r_{\alpha _1,\mathrm{},\alpha _m}\widehat{r}_{\alpha _1,\mathrm{},\alpha _m}`$ (A56)
$`+{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}\widehat{q}_{\alpha _1,\mathrm{},\alpha _m}{\displaystyle \underset{i=1}{\overset{N}{}}}S_i^{\alpha _1}\mathrm{}S_i^{\alpha _m}Z_i+{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}\widehat{r}_{\alpha _1,\mathrm{},\alpha _m}{\displaystyle \underset{j=1}{\overset{M}{}}}\tau _j^{\alpha _1}\mathrm{}\tau _j^{\alpha _m}Y_j]\}.`$ (A57)
In order to proceed further, one has to make an assumption about the order parameter symmetry. We adopt here the replica symmetric ansatz for the order parameters $`q`$, $`r`$, $`\widehat{q}`$ and $`\widehat{r}`$. This implies that the order parameters do not depend on the explicit indices but only on their number. It is therefore convenient to represent them as moments of random variables defined over the interval $`[1,1]`$
$`q_{\alpha _1,\mathrm{},\alpha _l}`$ $`=`$ $`q{\displaystyle 𝑑x\pi (x)x^l},r_{\alpha _1,\mathrm{},\alpha _l}=r{\displaystyle 𝑑y\rho (y)y^l},`$ (A58)
$`\widehat{q}_{\alpha _1,\mathrm{},\alpha _l}`$ $`=`$ $`\widehat{q}{\displaystyle 𝑑\widehat{x}\widehat{\pi }(\widehat{x})\widehat{x}^l},\widehat{r}_{\alpha _1,\mathrm{},\alpha _l}=\widehat{r}{\displaystyle 𝑑\widehat{y}\widehat{\rho }(\widehat{y})\widehat{y}^l},`$ (A59)
Then, each term in Eq.(A57) takes the form
$`{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}{\displaystyle \frac{q_{\alpha _1,\mathrm{},\alpha _m}^K}{K!}}{\displaystyle \frac{r_{\alpha _1,\mathrm{},\alpha _m}^L}{L!}}`$ $`=`$ $`{\displaystyle \frac{q^K}{K!}}{\displaystyle \frac{r^L}{L!}}{\displaystyle \underset{m=0}{\overset{n}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{n}{m}}\right){\displaystyle \underset{k=1}{\overset{K}{}}dx_k\pi (x_k)x_k^m\underset{l=1}{\overset{L}{}}dy_l\rho (y_l)y_l^m}`$ (A60)
$`=`$ $`{\displaystyle \frac{q^K}{K!}}{\displaystyle \frac{r^L}{L!}}{\displaystyle \underset{k=1}{\overset{K}{}}dx_k\pi (x_k)\underset{l=1}{\overset{L}{}}dy_l\rho (y_l)\left(1+\underset{k=1}{\overset{K}{}}x_k\underset{l=1}{\overset{L}{}}y_l\right)^n}`$ (A61)
$`{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}q_{\alpha _1,\mathrm{},\alpha _m}\widehat{q}_{\alpha _1,\mathrm{},\alpha _m}`$ $`=`$ $`q\widehat{q}{\displaystyle \underset{m=0}{\overset{n}{}}}\left({\displaystyle \genfrac{}{}{0pt}{}{n}{m}}\right){\displaystyle 𝑑x𝑑\widehat{x}\pi (x)\widehat{\pi }(\widehat{x})x^m\widehat{x}^m}`$ (A62)
$`=`$ $`q\widehat{q}{\displaystyle 𝑑x𝑑\widehat{x}\pi (x)\widehat{\pi }(\widehat{x})\left(1+x\widehat{x}\right)^n}`$ (A63)
$`{\displaystyle \underset{m=0}{\overset{n}{}}}{\displaystyle \underset{\alpha _1,\mathrm{},\alpha _m}{}}\widehat{q}_{\alpha _1,\mathrm{},\alpha _m}{\displaystyle \underset{i=1}{\overset{N}{}}}S_i^{\alpha _1}\mathrm{}S_i^{\alpha _m}Z_i`$ $`=`$ $`\widehat{q}{\displaystyle \underset{i=1}{\overset{N}{}}}Z_i{\displaystyle 𝑑\widehat{x}\widehat{\pi }(\widehat{x})\underset{m=0}{\overset{n}{}}\widehat{x}^m\underset{\alpha _1,\mathrm{},\alpha _m}{}S_i^{\alpha _1}\mathrm{}S_i^{\alpha _m}}`$ (A64)
$`=`$ $`\widehat{q}{\displaystyle \underset{i=1}{\overset{N}{}}}Z_i{\displaystyle 𝑑\widehat{x}\widehat{\pi }(\widehat{x})\underset{\alpha =1}{\overset{n}{}}\left(1+S_i^\alpha \widehat{x}\right)}.`$ (A65)
Substituting these into (A57), one obtains
$`Z(𝝃,𝜻,𝒟)^n_{𝝃,𝜻,𝒟}`$ (A66)
$`=`$ $`{\displaystyle \underset{𝑺^1\mathrm{}𝑺^n}{}}{\displaystyle \underset{𝝉^1\mathrm{}𝝉^n}{}}{\displaystyle \underset{i=1}{\overset{N}{}}}\mathrm{exp}\left(\xi F_s{\displaystyle \underset{\alpha =1}{\overset{n}{}}}S_i^\alpha \right)_\xi \times {\displaystyle \underset{j=1}{\overset{M}{}}}\mathrm{exp}\left(\zeta F_n{\displaystyle \underset{\alpha =1}{\overset{n}{}}}\tau _j^\alpha \right)_\zeta `$ (A67)
$`\times 𝒩^1{\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle \frac{dZ_i}{2\pi i}Z_i^{(C+1)}}\right\}{\displaystyle \underset{j=1}{\overset{M}{}}}\left\{{\displaystyle \frac{dY_j}{2\pi i}Y_j^{(L+1)}}\right\}`$ (A68)
$`\times \text{extr}_{\{\pi ,\widehat{\pi },\rho ,\widehat{\rho }\}}\{\mathrm{exp}[{\displaystyle \frac{1}{2^n}}\left\{{\displaystyle \frac{q^K}{K!}}{\displaystyle \frac{r^L}{L!}}{\displaystyle }{\displaystyle \underset{l=1}{\overset{K}{}}}dx_l\pi (x_l){\displaystyle }{\displaystyle \underset{l=1}{\overset{L}{}}}dy_l\rho (y_l)(1+{\displaystyle \underset{l=1}{\overset{K}{}}}x_l{\displaystyle \underset{l=1}{\overset{L}{}}}y_l)^n\right\}`$ (A69)
$`q\widehat{q}{\displaystyle 𝑑x𝑑\widehat{x}\pi (x)\widehat{\pi }(\widehat{x})(1+x\widehat{x})^n}r\widehat{r}{\displaystyle 𝑑y𝑑\widehat{y}\rho (y)\widehat{\rho }(\widehat{y})(1+y\widehat{y})^n}`$ (A70)
$`+\widehat{q}{\displaystyle \underset{i=1}{\overset{N}{}}}Z_i{\displaystyle }d\widehat{x}\widehat{\pi }(\widehat{x}){\displaystyle \underset{\alpha =1}{\overset{n}{}}}(1+S_i^\alpha \widehat{x})+\widehat{r}{\displaystyle \underset{j=1}{\overset{M}{}}}Y_j{\displaystyle }d\widehat{y}\widehat{\rho }(\widehat{y}){\displaystyle \underset{\alpha =1}{\overset{n}{}}}(1+\tau _j^\alpha \widehat{y})]\}.`$ (A71)
The term involving the spin variables $`S`$ is easily evaluated using the residue theorem
$`{\displaystyle \underset{𝑺^1\mathrm{}𝑺^n}{}}{\displaystyle \underset{i=1}{\overset{N}{}}}\mathrm{exp}\left(\xi F_s{\displaystyle \underset{\alpha =1}{\overset{n}{}}}S_i^\alpha \right)_\xi {\displaystyle \underset{i=1}{\overset{N}{}}}\left\{{\displaystyle \frac{dZ_i}{2\pi i}Z_i^{(C+1)}}\right\}\times \mathrm{exp}\left[\widehat{q}{\displaystyle \underset{i=1}{\overset{N}{}}}Z_i{\displaystyle 𝑑\widehat{x}\widehat{\pi }(\widehat{x})\underset{\alpha =1}{\overset{n}{}}(1+S_i^\alpha \widehat{x})}\right]`$ (A72)
$`=`$ $`\left({\displaystyle \frac{\widehat{q}^C}{C!}}{\displaystyle \underset{l=1}{\overset{C}{}}d\widehat{x}_l\widehat{\pi }(\widehat{x}_l)\underset{\alpha =1}{\overset{n}{}}\left\{e^{\xi F_s}\underset{l=1}{\overset{C}{}}(1+\widehat{x}_l)+e^{\xi F_s}\underset{l=1}{\overset{C}{}}(1\widehat{x}_l)\right\}_\xi }\right)^N,`$ (A73)
and similarly for the term involving the variables $`\tau `$. Substituting these into Eq. (A71), one obtains the $`n`$-th moment of partition function
$`𝒵(𝝃,𝜻,𝒟)^n_{\xi ,\zeta ,𝒟}`$ (A74)
$`=`$ $`\text{extr}_{\{\pi ,\widehat{\pi },\rho ,\widehat{\rho }\}}\{\mathrm{exp}[NC\{{\displaystyle }dxd\widehat{x}\pi (x)\widehat{\pi }(\widehat{x})\mathrm{ln}(1+x\widehat{x})^n1\}`$ (A75)
$`ML\{{\displaystyle }dyd\widehat{y}\rho (y)\widehat{\rho }(\widehat{y})\mathrm{ln}(1+y\widehat{y})^n1\}`$ (A76)
$`+{\displaystyle \frac{1}{2^n}}\{{\displaystyle \frac{NC}{K}}{\displaystyle }\left[{\displaystyle \underset{k=1}{\overset{K}{}}}dx_k\pi (x_k)\right]\left[{\displaystyle \underset{l=1}{\overset{L}{}}}dy_l\rho (y_l)\right]\mathrm{ln}[1+{\displaystyle \underset{k=1}{\overset{K}{}}}x_k{\displaystyle \underset{l=1}{\overset{L}{}}}y_l]^n1\}]`$ (A77)
$`\times \left({\displaystyle \left[\underset{k=1}{\overset{C}{}}d\widehat{x}_k\widehat{\pi }(\widehat{x}_k)\right]\left(\left[e^{F_s\xi }\underset{k=1}{\overset{C}{}}(1+\widehat{x}_k)+e^{F_s\xi }\underset{k=1}{\overset{C}{}}(1\widehat{x}_k)\right]\right)^n_\xi }\right)^N`$ (A78)
$`\times \left({\displaystyle }\left[{\displaystyle \underset{l=1}{\overset{L}{}}}d\widehat{y}_l\widehat{\rho }(\widehat{y}_l)\right]\left([e^{F_n\zeta }{\displaystyle \underset{l=1}{\overset{L}{}}}(1+\widehat{y}_l)+e^{F_n\zeta }{\displaystyle \underset{l=1}{\overset{L}{}}}(1\widehat{y}_l)]\right)^n_\zeta \right)^M\}.`$ (A79)
Finally, in the limit $`n0`$ one obtains
$`{\displaystyle \frac{1}{N}}\mathrm{ln}𝒵(𝝃,𝜻,𝒟)_{\xi ,\zeta ,𝒟}=\underset{n0}{lim}{\displaystyle \frac{𝒵(𝝃,𝜻,𝒟)^n_{\xi ,\zeta ,𝒟}1}{nN}}`$ (A80)
$`=`$ $`\text{extr}_{\{\pi ,\widehat{\pi },\rho ,\widehat{\rho }\}}\{{\displaystyle \frac{C}{K}}\mathrm{ln}2C{\displaystyle }dxd\widehat{x}\pi (x)\widehat{\pi }(\widehat{x})\mathrm{ln}(1+x\widehat{x}){\displaystyle \frac{CL}{K}}{\displaystyle }dyd\widehat{y}\rho (y)\widehat{\rho }(\widehat{y})\mathrm{ln}(1+y\widehat{y})`$ (A81)
$`+{\displaystyle \frac{C}{K}}{\displaystyle \left[\underset{k=1}{\overset{K}{}}dx_k\pi (x_k)\right]\left[\underset{l=1}{\overset{L}{}}dy_l\rho (y_l)\right]\mathrm{ln}\left[1+\underset{k=1}{\overset{K}{}}x_k\underset{l=1}{\overset{L}{}}y_l\right]}`$ (A82)
$`+{\displaystyle \left[\underset{k=1}{\overset{C}{}}d\widehat{x}_k\widehat{\pi }(\widehat{x}_k)\right]\mathrm{ln}\left[e^{F_s\xi }\underset{k=1}{\overset{C}{}}(1+\widehat{x}_k)+e^{F_s\xi }\underset{k=1}{\overset{C}{}}(1\widehat{x}_k)\right]_\xi }`$ (A83)
$`+{\displaystyle \frac{C}{K}}{\displaystyle }\left[{\displaystyle \underset{l=1}{\overset{L}{}}}d\widehat{y}_l\widehat{\rho }(\widehat{y}_l)\right]\mathrm{ln}[e^{F_n\zeta }{\displaystyle \underset{l=1}{\overset{L}{}}}(1+\widehat{y}_l)+e^{F_n\zeta }{\displaystyle \underset{l=1}{\overset{L}{}}}(1\widehat{y}_l)]_\zeta \}.`$ (A84)
## B Evaluation of the Magnetisation
Here, we derive explicitly Eqs.(31) and (32). After using the gauge transformation $`S_i\xi _iS_i`$, the magnetisation can be written as
$$m=\frac{1}{N}\underset{i=1}{\overset{N}{}}\text{sign}(m_i)_{𝝃,𝜻,𝒟},$$
(B1)
introducing the notation $`m_i=S_i_\beta \mathrm{}`$ (gauged average).
For an arbitrary natural number $`p`$, one can compute $`p`$-th moment of $`m_i`$
$$m_{i}^{}{}_{}{}^{p}_{𝝃,𝜻,𝒟}=\underset{n0}{lim}\underset{\beta \mathrm{}}{lim}\underset{\{𝑺^1,𝝉^1\},\mathrm{},\{𝑺^n,𝝉^n\}}{}S_i^1S_i^2\mathrm{}S_i^pe^{\beta _{a=1}^n_a}_{𝝃,\zeta ,𝒟},$$
(B2)
where $`_a`$ denotes the gauged Hamiltonian of the $`a`$-th replica. Decoupling the dynamical variables and introducing auxiliary functions $`\pi ()`$, $`\widehat{\pi }()`$, $`\rho ()`$ and $`\widehat{\rho }()`$, of a similar form to Eq. (A59), one obtains
$$m_{i}^{}{}_{}{}^{p}_{𝝃,𝜻,𝒟}=\underset{l=1}{\overset{C}{}}d\widehat{x}_l\widehat{\pi }(\widehat{x}_l)\mathrm{tanh}^p\left(F_s\xi +\underset{k=1}{\overset{C}{}}\mathrm{tanh}^1\widehat{x}_k\right)_\xi ,$$
(B3)
using the saddle point solution of $`\widehat{\pi }()`$.
Employing the identity
$$\text{sign}(x)=1+2\underset{n\mathrm{}}{lim}\underset{m=0}{\overset{n}{}}\left(\genfrac{}{}{0pt}{}{2n}{m}\right)\left(\frac{1+x}{2}\right)^{2nm}\left(\frac{1x}{2}\right)^m$$
(B4)
which holds for any arbitrary real number $`x[1,1]`$ and Eqs.(B3) and (B4) one obtains
$`\text{sign}(m_i)_{𝝃,𝜻,𝒟}`$ $`=`$ $`1+2{\displaystyle 𝑑z\varphi (z)\underset{n\mathrm{}}{lim}\underset{m=0}{\overset{n}{}}\left(\genfrac{}{}{0pt}{}{2n}{m}\right)\left(\frac{1+z}{2}\right)^{2nm}\left(\frac{1z}{2}\right)^m}`$ (B5)
$`=`$ $`{\displaystyle 𝑑z\varphi (z)\text{sign}(z)},`$ (B6)
where we introduced a new notation for the distribution
$$\varphi (z)=\underset{l=1}{\overset{C}{}}d\widehat{x}_l\widehat{\pi }(\widehat{x}_l)\delta (zF_s\xi \underset{k=1}{\overset{C}{}}\mathrm{tanh}^1\widehat{x}_k)_\xi ,$$
(B7)
thus reproducing Eqs.(31) and (32). |
warning/0003/quant-ph0003061.html | ar5iv | text | # Measurements in quantum physics: towards a physical picture of relevant processes
## I Measurements and ensembles
It took Albert Einstein, in 1935 , to formulate the basic conclusion to Heisenberg’s uncertainty relations : the systems, we are dealing with in quantum theory (QT), are not completely defined by our mathematical representations (In case this is new to you, please read Sheldon Goldstein’s recent articles in Physics Today ). However, given the proof by John von Neumann , that the linearity of the main equations in QT entails that no conventional ensemble can lie underneath the statistical description of measurements, we are left with a somewhat puzzling consequence: although there must be an ensemble, we do not know its structure. Most lucidly, this dilemma has been expressed by David Bohm :
Yet it is not immediately clear, how the ensembles, to which … probabilities refer, are formed and what their individual elements are. For the very terminology of quantum mechanics contains an unusual and significant feature, in that what is called the physical state of a quantum mechanical system is assumed to manifest itself only in an ensemble of systems.
By now the way Bohm proposed to amend this deficiency in the logical structure of QT is well researched: if the wavefunction at a given location and moment is a complex number defined by
$$\psi =Re^{iS}$$
(1)
then the Schrödinger equation gives rise to a ”quantum potential”
$$Q=\frac{^2R}{R}$$
(2)
which, in the same way a physical potential $`V`$ determines the solutions of Schrödinger’s equation, determines the mathematical form of the wavefunction $`\psi `$. There is an important difference, though: since $`Q`$ is proportional to the second derivative of $`R`$, the value of $`\psi `$ at one point of the system influences $`Q`$ throughout the whole system and vice versa: the very ansatz of Bohm thus gives rise to strongly non-local effects .
This non-locality of Bohm’s theory was, initially, taken as an argument against its theoretical soundness (see Goldstein’s article), until John Bell and Alain Aspect proved that the same applies to QT itself . Since then the community is divided: while the more orthodox faction believes in physical non-locality regardless of the contradiction with Special Relativity (see e.g. the recent article in Nature on ”quantum teleportation” ), the more cautious view is that we are dealing with a conceptual non-locality due to our representation of micro physical systems. It seems that only a clear perception of the ensemble structure in QT will allow a decision in this issue. As previous work revealed, the features of the ensembles in quantum theory could arise from intrinsic fields due to particle propagation . This approach goes beyond the proposed modifications of Everett , Ghirardi , Omnes , Hartle , Griffiths , Zurek and others, who, although pioneering new ways to solve the measurement problem in QT, confined themselves more or less to the question of wavefunction collapse.
The strategy employed in developing this new framework was roughly the following: since the mathematical description in QT yields correct results, it should be feasible to complete these - mathematical - expressions by a physical basis. It is evident that the conventional framework of QT allows for no such basis, since the wavefunction has no physical meaning. In the new framework, on the contrary, the wavefunction gains a double meaning: a physical as well as a statistical one, corresponding to the description of single particles (single elements in Bohm’s words) and ensembles of particles. The argument developed in the present paper will be that this double meaning is responsible for some of the most puzzling results in quantum theory. To see, where this double meaning comes from, let me briefly sketch the development of the theoretical framework, for a more thorough view please refer to the original paper .
## II Theoretical foundations
That the energy of a particle depends on its intrinsic frequency $`\omega `$ is the basis of Planck’s quantum hypothesis, thus:
$$E=\mathrm{}\omega $$
(3)
For photons, the frequency is equal to the frequency of its electromagnetic $`\stackrel{}{E}`$ and $`\stackrel{}{B}`$ fields. For massive particles like electrons there is no such connection, one either has to refer to de Broglie’s harmony of phases , or seek a solution in terms of intrinsic and rotational components of motion (e.g. in ). However, if the wave equation describes the intrinsic density of mass distribution $`\rho =\rho (\stackrel{}{r},t)`$, oscillating with the characteristic frequency $`\omega `$, then the total energy of the electron is split into two distinct components :
$$E=\mathrm{}\omega =:m_eu^2E_{QT}=E_K=\frac{m_e}{2}u^2E_F=\frac{m_e}{2}u^2$$
(4)
where $`u`$ is the velocity, $`m_e`$ the mass, and $`E_{QT}=E_K`$ the kinetic energy of the electron, equaling the expression in quantum theory. $`E_F`$, the field energy results from the intrinsic wave features of electron motion. While the latter is, initially, purely hypothetical, it can be justified, a posteriori, by calculating the solutions to Schrödinger’s equation under the condition that $`E_F`$ in QT is generally undefined. From the time-independent equation:
$$\left(\frac{\mathrm{}^2}{2m_e}^2+V\right)\psi =E\psi $$
(5)
and in a plane wave basis set the undefined field component of electron energy leads to an uncertainty in the range of the wavevector $`\stackrel{}{k}`$, and the consequence of this uncertainty can be expressed in Heisenberg’s uncertainty relations :
$$\mathrm{\Delta }P\mathrm{\Delta }X\frac{\mathrm{}}{2}$$
(6)
If the field energy $`E_F`$ is described by an intrinsic scalar field $`\varphi _F`$ and vector fields $`\stackrel{}{E}`$, $`\stackrel{}{B}`$ so that:
$$\varphi _F=\frac{1}{2}\left(\frac{1}{u^2}\stackrel{}{E}^2+\stackrel{}{B}^2\right)E_F=_V𝑑V\varphi _F\stackrel{}{E}\stackrel{}{u}=\stackrel{}{B}\stackrel{}{u}=\stackrel{}{E}\stackrel{}{B}=0$$
(7)
then it can be established that these $`\stackrel{}{E}`$ and $`\stackrel{}{B}`$ fields comply with the Maxwell equations . Please note that the shape of the particles or their volume do not enter the picture, they remain completely undefined. In addition, all the conventional Planck and de Broglie relations are valid, although the latter refer to phase velocity rather than group velocity. Based on these intrinsic features the physical meaning of the wavefunction is that the square of its real part $`\mathrm{}^2(\psi )`$, for a single particle, is proportional to its density of mass $`\rho `$. The statistical meaning is different for electrons and photons, since only the electron has an undefined intrinsic energy component.
In the following we shall simplify the model by assuming that the density of mass of any single particle is constant for a specific type (electron, photon), and described by $`\rho =|\psi |^2`$. The undefined field component of energy $`E_F`$ shall be accounted for by a range of particle energies rather than its development with time. The model treated in the next sections is therefore a static account of an essentially dynamical problem. However, as will be seen presently, this simplified account is sufficient to shed new light on some of the most interesting problems in measurement theory. The statistics in measurement processes result from the following unknown variables:
* Photons: unknown phase, unknown amplitude of wavefunction, unknown number of particles.
* Electrons: unknown phase, unknown amplitude of wavefunction, unknown intrinsic energy, unknown number of particles.
## III Quantum ensemble of free electrons
Due to periodic wave functions, the intrinsic potential at a moment $`t`$ can take any value, and the wave vector of the problem therefore is not exactly determined, but covers the whole range from $`k^2=0`$ to $`k^2=\frac{m}{\mathrm{}^2}E_T`$, where $`E_T=\mathrm{}\omega `$. Consider a point $`\stackrel{}{r}`$, where the external potential vanishes $`V(\stackrel{}{r})=0`$. Due to the disregard for intrinsic potentials the Schrödinger equation at this location applies for all wavelets described by:
$$\stackrel{}{k}^2(\stackrel{}{r})+\stackrel{}{k}_i^2(t)=\frac{m}{\mathrm{}^2}E_T0\stackrel{}{k}_i^2(t)\frac{m}{\mathrm{}^2}E_T$$
(8)
Here $`\stackrel{}{k}_i^2(t)`$ shall denote the intrinsic potential, not accounted for in QM. $`E_T`$ is the total energy of the particle. The two variables are given by:
$$E_T=mu^2\frac{\mathrm{}^2}{m}\stackrel{}{k}_i^2(t)=\varphi _i(t)V_p$$
(9)
where $`u`$ again is the velocity and $`V_P`$ the volume of a particle. The quantum ensemble shall be the integral over allowed wave states. Then the wave function $`\psi (\stackrel{}{r})`$ can be written as:
$$\psi (\stackrel{}{r})=\frac{1}{(2\pi )^{3/2}}_0^{k_0}d^3k\chi _0(\stackrel{}{k})e^{i\stackrel{}{k}\stackrel{}{r}}k_0=\sqrt{\frac{m}{\mathrm{}^2}E_T}$$
(10)
Using a Fourier transformation the amplitudes of the ensemble are:
$$\chi _0(\stackrel{}{k})=\frac{1}{(2\pi )^{3/2}}_{\mathrm{}}^+\mathrm{}d^3r\psi (\stackrel{}{r})e^{i\stackrel{}{k}\stackrel{}{r}}$$
(11)
In this case an ensemble of electrons is described by identical solutions of the Schrödinger equation, although in QT the mathematical representation seems to refer to one particle and one particle only. Bohm’s puzzlement with the individual system, which is in fact an ensemble of systems then is quite justified.
## IV Quantum ensemble in external potentials
An even more interesting consequence of the same feature is observed in external potentials. The external potential at a point $`\stackrel{}{r}`$ has two effects: the range of allowed wavelets and therefore the quantum ensemble will be changed, and the internal properties of single wavelets will be altered. If the potential at $`\stackrel{}{r}`$ equals $`V(\stackrel{}{r})`$, the allowed $`k`$–values will comply with:
$$k^2(\stackrel{}{r})+k_i^2(t)=\frac{m}{\mathrm{}^2}\left(E_TV(\stackrel{}{r})\right)0\stackrel{}{k}_i^2(t)\frac{m}{\mathrm{}^2}$$
(12)
$$\left(E_TV(\stackrel{}{r})\right)k_1^2=\frac{m}{\mathrm{}^2}(E_TV(\stackrel{}{r}))E_T=mu^2$$
(13)
For reasons of consistency the potential $`V(\stackrel{}{r})`$ is double the potential if only kinetic properties are considered. The range of allowed $`k`$–values in this case depends on the energy $`E_T`$ of a single particle as well as the potential applied. There are two distinct cases: $`E_TV(\stackrel{}{r})`$ is either a positive or a negative value, corresponding to wavelets in a potential or to exponential decay of single waves.
For $`E_TV(\stackrel{}{r})>0`$ the potential can either be a positive or a negative value, leading to an enhancement or a reduction of the quantum ensemble of valid solutions. The general solution for both cases is then:
$$\psi (\stackrel{}{r})=\frac{1}{(2\pi )^{3/2}}_0^{k_1}d^3k\chi _0(\stackrel{}{k})e^{i\stackrel{}{k}\stackrel{}{r}}k_1=\sqrt{\frac{m}{\mathrm{}^2}(E_T\pm |V(\stackrel{}{r})|)}E_T=mu^2$$
(14)
The range of individual wavelets defines, as in the case of a vanishing external potential, an ensemble of particles, which comply with the differential Schrödinger equation. Any integration of the equation therefore also contains a – hidden – manifold of individual wavelets. A positive potential essentially limits the number of individual waves contained in the ensemble, because it diminishes the range of $`k`$. A negative potential has the opposite effect: the number of waves in the ensemble is increased, and their statistical weight in the whole system is equally higher. Fig. 1 displays the quantum ensembles for different external potentials.
For energies $`E_T<V(\stackrel{}{r})`$ the mathematical formalism of Schrödinger’s equation allows for solutions with a negative square of $`\stackrel{}{k}`$, equivalent to an exponential decay of single wavelets:
$$k^2(\stackrel{}{r})+k_i^2(t)=\frac{m}{\mathrm{}^2}\left(E_TV(\stackrel{}{r})\right)00\stackrel{}{k}_i^2(t)\frac{m}{\mathrm{}^2}\left(E_TV(\stackrel{}{r})\right)$$
(15)
$$\psi (\stackrel{}{r})=\frac{1}{(2\pi )^{3/2}}_0^{k_1}d^3k\chi _0(\stackrel{}{k})e^{\stackrel{}{k}\stackrel{}{r}}k_1=\sqrt{\frac{m}{\mathrm{}^2}(V(\stackrel{}{r})E_T)}E_T=mu^2$$
(16)
The question in this case concerns not so much the mathematical formalism but the physical validity. Considering, that electrodynamics is a description of intrinsic properties of single particles, applicable to photons as well as electrons, the results of electrodynamics in different media should also have relevance for the wave properties of single particles. And considering, furthermore, that an exponential decay into a medium at a boundary is one type of solution, the same must generally hold for single wavelets. It is a basically classical solution to boundary value problems and, so far, no specific feature of a quantum system.
## V Wave function normalization
We have not yet defined the amplitude $`\chi _0(\stackrel{}{k})`$ of single components in the quantum ensemble. This can be done by requiring single wavelets to comply with the mass relations of particles. Since:
$$\chi ^{}(\stackrel{}{k})\chi (\stackrel{}{k}^{})=\frac{1}{(2\pi )^3}\chi _0(\stackrel{}{k})\chi _0(\stackrel{}{k}^{})e^{i\stackrel{}{r}(\stackrel{}{k}^{}\stackrel{}{k})}$$
(17)
an integration over infinite space yields the result:
$$_{\mathrm{}}^+\mathrm{}d^3r\chi ^{}(\stackrel{}{k})\chi (\stackrel{}{k}^{})=\delta ^3(\stackrel{}{k}\stackrel{}{k}^{})\chi _0(\stackrel{}{k})\chi _0(\stackrel{}{k}^{})m=_{\mathrm{}}^+\mathrm{}d^3r|\chi (\stackrel{}{k})|^2=\chi _0^2(\stackrel{}{k})$$
(18)
Using this amplitude the square of the wave function $`\psi (\stackrel{}{r})`$ in different external potentials can be calculated.
$`{\displaystyle _{\mathrm{}}^+\mathrm{}}d^3r|\psi (\stackrel{}{r})|^2`$ $`=`$ $`{\displaystyle \frac{m}{(2\pi )^3}}{\displaystyle d^3r_0^kd^3kd^3k^{}e^{i\stackrel{}{r}(\stackrel{}{k}^{}\stackrel{}{k})}}`$ (19)
$`=`$ $`m{\displaystyle _0^k}d^3kd^3k^{}\delta ^3(\stackrel{}{k}\stackrel{}{k}^{})={\displaystyle \frac{4\pi m}{3}}k^3`$ (20)
If we consider a system, where the potential $`V=V(\stackrel{}{r})`$, the wavefunction will depend, via the normalization condition, on the potentials in all parts of this system. And if we formalize two space-like separated events within a single system in QT (there is no way to avoid this in any experiment aimed at proving or refuting non-locality), these events are no longer logically (or physically) separate: non-locality is therefore an inherent conceptual feature of quantum theory.
## VI Boundary conditions
One of the easiest examples in quantum theory, which suffices to demonstrate the effect of boundary conditions, is the square potential well. We take a one dimensional potential well, the external potentials described by:
$$V=0|x|x_0V=V_0|x|x_0$$
(21)
To solve the problem we have two consider the behavior of single members of the quantum ensemble as well as the behavior of the whole ensemble. The limiting $`k`$ values can again be inferred from the solution of the one–dimensional Schrödinger equation, they will be for $`E_T<V_0`$:
$`k_0^2`$ $``$ $`{\displaystyle \frac{m}{\mathrm{}^2}}E_T|x|x_0`$ (22)
$`k_0^2`$ $``$ $`{\displaystyle \frac{m}{\mathrm{}^2}}(V_0E_T)|x|x_0`$ (23)
For incident, reflected and penetrating waves, the three components of the wave are a wave of positive and a wave of negative propagation in the region $`|x|<x_0`$, and a decaying component in the region $`|x|>x_0`$. Considering individual members of the ensemble, the lowest $`k`$ value should correspond to maximum decay in the potential, while the member with maximum total energy should display maximum penetration. The relation between an arbitrary wave vector $`k_1`$ and its corresponding member $`k_2`$ must therefore be:
$$k_1^2+k_2^2=\frac{m}{\mathrm{}^2}V_0$$
(24)
The wave functions in the three separate regions shall be described by standard solutions. Accounting for the boundary conditions for steady transition of the particle wave the coefficients can be determined and the solution for an individual wave is therefore, equivalent to the solution in quantum theory :
$$\chi _0e^{k_2x}xx_0$$
(25)
$$\chi (x)=\chi _0e^{k_2x_0}\frac{\mathrm{cos}k_1x}{\mathrm{cos}k_1x_0}|x|x_0$$
(26)
$$\chi _0e^{k_2x}xx_0$$
(27)
The normalization condition an ensemble member yields the amplitude of the particle wave:
$`\chi _0(k_1,k_2)=\sqrt{{\displaystyle \frac{mk_2}{1+k_2x_0}}}e^{k_2x_0}\mathrm{cos}k_1x_0`$ (28)
And the total ensemble can equally be calculated by integrating over the full range of allowed $`k`$ values.
$`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑x|\psi (x)|^2`$ $`=`$ $`2{\displaystyle _0^{\mathrm{}}}dx[\theta (x_0x){\displaystyle _0^{k_0}}dk_1\chi _0^2(k_1)e^{2k_1x_0}{\displaystyle \frac{\mathrm{cos}^2(k_1x)}{\mathrm{cos}^2(k_1x_0)}}+`$ (29)
$`+`$ $`\theta (xx_0){\displaystyle _0^{k_0^{}}}dk_2\chi _0^2(k_2)e^{2k_2x}]`$ (30)
The amplitude $`\chi _0(k)`$ must finally be renormalized, and the square of the wave function then describes the probability distribution of the whole ensemble. The procedure described is similar to the standard procedure in quantum theory, although in this model $`k`$ space is not unlimited, the cutoff is determined by the energy $`E_T`$. The structure of the ensembles in different environments provides a means to analyze the interplay between physical aspects (the intrinsic properties of single electrons) and statistical ones (the way quantum theory contains a hidden ensemble of single electrons).
## VII Spreading of a wave–packet
The effect is a common source of irritation, and concepts have been put forth to eliminate it in a modified version of quantum theory (see, for example, Mackinnon , or the GRW model ). In a one dimensional model the development of an amplitude $`\widehat{\psi }(k)`$ is described by the integral:
$$\psi (x,t)=𝑑ke^{i(kx\omega t)}\widehat{\psi }(k)$$
(31)
Two initial distributions are calculated: a Gaussian distribution, with a wave function centered around a value $`x_0=0`$ at $`t=0`$, and an ensemble with exactly defined energies (such an ensemble can be obtained by energy measurements, as shown further down):
$$\psi _1(x,t=0)=e^{\frac{x^2}{b^2}+ik_0x}\psi _2(x,t=0)=e^{ik_0x}$$
(32)
The evaluation of the integral yields:
$`\widehat{\psi }_1(k)={\displaystyle 𝑑x\psi _1(x,0)e^{ikx}}=e^{(kk_0)^2b^2/2}`$ (33)
$`\widehat{\psi }_2(k)={\displaystyle 𝑑x\psi _2(x,0)e^{ikx}}=\delta (kk_0)`$ (34)
The square of the two wave functions at $`t>0`$ is consequently:
$$\left|\psi _1(x,t>0)\right|^2=\left(1+\frac{\mathrm{}^2t^2}{m^2b^4}\right)^1\mathrm{exp}\left[b^2\left(1+\frac{\mathrm{}^2t^2}{m^2b^4}\right)^2\left(x\frac{\mathrm{}k_0}{m}t\right)^2\right]$$
(35)
$$\left|\psi _2(x,t>0)\right|^2=1$$
(36)
The development of the two ensembles is shown in Fig. 2. As it is possible to decompose an arbitrary distribution of initial $`k`$ values in Gaussian distributions, the result holds quite generally. There are two possibilities to account for this feature: (i) Either the restructuring is referred to some potential not covered by field theories, in this case we would have to recur to Bohm’s quantum potential . (ii) The initial conditions contain an assumption which affects the physical properties of particles. In the single particle case (see section II), where $`[\mathrm{}\psi (x)]^2\rho (x)\varphi (x)`$, an inhomogeneous distribution like $`\psi _1(x,0)`$ gives rise to an intrinsic potential $`\varphi (x)`$, described by:
$$\psi _1(x,0)=\psi _0e^{ik_0x}\psi _0=e^{x^2/2b^2}\varphi (x,0)=\psi _0^2=e^{x^2/b^2}$$
(37)
And the system is therefore not, as implied by the mathematical formulations, free of forces, but will experience a force along the direction $`x`$:
$$F_x=\frac{\varphi }{x}=\frac{2x}{b^2}e^{x^2/b^2}$$
(38)
Along this line of reasoning we may reconsider the question of quantum ensembles from the viewpoint of intrinsic potentials and forces. The most general form of a wave function is given by the integral:
$$\psi (\stackrel{}{r})=d^3k\psi _{0,\stackrel{}{k}}(\stackrel{}{r},\stackrel{}{k})e^{i\stackrel{}{k}\stackrel{}{r}}$$
(39)
The potentials due to the qualities of the amplitude are then responsible for intrinsic energy components in addition to the purely periodic fields of a plane wave. They are:
$$\varphi _i(\stackrel{}{r},\stackrel{}{k})=u^2\left|\psi _{0,\stackrel{}{k}}(\stackrel{}{r},\stackrel{}{k})\right|^2=\frac{\mathrm{}^2k^2}{m^2}\left|\psi _{0,\stackrel{}{k}}(\stackrel{}{r},\stackrel{}{k})\right|^2$$
(40)
The forces within the propagating wave are either periodic – the total energy density of the plane wave is constant –, or they are forces due to the properties of the amplitude. These forces will be:
$$\stackrel{}{F}=\frac{\mathrm{}^2k^2}{m^2}\left[\psi _{0,\stackrel{}{k}}^{}\psi _{0,\stackrel{}{k}}+\psi _{0,\stackrel{}{k}}\psi _{0,\stackrel{}{k}}^{}\right]$$
(41)
A stable state of the system can only be expected, if these forces vanish. The equilibrium condition for a system of particles described as plane waves is therefore:
$$\psi _{0,\stackrel{}{k}}^{}\psi _{0,\stackrel{}{k}}+\psi _{0,\stackrel{}{k}}\psi _{0,\stackrel{}{k}}^{}=0$$
(42)
The two ensembles defined, the ensemble with exact energies as well as the quantum ensemble (equal amplitude of all partial waves), comply with this condition since in both cases the amplitudes $`\psi _{0,\stackrel{}{k}}`$ do not depend on $`\stackrel{}{r}`$. But the distribution used for the calculation of the spreading wave packet is incompatible with this condition.
## VIII Collapse of the wave function
In his rather fundamental and comprehensive analysis of measurement processes in quantum theory Ballentine proceeded from two mutually exclusive statements on the quality of the state concept, i.e. that (i) a pure state provides a complete and exhaustive description of an individual system, and (ii) a pure (or mixed) state describes the statistical properties of an ensemble of similarly prepared systems. The subsequent analysis of measurement processes proved that any interpretation of the type (i) …is untenable.
If the state vector of a system, or the wave function in QT were an exhaustive information about the system, the logical problems seem indeed severe if not unsurmountable. The situation changes, though, if one concedes that the quantum mechanical description does not provide a full account of physical variables. To a greater or lesser extent all the proposed modifications of QT to account for the measurement problems, cited in the introduction, use this feature. As a simple example of the reduction of the wave function in a measurement process we consider a retarding field analyzer frequently employed in LEED (low energy electron diffraction) measurements. A retarding field analyzer is a positive potential, assumed rectangular for simplicity, which selects only electrons above a certain energy threshold. We equally assume, that the electrons initially are free, their energy shall be given by a value $`E_k`$ (see Fig. 3). From a strictly causal point of view, the electrons below an exactly defined threshold value $`E_{rfa}<E_k`$ cannot pass the filter and the number of electrons after the filter is therefore reduced to single particles with an energy above the threshold value. As the calculation in quantum theory integrates over all possible particle states at a given location $`\stackrel{}{r}`$, and since the range of allowed $`k`$–values depends on the level of kinetic energy, the total density $`\rho (\stackrel{}{r})`$ at any location before the analyzer will be:
$$\rho (\stackrel{}{r})=\psi ^{}(\stackrel{}{r})\psi (\stackrel{}{r})=_0^{k_0}d^3k\chi ^{}(\stackrel{}{k},\stackrel{}{r})\chi (\stackrel{}{k},\stackrel{}{r})k_0^2=\frac{2m}{\mathrm{}^2}E_k$$
(43)
while after the analyzer the wave function will be limited to states with energy values higher than the threshold:
$$\rho ^{}(\stackrel{}{r})=\psi ^{}(\stackrel{}{r})\psi ^{}(\stackrel{}{r})=_{k_1}^{k_0}d^3k\chi ^{}(\stackrel{}{k},\stackrel{}{r})\chi (\stackrel{}{k},\stackrel{}{r})k_1^2=\frac{2m}{\mathrm{}^2}(E_kE_{rfa})$$
(44)
Clearly the quantum ensemble has been reduced. The measurement of particle energies by retarding fields therefore leads to a reduction of the statistical ensemble, but in a causal and deterministic manner. The ensemble wave function is reduced due to the removal, in an equally causal and deterministic way, of partial waves. It should be noted that the collapse of the wave function in real space cannot, presently, be treated within the same model.
## IX The quantum eraser
The polarization of the intrinsic fields is decisive for interference measurements as can be demonstrated by an analysis of quantum eraser phenomena. In this case the which–path information of the photon is said to preclude interference. Conventionally, the measurement is formalized as follows . The amplitude of an incident photon of horizontal polarization is split coherently in two separate beams, described by the quantum state vector ($`1`$ denotes the first, $`2`$ the second of the two paths)
$$\psi _{12}^0=\frac{1}{\sqrt{2}}\left(\psi _{1,H}+\psi _{2,H}\right)$$
(45)
The square of $`\psi `$, or the probability density in this case contains an interference term $`\psi _{1,H}^{}\psi _{2,H}`$:
$`|\psi _{12}^0|^2={\displaystyle \frac{1}{2}}\left(|\psi _{1,H}|^2+|\psi _{2,H}|^2+\psi _{1,H}^{}\psi _{2,H}+\psi _{2,H}^{}\psi _{1,H}\right)`$
If a polarization rotator changing the polarization of the beam to vertical (V) orientation is placed in path 1, the interference pattern is no longer observable, and the measurement yields random results for the local probability density on the measurement screen. In quantum theory the result is referred to the orthogonality of the two states $`H`$, and $`V`$, and the state vector of the photon described by:
$$\psi _{12}^1=\frac{1}{\sqrt{2}}\left(\psi _{1,V}+\psi _{2,H}\right)|\psi _{12}^1|^2=\frac{1}{2}\left(|\psi _1|^2+|\psi _2|^2\right)$$
(46)
The result can be changed by inserting a diagonal polarizer into the path of the recombined beams, in this case the wave function and probability density will again show interference effects: the which–path information, connected to the polarization of the two separate beams is said to have been ”erased”.
$$\psi _{12}^2=\frac{1}{2\sqrt{2}}\left(\psi _1+\psi _2\right)\left(\psi _V+\psi _H\right)|\psi _{12}^2|^2=\frac{1}{4}\left(|\psi _1|^2+|\psi _2|^2+2\mathrm{}\left[\psi _1^{}\psi _2\right]\right)$$
(47)
In the context of intrinsic properties and polarizations of intrinsic fields, since the intensity of electromagnetic radiation is described by the electromagnetic potential $`\varphi _{em}`$ (for a general description the field vectors are assumed complex):
$$\varphi _{em}=\frac{1}{2}\left(\frac{1}{c^2}|\stackrel{}{E}|^2+|\stackrel{}{B}|^2\right)$$
(48)
If the beam of horizontal polarization (direction $`x`$) is split into two separate beams, the electric and magnetic fields after recombination will be:
$$\stackrel{}{E}_{12}^0=\frac{1}{\sqrt{2}}\left(E_1\stackrel{}{e}_x+E_2\stackrel{}{e}_x\right)\stackrel{}{B}_{12}^0=\frac{1}{\sqrt{2}}\left(B_1\stackrel{}{e}_y+B_2\stackrel{}{e}_y\right)$$
(49)
where the fields $`E_2,B_2`$ contain the phase information $`e^{i\phi }`$. The intensity measured after recombination will consequently contain interference terms:
$$\varphi _{em}^0=\frac{1}{2}\left(\frac{1}{c^2}|E_1|^2+|B_1|^2\right)\left(1+cos\phi \right)$$
(50)
A polarization rotator in path 1 changes the polarization of the electric and magnetic fields to $`\stackrel{}{e}_y`$ and $`\stackrel{}{e}_x`$ respectively, and the intensity after recombination is then not affected by the phase $`\phi `$:
$$\stackrel{}{E}_{12}^1=\frac{1}{\sqrt{2}}\left(E_1\stackrel{}{e}_y+E_2\stackrel{}{e}_x\right)\stackrel{}{B}_{12}^1=\frac{1}{\sqrt{2}}\left(B_1\stackrel{}{e}_x+B_2\stackrel{}{e}_y\right)$$
(51)
$$\varphi _{em}^1=\left(\frac{1}{c^2}|E_1|^2+|B_1|^2\right)$$
(52)
If the recombined beam is passing through a diagonal polarizer (plane of polarization in $`xy`$–direction), the electromagnetic fields after polarization are:
$$\stackrel{}{E}_{12}^2=\frac{1}{2}\left(E_1\stackrel{}{e}_{xy}+E_2\stackrel{}{e}_{xy}\right)\stackrel{}{B}_{12}^2=\frac{1}{2}\left(B_1\stackrel{}{e}_{yx}+B_2\stackrel{}{e}_{yx}\right)$$
(53)
And the intensity of the beam shows again the interference pattern of the phase $`\phi `$:
$$\varphi _{em}^2=\frac{1}{4}\left(\frac{1}{c^2}|E_1|^2+|B_1|^2\right)\left(1+cos\phi \right)$$
(54)
Mathematically, the description by way of intrinsic potentials and polarizations yields the same result as the conventional calculation in quantum theory. However, the interesting aspect of the effect is its interpretation. While in the conventional framework the which–path information (and its relation to the conception of complementarity) is seen as the ultimate reason for the experimental results, it is, in the new theory, the intrinsic information due to the electromagnetic fields and their vector features, which are held responsible.
## X Interaction–free measurements
An interaction–free measurement, which is based on a thought experiment by Renninger , provides information about the existence of an object in a closed system without necessarily interacting with this object. The essentials of such a measurement, recently undertaken by Kwiat et al. , can be seen in Fig. 4. Interaction free measurements, usually performed with down–converted photons, are interesting due to two features: the wave function of the system and consequently system energy is changed, even if no interaction occurs. And the results are seemingly incompatible with classical field theories, because the trajectories of single particles through the measurement apparatus can be identified.
The first feature was modeled by Dicke using a modified Heisenberg microscope and calculating the state vectors of photons and a non–interacting atom in first order perturbation theory. The result of Dicke’s calculation seemed to prove that even interaction free measurements correlate with an exchange of virtual photons or, in Dicke’s words: The apparent lack of interaction between the atom and the electromagnetic field is only illusionary. On the statistical basis developed in this paper, the result must be modified. The local modification of ensemble ranges means, in this context, that an interaction free measurement corresponds to a different ensemble, i.e. an ensemble which has zero probability in the range, where an interacting particle is appreciable. It is therefore the limitation imposed, the change of boundary conditions, which is the ultimate reason for the change of the wave function. And if this local range affects system energy like in Dicke’s model of an harmonic oscillator in a magnetic field , then the energy of the system changes.
The second feature of interaction free measurements, the assertion by Kwiat et al. that ”complementary is essential” to the experimental results achieved, requires a critical analysis. What the argument indicates, is the impossibility for a single photon in the interferometer to trigger the bomb and detector D2 (see Fig. 4). But as detection efficiency is only two percent, about 98 % of the incident energy (triggering a detector by one of the down–converted photons) is not accounted for. And in this case the argument of complementarity as well as the whole argumentation of interaction–free measurements seems questionable.
## XI Conclusion
We have shown in this paper that the intrinsic energy component plays a substantial role in the resolution of some of the most important paradoxes in quantum theory. In particular we found that: (i) the spreading of a wave packet is due to a peculiar choice of initial conditions which, in the present model, are physically problematic. (ii) The collapse of the wavefunction in $`k`$-space is due to a removal of partial waves from the full ensemble during the measuring process. (iii) The quantum eraser can be seen as an example, where the polarizations of intrinsic fields become decisive. (iv) And interaction-free measurements are not paradoxical, if the local extension of an ensemble is considered.
## Acknowledgements
Thanks are due to the Österreichische Forschungsgemeinschaft for their generous support to attend the Wigner Symposium. |
warning/0003/math0003063.html | ar5iv | text | # On the Intersection of Two Plane Curves
## 1. Introduction and Statement of Results
The following question was raised and partially answered by Geng Xu in \[X\].
###### Question 1.1.
Let $`D`$ be a general degree $`d`$ curve in $`^2`$. What is the minimal number $`i(d,m)`$ of points in the set-theoretical intersection $`CD`$ for any degree $`m`$ irreducible curve $`C`$ (suppose that $`C`$ and $`D`$ meet properly)?
This problem is related to a conjecture of Kobayashi and Zaidenberg which states that for a sufficiently general curve $`D^2`$ of degree $`d5`$, as general in the sense that $`D`$ lies in $`|𝒪_^2(d)|^{d(d+3)/2}`$ with countably many closed proper subvarieties removed, the affine variety $`^2\backslash D`$ is hyperbolic. One necessary condition for $`^2\backslash D`$ being hyperbolic is that there is no rational curve $`C^2`$ meeting $`D`$ set-theoretically at fewer than three points; otherwise, there is going to be a nonconstant holomorphic map $`C\backslash (CD)^2\backslash D`$. This property of $`^2\backslash D`$ was called “algebraic hyperbolic” in \[DSW\].
###### Definition 1.1.
A quasi-projective variety is called algebraic hyperbolic if it does not contain a curve whose normalization is an elliptic curve or a rational curve with two points removed, i.e., $`^1\backslash \{p,q\}^{}Spec[x,x^1]`$.
Obviously, hyperbolicity implies algebraic hyperbolicity for smooth quasi-projective varieties.
Using an elegant deformation-theoretical argument, Xu proved the following \[X, Theorem 1\].
###### Theorem 1.1 (Xu).
For $`d3`$, $`\mathrm{min}_{m>0}i(d,m)=d2`$.
He thus concluded that every curve $`C^2`$ meets $`D`$ at no less than three distinct points and hence $`^2\backslash D`$ is algebraic hyperbolic for a sufficiently general curve $`D`$ of degree $`d5`$. This bound is sharp for $`m=1`$ and it is achieved by a bitangent or flex line to $`D`$.
The purpose of this paper is two-fold. First, we will try to sharpen his bound with both $`d`$ and $`m`$ fixed. Second, we will try to extend his result to other surfaces.
By dimension count, one may expect that $`i(d,m)=dmr_{d,m}`$ where $`r_{d,m}`$ is the dimension of the linear series cut out on $`D`$ by all curves of degree $`m`$, namely, $`r_{d,m}=m(m+3)/2`$ for $`m<d`$ and $`r_{d,m}=dm(d1)(d2)/2`$ for $`md`$. However, this is simply false for $`md3`$ by the following construction.
Let $`L`$ be a bitangent (or flex) line to $`D`$. Since $`D`$ is general, $`L`$ meets $`D`$ at $`d2`$ distinct points. Let $`L(X,Y,Z)`$ and $`D(X,Y,Z)`$ be the homogeneous defining equations of $`L`$ and $`D`$, respectively. Then for any degree $`md`$ homogeneous polynomial $`G(X,Y,Z)`$, $`L^m(X,Y,Z)+D(X,Y,Z)G(X,Y,Z)=0`$ defines a degree $`m`$ curve $`C`$ which meets $`D`$ at $`d2`$ points, which are the intersections between $`L`$ and $`D`$. If we choose $`G(X,Y,Z)`$ general enough, $`C`$ is irreducible and actually smooth. Hence, by Xu’s result, $`i(d,m)=d2`$ for $`md3`$.
Nevertheless, we think that $`i(d,m)`$ has the expected value for $`d>m`$, i.e.,
###### Conjecture 1.1.
For $`d>m`$ and $`d3`$, $`i(d,m)=dmr_{d,m}`$.
Although we cannot prove the above conjecture, we have the following estimate for $`i(d,m)`$ when $`m<d`$.
###### Theorem 1.2.
For $`d>m`$,
$$i(d,m)\mathrm{min}(dm\frac{m(m+3)}{2},2dm2m^22).$$
An easy corollary of the above theorem is the following
###### Corollary 1.1.
For $`2d3m2`$ and $`d3`$, $`i(d,m)=dmm(m+3)/2`$, i.e., Conjecture 1.1 holds for $`2d3m2`$. In particular, it holds for $`m4`$.
In order to formulate Kobayashi type conjectures on surfaces other than $`^2`$, we need to study Question 1.1 in the following general setting.
###### Question 1.2.
Let $`S`$ be a smooth surface and let $`L`$ and $`M`$ be two line bundles on $`S`$. Let $`D`$ be a general member of $`|L|`$. What is the minimal number $`i(L,M)`$ of points in the set-theoretical intersection $`CD`$ for any irreducible curve $`C|M|`$ (suppose that $`C`$ and $`D`$ meet properly)?
We will work on rational ruled surfaces, although our method can be extended to other surfaces. By convention, let $`𝔽_n`$ be the rational ruled surface given by $`(𝒪_^1𝒪_^1(n))`$ over $`^1`$ and let $`C`$ and $`F`$ be the zero section and the fiber of $`𝔽_n^1`$, i.e., $`C^2=n`$, $`CF=1`$ and $`F^2=0`$. We have the following lower bound for $`i(L,M)`$ with $`L`$ ample.
###### Theorem 1.3.
Let $`L=𝒪(aC+bF)`$ be an ample line bundle on $`𝔽_n`$ with $`a2`$ and $`b2`$. Then $`\mathrm{min}_Mi(L,M)=\mathrm{min}(a1,ban,bn1)`$, where $`M`$ runs over all line bundles with irreducible general global sections.
It follows immediately from Theorem 1.3 that every curve on $`𝔽_n`$ meets $`D`$ at no less than three distinct points for a sufficiently general $`D|aC+bF|`$ with $`a4`$ and $`b\mathrm{max}(4+n,3+an)`$. Therefore
###### Corollary 1.2.
For a sufficiently general curve $`D|aC+bF|`$ on $`𝔽_n`$ with $`a4`$ and $`b\mathrm{max}(4+n,3+an)`$, the complement $`𝔽_n\backslash D`$ is algebraic hyperbolic.
Notice that the bound in Theorem 1.3 can be achieved by a curve in $`|C|`$ or $`|F|`$, which is necessarily a rational curve. So the lower bounds for $`a`$ and $`b`$ in the above corollary cannot be improved.
This enables us to formulate Kobayashi conjecture on $`𝔽_n`$.
###### Conjecture 1.2 (Kobayashi Conjecture on Rational Ruled Surfaces).
For a sufficiently general curve $`D|aC+bF|`$ on $`𝔽_n`$ with $`a4`$ and $`b\mathrm{max}(4+n,3+an)`$, the complement $`𝔽_n\backslash D`$ is hyperbolic.
The organization of this paper is as follows. Theorem 1.2 and 1.3 will be proved in the next two sections, respectively. At the end of the third section, we will also discuss some related problems.
Throughout this paper we work exclusively over the field of complex numbers $``$.
Acknowledgments. I am very grateful to M. Green, who introduced me to the subject of hyperbolic geometry and Kobayashi Conjecture, and to C. Hacon for helpful conversations.
## 2. Proof of Theorem 1.2
Let $`W_\delta |𝒪(m)|\times |𝒪(d)|`$ be the incidence correspondence defined by
$$\begin{array}{cc}\hfill W_\delta =\{& (C,D):C|𝒪(m)|\text{ is irreducible},D|𝒪(d)|\text{ is smooth},\hfill \\ & \text{and }C\text{ and }D\text{ meet set-theoretically at }\delta \text{ points}\}.\hfill \end{array}$$
Our proof of Theorem 1.2 is carried out by estimating the dimension of $`W_\delta `$ and show that $`dimW_\delta <dim|𝒪(d)|`$ if $`\delta <dmm(m+3)/2`$ and $`\delta <2dm2m^22`$ and hence it cannot dominate $`|𝒪(d)|`$ in this case.
Let $`\pi :W_\delta |𝒪(m)|`$ be the projection of $`W_\delta `$ to $`|𝒪(m)|`$ and let $`C`$ be a general point of $`\pi (W_\delta )`$ (by a general point, we mean a general point of an irreducible component of $`\pi (W_\delta )`$).
Let $`\pi _C`$ be the fiber of $`\pi :W_\delta |𝒪(m)|`$ over $`C`$ and let $`(C,D)`$ be a general point on $`\pi (C)`$. There exists a series of blowups of $`^2`$ such that the proper transforms $`\stackrel{~}{C}`$ and $`\stackrel{~}{D}`$ of $`C`$ and $`D`$ meet at smooth points on both curves (since we assume that $`D`$ is smooth, we only have to resolve the singularities of $`C`$ where $`D`$ passes through). Let $`\stackrel{~}{^2}`$ be the resulting blowup of $`^2`$ and $`E_i`$ ($`1i\alpha )`$ be the exceptional divisors. Suppose that $`\stackrel{~}{C}|𝒪(mH_{i=1}^\alpha r_iE_i)`$ and $`\stackrel{~}{D}|𝒪(dH_{i=1}^\alpha E_i)|`$, where $`r_i>1`$ and $`H`$ is the pull-back of the hyperplane divisor of $`^2`$.
Suppose that $`\stackrel{~}{C}`$ and $`\stackrel{~}{D}`$ meet at points $`p_1,p_2,\mathrm{},p_\beta `$ with multiplicities $`m_1,m_2,\mathrm{},m_\beta `$, respectively, where $`\beta \delta \alpha +\beta `$. Then by a deformation-theoretical argument, the tangent space $`T_{\pi _C,(C,D)}`$ of $`\pi _C`$ at $`(C,D)`$ is contained in
$$\begin{array}{cc}& H^0(𝒪_{\stackrel{~}{D}}(dH\underset{i=1}{\overset{\alpha }{}}E_i)𝒪_{\stackrel{~}{D}}(\underset{j=1}{\overset{\beta }{}}(m_j1)p_j))\hfill \\ & =H^0(𝒪_{\stackrel{~}{D}}((dm)H+\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i)𝒪_{\stackrel{~}{D}}(\underset{j=1}{\overset{\beta }{}}p_j)).\hfill \end{array}$$
Hence by Riemann-Roch
$$\begin{array}{cc}\hfill dim\pi _C& h^0(𝒪_{\stackrel{~}{D}}((dm)H+\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i+\underset{j=1}{\overset{\beta }{}}p_j))\hfill \\ & =\frac{d(d+3)}{2}+\underset{i=1}{\overset{\alpha }{}}(r_i1)+\beta dm\hfill \\ & +h^0(𝒪_{\stackrel{~}{D}}((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i)𝒪_{\stackrel{~}{D}}(\underset{j=1}{\overset{\beta }{}}p_j)).\hfill \end{array}$$
It is not hard to see that
(2.1)
$$\begin{array}{cc}& h^0(𝒪_{\stackrel{~}{D}}((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i)𝒪_{\stackrel{~}{D}}(\underset{j=1}{\overset{\beta }{}}p_j))\hfill \\ & h^0(𝒪_{\stackrel{~}{C}}((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i)𝒪_{\stackrel{~}{C}}(\underset{j=1}{\overset{\beta }{}}p_j))\hfill \\ & +\underset{i=1}{\overset{\alpha }{}}\frac{(r_i1)(r_i2)}{2}.\hfill \end{array}$$
This can be shown by the following argument.
We further blow up $`\stackrel{~}{^2}`$ at points $`p_1,p_2,\mathrm{},p_\beta `$ with corresponding exceptional divisors $`F_1,F_2,\mathrm{},F_\beta `$. We still denote the resulting surface by $`\stackrel{~}{^2}`$ and the proper transforms of $`C`$ and $`D`$ by $`\stackrel{~}{C}`$ and $`\stackrel{~}{D}`$. We have the following exact sequence on $`\stackrel{~}{^2}`$
$$\begin{array}{cc}\hfill 0& H^0((md3)H\underset{i=1}{\overset{\alpha }{}}(r_i2)E_i)\hfill \\ & H^0((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i\underset{j=1}{\overset{\beta }{}}F_j)\hfill \\ & H^0(𝒪_{\stackrel{~}{D}}((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i\underset{j=1}{\overset{\beta }{}}F_j))\hfill \\ & H^1((md3)H\underset{i=1}{\overset{\alpha }{}}(r_i2)E_i).\hfill \end{array}$$
Obviously, $`h^0((md3)H_{i=1}^\alpha (r_i2)E_i)=0`$ and
$$\begin{array}{cc}& h^2((md3)H\underset{i=1}{\overset{\alpha }{}}(r_i2)E_i)\hfill \\ & =h^0((dm)H+\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i+\underset{j=1}{\overset{\beta }{}}F_j)\hfill \\ & =\frac{(dm)(dm+3)}{2}+1.\hfill \end{array}$$
Hence by Riemann-Roch,
$$h^1((md3)H\underset{i=1}{\overset{\alpha }{}}(r_i2)E_i)=\underset{i=1}{\overset{\alpha }{}}\frac{(r_i1)(r_i2)}{2}.$$
Therefore
(2.2)
$$\begin{array}{cc}& h^0(𝒪_{\stackrel{~}{D}}((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i\underset{j=1}{\overset{\beta }{}}F_j))\hfill \\ & h^0((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i\underset{j=1}{\overset{\beta }{}}F_j)+\underset{i=1}{\overset{\alpha }{}}\frac{(r_i1)(r_i2)}{2}.\hfill \end{array}$$
Similarly, we have
(2.3)
$$\begin{array}{cc}& h^0(𝒪_{\stackrel{~}{C}}((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i\underset{j=1}{\overset{\beta }{}}F_j))\hfill \\ & =h^0((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i\underset{j=1}{\overset{\beta }{}}F_j).\hfill \end{array}$$
Combining (2.2) and (2.3), we obtain (2.1). Therefore
$$\begin{array}{cc}\hfill dim\pi _C& \frac{d(d+3)}{2}+\underset{i=1}{\overset{\alpha }{}}\frac{r_i(r_i1)}{2}+\beta dm\hfill \\ & +h^0(𝒪_{\stackrel{~}{C}}((m3)H\underset{i=1}{\overset{\alpha }{}}(r_i1)E_i)𝒪_{\stackrel{~}{C}}(\underset{j=1}{\overset{\beta }{}}p_j))\hfill \\ & =\frac{d(d+3)}{2}+\underset{i=1}{\overset{\alpha }{}}\frac{r_i(r_i1)}{2}+\beta dm+h^0(\omega _{\stackrel{~}{C}}𝒪_{\stackrel{~}{C}}(\underset{j=1}{\overset{\beta }{}}p_j))\hfill \end{array}$$
where $`\omega _{\stackrel{~}{C}}`$ is the dualizing sheaf of $`\stackrel{~}{C}`$.
By Clifford’s theorem (see for example \[ACGH, pp. 107-8\]), we have either
$$h^0(\omega _{\stackrel{~}{C}}𝒪_{\stackrel{~}{C}}(\underset{j=1}{\overset{\beta }{}}p_j))=0$$
or
$$h^0(𝒪_{\stackrel{~}{C}}(\underset{j=1}{\overset{\beta }{}}p_j))\beta /2+1.$$
Hence correspondingly, we have either
$$dim\pi _C\frac{d(d+3)}{2}+\underset{i=1}{\overset{\alpha }{}}\frac{r_i(r_i1)}{2}+\beta dm$$
or
$$dim\pi _C\frac{d(d+3)}{2}+\frac{\beta }{2}dm+\frac{(m1)(m2)}{2}.$$
Since $`C`$ is a general member of $`\pi (W_\delta )`$ and $`C`$ has singularities with multiplicities $`r_i`$ ($`1i\alpha `$), by Zariski’s theorem on the deformation of planary curve singularities \[Z\], we have
$$dim\pi (W_\delta )\frac{m(m+3)}{2}\underset{i=1}{\overset{\alpha }{}}\frac{r_i(r_i1)}{2}.$$
And hence we have either
$$dimW_\delta \frac{d(d+3)}{2}+\beta dm+\frac{m(m+3)}{2}$$
or
$$dimW_\delta \frac{d(d+3)}{2}+\frac{\beta }{2}dm+m^2+1\underset{i=1}{\overset{\alpha }{}}\frac{r_i(r_i1)}{2}.$$
Therefore, if $`W_\delta `$ dominates $`|𝒪(d)|`$, we necessarily have
$$\delta \beta \mathrm{min}(dm\frac{m(m+1)}{2},2(dmm^21)).$$
This finishes the proof of Theorem 1.2.
## 3. Intersections of Two Curves on Rational Ruled Surfaces
Our approach to Theorem 1.3 is different from that of Xu’s. A key ingredient of Xu’s proof of Theorem 1.1 is a map from the deformation space of the pair $`(D,E)`$, where $`D|𝒪_^2(d)|`$ and $`E|𝒪_^2(m)|`$ meet at no less than $`s`$ distinct points, to the cohomology group of a sheaf over $`D`$. More specifically, let $`(Z_0,Z_1,Z_2)`$ be generic homogeneous coordinates of $`^2`$ and let $`F_0H^0(𝒪_^2(d))`$ and $`G_0H^0(𝒪_^2(m))`$ be the defining equations of $`D`$ and $`E`$. A first order deformation of $`(D,E)`$ is given by $`F_1H^0(𝒪_^2(d))`$ and $`G_1H^0(𝒪_^2(m))`$ such that the curves $`\{F_0+tF_1=0\}`$ and $`\{G_0+tG_1=0\}`$ meet at no less than $`s`$ points over the ring $`[t]/(t^2)`$. It is observed by Xu that \[X, Lemma 1\]
(3.1)
$$\frac{F_0}{Z_i}G_1\frac{G_0}{Z_i}F_1H^0(D,𝒪_D(d+m1)𝒪_D(\underset{j=1}{\overset{s}{}}(\mu _j1)p_j))$$
for $`i=0,1,2`$, where $`D`$ and $`E`$ meet at $`p_1,p_2,\mathrm{},p_s`$ with multiplicities $`\mu _1,\mu _2,\mathrm{},\mu _s`$, respectively.
The relation (3.1) forms the basis of Xu’s proof of Theorem 1.1. If we were to prove Theorem 1.3 following Xu’s line of argument, we would have to come up with a relation similar to (3.1) on $`𝔽_n`$, which we are unable to do. So we find that Xu’s analysis, though ingenious on its own, is hard, if not impossible, to carry out on surfaces other than $`^2`$. Therefore, we will adopt a different approach to Theorem 1.3, which is based upon degeneration and induction.
Let $`\mathrm{\Delta }`$ be a disk parameterized by $`t`$ and let $`Y𝔽_n\times \mathrm{\Delta }`$ be a pencil of curves in $`|aC+bF|`$ whose central fiber $`Y_0=G\mathrm{\Gamma }`$ is reducible with two components $`G|(a1)C+(bn1)F|`$ and $`\mathrm{\Gamma }^1|C+(n+1)F|`$. Let $`X𝔽_n\times \mathrm{\Delta }`$ be a family of curves on $`𝔽_n`$ whose general fiber $`X_t`$ ($`t0`$) meets $`Y_t`$ at $`s`$ distinct points (a base change may be needed to ensure the existence of $`X`$). If $`X_0`$ meets $`Y_0`$ properly, we may deduce $`s\mathrm{min}(a1,ban,bn1)`$ by the induction hypothesis that
$$\begin{array}{cc}\hfill \mathrm{\#}(X_0G)& \mathrm{min}(a2,ban1,bn2)\hfill \\ & =\mathrm{min}(a1,ban,bn1)1\hfill \end{array}$$
and by the fact that $`\mathrm{\#}(X_0\mathrm{\Gamma })1`$, where we use the notation $`\mathrm{\#}(AB)`$ to denote the number of points in the set-theoretical intersection $`AB`$ between the two curves $`A`$ and $`B`$. Of course, some care has to be taken to make sure that $`X_0`$ meets $`\mathrm{\Gamma }`$ at at least one point outside of $`G\mathrm{\Gamma }`$ (see below). Unfortunately, $`X_0`$ may very well contain $`G`$ or $`\mathrm{\Gamma }`$ as a component. So we have to regard $`|𝒪_{Y_0}(X_0)|`$ as the limit linear series $`lim_{t0}|𝒪_{Y_t}(X_t)|`$ and, correspondingly, $`Y_0X_0`$ as the limit of the section $`Y_tX_t`$ in $`|𝒪_{Y_t}(X_t)|`$. For an introduction to the theory of limit linear series, please see, for example, \[E-H\] or \[H, Chap. 5\].
For the purpose of induction, we will prove Theorem 1.3 in the following slightly more general form.
###### Proposition 3.1.
Let $`L=𝒪(aC+bF)`$ be an ample line bundle on $`𝔽_n`$ with $`a2`$ and $`b2`$. Then for a sufficiently general curve $`D|L|`$,
1. $`\mathrm{\#}(DE)\mathrm{min}(a1,ban,bn1)`$ for any curve $`E𝔽_n`$ that meets $`D`$ properly;
2. in addition, there exists a set $`\mathrm{\Sigma }_D`$ consisting of countably many points on $`D`$ such that if $`\mathrm{\#}(DE)=\mathrm{min}(a1,ban,bn1)`$ for some $`E`$, $`(DE)\mathrm{\Sigma }`$.
We prove Proposition 3.1 by induction on $`\mathrm{min}(a1,ban,bn1)`$.
If $`\mathrm{min}(a1,ban,bn1)=1`$, we only need to verify the second part of the proposition. Notice that $`D`$ has genus $`g(D)=1+\frac{1}{2}(a2)(ban)+\frac{1}{2}a(bn2)1`$. If $`D`$ meets $`E`$ at a single $`p`$ for some $`E`$, $`𝒪_D(\mu p)=𝒪_D(E)`$, where $`\mu =DE`$. If we fix the divisor class of $`E`$, there are only finitely many points $`p`$ with this property since $`g(D)1`$. Therefore, there are only countably many points $`p`$ such that $`DE=\{p\}`$ for some $`E`$.
Suppose that $`\mathrm{min}(a1,ban,bn1)2`$. Notice that $`𝒪((a1)C+(bn1)F)`$ is ample under this assumption.
Let $`X,Y,G`$ and $`\mathrm{\Gamma }`$ be defined as before. Suppose that $`G`$ and $`\mathrm{\Gamma }`$ meet at points $`p_1,p_2,\mathrm{},p_l`$, where $`l=a+bn2`$. Let $`M=𝒪(X_t)`$ be the line bundle associated to $`X_t`$.
Let $`\sigma _t|𝒪_{Y_t}(X_t)|`$ be the section cut out by $`X_t`$ on $`Y_t`$ and let $`\sigma _0=lim_{t0}\sigma _t`$. Let $`\sigma _\mathrm{\Gamma }=\sigma _0|_\mathrm{\Gamma }`$ and $`\sigma _G=\sigma _0|_G`$ be the restrictions of $`\sigma _0`$ to $`\mathrm{\Gamma }`$ and $`G`$, respectively. Then $`\sigma _\mathrm{\Gamma }`$ is a section in
$$|𝒪_\mathrm{\Gamma }(\mu (p_1+p_2+\mathrm{}+p_l))M|=|𝒪_\mathrm{\Gamma }(\mu G)M|$$
and $`\sigma _G`$ is a section in
$$|𝒪_G(\mu (p_1+p_2+\mathrm{}+p_l))M|=|𝒪_G(\mu \mathrm{\Gamma })M|$$
where $`\mu `$ is an integer and $`\sigma _\mathrm{\Gamma }`$ and $`\sigma _G`$ are cut out by sections in $`|𝒪(\mu G)M|`$ and $`|𝒪(\mu \mathrm{\Gamma })M|`$, respectively.
Suppose that $`𝒪(\mu \mathrm{\Gamma })M`$ is nontrivial. Then by induction hypothesis $`\sigma _G`$ vanishes at no less than $`\mathrm{min}(a2,ban1,bn2)`$ distinct points. If $`\sigma _\mathrm{\Gamma }`$ vanishes at at least one point other than $`p_1,p_2,\mathrm{},p_l`$, we are done; if not, we have either $`𝒪(\mu G)M`$ is trivial and $`\sigma _\mathrm{\Gamma }`$ is nowhere vanishing or $`\sigma _\mathrm{\Gamma }`$ only vanishes at $`p_1,p_2,\mathrm{},p_l`$.
If $`𝒪(\mu G)M`$ is trivial and $`\sigma _\mathrm{\Gamma }`$ is nowhere vanishing, then for any two points among $`p_1,p_2,\mathrm{},p_l`$, say $`p_1`$ and $`p_2`$, the ratio $`\sigma _G(p_1)/\sigma _G(p_2)`$ is uniquely determined by the choice of the pencil $`Y`$. Actually we have the following very explicit relation
(3.2)
$$\frac{\sigma _G(p_1)}{\sigma _G(p_2)}=\left(\frac{f(p_1)}{f(p_2)}\right)^\mu $$
where $`f|L|`$ is the section which cuts out a general member $`Y_t`$ of the pencil $`Y`$. If $`\sigma _G`$ vanishes at more than $`\mathrm{min}(a2,ban1,bn2)`$ distinct points, there is nothing to prove; otherwise, $`\sigma _G`$ vanishes at exactly $`\mathrm{min}(a2,ban1,bn2)`$ distinct points. Then by induction hypothesis, there are only countably many possible choices of $`\sigma _G`$. However, by (3.2), the ratio $`\sigma _G(p_1)/\sigma _G(p_2)`$ can be made into an arbitrary complex value by a choice of $`f`$ (and thus a choice of the pencil $`Y`$). Contradiction.
If $`\sigma _\mathrm{\Gamma }`$ only vanishes at $`p_1,p_2,\mathrm{},p_l`$, since we have already taken care of the case that $`𝒪(\mu G)M`$ is trivial and $`\sigma _\mathrm{\Gamma }`$ is nowhere vanishing, we may assume that $`\sigma _\mathrm{\Gamma }`$ vanishes at at least one point among $`p_1,p_2,\mathrm{},p_l`$, say $`p_1`$. Then $`\sigma _G`$ must vanish at $`p_1`$ as well. Again, if $`\sigma _G`$ vanishes at more than $`\mathrm{min}(a2,ban1,bn2)`$ distinct points, there is nothing to prove; otherwise, $`\sigma _G`$ vanishes at exactly $`\mathrm{min}(a2,ban1,bn2)`$ distinct points. By induction hypothesis, $`p_1\mathrm{\Sigma }_G`$. But if we choose $`\mathrm{\Gamma }`$ generically, $`p_1\mathrm{\Sigma }_G`$. Contradiction.
Now suppose that $`𝒪(\mu \mathrm{\Gamma })M`$ is trivial. If $`\sigma _G=0`$, then $`\sigma _\mathrm{\Gamma }`$ vanishes at $`p_1,p_2,\mathrm{},p_l`$ and $`l=a+bn2>\mathrm{min}(a1,ban,bn1)`$; we are done. Otherwise, $`\sigma _G`$ is no where vanishing. The ratio $`\sigma _\mathrm{\Gamma }(p_i)/\sigma _\mathrm{\Gamma }(p_j)`$ for any two points $`p_i`$ and $`p_j`$ among $`p_1,p_2,\mathrm{},p_l`$, just as in (3.2), is uniquely determined by the choice of $`Y`$ and is given by
(3.3)
$$\frac{\sigma _\mathrm{\Gamma }(p_i)}{\sigma _\mathrm{\Gamma }(p_j)}=\left(\frac{f(p_i)}{f(p_j)}\right)^\mu .$$
The rational map $`|L|^{l1}`$ by sending $`f|L|`$ to
(3.4)
$$(f^\mu (p_1),f^\mu (p_2),\mathrm{},f^\mu (p_l))$$
is dominant due to the facts that $`H^0(𝔽_n,L)`$ surjects onto $`H^0(\mathrm{\Gamma },L)`$ and $`L𝒪_\mathrm{\Gamma }(_{ij}p_i)`$ is base point free on $`\mathrm{\Gamma }`$ for each $`1jl`$. On the other hand, the space
$$\{\sigma _\mathrm{\Gamma }|\sigma _\mathrm{\Gamma }\text{ vanishes at less than }l1\text{ distinct points}\}$$
has dimension $`l2`$ and hence cannot dominate $`^{l1}`$. So $`\sigma _\mathrm{\Gamma }`$ vanishes at at least $`l1>\mathrm{min}(a1,ban,bn1)`$ distinct points for a general choice of $`f`$ by (3.3).
This finishes the proof of the first part of the proposition.
Suppose that $`\sigma _0`$ vanishes at exactly $`\mathrm{min}(a1,ban,bn1)`$ distinct points. This can happen only when $`𝒪(\mu \mathrm{\Gamma })M`$ is nontrivial.
Suppose that $`\sigma _G`$ vanishes at exactly $`\mathrm{min}(a2,ban1,bn2)`$ distinct points. Our previous argument shows that $`\sigma _G`$ does not vanish at $`p_1,p_2,\mathrm{},p_l`$ for a general choice of $`G\mathrm{\Gamma }`$. Then $`\sigma _\mathrm{\Gamma }`$ must vanish at a single point $`p\{p_1,p_2,\mathrm{},p_l\}`$. Since $`\mathrm{\#}(G\mathrm{\Gamma })2`$, the natural map from $`Y_0\backslash \{p_1,p_2,\mathrm{},p_l\}`$ to $`Pic(Y_0)`$ is injective. So $`p`$ is determined up to finitely many possibilities by $`M`$ and the vanishing locus of $`\sigma _G`$. By induction, the vanishing locus of $`\sigma _G`$ is contained in some countable set $`\mathrm{\Sigma }_G`$ depending only on $`G`$. So the vanishing locus of $`\sigma _0`$ is also contained in some countable set $`\mathrm{\Sigma }_{G\mathrm{\Gamma }}`$ depending only on $`G\mathrm{\Gamma }`$.
Suppose that $`\sigma _G`$ vanishes at exactly $`\mathrm{min}(a1,ban,bn1)`$ distinct points and suppose that there is a one-parameter family of $`\sigma _0(u)`$ with this property, where $`\sigma _0(u)`$ is parameterized by $`uU`$ for some irreducible curve $`U`$.
Suppose that $`𝒪(\mu G)M`$ is trivial. There exists $`u_0U`$ such that $`\sigma _G(u_0)`$ vanishes at $`p_1`$. Since $`𝒪(\mu G)M`$ is trivial, $`\sigma _\mathrm{\Gamma }(u_0)=0`$ and hence $`\sigma _G(u_0)`$ vanishes at $`p_1,p_2,\mathrm{},p_l`$. But $`l>\mathrm{min}(a1,ban,bn1)`$. Contradiction.
Suppose that $`𝒪(\mu G)M`$ is nontrivial. Then $`\sigma _\mathrm{\Gamma }(u)`$ vanishes at at least one point among $`p_1,p_2,\mathrm{},p_l`$, say $`p_1`$. Hence $`\sigma _G(u)`$ vanishes at $`p_1`$ for all $`uU`$. As $`u`$ varies, another vanishing point of $`\sigma _G(u)`$ will approach $`p_1`$. So there exists $`u_0U`$ such that $`\sigma _G(u_0)`$ vanishes at $`\mathrm{min}(a2,ban1,bn2)`$ distinct points and among them vanishes at a general point $`p_1`$. Again this is impossible by induction. Contradiction.
This finishes the proof of Proposition 3.1.
The degeneration method we used can be applied to surfaces other than rational ruled surfaces. For example, we can give an alternative proof of Xu’s Theorem 1.1 by degenerating a degree $`d`$ curve to a union of a degree $`d1`$ curve and a line and arguing by induction.
###### A proof of Xu’s Theorem 1.1 via degeneration.
As in the case of Proposition 3.1, we need to add a clause to the theorem for the purpose of induction, i.e., we will prove the following statement by induction on $`d`$.
For a sufficiently general curve $`D`$ of degree $`d3`$ in $`^2`$, $`\mathrm{\#}(DE)d2`$ for any curve $`E^2`$ that meets $`D`$ properly. In addition, there exists a set $`\mathrm{\Sigma }_D`$ of countably many points on $`D`$ such that if $`\mathrm{\#}(DE)=d2`$ for some $`E`$, $`(DE)\mathrm{\Sigma }_D`$.
Let $`Y^2\times \mathrm{\Delta }`$ be a pencil of degree $`d`$ curves whose central fiber $`Y_0=G\mathrm{\Gamma }`$ is the union of a curve $`G`$ of degree $`d1`$ and a line $`\mathrm{\Gamma }`$ and let $`G\mathrm{\Gamma }=\{p_1,p_2,\mathrm{},p_l\}`$ where $`l=d1`$.
Let $`X,M,\sigma _t,\sigma _0,\sigma _G,\sigma _\mathrm{\Gamma }`$ and $`\mu `$ be defined as before. Almost nothing in the argument of Proposition 3.1 needs changing except in the case that $`𝒪(\mu \mathrm{\Gamma })M`$ is trivial and $`\sigma _G`$ is nowhere vanishing. In this case, following our previous argument, we can show that $`\sigma _\mathrm{\Gamma }`$ vanishes at no less than $`l1`$ points. The difference is that now we have $`l1=d2`$ and we have to verify that there are only finitely many $`\sigma _\mathrm{\Gamma }`$ that vanishes at exactly $`l1`$ distinct points. This is more or less obvious because the map from $`|L|`$ to $`^{l1}`$ given by (3.4) is dominant and the space
$$\{\sigma _\mathrm{\Gamma }|\sigma _\mathrm{\Gamma }\text{ vanishes at exactly }l1\text{ distinct points}\}$$
has dimension $`l1`$. ∎
Our degeneration method also works for Del Pezzo surfaces.
###### Theorem 3.1.
Let $`\stackrel{~}{^2}`$ be the blowup of $`^2`$ at $`2r6`$ general points and let $`L_1,L_2,\mathrm{},L_k,\mathrm{}`$ be all the smooth rational curves on $`\stackrel{~}{^2}`$ with self-intersection $`1`$. Let $`L`$ be an ample line bundle on $`\stackrel{~}{^2}`$. Then for a sufficiently general curve $`D|L|`$,
1. $`\mathrm{\#}(DE)\mathrm{min}_k(DL_k)`$ for any curve $`E\stackrel{~}{^2}`$ that meets $`D`$ properly;
2. in addition, there exists a set $`\mathrm{\Sigma }_D`$ of countably many points on $`D`$ such that if $`\mathrm{\#}(DE)=\mathrm{min}_k(DL_k)`$ for some $`E`$, $`(DE)\mathrm{\Sigma }_D`$.
Therefore, for a sufficiently general curve $`D|L|`$ with $`\mathrm{min}_k(DL_k)3`$, the complement $`\stackrel{~}{^2}\backslash D`$ is algebraic hyperbolic.
###### Proof.
Let $`K_{\stackrel{~}{^2}}`$ be the canonical divisor of $`\stackrel{~}{^2}`$. We argue by induction on $`\mathrm{min}_k(DL_k)`$.
For $`\mathrm{min}_k(DL_k)=1`$, we need to verify that $`g(D)1`$, which is more or less obvious.
Suppose that $`\mathrm{min}_k(DL_k)2`$. Let $`Y^2\times \mathrm{\Delta }`$ be a pencil of curves in $`|L|`$ whose central fiber $`Y_0=G\mathrm{\Gamma }`$ is a union of $`G\left|L𝒪(K_{\stackrel{~}{^2}})\right|`$ and $`\mathrm{\Gamma }\left|K_{\stackrel{~}{^2}}\right|`$ and let $`G\mathrm{\Gamma }=\{p_1,p_2,\mathrm{},p_l\}`$.
Let $`X,M,\sigma _t,\sigma _0,\sigma _G,\sigma _\mathrm{\Gamma }`$ and $`\mu `$ be defined as before. Again, the same argument for Proposition 3.1 goes through. We need only to check the following facts, all of which are routine exercises.
1. $`l>\mathrm{min}_k(DL_k)`$.
2. $`H^0(\stackrel{~}{^2},L)`$ surjects onto $`H^0(\mathrm{\Gamma },L)`$ and
$$L𝒪_\mathrm{\Gamma }(\underset{ij}{}p_i)=𝒪_\mathrm{\Gamma }(K_{\stackrel{~}{^2}})𝒪_\mathrm{\Gamma }(p_j)$$
is base point free on $`\mathrm{\Gamma }`$ for each $`1jl`$. Hence the map from $`|L|`$ to $`^{l1}`$ given by (3.4) is dominant.
3. In the case that $`𝒪(\mu \mathrm{\Gamma })M`$ is trivial and $`\sigma _G`$ is nowhere vanishing, we can prove that $`\sigma _\mathrm{\Gamma }`$ vanishes at no less than $`l1`$ distinct points as before. But actually, we can do better here since the space
$$\{\sigma _\mathrm{\Gamma }|\sigma _\mathrm{\Gamma }\text{ vanishes at less than }l\text{ distinct points}\}$$
has dimension $`l2`$ due to the fact that $`\mathrm{\Gamma }`$ is elliptic instead of rational. Therefore, $`\sigma _\mathrm{\Gamma }`$ vanishes at no less than $`l`$ distinct points.
When we go up in dimension, however, some essential difficulties present themselves. For example, in $`^3`$, fix a sufficient general surface $`S`$ of degree $`d`$ and it is expected that any curve meets $`S`$ at no less than $`d4`$ distinct points \[X, Question 2\]. Let $`Y`$ be a pencil of degree $`d`$ surfaces whose central fiber is a union of a degree $`d1`$ surface and a plane and let $`X`$ be family of curves in $`^3`$ meeting $`Y`$ fiberwise. To carry out the argument as in dimension two, we need to take the limit $`X_tY_t`$ as an element in $`A_0(Y_t)`$, the 0-dimension Chow ring of $`Y_t`$. Of course, we do not know how to do this at present. |
warning/0003/hep-lat0003006.html | ar5iv | text | # Determinant of a new fermionic action on a lattice - (II)
## I introduction
As is well known the lattice formulation of fermions has extra physical particles or breaks the chiral symmetry. This is unavoidable under a few plausible assumptions . Several methods have been proposed to deal with this difficulty. Wilson’s formulation , which is one of the most popular schemes, eliminates the unwanted particles with an additional term which vanishes in the naive continuum limit. However, this formulation sacrifices the chiral symmetry. An alternative scheme is the staggered fermion formulation proposed by Kogut and Susskind . This scheme preserves the discrete chiral symmetry and in this point the staggered fermion has an advantage over the Wilson fermion. But, the staggered formulation describes a theory with $`2^{\frac{1+D}{2}}`$ degenerate quark flavours ($`2^{1+D}`$ components) in $`(1+D)`$ dimensions, while there is no restriction on the flavour number in the Wilson formulation. Recently, it has been shown that lattice fermionic actions with the Ginsparg-Wilson relation have an exact chiral symmetry and are free from restriction on the flavour number. But, these actions cannot be ”ultralocal” , which makes numerical simulations complicated.
In the recent papers , we proposed a new type of fermionic action on a $`(1+D)`$-dimensional lattice. The action is ultralocal and has discrete chiral symmetry. On the Euclidean lattice the minimal number of fermion components is $`2^D`$, which should be compared with $`2^{1+D}`$ of the staggered fermion. When dynamical fermions are included, the numerical feasibility relies on the reality and positivity of the fermion determinant. In the previous paper we investigated, analytically and numerically, the fermion determinant of our new action in the $`(1+1)`$-dimensional U(1) lattice gauge theory. We showed the reality of our fermion determinant under the condition fixing the global phase of link variables along the temporal direction. By a similar discussion to the U(1) gauge group, we could also find the reality and the positivity of our fermion determinant in the $`(1+1)`$-dimensional SU(N) lattice gauge theory.
In this paper we analytically show that our fermion determinant with the SU(2) gauge fields is real and positive in $`(1+D)`$ dimensions. We also comment on the numerical results of the fermion determinant in the $`(1+D)`$-dimensional SU(3) gauge fields, and discuss the effectiveness of our new action for SU(2) and SU(3) lattice gauge theories.
## II new fermionic action
In the recent paper , we proposed a new fermionic action on the Euclidean lattice. Though the action keeps the discrete chiral symmetry like the staggered fermion action, the fermion field has $`2^D`$ components in $`(1+D)`$ dimensions. In this section we briefly sketch our formalism for later convenience.
The action can be written with a fermion matrix $`\mathrm{\Lambda }`$ as
$`S_f={\displaystyle \underset{n,m}{}}\psi _n^{}\mathrm{\Lambda }_{nm}\psi _m,`$ (1)
where the summation is over lattice points and spinor indices, and our fermion matrix is defined by
$`\mathrm{\Lambda }=1S_0^{}U_E.`$ (2)
Here $`U_E`$ is the Euclidean time evolution operator and $`S_\mu `$ is the unit shift operator defined by
$`S_\mu \psi (x^0,x^1,\mathrm{},x^\mu ,\mathrm{},x^D)=\psi (x^0,x^1,\mathrm{},x^\mu +1,\mathrm{},x^D)(\mu =0,1,\mathrm{},D).`$ (3)
We require that the propagator has no extra poles and find that $`U_E`$ has the form
$`U_E=1{\displaystyle \underset{i=1}{\overset{D}{}}}{\displaystyle \frac{r_E}{2}}\left\{iX_i\left(S_iS_i^{}\right)+\left(1Y_i\right)\left(S_i2+S_i^{}\right)\right\},`$ (4)
where $`r_E`$ is the ratio of the temporal lattice constant to the spatial one. The spinor matrices $`X`$’s and $`Y`$’s should satisfy the following algebra:
$`\{\begin{array}{ccc}\{X_i,X_j\}& =& {\displaystyle \frac{2}{r_E}}\delta _{ij},\hfill \\ \{X_i,Y_j\}& =& 0,\hfill \\ \{Y_i,Y_j\}& =& 2\left({\displaystyle \frac{1}{r_E}}\delta _{ij}+1\right),\hfill \end{array}`$ (8)
where $`i`$ and $`j`$ run from $`1`$ to $`D`$. The matrix $`2(\delta _{ij}/r_E+1)`$ is positive definite for any positive $`r_E`$, therefore $`X`$’s and $`Y`$’s can be assumed hermitian,
$`X_i^{}=X_i,Y_i^{}=Y_i.`$ (9)
The matrices $`X`$’s and $`Y`$’s can be expressed by the Clifford algebra:
$`\mathrm{\Gamma }_n^{}=\mathrm{\Gamma }_n,\{\mathrm{\Gamma }_n,\mathrm{\Gamma }_m\}=2\delta _{nm}(n,m=1,\mathrm{},2D)`$ (10)
in several ways. One is
$`X_i=\sqrt{{\displaystyle \frac{1}{r_E}}}\mathrm{\Gamma }_i,Y_i={\displaystyle \underset{j=1}{\overset{D}{}}}\alpha _{ij}\mathrm{\Gamma }_{D+j},`$ (11)
as was used in the previous paper. Another one is
$`X_i=\sqrt{{\displaystyle \frac{1}{r_E}}}\mathrm{\Gamma }_i,Y_i=\sqrt{{\displaystyle \frac{1}{r_E}}}\mathrm{\Gamma }_{D+i}+\mathrm{\Gamma }_{2D+1},`$ (12)
where $`\mathrm{\Gamma }_{2D+1}`$ is
$`\mathrm{\Gamma }_{2D+1}=(i)^D\mathrm{\Gamma }_1\mathrm{\Gamma }_2\mathrm{}\mathrm{\Gamma }_{2D}.`$ (13)
The latter is more convenient than the former for later use. The dimension of the irreducible representation for $`\mathrm{\Gamma }`$’s is $`2^D`$ and accordingly $`\psi `$ has $`2^D`$ components.
The interaction of the fermion with gauge fields is introduced by replacing the unit shift operators by covariant ones:
$`S_\mu S_\mu (x)U_{x,x+\widehat{\mu }}S_\mu ,`$ (14)
where $`\widehat{\mu }`$ is the unit vector along the $`\mu `$’th direction, and $`U_{x,y}`$ is a link variable connecting sites $`x`$ and $`y`$.
The fermion matrix Eq.(2) and the time evolution operator Eq.(4) become
$`\mathrm{\Lambda }(x)=1S_0^{}(x)U_E(x),`$ (15)
and
$`U_E(x)=1{\displaystyle \underset{i=1}{\overset{D}{}}}{\displaystyle \frac{r_E}{2}}\left\{iX_i\left(S_i(x)S_i^{}(x)\right)+\left(1Y_i\right)\left(S_i(x)2+S_i^{}(x)\right)\right\}.`$ (16)
## III fermion determinant for SU(2) case
In this section we analytically study the determinant of our fermion matrix in SU(2) gauge fields.
First, in the $`(1+1)`$-dimensional case, the complex conjugation of $`U_E(x)`$ is
$`U_E^{}(x)=1{\displaystyle \frac{r_E}{2}}\left\{iX_1^{}\left(S_1^{}(x)+S_1^{^{}}(x)\right)+\left(1Y_1^{}\right)\left(S_1^{}(x)2+S_1^{^{}}(x)\right)\right\},`$ (17)
where we can write
$`S_1(x)=\alpha _0(x)\text{1}+i{\displaystyle \underset{i=1}{\overset{3}{}}}\alpha _i(x)\tau _i,`$ (18)
since the link variables in $`S_1(x)`$ are SU(2) gauge group elements. Here $`\alpha _0(x)`$ and $`\alpha _i(x)`$ are real and depend on lattice points and $`\tau _{1,2,3}`$ are the Pauli-matrices:
$`\tau _1=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\tau _2=\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right),\tau _3=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right).`$ (25)
Then we have
$`S_1(x)\tau _2=\tau _2S_1^{}(x).`$ (26)
By the same discussion for other unit shift operators $`S_0^{}`$ and $`S_1^{}`$, we also have
$`S_0^{}(x)\tau _2=\tau _2S_{0}^{}{}_{}{}^{}(x),S_1^{}(x)\tau _2=\tau _2S_{1}^{}{}_{}{}^{}(x).`$ (27)
If we can find the matrix $`\mathrm{\Gamma }`$ such that
$`\{\begin{array}{ccc}\hfill X_1^{}\mathrm{\Gamma }& =& \mathrm{\Gamma }X_1,\hfill \\ \hfill Y_1^{}\mathrm{\Gamma }& =& \mathrm{\Gamma }Y_1,\hfill \end{array}`$ (30)
it is easily shown that
$`\left(\mathrm{\Gamma }\tau _2\right)\left(S_0^{}(x)U_E(x)\right)^{}\left(\mathrm{\Gamma }\tau _2\right)=S_0^{}(x)U_E(x).`$ (31)
For example, we make the following choice:
$`X_1=\sqrt{{\displaystyle \frac{1}{r_E}}}\tau _1,Y_1=\sqrt{{\displaystyle \frac{1}{r_E}}}\tau _2+\tau _3,`$ (32)
the matrix $`\mathrm{\Gamma }`$ acting on two components fermi fields defined by
$`\mathrm{\Gamma }=\tau _3`$ (33)
satisfies Eq.(30). The Eq.(31) implies that if $`\lambda `$ is some eigenvalue of our fermion matrix $`\mathrm{\Lambda }(x)`$, then $`\lambda ^{}`$ is an eigenvalue of $`\mathrm{\Lambda }^{}(x)`$ and thus also of $`\mathrm{\Lambda }(x)`$. Therefore eigenvalues of $`\mathrm{\Lambda }(x)`$ are either real or come in complex conjugate pairs. From the above discussion we can prove the reality of our fermion determinant for the SU(2) gauge groups.
Next we show its positivity. We define
$`\mathrm{\Gamma }^{}=\left(\mathrm{\Gamma }\tau _2\right)K,`$ (34)
where $`K`$ is complex-conjugation operator. We find
$`\mathrm{\Gamma }^{}\mathrm{\Gamma }^{}=\left(\tau _3\tau _2\right)K\left(\tau _3\tau _2\right)K=\left(\tau _3\tau _2\right)\left(\tau _3(\tau _2)\right)=1,`$ (35)
and from the relation Eq.(31) we can show
$`[\mathrm{\Gamma }^{},\mathrm{\Lambda }(x)]=0.`$ (36)
For a real eigenvalue $`\lambda _R`$ of $`\mathrm{\Lambda }(x)`$ and the eigenvector $`v_R`$ for this eigenvalue, from Eq.(36) we obtain
$`\mathrm{\Lambda }(x)\mathrm{\Gamma }^{}v_R=\mathrm{\Gamma }^{}\mathrm{\Lambda }(x)v_R=\mathrm{\Gamma }^{}\lambda _Rv_R=\lambda _R\mathrm{\Gamma }^{}v_R.`$ (37)
Suppose $`\mathrm{\Gamma }^{}v_R=cv_R`$, then we find
$`\mathrm{\Gamma }_{}^{}{}_{}{}^{2}v_R=\mathrm{\Gamma }^{}cv_R=c^{}\mathrm{\Gamma }^{}v_R=\left|c\right|^2v_R,`$ (38)
which is inconsistent with Eq.(35), so that $`\mathrm{\Gamma }^{}v_R`$ is different eigenvector for the same eigenvalue. Therefore the eigenvalues on real axis are degenerate in pairs and the determinant of $`\mathrm{\Lambda }(x)`$ is positive.
The above proof of the positivity of our fermion determinant for the SU(2) group can be expanded to higher dimensions. We can make a fundamental representation for the Clifford algebra with $`2D`$ elements $`\mathrm{\Gamma }_n`$ ($`n=1,\mathrm{},2D`$) using direct products of the Pauli-matrices:
$`\begin{array}{ccc}\hfill \mathrm{\Gamma }_1& =& \tau _1\underset{D1}{\underset{}{\tau _3\mathrm{}\tau _3}}\hfill \\ \hfill \mathrm{\Gamma }_2& =& \text{1}\tau _1\underset{D2}{\underset{}{\tau _3\mathrm{}\tau _3}}\hfill \\ \hfill \mathrm{}& & \\ \hfill \mathrm{\Gamma }_i& =& \underset{i1}{\underset{}{\text{1}\mathrm{}\text{1}}}\tau _1\underset{Di}{\underset{}{\tau _3\mathrm{}\tau _3}}\hfill \\ \hfill \mathrm{}& & \\ \hfill \mathrm{\Gamma }_D& =& \underset{D1}{\underset{}{\text{1}\mathrm{}\text{1}}}\tau _1\hfill \\ \hfill \mathrm{\Gamma }_{D+1}& =& \tau _2\underset{D1}{\underset{}{\tau _3\mathrm{}\tau _3}}\hfill \\ \hfill \mathrm{\Gamma }_{D+2}& =& \text{1}\tau _2\underset{D2}{\underset{}{\tau _3\mathrm{}\tau _3}}\hfill \\ \hfill \mathrm{}& & \\ \hfill \mathrm{\Gamma }_{D+i}& =& \underset{i1}{\underset{}{\text{1}\mathrm{}\text{1}}}\tau _2\underset{Di}{\underset{}{\tau _3\mathrm{}\tau _3}}\hfill \\ \hfill \mathrm{}& & \\ \hfill \mathrm{\Gamma }_{2D}& =& \underset{D1}{\underset{}{\text{1}\mathrm{}\text{1}}}\tau _2\hfill \end{array}`$ (51)
where $`i`$ runs from $`1`$ to $`D`$. It can be easily checked that $`\mathrm{\Gamma }_n`$’s satisfy the relation
$`\{\mathrm{\Gamma }_n,\mathrm{\Gamma }_m\}=2\delta _{nm}(n,m=1,\mathrm{},2D).`$ (52)
Moreover, we can see that the matrix $`\mathrm{\Gamma }_i`$ is real and the matrix $`\mathrm{\Gamma }_{D+i}`$ is pure imaginary:
$`\mathrm{\Gamma }_i^{}=\mathrm{\Gamma }_i,\mathrm{\Gamma }_{D+i}^{}=\mathrm{\Gamma }_{D+i}(i=1,\mathrm{},D).`$ (53)
From the anti-commutation relation, we find the hermite matrix $`\mathrm{\Gamma }_{2D+1}`$ which anti-commutes with all $`\mathrm{\Gamma }_n`$’s:
$`\mathrm{\Gamma }_{2D+1}`$ $`=`$ $`(i)^D\mathrm{\Gamma }_1\mathrm{\Gamma }_2\mathrm{}\mathrm{\Gamma }_{2D}`$ (54)
$`=`$ $`\tau _3\tau _3\mathrm{}\tau _3.`$ (55)
Clearly, the matrix $`\mathrm{\Gamma }_{2D+1}`$ is hermitian and the square of this matrix is equal to the unit matrix. Thus, we have
$`\begin{array}{ccc}\hfill \mathrm{\Gamma }_{2D+1}\mathrm{\Gamma }_i^{}\mathrm{\Gamma }_{2D+1}& =& \mathrm{\Gamma }_i,\hfill \\ \hfill \mathrm{\Gamma }_{2D+1}\mathrm{\Gamma }_{D+i}^{}\mathrm{\Gamma }_{2D+1}& =& \mathrm{\Gamma }_{D+i}.\hfill \end{array}`$ (58)
In $`(1+D)`$ dimensions, the relation Eq.(30) is rewritten as follows:
$`\{\begin{array}{ccc}\hfill X_i^{}\mathrm{\Gamma }& =& \mathrm{\Gamma }X_i,\hfill \\ \hfill Y_i^{}\mathrm{\Gamma }& =& \mathrm{\Gamma }Y_i.\hfill \end{array}`$ (61)
Then, for the representation of Eq.(12),
$`X_i=\sqrt{{\displaystyle \frac{1}{r_E}}}\mathrm{\Gamma }_i,Y_i=\sqrt{{\displaystyle \frac{1}{r_E}}}\mathrm{\Gamma }_{D+i}+\mathrm{\Gamma }_{2D+1},`$ (62)
we find
$`\mathrm{\Gamma }=\mathrm{\Gamma }_{2D+1}.`$ (63)
Since the eigenvalues of $`\mathrm{\Lambda }(x)`$ always consist of complex conjugate pairs and degenerated ones on real axis, we conclude the determinant of our fermion matrix is positive in SU(2) gauge fields in any dimensions.
Now we show a numerical evidence. Fig.1 shows the spectrum of our fermion matrix in a typical background configuration of link variables for SU(2) gauge group in $`(1+2)`$ dimensions. We find that their distribution is symmetric with respect to the real axis as expected. Similarly in $`(1+3)`$ dimensions we can numerically confirm the symmetry with respect to the real axis, and the positivity of the determinant.
## IV discussion and summary
In the previous paper we reported analytical and numerical results on the fermion determinant of our new action in $`(1+1)`$ dimensions. In the case of U(1) gauge group, we were faced with the problem of convergence in numerical simulations. The cause of the poorness of the convergence is that the summation of the $`det(1S_0^{}(x)U_E(x))=det(1e^{i\mathrm{\Theta }}\stackrel{~}{S}_0^{}(x)U_E(x))`$ over arbitrary phase angle $`\mathrm{\Theta }`$ is canceled out accidentally. The element $`e^{i\mathrm{\Theta }}`$ comes from the one sided time difference operator $`S_0^{}(x)`$ with $`\theta _0(x)`$ replaced by $`\theta _0(x)+\mathrm{\Theta }`$, i.e. $`S_0^{}(x)=e^{i\mathrm{\Theta }}\stackrel{~}{S}_0^{}(x)`$, where $`\theta _0(x)`$ is defined by $`U_{x,x+\widehat{0}}=e^{i\theta _0(x)}`$ . Therefore we must have control of the phase angle $`\mathrm{\Theta }`$ in order to get good convergence in the $`(1+1)`$-dimensional U(1) gauge theory. In fact we analytically showed that our fermion determinant is real for all configurations and positive for most configurations under the T-condition ($`\theta _0(x)=0`$), which corresponds to the temporal gauge condition on the infinite lattice, or the GT-condition ($`_x\theta _0(x)=n\pi `$ : $`n=\text{even}`$), which is achieved by a gauge transformation on the infinite lattice. It was also verified numerically. On the other hand we got good convergence without any conditions in $`(1+1)`$-dimensional SU(N) case, because the element like $`e^{i\mathrm{\Theta }}`$ does not belong to the SU(N) group.
The above discussion is applicable to higher dimensions to a certain extent. In SU(2) group our fermion determinant is analytically shown real and positive in any dimensions. In Fig.2(a) we give the numerical evidence that the determinant is real and positive in $`(1+2)`$ dimensions. In the case of SU(3) group, we cannot prove the reality of the determinant. But from Fig.2(b) we see that the distribution of the determinant is concentrated near the real axis without any conditions and the phase angle of the determinant is small. We have obtained similar results in $`(1+3)`$ dimensions. When the phase angle of the fermion determinant is small enough, we can neglect the phase factor and make use of $`|det\mathrm{\Lambda }(x)|`$ instead of $`det\mathrm{\Lambda }(x)`$. In the above numerical simulations link variables are updated by the Metropolis method and determinants are calculated by the LU decomposition. So there are no systematic errors in the determinants.
In conclusion, we believe that our new fermionic action is a profitable formulation for the numerical simulations of SU(2) and SU(3) lattice gauge theory. |
warning/0003/math0003015.html | ar5iv | text | # Semiinfinite cohomology of Tate Lie algebras
## 1. Introduction.
This note is a natural extension of the final part in \[Ar\] where a natural homological construction for graded Lie algebra semiinfinite cohomology and a natural explanation for the phenomenon of the critical cocycle was found. Here we propose a variant of the construction that works in the Tate Lie algebra case. Note that the standard complex for the computation of the Tate Lie algebra semiinfinite cohomology was written down by Beilinson and Drinfeld in \[BD\] and probably by some physicists. Still the construction was rather an indirect one and was not formulated in terms similar to the ones in \[Ar\]. In this note we spell out the construction of the complex in terms of some kind of quadratic-linear Koszul duality rather than Conformal Field Theory.
Let us say a few words about the contents of the note. In the second section we recall the notion of a differential graded Lie algebra with a curvature. We show that the standard Chevalley complex for computation of the Lie algebra cohomology can be modified in such a way that it still exists in this exotic case.
In the third section we recall a construction of an analogue of the Chevalley cohomological complex for a left module over a Lie algebroid $`A`$ over a commutative algebra $`R`$. Then we combine the setup from the previous section with the described one and obtain a picture consisting of a graded supercommutative algebra with a derivation, a graded (super) Lie algebroid over the algebra carrying an extension of the derivation of the basic algebra and finaly an analogue of curvature that “corrects” the fact that both derivations do not satisfy the constraint $`d^2=0`$. We construct an analogue of the cohomological Chevalley complex with coefficients in a left CDG-module over the differential graded Lie algebroid with curvature (CDG Lie algebroid).
In the fourth section we show that the standard complex for semiinfinite cohomology in the Tate Lie algebra case is a particular example of the described situation. Namely for a Tate Lie algebra $`𝔤`$ with a compact Lie subalgebra $`𝔟`$ we consider the graded supercommutative algebra $`\mathrm{\Lambda }^{}((𝔤/𝔟)^{})`$ and a graded Lie algebroid $`𝔟\mathrm{\Lambda }^{}((𝔤/𝔟)^{})`$ over it. We show that the components of the Lie bracket in $`𝔤`$ provide the derivations and the curvature.
Finally we show that for a discrete module $`M^{\mathrm{}}`$ over the extension of $`𝔤`$ with the help of the cricical cocycle of $`𝔤`$ the space $`M^{\mathrm{}}\mathrm{\Lambda }^{}(𝔤/𝔟)`$ carries a structure of a left CDG module over the above CDG Lie algebroid and that for this CDG-module the standard Chevalley complex from the third section coincides with the semiinfinite complex of $`𝔤`$ with coefficients in $`M^{\mathrm{}}`$.
In order to simplify the exposition we never use the language of derived categories in the note but rather work with concrete complexes.
## 2. Toy example.
The material of this section is based on a partly unpublished construction of A.Polishchuk and L.Positselsky (still see \[P\]).
We begin with the case of a differential graded Lie superalgebra with a curvature. The standard complex appearing in the case seems to be more understandable.
Let $`A=A_k`$ be a graded Lie algebra and $`d:A_kA_{k+1}`$ be a derivation of $`A`$ of order $`1`$. Note that at this point we do not put the constraint $`d^2=0`$. Instead we require an additional part of the data — an element $`hA_2`$ such that $`d^2(a)=[h,a]`$ for any $`aA`$ and $`d(h)=0`$.
### 2.1.
Definition: The data $`(A,d,h)`$ described above are called the differential graded Lie superalgebra with curvature or, for short, the CDG Lie algebra.
By definition a left (resp. a right) CDG-module over a CDG Lie algebra $`A`$ is a graded left (resp. right) module $`M=M_k`$ over the Lie algebra $`A`$ with the differential $`d:M_kM_{k+1}`$ satisfying the Leibnitz rule such that $`d^2(m)=hm`$ (resp. $`d^2(m)=mh`$) for any $`mM`$.
Denote the category of left (resp. right) CDG $`A`$-modules by $`CDG\mathrm{-}A\mathrm{-}\mathrm{mod}`$ (resp. $`CDG\mathrm{-}\mathrm{mod}\mathrm{-}A`$.
### 2.2. Construction of the standard complex.
For $`M^{}CDG\mathrm{-}A\mathrm{-}\mathrm{mod}`$ consider the bigraded vector space $`C^{}(A,M^{})`$ as follows: $`C^{}(A,M^{})=\mathrm{Hom}(\mathrm{\Lambda }^{}(A),M^{})`$, here the first grading comes from the number of wedges in the exterior product and
$$C^k(A,M^{})=\underset{p+q=k}{}\mathrm{Hom}((\mathrm{\Lambda }^{}(A))_p,M^q).$$
Consider the two differentials on the bigraded vector space. The first one of the grading $`(1,0)`$ is the usual Chevalley differential:
$$(d_1f)(a_1\mathrm{}a_m)=\underset{i}{}(1)^if(a_1\mathrm{}da_i\mathrm{}a_m)+d_M^{}f(a_1\mathrm{}a_m)$$
$$+\underset{i}{}(1)^ia_i^M^{}f(a_1\mathrm{}\widehat{a}_i\mathrm{}a_m)+\underset{i<j}{}(1)^{i+j}f([a_i,a_j]a_1\mathrm{}\widehat{a}_i\mathrm{}\widehat{a}_j\mathrm{}a_m).$$
The second differential of the grading $`(1,2)`$ is given by the formula that uses the curvature $`h`$:
$$(d_2f)(a_1\mathrm{}a_{m1})=f(ha_1\mathrm{}a_{m1}).$$
Consider the total grading on the bigraded space.
#### 2.2.1.
Lemma: The differential $`d=d_1+d_2`$ satisfies $`d^2=0`$.
Proof. The corresponding calculation repeats the one of Polishchuk and Positselsky. ∎
## 3. Standard complex for a CDG Lie algebroid.
### 3.1. Chevalley complex of a Lie algebroid.
Let $`R`$ be a commutative algebra, and let $`A`$ be a Lie algebroid over $`R`$, i.e. $`A`$ is a $`R`$-module carrying a Lie algebra structure over the base field and acting on $`R`$ by derivations such that $`[a,rb]=a(r)b+r[a,b]`$ for any $`a,bA`$ and $`rR`$.
By definition a $`A`$-module is a $`R`$-module $`M`$ with the Lie action of $`A`$ satisfying the constraint $`a(rm)=r(am)+(a(r))m`$ for any $`aA`$, $`rR`$ and $`mM`$.
#### 3.1.1.
For a left $`A`$-module $`M`$ consider the graded vector space $`C^{}(A,M)`$ as follows:
$$C^{}(A,M)=\underset{𝑘}{}C^k(A,M),C^k(A,M)=\mathrm{Hom}_R(\mathrm{\Lambda }_R^k(A),M)$$
We endow the graded vector space with the differential as follows:
$$(df)(a_1\mathrm{}a_m)=+\underset{i}{}(1)^ia_i^Mf(a_1\mathrm{}\widehat{a}_i\mathrm{}a_m)$$
$$+\underset{i<j}{}(1)^{i+j}f([a_i,a_j]a_1\mathrm{}\widehat{a}_i\mathrm{}\widehat{a}_j\mathrm{}a_m).$$
Lemma:
* The differential in the complex is well defined.
* The differential satisfies $`d^2=0`$.
Proof. (i) Let us perform a calculation showing that the differential $`d:C^kC^{k+1}`$ is well defined for $`k=1`$, the general case is quite similar. Note that
$$df(ra_1a_2)=ra_1f(a_2)a_2f(ra_1)f([ra_1,a_2])=ra_1f(a_2)a_2rf(a_1)+rf([a_1,a_2])+a_2(r)f(a_1)$$
$$=ra_1f(a_2)ra_2f(a_1)+rf([a_1,a_2])=df(a_1ra_2).$$
(ii) The calculation does not differ from the usual Lie algebra case. ∎
#### 3.1.2.
Remark: Note that there is no such standard complex for a right $`A`$-module $`M`$.
The constructed complex is called the (cohomological) Chevalley complex of the Lie algebroid $`A`$ with coefficients in the left $`A`$-module $`M`$.
### 3.2. CDG-algebroids.
Now let $`R=_kR_k`$ be a graded supercommutative algebra and let $`A=A_k`$ be a graded super Lie algebroid over $`R`$. Suppose also there are a super derivation $`d_R`$ on $`R`$ of degree $`1`$ and a super derivation $`d_A`$ of the Lie superalgebra $`A`$ also of degree $`1`$ satisfying Leibnitz rule with respect one to another. Again we put the constraint $`d^2=0`$ on neither of the derivations. Instead of that we fix the choice of an element $`hA_2`$ such that $`d_A^2(a)=[h,a]`$ and $`d_R^2(r)=h(r)`$.
#### 3.2.1.
Definition: The data $`(A,R,d_A,d_R,h)`$ are called the differential graded Lie algebroid with curvature or, for short, a CDG Lie algebroid.
The notion of a left (resp. right) CDG-module over a CDG algebroid is a natural combination of the previous definitions and we do not spell it out explicitly. The category of left (resp. right) CDG-modules ovea a CDG-algebroid $`A`$ is denoted by $`CDG\mathrm{-}A\mathrm{-}\mathrm{mod}`$ (resp. $`CDG\mathrm{-}\mathrm{mod}\mathrm{-}A`$.
### 3.3. Standard complex for a CDG Lie algebroid.
Now we sort of put togeather definitions of the standard complexes given in 2.2 and 3.1.1. For $`M^{}CDG\mathrm{-}A\mathrm{-}\mathrm{mod}`$ consider the bigraded vector space $`C^{}(A,M^{})`$ as follows: $`C^{}(A,M^{})=\mathrm{Hom}_R(\mathrm{\Lambda }_R^{}(A),M^{})`$, here the first grading comes from the number of wedges in the exterior product and
$$C^k(A,M^{})=\mathrm{Hom}_R^k((\mathrm{\Lambda }_R^{}(A)),M^{})$$
in the graded $`\mathrm{Hom}`$ sense. Consider the two differentials on the bigraded vector space. The first one of the grading $`(1,0)`$ is the usual Chevalley differential like in 3.1.1:
$$(d_1f)(a_1\mathrm{}a_m)=\underset{i}{}(1)^if(a_1\mathrm{}da_i\mathrm{}a_m)+d_M^{}f(a_1\mathrm{}a_m)$$
$$+\underset{i}{}(1)^ia_i^M^{}f(a_1\mathrm{}\widehat{a}_i\mathrm{}a_m)+\underset{i<j}{}(1)^{i+j}f([a_i,a_j]a_1\mathrm{}\widehat{a}_i\mathrm{}\widehat{a}_j\mathrm{}a_m).$$
The second differential of the grading $`(1,2)`$ is again given by the formula that uses the curvature $`h`$:
$$(d_2f)(a_1\mathrm{}a_{m1})=f(ha_1\mathrm{}a_{m1}).$$
Consider the total grading on the bigraded space.
#### 3.3.1.
Lemma:
* The differential $`d_1`$ is well defined and its square equals zero.
* The differential $`d_2`$ is well defined and its square equals zero.
* The differential $`d=d_1+d_2`$ satisfies $`d^2=0`$.
Proof. (i) Follows from Lemma 3.1.1. (ii) Follows from the obvious fact that $`hh=0\mathrm{\Lambda }_R^2(A)`$. (iii) Repeats the proof of Lemma 2.2.1. ∎
The obtained complex is called the (cohomological) Chevalley complex of the CDG Lie algebroid $`A`$ with coefficients in the left CDG $`A`$-module $`M`$.
## 4. Semiinfiite cohomology via CDG Lie algebroids.
In this section we show that the standard complex for the computation of semiinfinite cohomology of a discrete module over a Tate Lie algebra coincides with the Chevalley complex of the form 3.3.1 for a certain CDG Lie algebroid and a certain left module over it.
### 4.1. Tate Lie algebras.
Recall that a Tate space is a complete topological vector space having a base of neighbourhoods of $`0`$ consisting of commensurable vector spaces(i.e., $`dimU_1/(U_1U_2)<\mathrm{}`$ for any $`U_1`$ and $`U_2`$ from this base).
Recall also that a c-lattice in a Tate space $`V`$ is an open bounded subspace and a d-lattice $`LV`$ is a discrete subspace such that there exists a c-lattice $`P`$ with $`L+P=V`$. Note that the quotient of a Tate space by a c-lattice (resp. by a d-lattice) is discrete (resp. compact) in its natural topology. It is known that there is a natural duality on the category of Tate spaces and that $`V^{}\stackrel{~}{}V`$ for any Tate space V.
Recall also that by definition a Tate Lie algebra is a Tate vector space equipped with a Lie algebra structure continuous in the Tate topology.
### 4.2. Construction pf the CDG Lie algebroid.
Let $`𝔤`$ be a Tate Lie algebra with a subalgebra $`𝔟𝔤`$ that is a c-lattice In particular the space $`𝔠:=𝔤/𝔟`$ is discrete and its dual space is compact. Choose a section of the projection $`𝔤𝔠`$. Thus we fix a noncanonical decomposition $`𝔤=𝔟𝔠`$ and the Lie algebra structure on $`𝔤`$ is provided by the following collection of maps:
$$\mu _𝔟:𝔟𝔟𝔟,\mu _1:𝔟𝔠𝔠,\mu _𝔠:𝔠𝔠𝔠,\mu _2:𝔟𝔠𝔟,h^{}:𝔠𝔠𝔟.$$
The existence of $`h^{}`$ means exactly that $`𝔠`$ is not a Lie subalgebra in $`𝔤`$. The construction below is parallel to the one in \[Ar\].
#### 4.2.1.
First consider the graded supercommutative algebra $`\mathrm{\Lambda }^{}(𝔠^{})`$ or rather its Tate completion denoted in the same way. So as a vector spase it is Tate dual to the discrete coalgebra $`\mathrm{\Lambda }^{}(𝔠)`$. Consider the derivation $`d_R`$ on $`\mathrm{\Lambda }^{}(𝔠^{})`$ of degree one generated by the map dual to $`\mu _𝔠`$ and extended to the whole algebra by Leibnitz rule and by continuocity.
Next consider the graded Lie algebra $`𝔟\mathrm{\Lambda }^{}(𝔠^{})`$ where the tensor product is understood in the completed sense so that the whole space is Tate dual to the discrete space $`𝔟^{}\mathrm{\Lambda }^{}(𝔠)`$. The commutator map in the above Lie algebra is generated by the one on $`𝔟`$ and by the action of $`𝔟`$ on the space $`𝔠^{}=(𝔤/𝔟)^{}`$.
Denote the graded Lie algebra (resp. the graded supercommutative algebra) above by $`A`$ (resp. by $`R`$). Evidently $`A`$ is a $`R`$-module, moreover there is a natural adjoint action of $`A`$ on $`R`$. Lemma: $`A`$ is a graded Lie algebroid over the graded supercommutative algebra $`R`$. ∎
#### 4.2.2. Construction of the CDG Lie algebroid structure on $`(A,R)`$.
We extend the derivation of $`R`$ constructed above to the derivation of $`A`$. Namely consider the map $`𝔟𝔟𝔠^{}`$ dual to $`\mu _2`$. Denote the map by $`d_𝔟`$. Extend the sum of $`d_𝔟`$ and $`d_R`$ to the whole Lie algebra $`A`$ by Leibnitz rule and by continuocity. Denote the obtained derivation by $`d_A`$. Finally consider the element $`hA_2=𝔟\mathrm{\Lambda }^2(𝔠^{})`$ corresponding to the component $`h^{}`$ of the bracket in $`𝔤`$.
#### 4.2.3.
Proposition: The data $`(A,R,d_A,d_R,h)`$ form a CDG Lie algebroid. ∎
### 4.3. Construction of the standard semiinfinite complex.
First note that for any compact $`𝔤`$-module $`M`$ the (completed) tensor product $`M\mathrm{\Lambda }^{}(𝔠^{})`$ has a natural structure of a left CDG-module over the CDG algebroid $`(A,\mathrm{})`$ constructed in the previous subsection.
#### 4.3.1. Construction of the right CDG-module.
Now fix a discrete $`𝔤`$-module $`M`$ and consider the graded space $`M\mathrm{\Lambda }^{}(𝔠)`$. Lemma: The graded space $`M\mathrm{\Lambda }^{}(𝔠)`$ has a natural structure of the right CDG-module over the CDG Lie algebroid $`(A,\mathrm{})`$.∎
#### 4.3.2.
Here we come to the crucial point explaining the phenomenon of the critical cocycle in the semiinfinite cohomology of Tate Lie algebras. What we would like to do is to consider the standard complex of the CDG Lie algebroid $`(A,R,\mathrm{})`$ with coefficients in $`M\mathrm{\Lambda }^{}(𝔠)`$. Yet as noted in 3.1.2 there is no naive way to do it. Somehow we have to make $`M\mathrm{\Lambda }^{}(𝔠)`$ into a left CDG-module over our CDG Lie algebroid.
Happily the answer how to do this is known at least in the graded Lie algebra case. The idea works for Tate Lie algebra semiinfinite cohomology as well.
#### 4.3.3. Critical cocycle of $`𝔤`$.
It is known that the Lie algebra of continuous endomorphisms of $`𝔤`$ (and of any other Tate vector space as well) has a remarkable class in $`H^2`$ and the choice of the decomposition $`𝔤=𝔟𝔠`$ fixes its representative in the space of cocycles called the critical 2-cocycle and denoted by $`\omega _0`$. The adjoint action of $`𝔤`$ on itself provides the inverse image of the class denoted by $`\omega _0^𝔤`$ and called the critical 2-cocycle of $`𝔤`$.
Denote by $`𝔤^{\mathrm{}}`$ the central extension of $`𝔤`$ with the help of this cocycle. Note that $`𝔤^{\mathrm{}}`$ is a Tate Lie algebra with a fixed vector space decomposition $`𝔤=𝔟𝔠K`$, where $`K`$ denotes the central element.
#### 4.3.4.
Consider the CDG Lie algebroid $`(\stackrel{~}{A}^{\mathrm{}},R^{\mathrm{}},\mathrm{})`$ obtained from the pair $`𝔟K𝔤^{\mathrm{}}`$ by our main construction. Note that the algebra $`R^{\mathrm{}}\stackrel{~}{}R=\mathrm{\Lambda }^{}(𝔠^{})`$ and that the element $`K`$ is central in $`\stackrel{~}{A}^{\mathrm{}}`$. Denote by $`A^{\mathrm{}}`$ the quotient of $`\stackrel{~}{A}^{\mathrm{}}`$ by $`K1`$. Proposition: The CDG Lie algebroid $`(A^{\mathrm{}},R^{\mathrm{}},d_A^{\mathrm{}},d_R^{\mathrm{}},h^{\mathrm{}})`$ is isomorphic to the the CDG Lie algebroid $`(A,R,d_A,d_R,h)^{\mathrm{opp}}`$. ∎Remark: In particular for any discrete $`𝔤^{\mathrm{}}`$-module $`M^{\mathrm{}}`$ such that the central element $`K`$ acts on it by $`1`$ the right $`(A^{\mathrm{}},\mathrm{})`$ CDG-module $`M\mathrm{}\mathrm{\Lambda }^{}(𝔠)`$ becomes a left CDG-module over the CDG Lie algebroid $`(A,\mathrm{})`$.
#### 4.3.5.
Definition: For a discrete $`𝔤^{\mathrm{}}`$-module $`M^{\mathrm{}}`$ such that the central element $`K`$ acts on it by $`1`$ the complex
$$C^{}(A,M^{\mathrm{}}\mathrm{\Lambda }(𝔠))=\mathrm{Hom}_R(\mathrm{\Lambda }_R^{}(A),M^{\mathrm{}}\mathrm{\Lambda }^{}(𝔠))$$
is called the standard semiinfinite complex with coefficients in $`M^{\mathrm{}}`$ and is denoted by $`C^{\frac{\mathrm{}}{2}+}(𝔤,M^{\mathrm{}})`$.
#### 4.3.6.
Theorem:
* For a discrete $`𝔤^{\mathrm{}}`$-module $`M^{\mathrm{}}`$ as above the graded vector space $`C^{\frac{\mathrm{}}{2}+}(𝔤,M^{\mathrm{}})`$ is isomorphic to the graded vector space $`\mathrm{\Lambda }^{}(𝔟^{})\mathrm{\Lambda }^{}(𝔤/𝔟)M^{\mathrm{}}`$.
* The cohomology of the complex $`C^{\frac{\mathrm{}}{2}+}(𝔤,M^{\mathrm{}})`$ coincides with the semiinfinite cohomology of the Tate Lie algebra $`𝔤`$ with coefficients in the discrete module $`M^{\mathrm{}}`$ defined in \[BD\]. ∎ |
warning/0003/astro-ph0003334.html | ar5iv | text | # Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS 1footnote 11footnote 1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.
## 1 Introduction
Starburst galaxies constitute an important class of extragalactic objects. They contribute a significant fraction of the total high-mass star formation in the local universe (e.g. Soifer et al. 1987; Gallego et al. 1995), and about 25% of the high-mass star formation within 10 Mpc occurs in four starburst galaxies (Heckman 1998). At intermediate and high redshifts, important populations of galaxies are observed which exhibit properties indicating intense star formation activity (e.g. Colless et al. 1994; Babul & Ferguson 1996; Steidel et al. 1996; Lowenthal et al. 1997). A strong correlation between the most prodigious starbursts and interaction or merger events (e.g. Condon et al. 1982; Kennicutt et al. 1987; Telesco, Wolstencroft & Done 1988; Joseph 1990) further emphasizes the importance of starbursts in galaxy evolution.
Despite the obvious importance of starbursts, a canvas of recent starburst galaxy studies suggests that a full “prescription” for these events has yet to be written. Discrepancies exist even for the most massive stars formed, which represent the largest energy contributions. For instance, it has been suggested by some authors that the production of the highest mass stars may be suppressed in starburst galaxies, based on low measured values of diagnostic line ratios such as He I (2.06$`\mu `$m)/ HI(Br$`\gamma `$) and \[Ne II\]/\[Ar III\] detected in well-known starburst sources. These results have been interpreted as possibly indicating severe upper mass cutoffs to the initial mass function (IMF), some as low as $``$25-30$`M_{}`$ (e.g. Puxley et al. 1989; Doyon et al. 1994; Doherty et al. 1995; Achtermann & Lacy 1995; Beck, Kelly, & Lacy 1997).
A lack of massive stars in starbursts is difficult to reconcile with a growing body of evidence for the formation of very massive stars in nearby regions of active star-formation. In recent studies of local high-mass star-forming regions at both high and low metallicity, stars up to at least 100 $`M_{}`$ are observed; these regions include the Galactic Center (e.g. Krabbe et al. 1995; Najarro et al. 1997; Serabyn, Shupe, & Figer 1998; Figer et al. 1998), the Galactic star-forming region NGC 3603 (Drissen et al. 1995; Eisenhauer et al. 1998), and the R136 cluster at the center of 30 Doradus (e.g. Hunter et al. 1995, Massey & Hunter 1998). There is also a growing body of indirect evidence for the presence of very massive stars in starbursts. HST optical imaging has revealed the presence of young compact “super star clusters” in several nearby starburst galaxies, including M 82 (O’Connell et al. 1995), NGC 4038/4039 (Whitmore & Schweizer 1995), NGC 5253 (Meurer et al. 1995), He 2-10 (Conti & Vacca 1994), NGC 1569 and NGC 1705 (O’Connell, Gallagher, & Hunter 1994; Ho & Filippenko 1996; Sternberg 1998), and NGC 1140 (Hunter, O’Connell & Gallagher 1994). These super star clusters have extreme luminosities, and in some cases bright emission features characteristic of Wolf-Rayet stars (e.g., He II 4696Å emission); such properties are difficult to explain without a large contribution from very massive stars (M$``$60$`M_{}`$; see e.g., Gonzalez Delgado et al. 1997; Schaerer et al. 1997).
Several authors have recently argued that observed low-excitation nebular line ratios in starbursts may be due to aging effects rather than a severe upper mass cutoff (e.g. Rieke et al. 1993; Genzel, Hollenbach, & Townes 1994; Achtermann & Lacy 1995; Satyapal et al. 1997; Engelbracht et al. 1998; see also Vanzi & Rieke 1997). Studying a sufficiently large sample of starbursts is therefore valuable in breaking the degeneracy between age and upper mass cutoff. As we are not likely to catch all starbursts at a “late” age, and the ultraviolet, optical, and far-infrared luminosities decline with time after the peak star formation activity, we can determine the relative importance of variations in star formation parameters and aging effects by studying a larger sample.
In this work, we present results of a starburst modeling program to interpret mid-infrared (MIR) spectroscopy from the Short Wavelength Spectrometer (SWS; de Graauw et al. 1996) aboard the Infrared Space Observatory (ISO; Kessler et al. 1996). To this end, we have gathered atomic fine structure line fluxes (in particular for the \[Ne II\] 12.8$`\mu `$m and \[Ne III\] 15.6$`\mu `$m lines) for 27 starburst galaxies. This project is ideally suited for investigating the properties of the high-mass stellar population in heavily obscured star-forming regions of starburst galaxies, since the extinction at MIR wavelengths is only a few percent of the optical extinction. This investigation thus constitutes an important contribution to the understanding of star formation in starburst systems, a field that has been dominated by studies in the optical and ultraviolet regimes in the intervening years between the IRAS and ISO missions. As recent studies of the cosmic infrared background suggest a significant contribution from dusty starburst systems (e.g., Puget et al. 1996; Hughes et al. 1998; Hauser et al. 1998; Elbaz et al. 1998), the understanding of dusty starbursts is important in defining the picture of the most powerful star-forming events in the universe.
We present the sample data in §2. We outline our modeling procedure and compare our ISO-SWS observations with model predictions in §3. We examine additional constraints on this starburst modeling study and examine the robustness of the modeling analysis in §4. We discuss the implications of our findings and summarize this work in §5.
## 2 Neon Ratios in a Starburst Sample
### 2.1 Database selection
The database for this project was created by combining data from the MPE guaranteed-time program on bright galactic nuclei (15 objects) and an open-time observing proposal (12 objects), resulting in an ISO-SWS spectroscopic dataset for a total sample of 27 galaxies. The objects from the guaranteed-time program are all well-studied, infrared-bright starburst galaxies, and the open-time sample objects were selected on the basis of their 60$`\mu `$m flux densities (S<sub>60μm</sub> $`>`$20 Jy) and their observability by ISO in the last few months of the mission.
Within these programs, we have targeted observations at a specific set of spectral lines, particularly the \[Ne II\] 12.8$`\mu `$m and \[Ne III\] 15.6$`\mu `$m fine structure lines; in this work we will concentrate on the ratio of these two lines as a tracer of the massive star population, with additional lines introduced to constrain metallicities and gas densities. The neutral Ne atom and the $`\mathrm{Ne}^+`$ ion have ionization potentials of 21.56 eV and 40.95 eV respectively, making the \[Ne III\]/\[Ne II\] line ratio very sensitive to the spectral shape of the UV radiation field and thus to the properties of the most massive stars present. This information can be used to provide constraints on various star formation parameters, including the upper mass cutoff of the IMF and the age and duration of the starburst (see Kunze et al. 1996 and Rigopoulou et al. 1996 for earlier models using this and other MIR line ratios).
We have chosen the neon line ratio as the focus of this larger study, both for its sensitivity to massive, hot stars as well as the minimal uncertainties introduced in constructing the ratio of these two lines from ISO-SWS observations. The \[Ne II\] and \[Ne III\] lines span a wider range in ionization potential than other line ratios in this wavelength region, such as \[S IV\](10.5$`\mu `$m)/\[Ne II\](12.8$`\mu `$m), \[S IV\]/\[Ar III\](9.0$`\mu `$m), \[Ne II\]/\[Ar III\](e.g., Achtermann & Lacy 1995; Beck, Kelly, & Lacy 1997; Engelbracht et al. 1998), \[Ar III\](9.0$`\mu `$m)/\[Ar II\](7.0$`\mu `$m), and \[S IV\](10.5$`\mu `$m)/\[S III\](18.7$`\mu `$m)(e.g., Kunze et al. 1996; Rigopoulou et al. 1996). In addition, the \[S IV\]/\[Ne II\] ratio has inherent uncertainties due to the determination of metallicities, and all of the alternatives listed above are subject to corrections for aperture size variations and/or systematic uncertainties due to measurements in different ISO-SWS bandpasses. Systematic effects are minimized for the neon lines because they were observed with the same ISO aperture and bandpass. In addition, because the wavelengths of the two MIR neon lines are close, extinction effects can be neglected. For example, for a dust distribution in M82 which is well mixed with the stars and obeys a standard extinction law (e.g. Draine 1989) with A$`{}_{V}{}^{}`$50 mag (McLeod et al. 1993; Schreiber 1999), the observed line ratio is higher than the intrinsic line ratio by only a few percent. The same is true in the ultraluminous galaxy Arp 220 for a uniform foreground dust screen providing A$`{}_{V}{}^{}`$50 mag for the starburst emission (Sturm et al. 1996). By contrast, extinction corrections can change line ratios such as \[Ar III\]/\[Ar II\] and \[S IV\]/\[S III\] by 30-60% in similar conditions. Furthermore, the \[Ar III\] and \[S IV\] lines lie in the same spectral region as a broad silicate absorption feature at 9.8$`\mu `$m, further complicating the measurement of line fluxes.
### 2.2 Observations and Data Reduction
The majority of the data described herein was taken in grating line scan mode SWS02, with data from some galaxies taken from full-grating-scan SWS01 observations. Data reduction was completed using the ISO SWS Interactive Analysis (IA) package, which includes interactive tools for dark current subtraction, flat fielding, and removal of fringing caused by interference within the detectors themselves (see e.g., Schaeidt et al. 1996). The reduction completed
for this paper used post-mission calibration files (June 1998).
Line fluxes were determined by integration over the observed line profiles, which were generally close to Gaussian in shape. The systematic calibration of line fluxes has an uncertainty of $``$30% (Schaeidt et al. 1996), though the effect of this uncertainty is minimized by comparing two lines from the same bandpass. The uncertainties in the measured fluxes range from $``$10% for strong line sources to 30-40% for weaker sources (generally from the open-time proposal). For comparison with models (§4), we will assume uncertainties in the line ratios for the entire sample to be $`\pm `$50%. The aperture size for the bandpass containing the \[Ne II\] and \[Ne III\] lines is 14″ x 27″. Because of the weak sensitivity of this line ratio to extinction, no corrections for extinction were made (though see §4.3 for other effects of dust).
Table 1 lists the MIR neon line strengths for the galaxy sample. Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. shows the corresponding neon line spectra, and Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. shows the range of observed \[Ne III\]/\[Ne II\] ratios as a function of the infrared luminosity. The infrared luminosity $`L_{\mathrm{IR}}`$ is computed from the IRAS flux densities following the prescription of Sanders & Mirabel (1996, their Table 1): $`L_{IR}=L_{81000\mu \mathrm{m}}=5.6\times 10^5D_{\mathrm{Mpc}}^2(S_{100}+2.58S_{60}+5.16S_{25}+13.48S_{12})`$, where $`S_\lambda `$ is the flux density in Jy in the band centered at wavelength $`\lambda `$. For several of the galaxies, the sizes of the emission regions exceed the SWS field of view, so we have applied a aperture-size correction to $`L_{IR}`$ based on the spatial distribution of MIR or radio continuum emission, millimetric CO emission, or near-infrared hydrogen recombination line emission from the literature, assuming they all trace the same emission regions as the MIR neon lines. For eight galaxies, the observed $`L_{\mathrm{IR}}`$ was multiplied by a factor less than 0.5 to correct for the smaller SWS field of view, and for two galaxies there was insufficient data to determine an aperture-size correction. For the remaining galaxies, 11 required no correction, and seven galaxies were corrected by factors between 0.5 and 1.0.
In general, all of the sample objects have neon ratios between 0.05 and 1.0; however, the low-metallicity systems NGC 5253 and IIZw40 have neon ratios of 3.5 and 12, respectively. In §4.3, we discuss the effects of metallicity on model line ratios, in order to address the observed range of emission properties for the entire sample. The sample presented in this paper includes nine objects for which \[Ne II\] line fluxes were measured by Roche et al. (1991), who took observations from the ground in the 8-13$`\mu `$m atmospheric window. It is difficult to compare the fluxes derived from these two sets of observations due to differences in aperture size and the uncertain contribution of the broad 12.7$`\mu `$m PAH feature at the end of the 8-13$`\mu `$m band. This is particularly true for galaxies such as M82 which exhibit strong PAH features (e.g., Roche et al. 1991; Sturm et al. 2000); in the case of M82, we estimate from the SWS data that the broad 12.7$`\mu `$m feature could contribute a factor of 50% more flux to the apparent \[Ne II\] line at the resolution observed by Roche et al. . In addition, for some objects in the Roche et al. sample the \[Ne II\] line is redshifted out of the 8-13$`\mu `$m wavelength range and so cannot be measured from the ground. However, for the 7 objects for which comparisons could be made, the line fluxes are consistent to within these rather large uncertainties. NGC 5253 and IIZw40 were also singled out by Roche et al. (1991) as displaying high-excitation spectra, though the \[Ne II\] line was not detected. Roche et al. noted that these galaxies had little emission from broad PAH features, which is consistent with high excitation and UV energy density in a low-metallicity environment (see also Vigroux 1997; Thuan et al. 1999; and Madden 2000).
For further comparison, we include in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. the neon ratios for similar observations of W51, 30 Doradus, and the central parsec of the Galaxy. The neon ratios for W51 and 30 Doradus are higher than for the starburst galaxy sample, while that for the Galactic Center is lower (a further discussion of the Galactic Center can be found in §4.2). Our ISO-SWS data thus generally confirm findings from NIR and MIR spectroscopy that powerful dusty starbursts exhibit low nebular excitation in comparison to Galactic HII regions (e.g., Doherty et al. 1994; Cox et al. 1999). However, we find a range of values, with the most excited regions having neon ratios similar to those of young, Galactic HII regions.
## 3 Nebular emission from young stellar clusters
The starburst modeling method presented here predicts MIR spectral line ratios for evolving stellar clusters. Observations of nearby starburst galaxies reveal that starburst regions are in fact composed of many individual units. In addition to the HST detections of super star clusters in starburst galaxies, bright compact sources seen in high-resolution near-infrared broad-band images have been interpreted as young clusters of red supergiants (e.g. Tacconi-Garman, Sternberg & Eckart 1996; Satyapal et al. 1997). Furthermore, important substructure in the ISM on 10-parsec- or even parsec-scales is observed or inferred from modeling (e.g. Carral et al. 1994; Shen & Lo 1995; Achtermann & Lacy 1995).
In §3.4, we present a set of models in which the starburst region is composed of a heterogeneous ensemble of recently-formed star clusters. Therefore, we first lay the groundwork of modeling the emission from the HII region around stellar clusters.
### 3.1 The SED produced by a stellar cluster
We have used the evolutionary synthesis code STARS (Sternberg 1998) to model the integrated properties of individual stellar clusters. This code is based on the most recent Geneva stellar evolutionary tracks (Schaller et al. 1992; Schaerer et al. 1993a, 1993b; Charbonnel et al. 1993; Meynet et al. 1994), and is similar to other models developed by various investigators (e.g. Tinsley 1972; Huchra 1977; Bruzual 1983; Guiderdoni & Rocca-Volmerange 1987; Mas-Hesse & Kunth 1991; Bruzual & Charlot 1993; Leitherer & Heckman 1995; Leitherer et al. 1999). STARS follows the evolution in the H-R diagram (or, alternatively, in a $`\mathrm{log}T_{\mathrm{eff}}\mathrm{log}g`$ diagram) of a stellar population whose composition is specified by a time-independent IMF. The star formation rate (SFR) is assumed to decline exponentially as $`e^{t_\mathrm{b}/t_{\mathrm{sc}}}`$, where $`t_\mathrm{b}`$ is the age of the cluster and $`t_{\mathrm{sc}}`$ is the burst timescale. At a given age, the integrated SED is obtained by summing over the contributions of all stars present.
The SED assigned to each location in the $`\mathrm{log}T_{\mathrm{eff}}\mathrm{log}g`$ diagram is taken from a hybrid grid created from two separate libraries, in order to best represent the properties of all stars. For very hot stars, the effects of line blanketing and blocking in rapidly expanding atmospheres are not well represented by LTE models; thus for stars with temperatures above 25,000 K (M$`{}_{\mathrm{ZAMS}}{}^{}`$ 13$`M_{}`$), we have used new, non-LTE SEDs by Pauldrach et al. (1998). These models include the effects of stellar winds and mass loss, and cover effective temperatures up to 60,000 K (M$`{}_{\mathrm{ZAMS}}{}^{}`$ 120$`M_{}`$). The “cornerstone” grid of Pauldrach models consists of 8 main sequence models and 6 supergiant models, from which SEDs for intermediate values of Teff and log g are interpolated. Table 3 lists the (T<sub>eff</sub>, log g) combinations in the cornerstone grid, and Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. shows the sampling of the interpolated grid of solar-metallicity models. For stars with effective temperatures below 19,000 K (M$`{}_{\mathrm{ZAMS}}{}^{}<`$ 8$`M_{}`$), we use the Kurucz (1992) library of LTE SEDs. “Hybrid” SEDs for stars with intermediate effective temperatures were created by interpolating between the 19,000 K Kurucz model and the 25,000 K Pauldrach model. Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. compares examples of Kurucz, Pauldrach, and hybrid SEDs over a range of temperatures.
We have created SED grids for two metallicities, solar ($`Z_{}`$) and 0.2$`Z_{}`$. The low-metallicity grid allows us to better represent the stellar populations in NGC 5253 ($``$0.24$`Z_{}`$, Hunter, Gallagher, & Rautenkranz 1982) and IIZw40 ($``$0.15$`Z_{}`$, Masegosa, Moles, & Campos-Aguilar 1994), and provides a benchmark for the effects of metallicity variations on our models. The low-metallicity grid differs from the solar-metallicity grid only in the sampling of the cooler Kurucz models. Starburst models were computed using each grid.
Table 4 summarizes the model parameters used in this work. For each metallicity grid, only the upper mass cutoff ($`m_{\mathrm{up}}`$), burst age ($`t_\mathrm{b}`$), and burst timescale ($`t_{\mathrm{sc}}`$) were allowed to vary. The IMF was taken to have a Salpeter power-law index, $`\mathrm{d}N/\mathrm{d}mm^{2.35}`$ (Salpeter 1955), between a lower mass cutoff ($`m_{\mathrm{low}}`$) fixed at $`1\mathrm{M}_{}`$ and a variable $`m_{\mathrm{up}}`$. Though some studies suggest that the IMF in starburst galaxies is deficient in stars below a few solar masses (e.g. Rieke et al. 1980, 1993; Shier, Rieke & Rieke 1996; Engelbracht et al. 1998), variations in the low-mass IMF do not affect the ionizing continuum and the results presented here; thus we will not consider variations in $`m_{\mathrm{low}}`$.
### 3.2 Modeling the nebular emission around clusters
To model the nebular emission excited by a stellar cluster, we have used the photoionization code CLOUDY (version C90.04, Ferland 1996), with the source of ionizing radiation described by SEDs from STARS and the ionized nebula represented by a shell around a central stellar cluster. The nebular emission from young clusters will depend on the properties of the surrounding ISM. The nebular conditions are specified by the gas and dust composition, the hydrogen number density $`n_\mathrm{H}`$ of the gas, the distance $`R`$ between the ionizing source and the illuminated surface of the cloud, and the ionization parameter $`U`$. To estimate the appropriate inputs for these models, we have collected measurements from our data and from the literature which can be used to infer metallicities, $`n_\mathrm{H}`$, $`R`$, and $`U`$ for as many sample objects as possible.
Using our neon line measurements and published hydrogen recombination line measurements, we have calculated the abundance of neon relative to hydrogen for the sample using the strong-line method. This method provides abundances good to within factors of 2-3 (Osterbrock 1989), sufficient for our purposes here. Excluding NGC 5253 and IIZw40, the calculated metallicities are consistent with solar metallicity, to within the uncertainties: the average metallicity for the 13 objects for which metallicities could be determined was 1.9$`\pm `$$`Z_{}`$. The metallicities of these objects will be more rigorously derived by Spoon et al. (in prep).
We have computed the photoionization models assuming a solar gas-phase abundance for the majority of the sample, as well as models assuming a 0.2 $`Z_{}`$ gas-phase abundance for NGC 5253 and IIZw40. As changes in gas-phase abundances have relatively little effect on the neon ratio, we have not attempted to further tailor the gas-phase abundances for each galaxy. The comparison between solar and sub-solar metallicity models will be discussed further in §4.3.
Due to uncertainties about the dust properties in the starburst environments we have surveyed, we neglect the effects of dust grains mixed with the ionized gas. The gas density is assumed to be uniform, and we take $`n_\mathrm{H}=n_\mathrm{e}`$, where $`n_\mathrm{e}`$ is the electron density. The electron density was determined from the \[S III\] 18.7$`\mu `$m/33.5$`\mu `$m line ratio for the galaxies in the guaranteed-time sample, with the line fluxes corrected for the extinction determined in §4.1, assuming the extinction law recommended by Draine (1989). The typical uncertainties on the line ratios are 50%, and are dominated by extinction and aperture-size corrections (the \[S III\] 18.7$`\mu `$m and 33.5$`\mu `$m lines were measured through $`14^{\prime \prime }\times 27^{\prime \prime }`$ and $`20^{\prime \prime }\times 33^{\prime \prime }`$ apertures, respectively), and the absolute flux calibration. Within the uncertainties, the dereddened line ratios for this subsample lie in the low-density limit, and imply $`n_\mathrm{e}10^3\mathrm{cm}^3`$. We adopt a value of $`300\mathrm{cm}^3`$ for all models, inferred from the average of the line ratios for the galaxies in the guaranteed-time sample.
The ionization parameter (U) of a nebula is one of the most important parameters in photoionization models. A measure of $`U`$ gives the number of ionizing photons impinging at the surface of the nebula per hydrogen atom:
$$U\frac{Q_{Lyc}}{4\pi R^2n_\mathrm{H}c},$$
$`(1)`$
where $`Q_{Lyc}`$ is the rate of production of ionizing photons, $`R`$ is the radius of the shell, and $`c`$ is the speed of light. The determination of the ionization parameter $`U`$ from observed properties depends on the assumptions made about the geometry of the ionizing clusters and of the nebulae. As discussed above, entire starburst regions such as included in the SWS field of view likely do not consist of a gas shell illuminated by a single, centrally-concentrated cluster. If the ISM and a number of young star clusters are in a mixed distribution, gas clouds will shield each other partially from ionizing radiation from clusters distributed throughout a starburst region of radius R (e.g., Wolfire, Tielens, & Hollenbach 1990). Both representative geometries are shown schematically in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.. In the distributed cluster geometry, we can describe the conditions at the illuminated face of each cloud structure in terms of an effective ionization parameter, U<sub>eff</sub> which may be written
$$\mathrm{log}U_{\mathrm{eff}}=\mathrm{log}\left(\frac{Q_{Lyc}}{4\pi R^2n_\mathrm{H}c}\right)+\mathrm{log}\left\{\left(\frac{\lambda }{R}\right)\left[1e^{R/\lambda }\right]\right\},$$
$`(2)`$
The distributed cluster geometry is analogous to that in “mixed extinction” models, where R/$`\lambda `$ is a measure of the optical depth and $`\lambda `$ is the mean free path of photons. The parameter $`\lambda `$ depends on the number density and sizes of the clouds which reside in the starburst region. Therefore, the determination of $`U_{\mathrm{eff}}`$, and the corresponding effective radius $`R_{\mathrm{eff}}`$ (from replacing U with U<sub>eff</sub> in Eqn. 1), requires a knowledge of the relative distributions of gas clouds and ionizing clusters, as well as the properties of the clouds and clusters themselves. Such information is available for the archetypal starburst galaxy M 82: by combining new near- and mid-infrared data with the results of modeling of photodissociation regions by Lord et al. (1996) and observations of the molecular gas by Shen & Lo (1995), Schreiber (1999) reassessed the geometry of clusters and gas clouds in the starburst region of M82. The effective values, from an investigation of the ionization parameter in regions of varying size within the starburst core of M 82, are $`\mathrm{log}U_{\mathrm{eff}}=2.3`$ and $`R_{\mathrm{eff}}=25\mathrm{pc}`$ (Schreiber 1999).
The investigation of M 82 by Schreiber (1999) further shows that the effective ionization parameter is essentially independent of the location and the size of the regions studied, from individual burst sites $`20\mathrm{pc}`$ in radius to the entire starburst core extending over $`450\mathrm{pc}`$. This strongly suggests that the regions of intense starburst activity in M 82, which dominate the nebular line emission, have very similar properties over a wide range of physical sizes. On average, the nebular conditions of the gas in the starburst regions of M82 can thus be described locally and globally with the same parameters.
A similar analysis is possible for the entire starburst regions of NGC 3256 and NGC 253. By combining published data on the integrated Lyman continuum luminosity from Rigopoulou et al. (1996) and Engelbracht et al. (1998) with the results from PDR modeling by Carral et al. (1994), we infer $`\mathrm{log}U_{\mathrm{eff}}=2.3`$ and $`2.6`$, respectively. The value of $`\mathrm{log}U_{\mathrm{eff}}`$ we derive for NGC 253 is consistent with that derived by Engelbracht et al. 1998 (log U$`2.2`$ to $`2.5`$). The determination of log U for these three galaxies supports the suggestion by Carral et al. (1994) that the ISM properties in starburst galaxies are independent of the global luminosity and the triggering mechanism, and are endemic to starburst activity itself. We therefore use the $`\mathrm{log}U_{\mathrm{eff}}`$ and $`R_{\mathrm{eff}}`$ determined in M 82 as representative values for the entire sample.
In our models, $`n_\mathrm{e}`$ and $`R=R_{\mathrm{eff}}`$ are assumed to be time-independent. We first computed the neon line ratio for a very short-lived starburst, with $`t_{\mathrm{sc}}`$=1 Myr. In this model, $`U`$ varies proportionally with Q<sub>Lyc</sub> over time, via Equation (1). To set the absolute flux scale of the cluster SEDs, we choose the condition that when each cluster reaches its maximum Q<sub>Lyc</sub>, $`U_{\mathrm{max}}=U_{\mathrm{eff}}`$; the shape and relative flux scale as the clusters evolve are determined from the composite SEDs output by STARS. The computations are stopped when $`\mathrm{log}U`$ becomes smaller than $`5.5`$; at this point, the SFR has dropped down by several orders of magnitude compared to the initial SFR, and the stellar population produced by the starburst is expected to have faded away into the galaxy’s background population.
With the nebular parameters determined above as constraints, the SEDs produced by STARS are input into CLOUDY. Approximating a thin shell around the distributed cluster geometry, CLOUDY was run with plane-parallel geometry and an ionization parameter specified by Equation (1). Models for the burst timescales presented in this paper were computed by convolving the neon fluxes for the $`t_{\mathrm{sc}}`$=1 Myr burst with an exponential SFR with the appropriate value of $`t_{\mathrm{sc}}`$,
$$F_\lambda (t_b)=_0^{t_b}F_\lambda (t_{sc}=1Myr,\tau )e^{(t_b\tau )/t_{sc}}d(\tau ),$$
$`(3)`$
where F<sub>λ</sub> is the flux in the line at wavelength $`\lambda `$. This method inherently assumes that U($`t_\mathrm{b}`$) is proportional to the ionizing photon rate produced in the 1 Myr-timescale burst, rather than that indicated by the integrated Q<sub>Lyc</sub> for the longer burst timescales. Thus at later times, the \[Ne III\]/\[Ne II\] ratio will stay constant, as the line emission will be produced only by the most recently formed clusters. It is these recently-formed clusters which will dominate the ionizing radiation field at all ages, and are all assumed to have the same $`U_{\mathrm{max}}`$.
### 3.3 Results for a homogeneous sample of evolving clusters
The model neon line ratios output by CLOUDY are indicated by the solid lines in the upper panel of Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE., as a function of burst age and upper mass cutoff for various $`t_{\mathrm{sc}}`$. These ratios assume no dust, n$`{}_{e}{}^{}`$300 cm<sup>-3</sup>, and log U$`{}_{max}{}^{}=2.3`$. The model predictions for five different upper mass cutoffs ($`m_{\mathrm{up}}`$=25, 30, 35, 50, and 100 $`\mathrm{M}_{}`$) and three burst timescales ($`t_{\mathrm{sc}}`$=1, 5 and 20 Myr) are shown to represent a range of prescriptions for the formation of stars in clusters. There is a slight decline in the computed ratio for $`t_\mathrm{b}`$$``$5 Myr for the longest timescales, due to a build up of evolved clusters that soften the shape of the integrated ionizing radiation field. However, these evolved clusters must still be relatively young to affect the integrated ratio significantly. The asymptotic behaviour of the curves for timescales longer than $`5\mathrm{Myr}`$ can be attributed mainly to our choice of an exponentially decaying SFR, but also reflects the dominance of the youngest generations, for which $`\mathrm{log}U_{\mathrm{eff}}`$ is close to the maximum.
The vertical bar to the right of the plot represents the range of observed neon line ratios for all sample galaxies except the two low-metallicity blue dwarf galaxies NGC 5253 and IIZw40, which will be discussed separately in §4.3. From the comparison with models, values of $`m_{\mathrm{up}}<`$ $`25\mathrm{M}_{}`$ can be ruled out for the entire sample. In addition, the largest neon ratios in the sample are consistent with upper mass cutoffs at least as large as $`50\mathrm{M}_{}`$, independent of the burst age and timescale. Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. shows clearly the degeneracy between age and $`m_{\mathrm{up}}`$: the range of observed neon ratios may either be due to true variations in the population of the most massive stars, or to short bursts of star formation where $`m_{\mathrm{up}}`$ is always large but the SED is softened through aging. However, a similar modeling analysis of the nearby star formation sources plotted in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. (Thornley et al. , in prep.) and extensive modeling of M82 (Schreiber 1999) support the need for stars more massive than 50$`M_{}`$, in agreement with stellar census measurements in a variety of nearby star forming regions.
### 3.4 Modeling infrared emission from a heterogeneous ensemble of evolving clusters
The models described above are appropriate for individual clusters (and surrounding nebulae) or for a homogeneous collection of clusters, enclosed within a larger region. However, the starburst galaxies we are examining may contain evolving clusters with a wide range of total masses, and low-mass clusters may not have enough material to fully sample the IMF. Here, we explore the effect of an assumed cluster mass spectrum on the integrated \[Ne III\]/\[Ne II\] ratio. On average, the ionizing spectrum is softer if small clusters contribute a significant fraction of the total ionizing luminosity.
Such an ensemble of clusters can be conveniently described by a luminosity function (LF). Various studies have explored the LF of young clusters and HII regions in a wide range of star-forming environments, including the disk of our own Galaxy (McKee & Williams 1997), nearby spiral and irregular galaxies (e.g. Kennicutt, Edgar & Hodge 1989; Elson & Fall 1985; Gonzalez-Delgado et al. 1995), as well as merger systems like NGC 4038/4039 (Whitmore & Schweizer 1995). These studies show that the number distribution of clusters in optical light and in ionizing luminosity follows a power-law LF, $`\mathrm{d}N/\mathrm{d}(\mathrm{log}L)L^\beta `$, with remarkably similar indexes in the range $`\beta =0.51`$, down to the faintest luminosities observed. The typical completeness limits correspond to an absolute $`V`$-band magnitude of $`10\mathrm{mag}`$ or to a Lyman continuum luminosity of $`10^{50}\mathrm{s}^1`$. Hence only bright clusters are directly observed. For comparison, the Orion nebula is powered by stars which emit a total of $`10^{48.85}`$ Lyman continuum photons $`\mathrm{s}^1`$ (Kennicutt 1984).
In order to model the integrated emission properties of an ensemble of clusters, we must include the contribution from all clusters in a model starburst event. While this calculation is relatively straightforward in large, luminous clusters, it is more uncertain in lower-mass clusters where statistical fluctuations at the high-mass end of the IMF become important: for instance, the formation of a single 20$`M_{}`$ star instead of a single 15$`M_{}`$ star could change the cluster luminosity by an order of magnitude. To determine the shape of the LF consistently for the brightest clusters as well as clusters fainter than $``$10<sup>50</sup> s<sup>-1</sup>, we ran a Monte Carlo simulation to determine the stellar contents of clusters of different masses using Poisson statistics.
Our Monte Carlo simulation was run for a sample of 2x10<sup>5</sup> clusters, with the number of stars per cluster ranging from 100 to 10<sup>7.5</sup> and mass bins populated from $`m_{low}`$=0.1$`M_{}`$ upward in accordance with the specified Salpeter IMF. In this way, the upper mass cutoff of an individual cluster is dependent on both the total mass of the cluster and statistical variations in the population of the IMF. Using published values to represent the Q<sub>Lyc</sub>(m) relation for individual stars (Vacca, Garmany, & Shull 1996; Panagia 1973), we then determined the average cluster mass, m<sup>cl</sup>, as a function of ionizing luminosity Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ for the range of clusters falling in a given luminosity bin of width 0.1 dex. The resulting relation, m<sup>cl</sup>(Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$), can be approximated by two power laws, with m$`{}_{}{}^{\mathrm{cl}}`$ Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ for clusters with log Q$`{}_{Lyc}{}^{\mathrm{cl}}>`$49.5 and m$`{}_{}{}^{\mathrm{cl}}`$ (Q$`{}_{Lyc}{}^{\mathrm{cl}})^{0.19}`$ for clusters with log Q$`{}_{Lyc}{}^{\mathrm{cl}}<`$49.5. To convert from a function m<sup>cl</sup>(Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$) to a luminosity function, we assume a simple, power-law cluster mass function of the form dN/d(log m) $``$ m, such that dN/d(log Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$) $``$ (Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$)<sup>-αβ</sup>.
With these conditions, we derive a LF covering Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ from 10<sup>44</sup> to 10<sup>53</sup> photons s<sup>-1</sup>. For the cluster mass function, we assume the simple case $`\alpha `$=1, which produces a LF shape consistent with that determined by McKee & Williams (1997) for HII regions in the disk of our own Galaxy. The lower and upper limits for Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ in our LF were chosen to represent the smallest associations still capable of ionizing an H II region (i.e. containing only one early-B star), and the most luminous super star clusters detected (e.g., in NGC 4038/4039). The resulting LF is again represented by a broken power law, with $`\beta `$=0.19 at the lower end and $`\beta `$=1.0, with the break occuring at Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ $`10^{49.5}\mathrm{s}^1`$ as for the m<sup>cl</sup>(Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$) relation for individual clusters. Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. shows the derived LF, with the McKee & Williams LF overplotted for comparison.
We assume that the LF determined above describes the cluster ensemble at birth, and then follow the evolution of the ensemble of clusters for various star formation histories. The evolution of the LF with burst age for the ensemble of clusters is obtained by following the evolution of Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ for the clusters in each luminosity bin. For this purpose, we computed a library of cluster models for a $`t_{\mathrm{sc}}`$=1 Myr burst using STARS and CLOUDY. The Monte Carlo simulations described above provide the relation between Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ and $`m_{\mathrm{up}}^{\mathrm{cl}}`$, the mass of the most massive star in an individual cluster given a fixed IMF and an input stellar count (note that $`m_{\mathrm{up}}^{\mathrm{cl}}`$ should be distinguished from an intrinsic, galaxy-wide upper mass cutoff of the IMF, which we have thus far designated as $`m_{\mathrm{up}}`$).
Models were generated for $`t_{\mathrm{sc}}`$=1 Myr and $`m_{\mathrm{up}}^{\mathrm{cl}}`$=5-100 $`\mathrm{M}_{}`$, in steps of $`5\mathrm{M}_{}`$. The range of Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ was divided in logarithmic bins of $`\mathrm{\Delta }(\mathrm{log}`$Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$) = 0.05 dex, and the models for intermediate $`m_{\mathrm{up}}^{\mathrm{cl}}`$ were obtained by interpolation of the library models. At zero age, the clusters with different initial $`m_{\mathrm{up}}^{\mathrm{cl}}`$ (and Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$) were distributed in the luminosity bins according to the LF. As the burst age increases, the evolution of each cluster in Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ is followed (using the library models), and the new LF at each time step is determined from the distribution of the clusters in the different Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ bins after evolution has taken place.
The slope of the high-luminosity end of the LF ($`10^{50}\mathrm{s}^1`$) changes very little during this evolution; this is mainly due to the fact that these clusters contain stars with masses $`>`$50$`M_{}`$. All such massive stars have similarly short main-sequence lifetimes (e.g. Schaller et al. 1992), which implies that these luminous clusters move into lower Q$`{}_{}{}^{\mathrm{cl}}{}_{Lyc}{}^{}`$ bins at similar rates. The constant slope of the high-luminosity end of the LF is thus consistent with the observed LF in various sources, which have presumably a range of ages and starburst histories.
### 3.5 Results for a heterogeneous ensemble of evolving clusters
The neon ratio at any given age for the $`t_{\mathrm{sc}}`$=1 Myr ensemble of clusters was obtained by summing over the \[Ne II\] and \[Ne III\] line fluxes of all the model clusters still contributing; the integrated properties for longer burst timescales were again obtained by convolving those for the $`t_{\mathrm{sc}}`$=1 Myr burst. The neon ratios predicted for an ensemble of clusters distributed according to the derived cluster LF are indicated, for the longest and shortest timescales we explored, by the dashed lines in the upper panel of Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.. As expected, the main effect of accounting for a cluster size distribution is that the smaller, less luminous clusters soften the integrated ultraviolet radiation field. In particular, the predicted \[Ne III\]/\[Ne II\] for an ensemble of clusters is lower than for a single cluster, assuming the same galaxy-wide $`m_{\mathrm{up}}`$. The effect of the LF is larger for higher $`m_{\mathrm{up}}`$, with a reduction in predicted ratios by a factor of $``$2 for a galaxy-wide $`m_{\mathrm{up}}`$=100 $`\mathrm{M}_{}`$, and $``$1.2 for $`m_{\mathrm{up}}`$=25 $`\mathrm{M}_{}`$. This dependence on $`m_{\mathrm{up}}`$ reflects primarily the stellar properties themselves: the hardness of the ionizing spectrum decreases steeply as the stellar mass decreases. The magnitude of the effect of incorporating a cluster LF depends mainly on the power-law index for the LF and on the upper limit in Q$`{}_{Lyc}{}^{}{}_{}{}^{\mathrm{cl}}`$. The steep power-law we have adopted maximizes the differences between single and ensemble cluster models, in comparison with other plausible LFs with indices above the break of $`\beta =0.51`$. The choice of a high upper limit in Q$`{}_{Lyc}{}^{}{}_{}{}^{\mathrm{cl}}`$ is justified by observations in some starburst galaxies, notably NGC 4038/4039 and M 82 (Whitmore & Schweizer 1995; O’Connell et al. 1995).
Although the “down-weighting” effects of the smaller clusters, which produce softer ionizing radiation, are measurable, they do not require a significantly different conclusion than that reached by comparison of our data with models of homogeneous clusters. Accounting for a LF of the ionizing clusters which excite the observed nebular emission lines, we confirm that very high-mass stars can form in starburst galaxies, allowing $`m_{\mathrm{up}}=50100\mathrm{M}_{}`$ even more easily in the sources with the highest measured ratios.
## 4 Additional considerations of the starburst scenario
### 4.1 Insights from the L<sub>bol</sub>/L<sub>Lyc</sub> ratio
If we posit that all starbursts form stars in the manner that nearby, massive star forming regions do, then the most plausible explanation for low observed line ratios is that starbursts are events of short duration (Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. suggests $`t_{\mathrm{sc}}`$$``$1 Myr) which produce very massive stars, but whose aging rapidly softens diagnostic ratios such as the ones we use here. Our models show that as long as very massive stars are formed, even in small numbers, they strongly dominate the ionizing radiation field and thus maintain high neon line ratios. The ratios can only decrease to the observed range long enough after the exhaustion of starburst activity for the most massive stars to have evolved off the main sequence. For ratios near 0.1, corresponding to stars with initial masses near $`30\mathrm{M}_{}`$, this will take about 5 Myr. In principle, the neon ratios will start declining very rapidly after the last massive stars have formed in the burst, so that the neon ratio alone does not fully constrain the timescale. However, by combining with other measurements which are indicative of starburst properties, we can explore the robustness of our conclusion that starburst timescales are generally quite short.
The ratio of the bolometric to Lyman continuum luminosities ($`L_{\mathrm{bol}}`$/$`L_{Lyc}`$) is a useful contrasting probe of the properties of massive stars: since this ratio is sensitive to a somewhat lower mass range than that to which \[Ne III\] 15.6 $`\mu \mathrm{m}`$/\[Ne II\] 12.8 $`\mu \mathrm{m}`$ is sensitive, it varies significantly with time even for longer burst timescales. This reflects the buildup of a population of stars in a lower mass range which contribute more to $`L_{\mathrm{bol}}`$, and which have longer main-sequence lifetimes. Therefore, the combination of both diagnostics can help address the degeneracy between aging effects and variations in the upper mass cutoff.
As we are characterizing the properties of massive star forming regions, we assume $`L_{\mathrm{bol}}=L_{\mathrm{IR}}`$ (as defined in §2.2). This is usually a good approximation since a large fraction of the energy output of OB stars is absorbed by the surrounding interstellar dust, present in large amounts in starbursts, and re-emitted in the thermal infrared. We derived L<sub>Lyc</sub> using two different methods. First, we use measurements of hydrogen recombination line fluxes and thermal radio continuum emission, when such data were available from observations with the SWS and in the literature. This allowed us to constrain simultaneously the extinction toward the sources from the comparison of observed and theoretical relative line fluxes. Second, we used our own neon line fluxes corrected for extinction and assuming all neon atoms are either singly- or doubly-ionized.
We assumed case B recombination coefficients and line emissivities (Hummer & Storey 1987), with an electron density of $`n_\mathrm{e}=100\mathrm{cm}^3`$ and temperature of $`T_\mathrm{e}=5000\mathrm{K}`$, except when individual determinations were available in the literature. For the extinction correction, we adopted a composite extinction law made up of the Rieke & Lebofsky (1985) law for $`\lambda `$ 0.9$`\mu `$m and the Draine (1989) law for 0.9$`\mu `$m$`<\lambda <`$40$`\mu `$m. The effects of obscuration were neglected for $`\lambda >`$ 40$`\mu `$m. If sufficient data were available, we constrained the geometry of the emission sources and obscuring dust as well. Two models were considered: a uniform foreground screen (UFS) of dust and a homogeneous mixture of dust and sources (MIX). Otherwise, we considered only the UFS model.
For the sources for which the determination from recombination lines was possible, the estimates from the hydrogen and neon lines agree to within a factor of three or better, and we adopted the average as the final L<sub>Lyc</sub>. The cases where the extinction could not be reliably constrained yield lower limits on L<sub>Lyc</sub>, and thus upper limits on L<sub>IR</sub>/L<sub>Lyc</sub>. The lower panel of Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. shows the models for L<sub>IR</sub>/L<sub>Lyc</sub> obtained with STARS. As in the upper panel, the ratios expected for a homogeneous cluster population are shown as solid lines and those expected for a cluster ensemble defined by our LF are shown as dashed lines. The effect of including a cluster LF is of similar magnitude for $`L_{\mathrm{IR}}`$/$`L_{Lyc}`$ as for the neon ratio. It is clear that the average neon and $`L_{\mathrm{IR}}`$/$`L_{Lyc}`$ ratios are consistent with conditions where the clusters have high upper mass cutoffs, $`m_{\mathrm{up}}50100\mathrm{M}_{}`$, and short burst timescales of a few million to $``$10<sup>7</sup> years. In fact, by plotting the neon ratios against the $`L_{\mathrm{IR}}`$/$`L_{Lyc}`$ ratios in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE., we see that the models suggest very short timescales, so short as to be difficult to produce even with the $`t_{\mathrm{sc}}`$=1 Myr burst, the burst with the shortest timescale considered in this study. Though this discrepancy is model-dependent, it is generally clear that short timescales are needed to reproduce the range of ratios observed in our starburst sample.
### 4.2 The Galactic Center as a template for a short starburst
The short timescales and ages inferred for the star forming activity in starburst galaxies are reminiscent of the ones in the young stellar clusters in the central parsec of our Galaxy, and can be checked there on the basis of the existing stellar census. The observed \[Ne III\]/\[Ne II\] ratio is even lower than in most starbursts, despite the high ionization parameter log U = $`1`$ derived for the spatially resolved ionized region ($``$ 1 pc, Lutz et al. 1996). The picture of a short but aged burst is supported by the direct stellar census (Genzel et al., 1994; Krabbe et al., 1995; Najarro et al., 1997) which suggests a star formation event of age $``$7 Myr and duration approximately 4 Myr. This event is most directly indicated by the presence of both cool red supergiants like IRS7 and massive, moderately hot (20,000 to 30,000 K) blue helium-rich supergiants or Wolf-Rayet (WR) stars.
We have computed models optimized for this Galactic Center region, assuming a single cluster with a Salpeter IMF, $`t_{\mathrm{sc}}`$=1 Myr, $`m_{\mathrm{up}}`$=100, solar metallicity, and fixed log U=$`1`$, n=3000 cm<sup>-3</sup> as derived by Lutz et al. (1996). The burst timescale is somewhat shorter than that suggested by the stellar census and was chosen as a conservative assumption, since a shorter burst will produce softer radiation fields at late ages and thus minimize any need to invoke other effects for explaining the observed soft radiation. The decay of line ratios with time is similar to Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE., but the ratios at any given time are higher due to the higher log U in this smaller region. At the age of 7 Myr preferred by the stellar census, the \[Ne III\]/\[Ne II\] ratio is still about 1 to 2, well above the value of 0.05 observed by Lutz et al. (1996). The model reaches the observed value only after more than 13 Myr, which is difficult to reconcile with the presence of evolved massive stars approaching 100 M (Najarro et al. 1997; Ott, Eckart, & Genzel 1999). While aging effects push the Galactic Center neon ratio more closely into agreement with models, they are not enough to fully account for the low observed neon ratio.
The stellar census for the Galactic Center gives an independent view through direct analysis of the contributions of different stellar types to the ionizing continuum. Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. presents an Hertzsprung-Russell diagram weighted by Lyman continuum luminosity, derived from the same starburst model for an age of 7 Myr. It is evident that most of the ionizing luminosity still originates in stars close to the main sequence at log T= 4.5 to 4.6, with smaller contributions by stars somewhat evolved towards lower temperatures and a population of hot WR stars to the left of the main sequence. This is in stark contrast to the finding by Najarro et al. (1997) that 7 of their stars, found at log T $`<`$ 4.5 and log L $`>`$ 5.75, contribute half of the ionizing luminosity of the Galactic Center. The same region holds less than 1% of the ionizing luminosity for the model in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.. At all other model ages, this fraction does not exceed about 1%.
The fact that the discrepancy between ‘hard’ models and ‘soft’ observations is seen both in the photoionization modelling and the stellar census leads us to the conclusion that this discrepancy is not primarily due to uncertainties in the hot star SEDs used for the photoionization models. Such uncertainties affect mainly the shape of the ionizing flux and thus the fine structure line ratios, but much less the total ionizing flux of the star and the census. Another possibility is that the adopted Geneva stellar evolutionary tracks Schaller et al. (1992) do not predict the large number of 20,000 K to 30,000K supergiants seen in the Galactic Center, or that there exists a mismatch between the stellar effective temperatures from evolutionary tracks and those from stellar atmosphere models. The corresponding region is populated by these tracks but only for short intervals, possibly because massive, post-main-sequence stars rapidly move off to hotter parts of the tracks. The disagreement between tracks and observations may be related to the difficulty of defining mass loss, atmospheres, and effective temperatures for late stages of massive star evolution governed by strong winds. Our conclusions are unchanged when using evolutionary tracks with twice solar metallicity Schaller et al. (1992) and higher mass loss Meynet et al. (1994), which might be adequate for the somewhat higher metallicity in the Galactic Center. The latter tracks in fact increase the disagreement by adding more very hot WR stars. In fact, WR stars may contribute to the uncertainties in the evolutionary tracks, as our knowledge of the ionizing spectra of WR stars is uncertain. Indeed, there is both observational and theoretical evidence Crowther et al. (1999); Hillier & Miller (1998) which suggests that the ionizing spectra of WR stars may be much softer than commonly assumed.
The case of the Galactic Center has obvious implications for aging starbursts with similar populations. If the current Geneva tracks, interpreted in the fashion described above, indeed mispredict the post-main-sequence evolution of massive stars by postulating very hot stars where the Galactic Center shows us a cooler luminous population, then starburst models will tend to predict too high values for ratios like \[Ne III\]/\[Ne II\]. The resultant “need” to invoke low upper mass cutoffs (or aging effects) to explain soft radiation fields may be due to this effect. Corrections for this effect would likely lower the neon ratios, and thus relax the stringent conditions of short burst time scales, leading to more plausible timescales of $``$10<sup>7</sup> years. While a careful study is needed to test the reliability of current evolutionary tracks, this analysis of the Galactic Center supports our general conclusion that starbursts are “normal” star-forming environments, in the sense that they produce very massive stars just as local star forming regions do.
### 4.3 The effects of parameter variations
Starburst modeling is, by the very nature of the star formation process, a multi-parameter problem. To assess the robustness of the results we have presented, we now discuss the effects of varying the possible input parameters of our model, considering variations in starburst SEDs and nebular parameters. The most significant changes in a starburst SED will arise because of our choice of metallicity, model SED libraries for individual stars, or the shape of the IMF. Changes in nebular parameters have effects of similar magnitude as those of changes in SEDs. The most significant effects will be due to variations in the ionization parameter, though variations in the dust population as well as the density of the gas in these regions also have minor effects. For simplicity, these variations are illustrated for single-cluster models with $`m_{\mathrm{up}}`$=100$`M_{}`$ and $`t_{\mathrm{sc}}`$=1 Myr in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE..
#### 4.3.1 Metallicity
Of the above effects, those of metallicity are the most significant. Metallicity effects are twofold: at sub-solar metallicities, the stellar SEDs are harder due to reduced line blanketing and blocking (e.g., Pauldrach et al. 1998), and the evolutionary track each star follows through the H-R diagram changes, producing a ”hotter” main sequence. Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.a compares the neon ratios for two models, the hybrid SED grid at solar metallicity, and the corresponding hybrid grid with stars and gas at 0.2 $`Z_{}`$, which is more appropriate for the two low-metallicity dwarf galaxies in our sample. Lowering the metallicity results in considerably increased neon ratios: for Z=0.2$`Z_{}`$, the predicted neon ratios increase by factors of $``$4-10. The increase in neon ratios at low metallicity is dominated by changes in the evolutionary tracks and SEDs of model stars, with the corresponding changes to gas-phase abundances playing a relatively minor role. NGC 5253 and IIZw40 have neon ratios 3 and 13 times higher, respectively, than in any of the other starbursts in our sample; the observed ratio values are also plotted as horizontal grey lines in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.a. Therefore, even accounting for low metallicity, the neon ratios measured for these systems are consistent with a stellar population with $`m_{\mathrm{up}}`$$`>`$50-100$`M_{}`$. In contrast with the example of Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.a, models with Z$`>Z_{}`$ will have correspondingly lower predicted neon ratios; this factor is worth considering if more accurate determinations of abundances become available.
#### 4.3.2 SED libraries
If we were to choose Kurucz instead of Pauldrach SEDs to represent the most massive stars, the contribution of the softer Kurucz spectra for high-mass stars would produce lower neon ratios (cf. Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.). The significance of this effect can be seen in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.b, which shows the neon ratio for two input stellar grids: the interpolated, hybrid grid we are using, and a standard Kurucz model grid. At a burst age of $``$3 Myr, the predicted neon ratio using Kurucz SEDs is a factor of $``$2 lower than that predicted by our hybrid grid. Comparison with our ISO-SWS data would suggest $`m_{\mathrm{up}}`$$`>`$50$`M_{}`$ for more than half the sample, even at zero-age.
#### 4.3.3 IMF slope
We have taken a Salpeter IMF to be the most representative form of the initial mass function; however, the universality of the IMF is still debated (see, e.g., Massey 1998 and Scalo 1998 for contrasting views). Some authors have suggested that the IMF in active star-forming environments is well-represented by a Salpeter form (Hunter 1995; Massey & Hunter 1998). While the alternative model favored by Scalo (1998) exhibits a steeper function at intermediate masses, it is similar to Salpeter at the high-mass end (M$``$10$`M_{}`$), where the neon ratio is most sensitive. To illustrate the effects of changing the IMF, we show the resulting neon ratios for a Miller & Scalo (1979) IMF in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.b, where this IMF is represented as a power law with index $`1.4`$ for M=1-10$`M_{}`$ and $`2.5`$ for M=10-100$`M_{}`$. The Miller-Scalo IMF is generally steeper than Salpeter, and thus there are fewer massive stars formed relative to a given number of low-mass stars. The net result is a softer composite SED and a prediction of lower neon line ratios. The decrease in the predicted neon ratio caused by changing the IMF in this way is smaller than that caused by changing SED libraries.
#### 4.3.4 Ionization parameter
Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.c shows the predicted neon ratios for two alternate values of the ionization parameter: log U$`{}_{max}{}^{}=1.5`$, the upper limit for U assuming that all clusters lie at the center of the starburst nebular emission region of M82 (Schreiber 1999), and log U$`{}_{max}{}^{}=3.5`$, a value more similar to that derived by Wang, Heckman, & Lehnert (1997) for the diffuse ionized medium (DIM). An ionization parameter of log U$`{}_{max}{}^{}=1.5`$ would result in predicted neon line ratios $``$3 times higher out to $``$7 Myr for $`t_{\mathrm{sc}}`$=1 Myr. Such an increase would imply upper mass cutoffs between 25 and 50 $`M_{}`$ for any bursts with $`t_{\mathrm{sc}}`$$``$5 Myr, without accounting for any aging effects. This value of U<sub>max</sub> is a factor of $``$3 greater than the highest value which is consistent with the M82 analysis. The effect of reducing the ionization parameter to log U$`{}_{max}{}^{}=3.5`$ is larger, causing a decrease in the predicted neon ratios by an order of magnitude. The highest observed ratios in our starburst sample would not be reproduceable, even for the most massive stars for which we have models, if log U$`{}_{max}{}^{}=3.5`$. Previous starburst modeling studies along these same lines (see, e.g., Kunze et al. 1996, Rigopoulou et al. 1996) have used log U=$`2.5`$, similar to the M82-based value used in this study.
#### 4.3.5 Dust population and gas density
The presence of dust within HII regions can affect the efficiency of nebular photoionization, but introducing a dust component has a relatively small effect on our models. Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.d shows the effect of adding dust grains similar to those in Orion (we use the Orion dust population that is incorporated in CLOUDY, from Baldwin et al. 1991). Adding such a dust component causes a $``$15% increase in the predicted line ratios. This variation is smaller than the uncertainties in the measured neon line ratios, and negligible compared to the other parameter variations we have explored. We conclude therefore that we have introduced no significant uncertainties by neglecting dust in earlier sections. Note, however, that the uncertainties are larger for determinations of L<sub>bol</sub>/L<sub>Lyc</sub>, where dust could have a much more significant effect.
Due to the large uncertainties in inferring the gas density from the SIII (18.7/33.5$`\mu `$m) ratio, we also show in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.d the effects of increasing the gas density to 10<sup>3</sup> cm<sup>-3</sup>. This is the highest density consistent with the range of measured sulfur ratios and their uncertainties. For the model shown, the neon line ratio increases by less than five percent over the entire age range, indicating that variations in gas densities have a negligible effect on the output neon ratios. However, we note that the ionization parameter changes inversely with gas density, such that this increase in gas density would also imply a drop in the ionization parameter to approximately log U$`sim2.8`$, thus solidifying the case for very massive stars being present in the starbursts in our sample.
### 4.4 Extra-starburst contributions?
The models we have presented thus far assume that the only contribution to the MIR neon line fluxes comes from direct photoionization by stars in the starburst region itself. In this section we examine constraints, from ISO spectroscopy, on the possible contributions of two other processes: excitation by active galactic nuclei (AGNs), and contributions from a diffuse ionized medium (DIM) in the surrounding galaxy (e.g., Lehnert & Heckman 1994; Wang et al. 1997).
#### 4.4.1 AGN contributions
The MIR spectra of AGNs are distinctive in their display of strong emission lines from highly ionized species, such as \[Ne V\] and \[O IV\], which require higher excitation that can be produced even by the hottest stars. The absence of these lines, or their weakness relative to lower excitation lines, has been used to demonstrate the dominant contribution of star formation to the power produced in ultraluminous IR galaxies (Lutz et al. 1996; Genzel et al. 1998). Though some starbursts show very weak \[O IV\] emission, the most plausible explanation is ionizing shocks from supernovae or superwinds (Lutz et al. 1998). The sample presented here significantly overlaps with the Lutz et al. (1998) sample, and strong \[O IV\] emission is generally not seen; furthermore, the shock models which reproduce the weak \[O IV\] fluxes show negligible contributions to the \[Ne II\] and \[Ne III\] lines analyzed here. There are two possible exceptions. In NGC 6240, faint \[O IV\] emission is detected but it is stronger than that measured in “normal” starbursts. In NGC 7469, a comparison of \[O IV\] and \[Ne III\] line profiles suggests some AGN contribution to the \[Ne III\] emission. we therefore consider the measured neon ratios for NGC 6240 and NGC 7469 to be upper limits to the emission arising from the starburst region. With these exceptions in mind, we consider the contribution from an AGN to be unlikely across the sample, and any contribution must have a neglible effect on the results presented here.
#### 4.4.2 DIM contributions
Studies of the Milky Way and other nearby galaxies suggest the presence of a “diffuse ionized medium” (DIM), a gas component with a relatively low ionization state and large scale height which permeates the galaxy disk (e.g., Reynolds 1990; Dettmar 1992). It may be difficult to exclude a DIM component as a contributor to the neon line fluxes measured for this sample. Several studies have shown the existence of diffuse, ionized emission in the disks of galaxies, which is generally not associated with individual star forming regions. For the more distant objects in our sample, the aperture covers a large physical area (the long axis of the SWS aperture corresponds to linear diameters of 0.3-14 kpc for the galaxies observed), which may encompass non-starburst emitting regions elsewhere in the galaxy. Thus, we must consider the possibility that some form of DIM emission may influence the integrated line fluxes. Lehnert & Heckman (1994) and Ferguson et al. (1996) showed evidence that the DIM may be produced by ionizing starlight escaping from HII regions in the disks of galaxies; Wang et al. (1997) measured its effects on large scale measurements of optical excitation ratios such as \[NII/H$`\alpha `$\] and \[SII/H$`\alpha `$\]. These studies suggest that the DIM may contribute as much as 50% of the global, integrated H$`\alpha `$ flux in spiral galaxies. If we assume that the DIM consists of ionizing radiation escaping from young star clusters, with the same average parameters as derived by Wang et al. (log U$`4`$, n$`{}_{e}{}^{}`$1 cm<sup>-3</sup>), then the model neon ratios could drop by a factor of two, making the effect of a DIM contribution similar in magnitude to that of variations in the ionization parameter that were discussed in §4.3.
We cannot exclude the contribution from a low-density component, as the \[S III\] ratio generally provides only an upper limit to the density. However, we have a qualitative constraint supplied by the range of neon line ratios observed in our sample: the observed neon ratios show no correlation with distance, as seen in Figure Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE.. If the DIM were a significant contributor, we might expect the neon ratio to decrease with increasing source distance, as the aperture encloses an ever larger physical area. In the nearest galaxy in our sample, IC 342, the aperture encloses an area $``$200 pc in diameter, and for galaxies at a distance of 30 Mpc the aperture still covers a region less than 5 kpc in diameter. Thus, unless we are observing objects in which the size of the starburst area grows in proportion to its distance, we conclude that any DIM-like component in the sample galaxies does not make a significant contribution. Any DIM contributions will result in lower predicted neon ratios; thus, our conclusion that the most massive stars are generally formed in all starburst environments is not affected.
## 5 Discussion and Summary
The cluster models that we have presented in this paper support the formation of very massive stars (50-100$`M_{}`$) in starburst galaxies. While the quantitative estimate of $`m_{\mathrm{up}}`$ for each galaxy is model-dependent, it is clear that the formation of very massive stars is necessary to explain the ionized line diagnostics observed in this starburst sample. This result suggests that while starbursts produce prodigious amounts of energy and stars, the high-mass stellar populations in starburst galaxies are not radically different than those in high-mass star-forming regions observed locally.
### 5.1 Short timescales for starburst activity
As illustrated in Figures Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE. and Massive star formation and evolution in starburst galaxies: mid-infrared spectroscopy with ISO-SWS <sup>1</sup><sup>1</sup>1Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, the Netherlands, and the United Kingdom) and with the participation of ISAS and NASA. The SWS is a joint project of SRON and MPE., the combination of the neon line ratio with the $`L_{\mathrm{IR}}`$/$`L_{Lyc}`$ ratio strongly favors the scenario for starburst activity where very massive stars form, as in local smaller-scale starburst templates, and where the burst last typically a few million to $``$10<sup>7</sup> years. Such timescales are shorter than previously thought (10$`{}_{}{}^{7}10^8\mathrm{yr}`$; e.g., Thronson & Telesco 1986; Heckman 1998). It is clear that detailed modeling is required to secure this result, including additional constraints (e.g. $`K`$-band luminosity, the rate of supernova explosions, the depth of the near-infrared CO bandheads) and spatially resolved information. It is nonetheless in agreement with other recent detailed studies of a few starburst galaxies, some of which are also included in our sample (e.g. M 82, Schreiber 1999; NGC 253, Engelbracht et al. 1998). As a result of instrumental progress, there is now growing evidence that starburst activity occurs in individual burst sites on physical scales of a few tens of parsecs or less. Short timescales are therefore naturally understandable locally. Our data, in conjunction with the other studies cited above, provide evidence for short timescales on much larger scales, suggesting that starburst activity also occurs globally on short timescales, presumably as a result of one brief gas compression event, or of successive episodes of such events separated by more than one typical timescale. Short burst durations thus imply strong negative feedback effects of starburst activity, globally as well as locally.
A simple argument can be invoked to explain the physical arguments behind this result. We can compare the cumulative kinetic energy injected in the ISM by the supernova explosions over time ($`E_{\mathrm{kin}}`$) with the binding energy of the gas ($`E_{\mathrm{grav}}`$), and assume the starburst activity will stop when $`E_{\mathrm{kin}}`$ just balances $`E_{\mathrm{grav}}`$. This is a simplistic way of expressing the conditions for a starburst wind to break out of the galaxy (e.g. Heckman, Armus & Miley 1990), but it is sufficient for order-of-magnitude estimates.
In order to relate $`E_{\mathrm{kin}}`$ to observed quantities, we have considered the relationship between the rate of supernova explosions $`\nu _{\mathrm{SN}}`$ and the bolometric luminosity $`L_{\mathrm{bol}}`$. Model predictions obtained with STARS for a variety of star formation histories and upper mass cutoffs of the IMF indicate that
$$10^{12}\left(\frac{\nu _{\mathrm{SN}}}{\mathrm{yr}^1}\right)\left(\frac{L_{\mathrm{bol}}}{\mathrm{L}_{}}\right)^11$$
$`(4)`$
as soon as the massive stars start to explode as supernovae, and as long as substantial star formation takes place. It thus seems reasonable to assume that Equation (4) holds for the bulk of the sample, likely having a range in age and timescale but all exhibiting signs of significant, recent starburst activity.
For simplicity, we here assume a spherical geometry for the starbursts, with uniform mass distribution. In addition, we assume that each supernova explosion liberates $`E_{\mathrm{mech}}^{\mathrm{SN}}=10^{51}\mathrm{erg}`$ of mechanical energy, transferred as kinetic energy to the ISM with an efficiency $`\eta `$. The timescale $`\tau `$ for our condition above satisfies :
$$\eta \left(\frac{\mathrm{d}E_{\mathrm{mech}}^{\mathrm{SN}}}{\mathrm{d}t}\right)\tau \frac{GM_{\mathrm{dyn}}^2}{R},$$
$`(5)`$
where $`\mathrm{d}E_{\mathrm{mech}}^{\mathrm{SN}}/\mathrm{d}t`$ is the rate of mechanical energy injection from the supernovae, $`G`$ is the gravitational constant, $`M_{\mathrm{dyn}}`$ is the dynamical mass of the system, and $`R`$ is the radius of the starburst region. Equation (5) can be re-written as
$$\frac{\tau }{\mathrm{Myr}}\left(\frac{8.5}{\eta }\right)\left(\frac{M_{\mathrm{dyn}}}{10^9\mathrm{M}_{}}\right)^2\left(\frac{R}{\mathrm{kpc}}\right)^1\left(\frac{L_{\mathrm{IR}}}{10^{10}\mathrm{L}_{}}\right)^1$$
$`(6)`$
where we have substituted $`L_{\mathrm{IR}}`$ for $`L_{\mathrm{bol}}`$, appropriate for our sample galaxies. Application of Equation (6) to M82 ($`M_{\mathrm{dyn}}=8\times 10^8\mathrm{M}_{}`$, $`R=0.25\mathrm{kpc}`$, $`L_{\mathrm{IR}}=4\times 10^{10}\mathrm{L}_{}`$), yields $`\tau 5\eta ^1\mathrm{Myr}`$, so for efficiencies $`10\%`$, the estimated timescales are $`10^610^7\mathrm{yr}`$.
Our argument above is based on “gas-disruption timescale” estimates. This differs from the conventional “gas-consumption” arguments, which compare the star formation rates with the mass of the gas reservoir. In such estimates, the star formation rates are based on comparison of absolute fluxes (e.g. H$`\alpha `$ fluxes, $`L_{\mathrm{IR}}`$) with predictions from starburst models. The estimates are thus very sensitive to the assumed age and history of the starburst. Our estimates of the gas-disruption timescales are also model-dependent, but have the advantage of being based on a quantity ($`10^{12}\nu _{\mathrm{SN}}/L_{\mathrm{bol}}`$) which varies by smaller factors. Neither point of view accounts for further fueling processes, or dynamical evolution of the systems (e.g. starbursts in barred galaxies, interacting/merging systems, etc.). However, the discussion presented here gives an alternative perspective to the issue of global burst timescales, and provides a plausible explanation for our results.
### 5.2 Summary
Starburst models predicting the \[Ne III\]/\[Ne II\] ratio from ISO-SWS spectra of 27 starburst galaxies show that the observed data are consistent with the formation of very massive stars in starbursts, thus precluding the need for the restrictive upper mass cutoffs suggested by some earlier studies ($`m_{\mathrm{up}}`$$``$25-30$`M_{}`$). Combining the neon line ratios with starburst modeling and the consideration of the stellar content measured in local star forming regions, we find that starburst events may be generally described as short bursts of star formation which produce very massive stars, and which exhibit relatively soft integrated line ratios as a result of aging the stellar population.
In particular, our modeling of neon and $`L_{\mathrm{IR}}`$/$`L_{Lyc}`$ ratios, together with results on local high-mass star-forming regions, suggest:
$``$ very massive stars ($`m_{\mathrm{up}}`$$``$50$`M_{}`$) form in typical starbursts.
$``$ starbursts have short global timescales, $`t_{\mathrm{sc}}`$$``$10<sup>7</sup> years.
These results suggest strong negative feedback from starburst activity; the galactic superwinds frequently observed in starburst galaxies are particularly striking examples of the consequences of such feedback.
In our analysis, we have examined the degeneracy between aging effects and model parameter variations in the assessment of upper mass cutoffs to the IMF. There is still room for significant improvements in modeling the properties of starbursts: determination of metallicities and the radiation environment (e.g., for measurements of the ionization parameter U) compete with the characterization of stellar properties (SEDs, evolutionary tracks) as the largest contributors to uncertainty in the modeling of star formation properties such as the upper mass cutoff. Other datasets, such as additional MIR line ratios (e.g., Kunze et al. 1996; Rigopoulou et al. 1996; Engelbracht et al. 1998) or K-band luminosities and near-infrared spectroscopy (e.g., Forbes et al. 1993; van der Werf et al. 1993; Genzel et al. 1995; Tacconi-Garman et al. 1996; Böker, Förster-Schreiber, & Genzel 1997; Engelbracht et al. 1998; Schreiber 1999) would be very useful in further constraining the properties of starbursts. However, it will be important to compile such additional data for a large sample in order to proscribe further, general constraints on the way in which starbursts form stars. Observations with higher spatial resolution would better isolate regions of active star formation, making it possible to confirm whether high- and low-excitation lines arise from the same region; the spectroscopic capabilities of SIRTF will be well-suited to addressing this issue. By accounting for a reasonable range of uncertainties which constrain the present observations, we find that the observed MIR neon ratios are generally consistent with the formation of very massive stars in starburst events; we offer this hypothesis up to future datasets for increasingly rigorous testing.
We would like to thank A. Pauldrach and R.-P. Kudritzki for providing model atmospheres, Tal Alexander for assistance in introducing low-metallicity SEDs into STARS, and Eckhard Sturm for providing the ISO-SWS spectra of Arp 220. We would also like to thank Jack Gallimore and Dan Tran for interesting discussions. MDT would like to thank the Alexander von Humboldt-Stiftung and the NRAO<sup>2</sup><sup>2</sup>2The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. for support during the production of this work. This research also received support from the German-Israeli Foundation under grant I-0551-186.07/97. SWS and the ISO Spectrometer Data Center at MPE are supported by DARA under grants 50 QI 8610 8 and 50 QI 9402 3. |
warning/0003/hep-ph0003260.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The study of the electroweak single top-quark physics is a very important part of research programs at future TeV energy colliders. Such a study allows to investigate with high enough accuracy properties of the top-quark and to measure a $`Wtb`$ coupling structure. It may shed a light on the underlying theory which probably stands beyond the Standard Model . Besides, single top production at LHC has a large rate of the order of 300 pb and therefore it gives an important part of the background to various ”new physics” processes.
In this paper we concentrated on the $`pptW+X`$ production process at the LHC. This process was the subject of the previous studies . In the paper we have calculated $`tW+X`$ process among the others processes important at the Tevatron and LHC colliders. In order to separate $`tWb`$ process of the single top-quark production from the $`t\overline{t}`$ pair production we have introduced a cut on the invariant $`Wb`$ mass window $`(^{}approachI^{})`$. In the paper the cross section of $`tW`$ process was calculated in a different approach where $`t\overline{t}`$ contribution was subtracted from $`tWb`$ process in the narrow width approximation ($`{}_{}{}^{}approachII^{}`$). However, both approaches have different drawbacks and have some aspects which have been treated not quite correctly. Therefore it deserves closer look at the process.
In $`{}_{}{}^{}approachI^{}`$ results are formally $`|m_{WB}m_{top}|`$ cut dependent. In the paper this cut was chosen too modest from the experimental point of view ($`|m_{WB}m_{top}|<3\mathrm{\Gamma }_t`$) since the real experimental mass window for the top-quark is of the order of 20 GeV $`1015\mathrm{\Gamma }_t`$. This was pointed out in the paper . However, in the present study we give the pure theoretical arguments how the cut should be chosen and explain why it has to be significantly larger than the top width. This cut should be of the order of $`20`$ GeV even in case of an ideal detector. After that it will be clear that a formal cut dependence is significantly reduced. Specially one should stress that the ($`{}_{}{}^{}approachI^{}`$) reproduces $`correctly`$ kinematical distributions which is the crucial point for a phenomenological analysis.
The $`{}_{}{}^{}approachII^{}`$ in it’s turn is cut independent but it does not have receipt how to simulate $`Wtb`$ events at all. In the paper the only $`22`$\[$`Wt`$\] process ($`bgWt`$) has been used for an event simulation. However in this case an important part of $`Wtb`$ events are absent and such an implementation of the $`{}_{}{}^{}approachII^{}`$ leads to the wrong kinematical distributions for $`tW+X`$ process. For the numerical values of the cross section the QCD scale $`\widehat{s}`$ has been used. But one should point out that such a scale is too large and leads to significantly lower rate even for top quark pair production as we know from NLO calculations . For single top $`Wt`$ process NLO results have not been obtained yet, however one would expect lower characteristic scale comparing to top pair.
In this paper we have been developed the $`{}_{}{}^{}approachI^{}`$. We apply the reasonable $`|m_{WB}m_{top}|`$ cut consistent with theoretical arguments and an experimental mass resolution for top-quark. We have calculated also the cross sections using $`{}_{}{}^{}approachII^{}`$ for the cross check.
For both methods one needs to apply subtraction procedure to avoid double counting. We have suggested the new procedure of the combining of the two different processes. This method implies the correct subtraction procedure, reproduces the correct kinematical properties in the whole phase space region, and gives stable results.
Our paper is organised as follows. In the section II we compare two different approaches of a subtraction of $`t\overline{t}`$ pair production from the process with $`tWb`$ final state. In section III we develop the new approach for a treatment of double counting and combining two signal processes together. In section IV we present the final results and draw the conclusions.
## 2 Leading order $`22`$ and $`𝒪`$($`1/logm_t^2/m_b^2`$) $`23`$ process: subtraction of $`t\overline{t}`$ pair production
In Fig. 1 we present the complete gauge invariant set of leading order and $`𝒪`$($`1/logm_t^2/m_b^2`$) diagrams contributing to the $`tW+X`$ final state.
There are two problems one should avoid in order to combine correctly different contributions to the $`tW`$ single top production: one should remove the contribution from $`t\overline{t}`$ production, giving the same $`tWb`$ final state and take care about the double counting which takes place when one simply adds contribution from two $`22`$ and $`23`$ processes.
This happens because $`23`$ process is singular in the region of collinear b-quarks coming from gluon splitting. The same singularity has been resolved for $`22`$ process when b-quark PDF was defined and collinear contributions of b-quark was resumed. The contribution from the collinear region should be taken only once, and therefore one should apply a subtraction procedure. It should be noticed that $`22`$ and $`23`$ processes have overlapping only for the leading log gluon splitting term.
In this section we would like to compare two different approaches of solving the first problem – subtracting the contribution from the top-quark pair production. As for double counting, we use the conventional solution in this section , namely, we use the subtraction of gluon splitting term:
$$\sigma (gb+ggtW+X)_{real}=\sigma (gbtW)+\sigma (ggtW\overline{b})\sigma (gb\overline{b}gbtW)$$
(1)
As it was mentioned in the introduction, there are two basic approaches to remove the contribution from top-quark pair production in a gauge invariant way. The first ($`{}_{}{}^{}approachI^{}`$), is the application of the cut on the invariant mass of $`Wb`$– pair in order to remove the resonant $`t\overline{t}`$ contribution. This procedure is cut dependent, but it has the straightforward receipt how to simulate single-top quark production events with the proper kinematics. One should note however that the cut dependence is not arbitrary since the cut should be applied according to well known mass resolution which is typically 10-15 GeV. In terms of the top-quark width the window cut should be applied to remove the $`t\overline{t}`$ contribution would be of the order of $`\pm 20`$ GeV $``$ $`1015\mathrm{\Gamma }_{top}`$.
The second ($`{}_{}{}^{}approachII^{}`$) way of subtraction of $`t\overline{t}`$ contribution is the narrow width limit approach :
$$\sigma (ggtWb)_{singletop}=\sigma (ggtWb)_{total}\sigma (ggt\overline{t})Br(tWb)interf[t\overline{t}tWb],$$
(2)
where $`interf[t\overline{t}tWb]`$ means interference of $`t\overline{t}`$ diagrams with the non-resonant ones. This procedure formally should reproduce the correct production rate for the single top quark. But from the practical point of view, it does not give any receipt how to simulate events of the single top-quark production which is crucial for the further kinematical studies.
Table 1 shows the results for two methods of subtraction of the $`t\overline{t}`$ contribution. In order to give the idea how strong is the dependence on the $`Wb`$ invariant mass we present numbers for two – $`10`$ and $`15\mathrm{\Gamma }_{top}`$ window cuts which corresponds to $`\pm 16`$ and $`24`$ GeV mass windows respectively. All the numerical results have been obtained by means of the program CompHEP .
Results in the table are shown for several characteristic QCD factorisation/normalisation scales: $`Q=m_W,m_{top},m_{top}+m_W`$ GeV which give a natural scale interval for the process under study. We also show in the last column the results at $`Q^2=\widehat{s}`$ for a comparison to previous calculations. From the table one can see that the subtraction term $`gb\overline{b}(gb+bg)tW^{}`$ is of the order of $`80\%`$ of $`(gb+bg)W^{}t`$ cross section. One can also see that $`{}_{}{}^{}approachII^{}`$ of subtraction of $`t\overline{t}`$ contribution gives cross section for $`W^{}t+X`$ process consistent with the $`{}_{}{}^{}approachI^{}`$ for $`\pm 15\mathrm{\Gamma }_{top}`$ $`W^{}b`$ mass window cut. For example, one has 31.0 and 28.9 pb respectively for these two methods at $`\mu =m_{top}`$. Since the $`{}_{}{}^{}approachI^{}`$ is more physical in the sense that it allows to reproduce the correct kinematical distribution we use it with $`\pm 15\mathrm{\Gamma }_{top}`$ $`W^{}b`$ mass window cut for the final results. One could also take a look at the $`Wb`$ invariant mass distribution at the parton level which is presented in Fig. 2. From this figure one can clearly see that $`25GeVWb`$ mass window would completely remove $`t\overline{t}`$ contribution with its interference to the $`ggtW^{}\overline{b}`$ processes.
One could easily perform the fitting procedure which leads to the more quantitative answer about the size of this window and gives those 25 GeV. This procedure significantly reduce the ambiguity of the choice of this window cut which is also of the order of the experimental mass resolution mentioned above.
Contribution from $`qqtWb`$ process has been also taken into account in our study. The contribution from this process is not negligible and is of the order of 7% to the $`tWb`$ final state after the removing $`t\overline{t}`$ contribution.
One should also notice that cross sections are quite scale dependent. For three QCD scales: $`Q=m_W,m_{top},m_{top}+m_W`$ GeV the uncertainty due to the choice of different scales are of the order of 25-30%. The choice of the scale $`Q=\sqrt{\widehat{s}}`$ gives significantly lower results. But we should stress that high QCD scale $`Q=\sqrt{\widehat{s}}`$ seems to be unphysical since it gives almost factor two lower cross section even for $`t\overline{t}`$ production at tree level in comparison with the next-to-leading(NLO) order result . For $`Q=m_{top}=175`$ GeV tree level result is much close to NLO one. Therefore it is quite reasonable to use $`Q=m_{top}=175`$ GeV choice for the processes involving single top-quark production for which the physical scale could be even smaller then for the $`t\overline{t}`$ pair production.
## 3 Treatment of the double counting: comparison and combining of the $`Wt`$+ISR and complete $`tWb`$ processes
In this section we would like make close look at the solution of the double counting problem.
Results from in Table 1 have been obtained using ’conventional’ subtraction procedure. However one need to study how $`22`$ and $`23`$ processes should be combined in order to reproduce not only the total cross section but also the correct event kinematics.
We have used the following procedure to work out the receipt for this. We have compared various kinematical distributions of $`pp(bg)tW^{}`$\+ $`b_{ISR}`$$`22`$ process with an additional b-quark from the initial state radiation and complete $`23`$$`pp(gg+q\overline{q})tW^{}\overline{b}`$ process. In this way one can try to find a proper matching between resumed contribution at the collinear region for the b-quark and complete tree level contribution at the hard region. Figure 3 shows transverse momenta and rapidity distributions for all three particles in the final state. As expected one can see the difference in the b-quark distributions. For $`pp(bg)tW^{}`$\+ $`b_{ISR}`$ process it is much softer and less central in comparison to the $`pp(gg+q\overline{q})tW^{}\overline{b}`$ process. In the same time one can see that $`Wboson`$ and $`t`$quark distributions are nearly the same.
Since we know the absolute value of the combined cross section we propose the following method to match collinear and hard kinematical regions. One can use kinematical $`p_T^b`$ separation of $`pp(bg)tW^{}`$\+ $`b_{ISR}`$ and $`pp(gg+q\overline{q})tW^{}\overline{b}`$ in the regions $`p_T^b<P_T^{cut}`$ and $`p_T^b>P_T^{cut}`$ respectively.
Now one can move the cut and try to satisfy two requirements, namely:
1) the common rate of $`pp(bg)tW^{}`$\+ $`b_{ISR}`$ with $`p_T^b<P_T^{cut}`$ and $`pp(gg+q\overline{q})tW^{}\overline{b}`$ with $`p_T^b>P_T^{cut}`$ gives the combined total rate computed in previous section, in other words one can normalise a rate in a collinear region on the $`\sigma _{total}\sigma [pp(gg+q\overline{q})tW^{}\overline{b},p_T^b>P_T^{cut}]`$;
2) the overall $`p_T^b`$ distribution should be smooth.
The result is illustrated in Fig. 4 where we show several variants of combining those two processes for various values of $`P_T^{cut}`$. We have found that the optimal $`P_T^{cut}`$ providing the smooth sewing for these two processes at the LHC is equal to 20 GeV. This value gives physically reasonable answer in which regions $`pp(bg)tW^{}+b_{ISR}`$ and $`pp(gg+q\overline{q})tW^{}\overline{b}`$ processes should be considered.
We conclude that the method of combining of the $`p_T^b`$ distribution of $`Wt`$+ISR gluon and complete tree level $`tWb`$ process allows to find the physically motivated $`p_T`$ cut on the $`bquark`$ which allows us to treat together those processes and simulate them in different kinematical regions of $`p_T^b`$.
We have estimated uncertainties due to a the choice of the QCD scale within the range $`M_W<\mu <M_{TOP}+M_W`$ taking the central value of $`\mu =M_{TOP}`$. The total cross section presented in Table 1 is $`31.0_{1.8}^{+8.3}`$ pb within the QCD scale mentioned above.
## 4 Final results and conclusions
We have reexamined the $`tW+X`$ single top-quark production process which is important at the LHC. We have shown that $`23`$\[$`Wtb`$\] process has to be correctly taken into account with a proper subtraction of the top pair contribution and that it has qualitatively different kinematical distributions from the $`22`$\[$`Wt`$\] process.
We suggest the new method of ’kinematical’ sewing of two different processes contributing to the $`tW+X`$ productions using the transverse b-quark momenta distribution. This method allows unambiguously simulate correct kinematical distribution of the total process of $`tW+X`$ production in the whole kinematical region.
We have estimated the cross section of the single top $`tW+X`$ production taking into account uncertainties due to the choice of the QCD scale. The cross section for single top and single anti-top quark production – $`tW^{}+X`$ and $`\overline{t}W^++X`$ at the LHC are equal to each other in contrary to other processes of the single top-quark production. So, combined \[$`tW^{}+X`$ \+ $`\overline{t}W^++X`$\] cross section is $`62.0_{3.6}^{+16.6}`$ pb.
## Acknowledgements
We thank C.-P. Yuan for useful remarks and discussions. E.B. is grateful to the Russian Ministry of Science and Technologies, Russian Foundation for Basic Research (grant 99-02-04011), INTAS-CERN Foundation (grant No 377), and the German Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie (BMBF) (project no. HTE0499 Manakos) for partial financial support. E.B. would like to thank P. Manakos and Th. Ohl for their kind hospitality during a visit to Darmstadt University of Technology where the paper has been completed. |
warning/0003/hep-th0003116.html | ar5iv | text | # LMU-TPW 00-9UAHEP 00-4hep-th/0003116 A note on gravity-scalar fluctuations in holographic RG flow geometries
## 1 Introduction
The AdS/CFT correspondence provides a powerful method for computing correlation functions of gauge invariant operators in $`𝒩=4`$ supersymmetric Yang-Mills theory (YM) in four dimensions at large $`N`$ and at strong ’t Hooft coupling. A question of obvious interest is to extend the correspondence and in particular the prescription for computing Green functions to a general class of theories without (or with spontaneously broken) superconformal invariance. Such theories have a renormalization group flow and they generically arise either from deforming YM away from the conformal point by adding to the Lagrangian IR relevant perturbations - or by giving vacuum expectation values to scalar fields -. In the gauged supergravity description the corresponding bulk geometries are supported in both cases by 4-dimensional Poincaré invariant kink solutions approaching asymptotically AdS space (see for a recent extensive review of the subject).
Recently a gravity computation of some two-point correlation functions in field theories with renormalization group flow was presented in . This computation amounts to the study of the coupled gravity-scalar equations describing fluctuations of the scalars and the metric around a kink solution. The background scalars naturally fall into two classes: the scalars with a non-trivial dependence on the fifth dimension which, in the terminology of , are called “active” and the constant (or vanishing) “inert” scalars. In striking differences between the expected correlation functions of YM operators dual to inert and to active scalars were observed. The correlation functions of the operators dual to the inert scalars are straightforward to compute and they provide a consistent description of the spectrum in the boundary field theory. Quite opposite, the correlation functions of the operators dual to active scalars were then found to be physically unreasonable. The authors of suggested that this might be due to the presence of the metric singularity in the interior, which in turn might invalidate the standard AdS prescription for computing correlation functions. Adding the standard supplementary gravity boundary terms did not clarify the situation.
In this note we reconsider the two-point correlation functions for operators dual to active scalars. We use the Hamiltonian formulation of the AdS/CFT correspondence to deal with boundary terms and the standard prescription for computing correlation functions in the gravity approximation . Another point of deviation from the analysis in is our choice of gauge.
We start by analyzing the quadratic action for fluctuations of the scalar fields and the metric near their background values. All the boundary terms one could add to the gravity action are unambiguously fixed by the Hamiltonian version of the AdS/CFT correspondence that requires the gravity action to be schematically of the form $`𝑑t(p\dot{q}H(p,q))`$, where $`H`$ is the Hamiltonian and $`t`$, the coordinate in the bulk direction, plays the role of the time. For the quadratic action this prescription implies that it does not contain any gravity or scalar fields with second derivative in the bulk direction. All such terms are integrated by parts and arising boundary terms are simply discarded.
For the sake of simplicity we consider the case of a unique active scalar $`\overline{\varphi }`$ and pick up the “almost” radial gauge for which the scalar fluctuation $`\varphi `$ is set to zero. The residual gauge symmetry is enough to decouple the transverse components of the graviton and it leads to the conservation law for the stress-energy tensor in the boundary field theory. Except for the traceless transverse graviton the only physical degree of freedom is then the trace of the graviton $`h`$ for which we obtain a very simple action, basically due to the existence of a superpotential.
The trace of the graviton is usually viewed as the gravity field dual to the trace of the stress-energy tensor in the field theory away from the conformal point. However, we show that by using a field redefinition one can recast the action for $`h`$ into the standard action for a scalar field $`s`$ in the kink background with some complicated potential. Depending of the form of the kink solution, the scalar $`s`$ may then be naturally interpreted as the dual either to the YM operator invoking RG flow or to the corresponding field theory operator with a non-trivial vacuum expectation value. The relation between $`h`$ and $`s`$ appears to be $`h\beta ^1s`$, where $`\beta `$ is the holographic beta function introduced in . We believe that a similar relation also occurs for general backgrounds with many active scalars .
We then consider the same two kink solutions as in . The first one corresponds to $`𝒩=4`$ SYM, perturbed by an operator of dimension 3, that flows in the IR to $`𝒩=1`$ SYM. The second kink involves one active scalar from the $`\mathrm{𝟐𝟎}`$ of $`SO(6)`$ and the corresponding dual flow describes the states on the Coulomb branch of $`𝒩=4`$ SYM, which is parametrized by the vacuum expectation value of one of the operators $`O_2=\mathrm{tr}(X^{(I}X^{J)})`$, where $`X^I`$ are YM scalars.
Evaluating the on-shell action for these supergravity solutions we find that in both cases the two-point functions exhibit the same behaviour as was found for the inert scalars in . Indeed, in the first case we get a discrete spectrum while in the second example the spectrum is continuous with a mass gap.
## 2 Gravity/active scalar system
Consider $`(d+1)`$-dimensional gravity, described by a metric $`G_{\mu \nu }`$ with signature $`(1,1,\mathrm{},1,1)`$, minimally coupled to scalar fields $`\phi ^I`$. The action is of the form <sup>1</sup><sup>1</sup>1Our conventions are: $`[_\mu ,_\nu ]V_\rho =R_{\mu \nu \rho }{}_{}{}^{\sigma }V_{\sigma }^{}`$ and $`R_{\mu \rho }=R_{\mu \nu \rho }^\nu `$.
$`S={\displaystyle d^{d+1}x\sqrt{G}\left(\frac{1}{4}R\frac{1}{2}G^{\mu \nu }_\mu \phi ^I_\nu \phi ^IV(\phi )\right)};`$ (2.1)
$`V(\phi )`$ is the scalar potential. The equations of motion that follow from this action are
$`R_{\mu \nu }{\displaystyle \frac{1}{2}}G_{\mu \nu }R2_\mu \phi ^I_\nu \phi ^I+G_{\mu \nu }_\rho \phi ^I^\rho \phi ^I+2G_{\mu \nu }V=0,`$ (2.2)
$`^\mu _\mu \phi ^I{\displaystyle \frac{V}{\phi ^I}}=0.`$ (2.3)
They imply
$`R2_\rho \phi ^I^\rho \phi ^I=4{\displaystyle \frac{d+1}{d1}}V,`$ (2.4)
$`R_{\mu \nu }2_\mu \phi ^I_\nu \phi ^I={\displaystyle \frac{4V}{d1}}G_{\mu \nu }.`$ (2.5)
Let $`g_{\mu \nu }`$ and $`\overline{\varphi }^I`$ be solutions of the equations of motion and decompose $`G_{\mu \nu }`$ and $`\phi `$ around their background values
$$G_{\mu \nu }=g_{\mu \nu }+h_{\mu \nu },\phi ^I=\overline{\varphi }^I+\varphi ^I.$$
Discarding total derivative terms, we find the following quadratic action for the fluctuations
$`S_2={\displaystyle d^{d+1}x\sqrt{g}}`$ $`[`$ $`{\displaystyle \frac{1}{4}}({\displaystyle \frac{1}{4}}_\rho h_{\mu \nu }^\rho h^{\mu \nu }+{\displaystyle \frac{1}{2}}_\rho h_{\mu \nu }^\mu h^{\rho \nu }`$ (2.6)
$``$ $`{\displaystyle \frac{1}{2}}^\nu h^\mu h_{\mu \nu }+{\displaystyle \frac{1}{4}}_\mu h^\mu h+{\displaystyle \frac{2V}{d1}}h_{\mu \nu }^2{\displaystyle \frac{V}{d1}}h^2)`$ (2.7)
$`+`$ $`h^{\mu \nu }_\mu \varphi ^I_\nu \overline{\varphi }^I+{\displaystyle \frac{1}{2}}_\mu h\varphi ^I^\mu \overline{\varphi }^I`$ (2.8)
$``$ $`{\displaystyle \frac{1}{2}}_\mu \varphi ^I^\mu \varphi ^I{\displaystyle \frac{1}{2}}V_{IJ}\varphi ^I\varphi ^J].`$ (2.9)
The covariant derivative $`_\mu `$ is with respect to the background metric $`g_{\mu \nu }`$ which is also used to raise and lower indices, $`h=h_\mu ^\mu `$, and we have introduced the notation
$$V_I=\frac{V}{\phi ^I}|_{\phi =\overline{\varphi }},V_{IJ}=\frac{^2V}{\phi ^I\phi ^J}|_{\phi =\overline{\varphi }}.$$
Within the context of the AdS/CFT correspondence, we are interested in background solutions which respect $`d`$-dimensional Poincaré invariance. Thus, we make an ansatz for the background solving (2.3) of the form
$`ds^2=dx_0^2+e^{2A(x_0)}\eta _{ij}dx^idx^j,`$ (2.10)
where $`\eta _{ij}`$ is the Minkowski metric and take $`\overline{\varphi }`$ to depend only on $`x_0`$.
The analysis of the action and the equations of motion is complicated by the fact that the scalar fields couple to the graviton already in the quadratic action. Therefore, the spectrum of the theory on this background cannot be readily read off the action or the equations. However, the problem of finding the spectrum of the theory can be simplified by an appropriate gauge choice. The quadratic action and the equations of motion are invariant under gauge transformations induced by reparametrizations:
$`\delta h_{\mu \nu }=_\mu \zeta _\nu +_\nu \zeta _\mu ,\delta \varphi ^I=\zeta ^\mu _\mu \overline{\varphi }^I.`$ (2.11)
One usually imposes the temporal gauge $`h_{0\mu }=0`$ and solves the constraints, i.e. the equations of motion for $`h_{0\mu }`$. This was done in . However, this gauge choice leads to a very complicated system of equations and, moreover, it was only possible to derive a third-order equation for scalar fields. On the other hand, the gauge transformations (2.11) of scalar fields show that on backgrounds with at least one nonvanishing field $`\overline{\varphi }^1\overline{\varphi }`$ one can impose the almost temporal gauge $`h_{0i}=0=\varphi `$. This gauge choice is very natural, because the combination $`\stackrel{~}{h}_{00}h_{00}2_0(\varphi /_0\overline{\varphi })`$ is gauge invariant, and one could express the action and the equations of motion in terms of $`\stackrel{~}{h}_{00}`$ rather than $`h_{00}`$. Then only the scalar fields change under gauge transformations.
For the sake of simplicity we restrict ourselves to considering the simplest case of one active scalar. Then the action depends only on $`h_{ij}`$ and $`h_{00}`$. It is convenient to introduce $`t_{ij}`$ via
$$h_{ij}=e^{2A}t_{ij},h^{ij}=e^{2A}t^{ij},tt_i^i=t_{ij}\eta ^{ij}.$$
In what follows we will not distinguish upper and lower indices. It is straightforward to derive
$`_0h_{00}=_0h_{00},_ih_{00}=_ih_{00},_0h_{0i}=0,`$ (2.12)
$`_ih_{0j}=e^{2A}_0A(\eta _{ij}h_{00}t_{ij}),_0h_{ij}=e^{2A}_0t_{ij},_ih_{kl}=e^{2A}_it_{kl}`$
and to cast the action (2.9) in the form
$`S_2`$ $`=`$ $`{\displaystyle }d^{d+1}x{\displaystyle \frac{1}{4}}e^{dA}[{\displaystyle \frac{1}{4}}(_0t_{ij})^2+{\displaystyle \frac{1}{4}}(_0t)^2{\displaystyle \frac{1}{4}}e^{2A}(_it_{kl})^2+{\displaystyle \frac{1}{2}}_0^2At_{ij}^2`$ (2.13)
$`+`$ $`{\displaystyle \frac{d}{2}}(_0A)^2t_{ij}^2+{\displaystyle \frac{1}{2}}e^{2A}_it_{kl}_kt_{il}{\displaystyle \frac{1}{4}}_0^2At^2{\displaystyle \frac{d}{4}}(_0A)^2t^2`$ (2.14)
$`+`$ $`{\displaystyle \frac{1}{2}}e^{2A}t_i_jt_{ij}+{\displaystyle \frac{1}{4}}e^{2A}(_it)^2+{\displaystyle \frac{2V}{d1}}t_{ij}^2{\displaystyle \frac{V}{d1}}t^2`$ (2.15)
$`+`$ $`h_{00}\left({\displaystyle \frac{1d}{2}}_0A_0t{\displaystyle \frac{1}{2}}_0^2At{\displaystyle \frac{d}{2}}(_0A)^2t+{\displaystyle \frac{1}{2}}e^{2A}_i_jt_{ij}{\displaystyle \frac{1}{2}}e^{2A}_i^2t{\displaystyle \frac{2V}{d1}}t\right)`$ (2.16)
$`+`$ $`h_{00}({\displaystyle \frac{d}{4}}_0^2A+{\displaystyle \frac{d^2}{4}}(_0A)^2+{\displaystyle \frac{V}{d1}})h_{00}].`$ (2.17)
We see from this action that $`h_{00}`$ is a non-dynamical field, and can thus be integrated out. On the other hand we have the constraints that follow from the $`h_{0i}`$ equations of motion and from the equation for $`\varphi `$. They are
$`(1d)_0A_ih_{00}+_0(_it_jt_{ji})=0,`$ (2.18)
$`_0h_{00}_0\overline{\varphi }+2(d_0A_0\overline{\varphi }+_0^2\overline{\varphi })h_{00}_0t_0\overline{\varphi }=0.`$ (2.19)
These constraints allow us to express $`h_{00}`$ through $`t_{ij}`$ and therefore, they should be compatible with the equation of motion for $`h_{00}`$ that follows from (2.17).
From here on we restrict ourselves to the most interesting case $`d=4`$, and to a potential $`V(\phi )`$ which can be derived from a superpotential $`W(\phi )`$,
$$V(\varphi )=\frac{g^2}{8}\left(\frac{W}{\varphi }\right)^2\frac{g^2}{3}W^2.$$
(2.20)
All explicitly known backgrounds are obtained from such a potential. One can show , that any solution to the equations
$$_0A=\frac{g}{3}W,_0\overline{\varphi }=\frac{g}{2}\frac{W}{\overline{\varphi }},g=\frac{2}{L},$$
(2.21)
also satisfies the equations of motion (2.3). The length scale $`L`$ is related to the cosmological constant $`\mathrm{\Lambda }`$ via $`\mathrm{\Lambda }=12/L^2=4V(\phi =0)`$. It is not difficult to verify that these relations lead to the identity
$`_0^2A+4(_0A)^2+{\displaystyle \frac{4}{3}}V=0.`$ (2.22)
This identity simplifies the action (2.17) considerably and it now takes the form
$`S_2`$ $`=`$ $`{\displaystyle }d^5x{\displaystyle \frac{1}{4}}e^{4A}[{\displaystyle \frac{1}{4}}(_0t_{ij})^2+{\displaystyle \frac{1}{4}}(_0t)^2{\displaystyle \frac{1}{4}}e^{2A}(_it_{kl})^2+{\displaystyle \frac{1}{2}}e^{2A}_kt_{kl}_it_{il}`$ (2.23)
$`+`$ $`{\displaystyle \frac{1}{2}}e^{2A}t_i_jt_{ij}+{\displaystyle \frac{1}{4}}e^{2A}(_it)^2`$ (2.24)
$`+`$ $`h_{00}({\displaystyle \frac{3}{2}}_0A_0t+{\displaystyle \frac{1}{2}}e^{2A}_i_jt_{ij}{\displaystyle \frac{1}{2}}e^{2A}_i^2t)Vh_{00}^2].`$ (2.25)
It is well-known (see, e.g. ) that the transverse traceless components of the metric fluctuations decouple, and are described by the same equation as a free minimally-coupled massless scalar in the background (2.10). The physical reason for the decoupling is that due to the Lorentz invariance the boundary stress tensor is conserved and, therefore, only the transverse traceless components can couple to it. For this reason it is convenient to introduce the following decomposition of the graviton
$$t_{ij}=t_{ij}^{}+t_{ij}^{||}+\frac{1}{4}h\eta _{ij}_i_jH.$$
Here $`t_{ij}^{}`$ is the traceless transverse part and $`t_{ij}^{||}`$ is a traceless longitudinal part given by
$`t_{ij}^{||}={\displaystyle \frac{_i}{\mathrm{}}}_kt_{jk}+{\displaystyle \frac{_j}{\mathrm{}}}_kt_{ik}2{\displaystyle \frac{_i_j}{\mathrm{}^2}}_k_mt_{km};`$ (2.26)
it satisfies $`_i_jt_{ij}^{||}=0`$.
Substituting this decomposition in the action one can easily see that the transverse traceless components do decouple, and in what follows we will drop them. Moreover, the longitudinal traceless components also decouple and the only remaining coupled fields are $`h`$ and $`H`$.
To analyze their action we introduce their Fourier transforms
$$t_{ij}(x_0,x)=\frac{1}{4\pi ^2}d^4pe^{ipx}t_{ij}(x_0,p),h_{00}(x_0,x)=\frac{1}{4\pi ^2}d^4pe^{ipx}h_{00}(x_0,p).$$
In momentum space the constraints (2.18) take the form
$`3_0Ah_{00}p_i+{\displaystyle \frac{3}{4}}p_i_0hp_j_0t_{ij}^{||}=0.`$
From here one finds that
$$_0Ah_{00}=\frac{1}{4}_0h.$$
(2.27)
and $`t_{ij}^{||}`$ does not depend on $`x_0`$. Thus, $`t_{ij}^{||}`$ are not dynamical modes and do not couple to any operators in the boundary theory. They may therefore be omitted.
We are left with the graviton modes $`t_{ij}`$ of the form
$$t_{ij}=\frac{1}{4}\eta _{ij}h+p_ip_jH,$$
whose dynamics is described by the action
$`S_2`$ $`=`$ $`{\displaystyle }dx_0d^4p{\displaystyle \frac{1}{4}}e^{4A}[{\displaystyle \frac{3}{16}}(_0h)^2+{\displaystyle \frac{3}{8}}p^2_0h_0H+{\displaystyle \frac{3}{32}}e^{2A}p^2h^2`$ (2.28)
$`+`$ $`h_{00}({\displaystyle \frac{3}{2}}_0A(_0h+p^2_0H)+{\displaystyle \frac{3}{8}}e^{2A}p^2h)Vh_{00}^2].`$ (2.29)
Making the following shift of $`h_{00}`$
$$h_{00}h_{00}+\frac{_0h}{4_0A},$$
(2.30)
we rewrite (2.29) as
$`S_2`$ $`=`$ $`{\displaystyle }dx_0d^4p{\displaystyle \frac{1}{4}}e^{4A}[{\displaystyle \frac{3}{16}}(_0h)^2{\displaystyle \frac{3}{2}}p^2_0Ah_{00}_0H+{\displaystyle \frac{3}{32}}e^{2A}p^2h^2`$ (2.31)
$`+`$ $`(h_{00}+{\displaystyle \frac{_0h}{4_0A}})({\displaystyle \frac{3}{2}}_0A_0h+{\displaystyle \frac{3}{8}}e^{2A}p^2h)V(h_{00}+{\displaystyle \frac{_0h}{4_0A}})^2].`$ (2.32)
Thus $`h_{00}`$ is a momentum for $`H`$ and the constraint (2.27), which, after the shift (2.30) reads $`h_{00}=0`$, shows that $`H`$ is not a dynamical field. So we can set $`h_{00}=0`$ and obtain, using again (2.22), the final action for $`h`$
$`S_2={\displaystyle 𝑑x_0d^4p\frac{1}{4}e^{4A}\frac{3}{64}\frac{_0^2A}{(_0A)^2}\left((_0h)^2+e^{2A}p^2h^2\right)}.`$ (2.33)
Some comments are in order. The action (2.33) has the correct overall sign as $`_0^2A`$ is always negative . The terms in parentheses are exactly the same as for the transverse traceless components. Actually the only difference between the action (2.33) and the action for the transverse traceless components is in the factor
$$\frac{3}{16}\frac{_0^2A}{(_0A)^2}=\frac{9}{32}(\frac{}{\overline{\varphi }}\mathrm{log}W)^2=\frac{1}{8}\left(\frac{d\overline{\varphi }}{dA}\right)^2=\frac{1}{8}\beta ^2,$$
where $`\beta =\frac{d\overline{\varphi }}{dA}`$ is the holographic beta function introduced in . One can remove this factor by rescaling $`h`$
$$h=\frac{8}{\beta }s=8\sqrt{\frac{2(_0A)^2}{3_0^2A}}s.$$
The new field $`s`$ is a scalar with proper transformation properties under reparametrizations, however, it has a complicated potential $`U(x_0)`$, and is described by the action
$`S_2={\displaystyle 𝑑x_0d^4p\frac{1}{2}e^{4A}\left((_0s)^2+(e^{2A}p^2+U(x_0))s^2\right)},`$ (2.34)
where
$`U(x_0)=2\left({\displaystyle \frac{A^{\prime \prime }}{A^{}}}\right)^22{\displaystyle \frac{A^{\prime \prime \prime }}{A^{}}}{\displaystyle \frac{1}{4}}\left({\displaystyle \frac{A^{\prime \prime \prime }}{A^{\prime \prime }}}\right)^2+{\displaystyle \frac{1}{2}}{\displaystyle \frac{A^{(iv)}}{A^{\prime \prime }}}+2{\displaystyle \frac{A^{\prime \prime \prime }A^{}}{A^{\prime \prime }}}4A^{\prime \prime }.`$ (2.35)
The appearance of the relative factor between $`h`$ and $`s`$ is indeed very natural, because the trace of the graviton $`h`$ is dual to the trace of the stress tensor and, as was discussed in , in a deformed conformal field theory we expect to have an operator relation
$$T_i^i(x)=\beta 𝒪(x),$$
where $`𝒪`$ is the operator responsible for the deformation of the conformal field theory. In other words, the coupling of $`h`$ to the trace of the stress-energy tensor is $`_\epsilon hT=_e\frac{1}{\beta }sT_\epsilon s𝒪`$. Therefore, we see that if $`h`$ is dual to the trace of the stress tensor then the scalar $`s`$ is dual to the operator $`𝒪`$.
It is worth noting that we have no linear term in the final action. Linear terms do not appear because they can come only from total-derivative bulk terms, but we omit any such a term following the prescription of . Moreover, according to , we do not need to add any boundary term to the actions (2.33) or (2.34), and these actions are appropriate for computing the 2-point function of the operator $`𝒪`$. The absence of linear terms leads to the vanishing of the one-point correlator of the operator $`𝒪`$ dual to the scalar $`s`$. This may look strange because one usually says that in a perturbed conformal field theory an operator dual to an active scalar has a nonvanishing one-point function. Nevertheless, we work in flat 4-dimensional space, and, therefore, we can always choose such a substraction scheme that a one-point function of any local operator vanishes. This seems to mean that the scalar $`s`$ is actually dual to the operator $`𝒪𝒪`$.
Deriving the action (2.33), we have not used the constraint (2.19) yet. One may wonder if this constraint imposes additional restrictions on admissible configurations of the gravity fields. In the appendix we show that this constraint follows from the equations of motion for $`h`$ and $`H`$, and from the $`h_{0i}`$-constraints (2.18), and, therefore, can be omitted.
Next we use the action (2.33) to compute 2-point functions in the two cases recently studied in . Recall that the 2-point functions of active scalars obtained in appear to be problematic. We will see that action (2.33) in both cases leads to reasonable 2-point functions.
We begin with the case considered in section 3 of . The supergravity solution discussed there was found in , and describes the renormalization group flow of $`𝒩=4`$ SYM theory to $`𝒩=1`$ SYM theory at long distances. We refer the reader to for details.
To simplify the equations of motion of the scalar field and its solution, a new coordinate $`u`$ was introduced in such that
$$e^{2A}=\frac{u}{1u},\frac{du}{dx_0}=\frac{2}{L}(1u),\frac{dA}{dx_0}=\frac{1}{Lu}.$$
In this coordinate the boundary of the 5-d space is at $`u=1`$, and there is a singularity at $`u=0`$. With these formulas we rewrite the action (2.33) as
$`S_2={\displaystyle \frac{3}{64L}}{\displaystyle 𝑑ud^4pu^2\left((_uh)^2+\frac{p^2L^2}{4u(1u)}h^2\right)}={\displaystyle \frac{3}{64L}}{\displaystyle 𝑑ud^4p_u(u^2h_uh)},`$ (2.36)
where in the second step the equation of motion for $`h`$
$$_u^2h+\frac{2}{u}_uh\frac{p^2L^2}{4u(1u)}h=0$$
(2.37)
has been used. To compute the 2-point function we follow the standard AdS/CFT prescription . Imposing Dirichlet boundary conditions at $`u=1\epsilon ^2`$, we obtain the solution to (2.37), regular at $`u=0`$
$$h(u,p)=\frac{F(a_{},a_+;2;u)}{F(a_{},a_+;2;1\epsilon ^2)}h(p),$$
(2.38)
where
$$a_\pm =\frac{1}{2}(1\pm \sqrt{1p^2L^2})$$
and $`F(a,b;c,u)`$ is the hypergeometric function $`{}_{2}{}^{}F_{1}^{}`$. The 2-point function of $`h`$ in momentum space is given by the familiar formula
$`O(p)O(p)=\mathrm{lim}_{\epsilon 0}{\displaystyle \frac{3}{32L}}{\displaystyle \frac{_uF(a_{},a_+;2;u)}{F(a_{},a_+;2;u)}}|_{u=1\epsilon ^2}.`$ (2.39)
To find the 2-point function we need
$$\frac{d}{du}(uF(a,b;2;u))=F(a,b;1;u),F(a_{},a_+;2;1)=\frac{1}{\mathrm{\Gamma }(2a_{})\mathrm{\Gamma }(2a_+)},$$
and the expansion
$$F(a_{},a_+;1;u)=\frac{1}{\mathrm{\Gamma }(a_{})\mathrm{\Gamma }(a_+)}\left(2\mathrm{\Psi }(1)\mathrm{\Psi }(a_{})\mathrm{\Psi }(a_+)2\mathrm{l}\mathrm{o}\mathrm{g}\epsilon +o(\epsilon )\right).$$
Then, omitting all terms polynomial in $`p`$, we find in the limit $`\epsilon 0`$
$`O(p)O(p)={\displaystyle \frac{3}{32L}}{\displaystyle \frac{p^2L^2}{4}}\left(\mathrm{\Psi }({\displaystyle \frac{1}{2}}(1\sqrt{1p^2L^2}))+\mathrm{\Psi }({\displaystyle \frac{1}{2}}(1+\sqrt{1p^2L^2}))\right).`$ (2.40)
The correlator has a discrete spectrum of poles at $`p^2=4n(n+1)/L^2`$, $`n=0,1,\mathrm{}`$. A similar discrete spectrum was found in by studying 2-point correlators of inert scalars. Thus contrary to the result in , we obtain for the active scalar a 2-point correlator with the expected behaviour.
Next we consider the second case, c.f. section 4 of , where the flow obtained in was used to analyze the 2-point function of an active scalar. This flow changes the vacuum of $`𝒩=4`$ SYM which is now on the Coulomb branch.
A new coordinate $`v`$ was introduced such that
$`e^{2A}={\displaystyle \frac{l^2}{L^2}}{\displaystyle \frac{v^{2/3}}{1v}},{\displaystyle \frac{dv}{dx_0}}={\displaystyle \frac{2}{L}}v^{2/3}(1v),{\displaystyle \frac{dA}{dx_0}}={\displaystyle \frac{v+2}{3Lv^{1/3}}},`$ (2.41)
$`l`$ is an additional length scale. As in the first example, the boundary of the 5-d space is at $`v=1`$, and there is a singularity at $`v=0`$.
First we check that our interpretation of the field $`s`$ as the dual to the operator $`O_2`$ in the YM is consistent. For the kink solution (2.41) there exists a limiting procedure that allows one to remove the flow and to restore the AdS solution. This is a limiting case when $`\frac{l^2}{L^2}\xi ^2`$ becomes small while $`\frac{z}{L}=x`$ is kept fixed. The variable $`z`$ is introduced via
$$v=\mathrm{sech}^2\left(\frac{zl}{L^2}\right)=\frac{1}{\mathrm{ch}^2(\xi x)}=1\xi ^2x^2+\frac{2}{3}\xi ^4x^4+\mathrm{}$$
Clearly, in this limit
$$2A=\mathrm{log}\left(\xi ^2\frac{v^{2/3}}{1v}\right)\mathrm{log}(x^2),$$
and from $`\frac{dv}{dx_0}=\frac{2}{L}v^{2/3}(1v)`$ it follows that
$`dx_0={\displaystyle \frac{L}{x}}dx,`$ (2.42)
i.e. one recovers the standard AdS metric.
By using $`A^{}=dA/dx_0`$ from (2.41) together with (2.42) it is easy to find the leading terms of the derivatives
$$A^{}\frac{1}{L}+\frac{1}{9L}\xi ^4x^4;A^{\prime \prime }\frac{4}{9L^2}\xi ^4x^4;A^{\prime \prime \prime }\frac{4^2}{9L^3}\xi ^4x^4;A^{(IV)}\frac{4^3}{9L^4}\xi ^4x^4;$$
(2.43)
Since $`\beta \xi ^2`$, it vanishes in this limit.
Upon substituting (2.43) into (2.35) one gets $`U(x_0)=\frac{4}{L^2}+𝒪(\xi ^4)`$. Thus, in the limit $`\xi 0`$ the action (2.34) becomes
$`S_2={\displaystyle \frac{L^3}{2}}{\displaystyle 𝑑zd^4p\sqrt{g_a}\left(z^2(_zs)^2+z^2p^2s^24s^2\right)},`$ (2.44)
where $`g_a`$ is the determinant of the standard $`AdS`$ metric. Eq. (2.44) is the familiar action for the scalar field on the AdS space with mass $`m^2=4`$ that is dual to the YM operator $`O_2`$ of conformal weight $`\mathrm{\Delta }=2`$. It is worth noting that from the point of view of the 2-point correlator of the operator dual to the scalar $`s`$, the limit $`\xi 0`$ is equivalent to taking $`p^2`$ to infinity, i.e. to the UV limit. Therefore, this consideration shows that the 2-point correlator does not vanish, and behaves itself in the UV as expected from an operator of the UV conformal dimension $`\mathrm{\Delta }=2`$.
With the help of (2.41) we rewrite (2.34) as
$`S_2={\displaystyle \frac{l^4}{L^5}}{\displaystyle }dvd^4p{\displaystyle \frac{v^2}{(1v)}}((_vs)^2+{\displaystyle \frac{p^2L^4}{4l^2v^2(1v)}}s^2{\displaystyle \frac{3(42v+v^2)}{(v+2)^2v(1v)^2}}s^2.)`$ (2.45)
This action leads to the following equation of motion for $`s`$
$$_v^2s+\frac{2v}{v(1v)}_vs\frac{p^2L^4}{4l^2v^2(1v)}s+\frac{3(42v+v^2)}{(v+2)^2v(1v)^2}s=0.$$
(2.46)
The on-shell value of the action is
$`S_2={\displaystyle \frac{l^4}{L^5}}{\displaystyle d^4p\left(\frac{v^2}{1v}s_vs\right)_{v=1\epsilon ^2}}.`$ (2.47)
Equation (2.46) has four regular singular points, $`(2,0,1,\mathrm{})`$ and a closed form solution does not exist. But we can nevertheless analyze the behaviour of a solution in the neighbourhood of the physically relevant points at $`v=1`$ and at $`v=0`$. At $`v=1`$, the power series Ansatz $`s(v)=(1v)^\rho [1+a_1(1v)+𝒪((1v)^2)]`$ leads to the indicial equation $`\rho ^22\rho +1=0`$ with two degenerate solutions $`\rho =1`$. Thus, the general solution is of the form
$$s(v)=(1v)\left(c_1f_1(v)+c_2f_2(v)\mathrm{log}(1v)\right),$$
(2.48)
where $`f_1`$ and $`f_2`$ are power series in $`(1v)`$ with leading term 1.
A similar analysis at $`v=0`$ leads to an indicial equation with the two solutions
$$\rho _\pm =\frac{1}{2}\pm \frac{1}{2}\sqrt{1+\frac{p^2L^4}{l^2}}.$$
For generic parameters the two independent solutions are pure power series. Of these, the one for $`\rho _{}`$, is forbidden by regularity. Thus, the solution we are interested in has the general form
$$h(v)=cv^{\rho _+}(1+a_1v+a_2v^2+\mathrm{}).$$
It is regular at $`v=0`$ for space-like momenta ($`p^2>0`$). Note however that although the solution is not regular at the singularity $`v=0`$ if the momentum obeys $`0p^2<l^2/L^4`$, the $`v`$-integration, c.f. (2.47), gives a vanishing contribution at $`v=0`$ due to the factor $`v^2`$ in the numerator.
To find the 2-point function we need to analytically continue this solution to the neighbourhood of $`v=1`$. By comparing with the known solution of (2.46) in the UV limit $`p^2\mathrm{}`$ ($`\xi 0`$) we conclude that both constants $`c_1`$ and $`c_2`$ in (2.48) are nonvanishing. It is not difficult to compute the 2-point function in terms of these constants. By using the conventional rules of the AdS/CFT correspondence, we get
$`O(p)O(p)=\mathrm{lim}_{\epsilon 0}{\displaystyle \frac{2l^4}{L^5}}{\displaystyle \frac{v^2}{1v}}_vs_\epsilon (v)|_{v=1\epsilon ^2},`$ (2.49)
where
$$s_\epsilon (v)=\frac{(1v)\left(c_1f_1(v)+c_2f_2(v)\mathrm{log}(1v)\right)}{\epsilon ^2\left(c_1f_1(1\epsilon ^2)+c_2f_2(1\epsilon ^2)\mathrm{log}(\epsilon ^2)\right)}$$
is the solution of the equation of motion normalized to be 1 at $`v=1\epsilon ^2`$. The leading term in $`\epsilon `$, non-analytic in $`p^2`$, is
$`O(p)O(p)={\displaystyle \frac{1}{\epsilon ^4\mathrm{log}^2\epsilon }}{\displaystyle \frac{l^4c_1}{2L^5c_2}}.`$ (2.50)
The factor $`\frac{1}{\epsilon ^4\mathrm{log}^2\epsilon }`$ is the one that one expects in the 2-point correlator of a scalar field with the UV conformal weight 2 (see, e.g. or appendix of ). Although we do not know the constants $`c_1`$ and $`c_2`$ explicitly, their $`p^2`$-dependence will be through $`\rho _+`$. We thus conclude, as in for the case of the inert scalar, that the correlator has a continuous spectrum with a mass gap, $`m^2l^2/L^4`$, which vanishes in the limit $`\xi 0`$.
## 3 Conclusion
In this paper we studied graviton-scalar fluctuations in $`d=5`$ flow geometries which are dual to boundary field theories with RG flow. We showed that the analysis of the coupled gravity-scalar sector drastically simplifies by a choice of the almost radial gauge, where one of the active scalars vanishes, but the trace of the graviton $`h`$ remains dynamical. We considered in detail the simplest case of one active scalar, decoupled the graviton trace from the transverse traceless components, and obtained a very simple quadratic action for it. The Lagrangian of the graviton trace differs from the one of a minimally-coupled massless scalar field (e.g. dilaton) only by a factor which coincides with the square of the holographic beta function of the operator responsible for the deformation of the conformal field theory. This is a very natural result due to the operator relation $`T_i^i=_I\beta ^I𝒪^I`$ in a deformed CFT. We expect that a similar relation between the action for $`h`$ and the dilaton action also holds for the general case of many active scalars, where the factor would be given by the square of a “weighted” holographic beta function.
We fixed the form of the quadratic action by means of the Hamiltonian prescription and used the action to compute 2-point functions of the graviton trace in two cases of flow geometries. In both cases we obtained physically reasonable functions which have the same momentum dependence as those of inert scalars. Thus, we successfully resolved the problem recently raised in .
It is worth noting that the geometries we considered have a curvature singularity in the interior and one would expect large string corrections to the 2-point functions, which could drastically change the behaviour of the correlators. Nevertherless, this seems not to happen because the curvature is small near the boundary, and the contribution of the vicinity of the singularity to the on-shell gravity action vanishes.
## Appendix AThe second constraint
Here we analyze the constraint (2.19) and show that it follows from (2.18) and the equations of motion. Since the equation of motion for $`\overline{\varphi }`$ is
$$\frac{V}{\overline{\varphi }}=_0^2\overline{\varphi }+4_0\overline{\varphi }_0A,$$
eq.(2.19) can be written as
$`C=_0h_{00}_0\overline{\varphi }+2{\displaystyle \frac{V}{\overline{\varphi }}}h_{00}_0t_0\overline{\varphi }.`$ (A.1)
By using the equation of motion for $`h_{00}`$ together with (2.18) we find
$$p^2_0H=_0h\frac{V}{3}\frac{_0h}{(_0A)^2}+\frac{1}{4}\frac{e^{2A}p^2h}{_0A}$$
and, therefore,
$$_0t=\frac{V}{3}\frac{_0h}{(_0A)^2}+\frac{1}{4}\frac{e^{2A}p^2h}{_0A}.$$
Taking into account that
$`_0h_{00}={\displaystyle \frac{1}{4}}{\displaystyle \frac{_0^2h}{_0A}}{\displaystyle \frac{1}{4}}{\displaystyle \frac{_0^2A}{(_0A)^2}}_0h`$
one finds
$`C=\left({\displaystyle \frac{1}{4}}{\displaystyle \frac{_0^2h}{_0A}}{\displaystyle \frac{1}{4}}{\displaystyle \frac{_0^2A}{(_0A)^2}}_0h{\displaystyle \frac{1}{4}}{\displaystyle \frac{e^{2A}p^2h}{_0A}}+{\displaystyle \frac{V}{3}}{\displaystyle \frac{_0h}{(_0A)^2}}\right)_0\overline{\varphi }+{\displaystyle \frac{1}{2}}{\displaystyle \frac{V}{\overline{\varphi }}}{\displaystyle \frac{_0h}{_0A}}`$ (A.2)
From (2.33) one finds the equation of motion for $`h`$:
$`{\displaystyle \frac{_0^2h}{_0A}}2{\displaystyle \frac{_0^2A}{(_0A)^2}}_0h{\displaystyle \frac{e^{2A}p^2h}{_0A}}=4_0h{\displaystyle \frac{_0^3A}{_0A_0^2A}}_0h.`$ (A.3)
Therefore, by virtue of this equation one obtains
$`C=\left({\displaystyle \frac{1}{4}}{\displaystyle \frac{_0^2A}{(_0A)^2}}_0h_0h{\displaystyle \frac{_0^3A}{4_0A_0^2A}}_0h+{\displaystyle \frac{V}{3}}{\displaystyle \frac{_0h}{(_0A)^2}}\right)_0\overline{\varphi }+{\displaystyle \frac{1}{2}}{\displaystyle \frac{V}{\overline{\varphi }}}{\displaystyle \frac{_0h}{_0A}}.`$ (A.4)
Recalling (2.22), one then finds
$`C={\displaystyle \frac{1}{4}}\left(8{\displaystyle \frac{_0^3A}{_0A_0^2A}}\right)_0\overline{\varphi }_0h+{\displaystyle \frac{1}{2}}{\displaystyle \frac{V}{\overline{\varphi }}}{\displaystyle \frac{_0h}{_0A}}.`$ (A.5)
Differentiating (2.22), one gets
$$_0^3A+8_0A_0^2A+\frac{4}{3}\frac{V}{\overline{\varphi }}_0\overline{\varphi }=0$$
and, as a consequence,
$$C=\frac{1}{3}\frac{V}{\overline{\varphi }}\frac{(_0\overline{\varphi })^2_0h}{_0A_0^2A}+\frac{1}{2}\frac{V}{\overline{\varphi }}\frac{_0h}{_0A}=\frac{1}{2}\frac{V}{\overline{\varphi }}\frac{_0h}{_0A_0^2A}\left(\frac{2}{3}(_0\overline{\varphi })^2+_0^2A\right).$$
Finally, we have the relation
$`_0^2A={\displaystyle \frac{g}{3}}{\displaystyle \frac{W}{\varphi }}_0\overline{\varphi }={\displaystyle \frac{2}{3}}(_0\overline{\varphi })^2.`$ (A.6)
Upon substituting this relation into the previous formula we find $`C=0`$. Thus, constraint (2.19) is compatible with the dynamics.
ACKNOWLEDGMENT We would like to thank A. Tseytlin for useful comments. The work of G.A. was supported by the Alexander von Humboldt Foundation and in part by the RFBI grant N99-01-00166, and the work of S.F. was supported by the U.S. Department of Energy under grant No. DE-FG02-96ER40967 and in part by RFBI grant N99-01-00190. S.T. is supported by GIF – the German-Israeli Foundation for Scientific Research by the TMR programme ERBRMX-CT96-0045. |
warning/0003/astro-ph0003443.html | ar5iv | text | # High Resolution Near-Infrared Spectra of Protostars
## 1 Introduction
The physical natures, evolutionary states, and circumstellar disks of classical T Tauri stars are becoming better understood due to recent spectroscopic observations, high resolution imaging, and advances in the theory of pre-main-sequence (PMS) evolution. However, the natures of the central stars and inner circumstellar environments of protostars are still not very well known. This is primarily because they are so heavily extinguished that they are difficult to observe even with modern instruments and detectors. For example, it is not known whether the central stars in protostellar objects differ substantially in effective temperature, radius, or rotation properties from classical T Tauri stars (CTTSs). The presence of protostellar envelopes, significantly higher accretion rates, and more powerful outflows suggest that the photospheres of protostellar cores may indeed be physically different from CTTSs. Moreover, it is not known whether the physical natures of protostellar photospheres are consistent with the predictions of protostellar evolution and PMS stellar theory.
Flat-spectrum and Class I young stellar objects (YSOs) are the best low-mass protostellar candidates for spectroscopic study because they have relatively well-developed central stars and are detectable at near-IR wavelengths. Several pioneering studies have recently been undertaken to begin investigating the physical natures of these objects. A few flat-spectrum protostars have been observed in low resolution spectroscopic surveys and thus far they appear to be characterized by late-type photospheres with high continuum veilings and sub-giant surface gravities (Luhman & Rieke, 1999; Kenyon et al., 1998; Casali & Matthews, 1992; Greene & Lada, 1996, hereafter Paper I). Even fewer have been observed at high spectroscopic resolution, and these observations suggest that the flat-spectrum protostars rotate significantly faster than the more evolved CTTSs (Greene & Lada, 1997, hereafter Paper II).
These investigations have provided some interesting clues to the natures of these objects, but more observations are clearly needed to make sense of these late-phase protostars. It would be most interesting to determine if the less evolved, more heavily embedded and veiled Class I protostars also show near-IR photospheric absorption lines when observed at high resolution with high signal-to-noise. Such observations could directly constrain the effective temperatures, gravities, veilings, and rotations of these objects, providing further evidence as to whether they are dominated by near-IR stellar, disk, or envelope emission and whether they are physically similar to flat-spectrum YSOs. More flat spectrum YSOs should also be observed with high-resolution near-IR spectroscopy to confirm that they are indeed late-type rapid rotators. In a sense the flat-spectrum YSOs provide a link to the well-known PMS stars, and this must be better developed so that they can in turn serve as a link to the less well-known Class I objects.
Therefore we have undertaken a new high-resolution, near-IR spectroscopic study of flat-spectrum and Class I YSOs in the $`\rho `$ Ophiuchi cloud core. We describe these new observations in §2 and present a rotation and veiling analysis of these data in §3. In §4 we discuss the likely natures of each of these objects and suggest further observational work.
## 2 Observations and Data Reduction
Near-IR spectra were acquired in 1997 May – June and 1998 July with the 3.0 m NASA Infrared Telescope Facility on Mauna Kea, Hawaii, using the CSHELL facility single-order cryogenic echelle spectrograph (Tokunaga et al., 1990; Greene et al., 1993). Spectra were acquired with a 1$`\stackrel{}{\mathrm{.}}`$0 (5 pixel) wide slit on the dates indicated in Table 1, providing a spectroscopic resolution $`R\lambda /\delta \lambda `$ = 21,000 (14 km s<sup>-1</sup>). The spectrograph was fitted with a 256 $`\times `$ 256 pixel InSb detector array, and custom circular variable filters (CVFs) manufactured by Optical Coating Laboratories Incorporated were used for order sorting. These filters successfully eliminated the significant interference fringing normally produced in CSHELL and other echelle spectrographs which use CVFs for order sorting. The plate scale was 0$`\stackrel{}{\mathrm{.}}`$20 pixel<sup>-1</sup> along the 30$`\mathrm{}`$ long slit (oriented east – west on the sky), and all spectra were acquired at a central wavelength setting of 2.29353 $`\mu `$m corresponding to the v = 0 – 2 CO band head. Each exposure had a spectral range $`\mathrm{\Delta }\lambda \lambda /400`$ ($`\mathrm{\Delta }v`$ 700 km s<sup>-1</sup>). Total integration times for each YSO are given in Table 1.
Data were acquired in pairs of exposures of up to 400 s duration each, with the telescope nodded $`10\mathrm{}`$ east or west between exposures so that object spectra were acquired in all exposures. The B0V star HR 6165 ($`\tau `$ Sco) was observed periodically for telluric corrections. The telescope was guided with the CSHELL internal CCD autoguider during exposures of these telluric correction stars, while the telescope tracking rates were adjusted for minimum drift while observing the optically invisible $`\rho `$ Oph YSOs. Spectra of the internal CSHELL continuum lamp were taken for flat fields, and exposures of the internal CSHELL Ar and Kr lamps were used for wavelength calibrations.
All data were reduced with IRAF. First, object and sky frames were differenced and then divided by flat fields. Next, bad pixels were fixed via interpolation, and spectra were extracted with the APALL task. Extracted spectra were typically 5 pixels (1″) wide along the slit (spatial) direction at their half-intensity points. Spectra were wavelength calibrated using low-order fits to lines in the arc lamp exposures, and spectra at each slit position of each object were co-added. Instrumental and atmospheric features were removed by dividing wavelength-calibrated object spectra by spectra of early-type stars observed at similar airmass at each slit position. Final spectra were produced by combining the spectra of both slit positions for each object.
## 3 Data Analysis and Results
### 3.1 Object Sample
The object sample was selected from the Class I and flat-spectrum YSOs observed at low spectral resolution ($`R500`$) in Paper I which were not subsequently observed at high spectral resolution in Paper II. None of the newly observed sources (listed in Table 1) showed any absorption features in their low resolution $`K`$-band spectra (Paper I), and they are also relatively bright, 7 mag $`K`$ 10 mag. Table 1 shows that the brighter point sources (i.e. Elias 29, IRS 54) were observed with higher signal-to-noise ratios than the fainter ones (i.e. IRS 43, WL 6). This is useful for analyzing the veiling in these objects if the brightness differences among the sources are mostly due to different amount of IR excess emission from circumstellar regions. If this is true, then the brighter objects have greater near-IR veiling and greater signal-to-noise is required to detect their photospheric absorption lines. On the other hand, it may be possible that some of the bright protostars are featureless because they are of relatively early spectral type (G or earlier) and possess intrinsically weak $`K`$-band absorption lines (besides H and He).
### 3.2 Veiling and Rotation Analysis
The new flat-spectrum and Class I YSO spectra are shown in Figures 1 and 2, respectively. None of the five Class I spectra show any evidence of CO absorption. Two of the flat-spectrum YSOs also show no evidence of any CO absorption (GSS 26 and YLW 13B), while the other two show evidence of weak, broad band heads and perhaps some overlapping rotation-vibration lines as well (Figure 1).
We now analyze these spectra to constrain the veilings and physical natures of these sources. Our previous high resolution study (Paper II) showed that late-type Class II YSOs (pre-main-sequence stars) rotate slowly, $`v\mathrm{sin}i<20`$ km s<sup>-1</sup>, while flat-spectrum ones rotate quickly, $`v\mathrm{sin}i>20`$ km s<sup>-1</sup>. This study also showed that in addition to broadening the band head, high rotation also decreased the maximum absorption depth of the band head and adjacent individual rotation-vibration lines. Thus rotation as well as veiling can reduce the detectability of CO absorption in finite signal-to-noise spectra.
We now explore the limits of $`K`$-band veiling and rotation in our new protostar sample by comparing their spectra to those of flat-spectrum and Class II YSOs to which we have artificially added continuum veiling. We chose VSSG 25 to be representative of a slowly rotating Class II YSO ($`v`$ sin $`i=5`$ km s<sup>-1</sup>) and VSSG 17 to be representative of a quickly rotating ($`v`$ sin $`i=47`$ km s<sup>-1</sup>) flat-spectrum source (see Paper II). The spectral types and $`K`$-band veilings have been measured for both of these objects. VSSG 25 has a spectral type M0IV/V with $`r_k`$ = 0.25 (Luhman & Rieke, 1999) while VSSG 17 is M0IV/V with $`r_k`$ = 0.9 (Luhman & Rieke; Paper I). The $`K`$-band veiling is defined as $`r_k=F_{Kex}/F_K`$ where $`F_{Kex}`$ is the $`K`$ band excess flux, and $`F_K`$ is the $`K`$ band stellar flux. We added veiling (a constant positive offset) to the template spectrum of VSSG 17 (taken from Paper II) and degraded its signal-to-noise so that its CO absorption equivalent width and detectability were weakened. We then used a series of these VSSG 17 templates with various veilings and signal-to-noise ratios to estimate the veilings of our observed sources.
We estimated the veiling for sources which show CO band head absorptions (IRS 51 and IRS 63) by matching their spectra to veiled templates with identical signal-to-noise and similar features (CO band head depth and slope). The band head profiles of these objects also matched the shape of VSSG 17 and matched those of observed slowly rotating late-type stars which had been artificially broadened with a stellar rotation profile of $`v`$ sin $`i=50`$ km s<sup>-1</sup> (see Paper II). Thus it is likely that their CO absorptions arise in rapidly rotating stellar photospheres. However, the absorption features of these highly veiled sources are very weak, and this limits the derived rotation velocities to uncertainties of approximately 40% and the derived veilings to uncertainties of at least 20%.
The minimum likely veilings of the Class I and flat-spectrum sources without detectable CO absorptions were estimated by assuming that their CO absorptions were intrinsically similar to VSSG 17 but had increased continuum veiling. We derived an analytical relation for veiling based on the principle that an object’s CO band head is undetectable when its maximum absorption depth is less than 3.0 times the RMS noise over 1 resolution element (5 pixels) in the spectrum. This is a robust minimum likely veiling criterion because less veiling would ensure that the band head would be definitely detected, and considering more resolution elements would increase the amount of veiling derived. The resultant derived minimum likely $`r_k`$ values matched those of templates which were veiled to the point where their CO absorptions just disappeared visually.
Possible differences between the actual spectral types of these sources and the M0 template also cause uncertainties in the derived veilings of up to about 50%, but most deviations are expected to be smaller than this. All observed Class I sources except Elias 29 are relatively low luminosity, 2.4 L$`{}_{}{}^{}`$ L$`{}_{\mathrm{bol}}{}^{}13`$ L (Wilking, Lada, & Young, 1989, hereafter WLY). All observed flat-spectrum YSOs have 1 L$`{}_{}{}^{}`$ L$`{}_{\mathrm{bol}}{}^{}<3`$ L (WLY; Greene et al. 1994). These luminosities are consistent with the sources being low-mass PMS YSOs (M $`<`$ 1 M) if the Class I objects are powered by accretion, an assumption which is consistent with their high veilings (see also §4.2). In using VSSG 17 as a spectral type template to measure veilings, we have implicitly assumed that all sources have spectral types near M0IV/V, a typical value for T Tauri stars. A young star of this spectral type has a mass of approximately 0.4 M if on the birthline of the H–R Diagram (see Stahler, 1988). If an observed YSO is really 1 M, then it would have a spectral type of K3–4IV/V (see D’Antona & Mazzitelli, 1997; Stahler, 1988) and thus an intrinsic stellar CO equivalent width of only about 0.6 times that of an M0IV/V star (see Paper I). This would mean that its actual veiling is $`r_k^{}=0.6r_k0.4`$ (see §4.2 of Paper I) where $`r_k`$ is the veiling estimate made assuming an M0IV/V spectral type. Likewise, if an observed object were really a lower mass star near the brown dwarf limit, then its true birthline spectral type would be near M6IV/V and its intrinsic CO absorption would be about 40% greater than that of the M0IV/V template. Thus the true veilings of observed sources may differ from our derived ones by as much as 50%. However, it is unlikely that many of the observed objects have such large deviations (in either direction) because their luminosities are consistent with most having masses of approximately 0.5 M, similar to that expected for our M0IV/V template. We have used other published information on the observed objects when available to estimate their intrinsic spectral types before deriving veiling estimates (see §4.1 and 4.2).
Finally, we studied how rotation decreases the apparent CO band head absorption depths. This effect can be separated from veiling by measuring the slope of the CO band head, but this is not possible for sources which do not show this absorption feature. The majority of sources in our sample do not show this feature, so we do not know their rotation velocities. Our earlier study (Paper II) showed that flat-spectrum sources rotate significantly faster than Class II YSOs, so our current sample of more highly embedded flat-spectrum and Class I YSOs may rotate even faster still (perhaps $`v`$ sin $`i>50`$ km s<sup>-1</sup>). Indeed, recent X-ray observations with the ASCA satellite indicate that one Class I source in our sample (IRS 43) is rotating at least this rapidly (Montmerle et al., 2000). We estimate that the equatorial rotational breakup velocities of these young stars ($`M0.5M_{\mathrm{}}`$ and $`R3R_{\mathrm{}}`$) are approximately $`v180`$ km s<sup>-1</sup>, or $`v`$ sin $`i150`$ km s<sup>-1</sup> for a mean inclination $`i=57\mathrm{deg}`$. Next, we studied how rotation decreases the maximum CO band head absorption in YSOs by artificially rotating our WL 5 template by convolving its observed spectrum with limb-darkened stellar broadening profiles for 25 km s$`{}_{}{}^{1}v`$ sin $`i175`$ km s<sup>-1</sup>. See Paper II for more details and examples of artificially rotated spectra.
These experiments showed that the maximum CO absorption depth of the VSSG 25 template with $`v`$ sin $`i=150`$ km s<sup>-1</sup> is a factor of 1.37 weaker than the one with $`v`$ sin $`i=50`$ km s<sup>-1</sup>. Therefore if VSSG 17 were rotating at breakup, its maximum CO absorption depth would be only 73% as deep as now seen in its spectrum. Thus if rotating near breakup, the featureless objects in our sample would have somewhat lower continuum veilings than those calculated based on the observed VSSG 17 spectrum. We have calculated these reduced values, and in Table 2 we present all of these estimated veilings for the observed sources, using our best estimates of their intrinsic spectral types.
## 4 Discussion and Conclusions
We now discuss how the results of this veiling / rotation analysis and pre-existing data constrain the possible physical natures of these sources.
### 4.1 Flat-Spectrum Objects
The flat-spectrum YSOs IRS 63 and IRS 51 were both found to have broad, weak CO absorptions which matched those expected for late-type stellar photospheres rotating at $`v`$ sin $`i50`$ km s<sup>-1</sup>. The weak CO absorptions of these two YSOs are consistent with their not being detected in our initial low-resolution survey (Paper I). We estimate the continuum veiling of IRS 63 to be $`r_k4`$ provided that it is a PMS YSO near M0 spectral type. Luhman & Rieke find that the spectral type of IRS 51 is G5 – K7, earlier than our M0 template VSSG 17. Thus their derived veiling $`r_k`$ = 1 – 3 is lower than ours because a G5 – K7 PMS star has less intrinsic CO absorption than a M0 one (see §3.2). It is likely that IRS 51 is indeed an embedded low-mass YSO because its bolometric luminosity is only 1.4 L (WLY). The birthline mass for this luminosity is approximately 0.5 M, corresponding to a spectral type of K5–7 and a true veiling of $`r_k3`$. Thus it is likely that both IRS 63 and IRS 51 are similar to the quickly rotating flat-spectrum YSOs which we analyzed in Paper II, but these new objects have even greater veiling (i.e. $`r_k`$ = 3 – 4 versus $`r_k1`$ for the Paper II YSOs).
Luhman & Rieke found GSS 26 to have variable veiling, $`r_k`$ = 0.75 and $`r_k`$ = 4 at epochs of 1994 July and 1996 May, respectively. Our spectrum of GSS 26 (in Figure 1) was taken in 1997 May, one year after the latest Luhman & Rieke spectrum. We estimate that $`r_k11`$ when our spectrum was acquired, and our assumption of a M0 spectral type is consistent with the Luhman & Rieke determination of K5 – M2. This rapid increase in veiling - a factor of 2 each year - is perhaps suggestive of a similarly rapid increase in accretion. Luhman & Rieke also note that this source increased in brightness by $`K`$ 1.2 mag between epochs. Our spectra are not photometrically calibrated, but comparisons with other objects support that this source was at least as bright as when observed by Luhman & Rieke the previous year. This is one of the YSOs Luhman & Rieke observed whose HI Br $`\gamma `$ emission line flux increased as its veiling increased, implying that the excess continuum emission is associated with a circumstellar accretion disk if the Br $`\gamma `$ emission arises from disk accretion. The rapid variability of this object’s veiling also suggests that its $`K`$-band veiling is produced by accretion from inner disk distances (several AU) and not from an outer disk or an outer circumstellar envelope.
The high $`r_k`$ values of these objects also constrain the physical origins of their veilings. In Paper I we showed that veilings $`r_k>1`$ cannot be produced by a simple optically thick, geometrically thin reprocessing disk around a low-mass PMS star. Consequently we argued that these high veilings are most likely produced by either actively accreting circumstellar disks or circumstellar envelopes associated with these objects. Furthermore we found that veilings in the range measured for IRS 51 and IRS 63, $`r_k`$ = 3 – 4, can be caused by luminous accretion disks ($`L_{disk}/L_{}3`$). Veilings in the range observed for GSS 26, $`r_k510`$, could be explained by extremely luminous accretion disks ($`L_{disk}/L_{}3`$). In either case, these accretion disks would have to have relatively large central holes to avoid producing strong CO absorption-line systems in the disk photosphere itself. However, because there is no obvious physical mechanism for producing central holes of the needed size, Calvet, Hartmann, & Strom (1997) suggested that the veiling flux must originate in some other circumstellar structure such as the inner regions of the protostellar envelope. On the other hand, this seems to be inconsistent with the observation by Luhman & Rieke that the $`K`$-band excesses of flat-spectrum YSOs are correlated with their HI Br $`\gamma `$ line fluxes which in turn suggests that the veiling flux should originate in the disk. More detailed knowledge of the conditions required to produce CO absorption line systems in an accretion disk may be needed to resolve this issue.
The flat-spectrum source YLW 13B was found to have H Br $`\gamma `$ absorption by Luhman & Rieke, who estimate its spectral type to be earlier than K0. However, they also find it to be significantly veiled with $`r_k>1`$, so it is possibly an intermediate mass PMS cloud population member. Our non-detection of CO absorption does not constrain this source further.
### 4.2 Class I Objects
We do not detect CO absorptions in any of the Class I YSOs which we observed (Figure 2), confirming earlier low resolution spectroscopic observations that found all these objects to be featureless and highly veiled (Paper I; Luhman & Rieke). Consequently their spectral types are unknown, however all of these Class I sources exhibit HI Br $`\gamma `$ emission (Paper I; Luhman & Rieke).
Analysis of our new data constrains the natures of these objects. Table 2 shows that they all have large veilings, $`r_k>4`$, if they are late-type ($``$ M0) stars. GSS 30, IRS 43, and WL 6, all have estimated minimum veilings of $`r_k58`$, overlapping with the flat-spectrum sample. These sources all have bolometric luminosities L$`{}_{\mathrm{bol}}{}^{}13`$ L (WLY). This is consistent with their being low-mass (M $`<`$ 1 M) protostars accreting matter at rates $`\dot{M}5\times 10^6`$ M yr<sup>-1</sup>, the value expected for the T $``$ 20 K gas temperatures in the $`\rho `$ Oph cloud (see Adams, Lada, & Shu, 1987, hereafter ALS). Veilings in the observed range, $`r_k>410`$, are also predicted by theoretical models of Class I circumstellar envelopes (Paper I; Calvet et al.).
Montmerle et al. note that the maximum possible mass of IRS 43 (also known as YLW 15) is 2.2 M which is derived from PMS models given its bolometric luminosity (L $``$ 10 L) and assuming it is on the birthline. Likewise, they calculate that the maximum likely mass of WL6 is approximately 0.4 M; this increases somewhat if the L<sub>bol</sub> = 2.4 L of WLY is adopted. These maximum calculated masses assume that essentially all luminosity is due to photospheric thermal radiation ($`L=4\pi R^2\sigma T^4`$) and essentially none is due to accretion ($`L=GM\dot{M}/R`$). These sources are discussed further in $`\mathrm{\S }4.3`$.
We estimate that Elias 29 and IRS 54 have very high veilings if they are late-type low-mass stars, $`r_k>1434`$. The bolometric luminosity of IRS 54 is estimated to be only 12 L (WLY), also consistent with this object being a low-mass protostar which is accreting its envelope at the rate prescribed by the $`\rho `$ Oph cloud’s gas temperature. However, its continuum veiling must be $`r_k>1420`$ if it has spectral type M0 and is rotating rapidly. This is about a factor of 2 higher than the model predictions of Calvet et al., but those calculations were done for a hypothetical $`\rho `$ Oph Class I YSO with L = 5 L. The model may predict greater veiling for IRS 54 if its higher luminosity is taken into consideration. It is also possible that this source may be a somewhat earlier type protostar which is less veiled.
Elias 29 has the highest derived veiling of the sample, $`r_k>2534`$, assuming an intrinsic M0 photosphere and a high rotation rate. WLY estimate its luminosity to be L<sub>bol</sub> = 48 L, and ALS have modeled it as a 1 M protostar which is accreting its circumstellar envelope. Such a star would have a spectral type of K3–4 if on the birthline (see D’Antona & Mazzitelli, 1997; Stahler, 1988), with an intrinsic CO absorption approximately 60% as strong as that of an M0 star (see §3.2). Thus we revise our estimate of the likely veiling of this YSO to $`r_k>1520`$ if it is indeed a 1 M protostar. In Paper I we analyzed the ALS model for Elias 29 and showed that the predicted emission from the inner protostellar envelope of this source would produce a veiling of $`r_k20`$, assuming that its disk luminosity is 0.75 L<sub>bol</sub>. Thus our new measurement is consistent with our earlier prediction based on the ALS model.
The high luminosity of Elias 29 also allows for it being a more massive, earlier spectral type YSO that has higher stellar luminosity and less accretion luminosity than assumed by the ALS model. However, it is unlikely to be very different because the observed near-to-far IR energy distribution and the 10 $`\mathrm{\mu m}`$ silicate absorption of Elias 29 are fit well by the ALS model, and there are no clues which indicate that Elias 29 is an early-type object. For example, WL 16, which is likely an early A type star (Biscaya Holzbach et al., 2000), has mid-IR aromatic hydrocarbon emission features which indicate a UV radiation field (Hanner, Tokunaga, & Geballe, 1992). However, Elias 29 shows no evidence for IR hydrocarbon emission (Hanner, Brooke, & Tokunaga, 1995; Boogert, 1999) and thus no evidence for a UV radiation field.
### 4.3 Protostellar Rotation
Our observations, specifically the broad band head shapes of IRS 51 and IRS 63, strengthen the earlier findings of Paper II which suggested that flat-spectrum protostars rotate more rapidly than Class II sources (CTTSs). Since flat-spectrum protostars are believed to be evolutionary precursors of CTTSs, then this finding may indicate that certain physical conditions characteristic of protostellar evolution (e.g., high accretion rates) may result in their higher rotation rates. It is therefore interesting to ask whether the less evolved Class I sources might rotate even more rapidly than the flat spectrum sources. Indeed, some models of protostar development predict that such objects should be rotating near breakup (Shu, 1991).
Recently, strong hard X-ray flares have been observed with the ASCA satellite from two of the Class I protostars in our sample – IRS 43 and WL 6 (Montmerle et al.). These remarkable observations reveal relatively rapid periodicities in the X-ray emission from these sources, enabling the derivation of their photometric rotation rates. Montmerle et al. find that WL 6 is rotating with a period of about 3 days ($`v`$ sin $`i40`$ km s<sup>-1</sup> for a 0.5 M star on the birthline), comparable to the rotation rates of the flat-spectrum sources observed here and in Paper II. This source also has a weak outflow (Sekimoto et al., 1997), undetected millimeter emission from its envelope (André & Montmerle, 1994), and an IR energy distribution which can be modeled as a highly extinguished flat spectrum YSO (Montmerle et al.). All of these properties indicate that WL 6 may indeed be very similar to the flat spectrum YSOs for which we have detected absorption lines and have found to be rotating more rapidly than CTTSs (Class II YSOs).
Montmerle et al. also argue that the central star of IRS 43 has a 20 h rotation period, the observed period of its X-ray variability. This requires that its mass be greater than or equal to 1.8 M in order for it to be rotating below breakup velocity if it is on the birthline (Montmerle et al.). Thus the mass of IRS 43 is constrained to lie in the range 1.8 – 2.2 M by its X-ray emission and bolometric luminosity (see $`\mathrm{\S }4.2`$) if its rotation period is indeed equal to its X-ray variability period of 20 h. IRS 43 does have a more steep mid-IR energy distribution (clearly Class I) than WL 6, and it also has spatially-resolved (r $``$ 3000 AU) millimeter emission with a derived envelope mass of approximately 0.1 M (André & Montmerle). Therefore it is likely to be in an earlier evolutionary state than WL 6.
The slow rotation velocities of CTTSs have been explained by angular momentum regulation of these stars by magnetic coupling to their disks. Edwards et al. (1993) found that late-type T Tauri stars (TTSs) with large $`HK`$ IR excesses (CTTSs) had slow rotation periods, P $`>`$ 4 d. They also found that TTS with small $`HK`$ IR excesses had a broad range of periods, including a significant number with P $`<`$ 4 days. Edwards et al. interpreted this correlation to arise because the magnetic fields of the CTTSs were were coupled to their disks, providing stellar angular momentum regulation and therefore long stellar rotation periods. The low-excess TTSs had already dissipated their disks and so were not subject to this regulation mechanism. More recent studies of larger TTS samples have both disputed that the correlation between IR excess and rotation period exists (Stassun et al., 1999) and have provided evidence that it exists but is weak (Herbst et al., 2000).
Our studies of YSO rotation (this paper and Paper II) have shown that flat-spectrum protostars rotate significantly more rapidly than Class II YSOs or CTTSs, suggesting that rotation velocities decrease as stars evolve past the protostellar state. This scenario has been bolstered and expanded further by the recent X-ray results of Montmerle et al.. Taken together (and along with the many rotation studies of optically visible CTTSs), these works suggest that heavily embedded protostars (Class I) rotate very rapidly, in some cases near breakup velocity, while less embedded ones (flat–spectrum YSOs) rotate somewhat less rapidly, at about 1/3 breakup velocity ($`v`$ sin $`i50`$ km s<sup>-1</sup>), and Class II YSOs / CTTSs rotate slowly, $`v`$ sin $`i<20`$ km s<sup>-1</sup>. This finding would be strengthened considerably by further cross-checking of observational techniques; the X-ray protostars should be observed at higher signal-to-noise in the near-IR to search for rotationally broadened lines, while the flat spectrum and Class II YSOs with IR-derived rotation velocities should be observed for periodic X-ray variability.
If the angular momenta of low-mass YSOs are indeed regulated by star–disk coupling, then the fact that flat-spectrum (and at least one Class I) YSOs rotate significantly more rapidly than CTTS implies that either the flat-spectrum / protostellar YSOs are coupled to faster rotating disk regions than CTTSs, or else that stars and disks do not become rotationally locked until the CTTS evolutionary phase. In the first case, flat-spectrum and Class I YSOs may couple to their disks at smaller radii (and hence have higher rotation velocities) because their accretion rates are much higher than CTTS. The veilings and luminosities of Class I and flat-spectrum YSOs are considerably higher (by about an order of magnitude) than CTTS, supporting the notion that they have higher accretion rates also. In support of the second case, Montmerle et al. have posited that protostars are not initially magnetically coupled to their disks but rather spin-down and become coupled over a magnetic braking time on the order of 10<sup>5</sup> yr which is nearly linearly proportional to stellar mass. This is comparable to the lifetime of the Class I and flat spectrum phases, so this would account for the higher rotation velocities of flat-spectrum and Class I protostars. This latter magnetic braking scenario of velocity evolution from Class I to flat-spectrum to CTTS YSOs may be somewhat complicated by mass effects; Montmerle et al. predict that at the same age more massive protostars will rotate more quickly than less massive ones. Protostellar masses must be measured much more accurately before this effect can be verified, however.
Obtaining new high resolution, high signal-to-noise spectra over the entire 1.5 – 2.4 $`\mathrm{\mu m}`$ region will likely be the best method for obtaining more definitive information on the masses (spectral types) and rotational characteristics of Class I protostars. This wide spectral range is required in order to be sensitive to a wide range of spectral types. Even intermediate-to-high mass stars with HI line emission may show $`H`$-band HI Br absorption lines which may strongly constrain spectral types and masses (e.g., Biscaya Holzbach et al.), while $`K`$-band data are required to determine the properties of very red late-type YSOs. Detecting and resolving near-IR lines in WL 6 and IRS 43 would allow determination of their masses and photospheric rotation rates, providing a good test of the emerging scenario of protostellar rotational evolution.
We thank the referee John Lacy and also Pat Cassen for providing comments which improved this paper. We also thank W. Golisch, D. Griep, and C. Kaminski for assistance with the observations. We acknowledge the National Science Foundation for funding grant AST–9420506 to develop the fringe-free CVFs used to acquire these data with CSHELL. TPG acknowledges a grant from the NASA Ames Research Center Director’s Discretionary Fund. All data were reduced with IRAF, which is distributed by the National Optical Astronomy Observatories, which is operated by the Association of Universities for Research in Astronomy, Inc., under contract to the National Science Foundation. |
warning/0003/astro-ph0003311.html | ar5iv | text | # Calibration of Nebular Emission-Line Diagnostics: II. Abundances
## 1. Introduction
The emission-line signatures of H ii regions are a powerful and widely-used indicator of galactic abundances. This is especially true in distant galaxies where most stellar abundance probes cannot be employed. Furthermore, the spectral properties of H ii regions also give important diagnostics of the ionizing stellar population, such as effective temperature and numbers of stars. Given the wide use of such nebular diagnostics, it is vital to test and calibrate them using H ii regions with known characteristics and ionizing stellar populations.
Along with a companion paper (Oey et al. 2000; hereafter Paper I), we report here on a detailed study for this purpose, of four H ii regions in the Large Magellanic Cloud (LMC). The OB associations in all four of these H ii regions have been examined in detail and classified, thereby strongly constraining the ionizing energy distributions. In addition, the LMC’s proximity and orientation with respect to the Galaxy also permit a detailed understanding of the nebular morphology. Thirdly, the abundances can be accurately determined. We therefore have high-quality information on the three principal parameters that determine the nebular emission: stellar effective temperature ($`T_{}`$), the ionization parameter ($`U`$) that relates ionizing flux to gas density, and metallicity ($`Z`$).
Table 1 gives a brief summary of our objects, which can be examined in greater detail in Paper I. The first two columns in Table 1 give the H ii region identification in the Davies, Elliott, & Meaburn (1976) and Henize (1956) H$`\alpha `$ catalogs, respectively; the third column identifies the parent OB association from Lucke & Hodge (1970). Column four shows the spectral type of the dominant ionizing stars, as classified by the references in column 8. Column five lists the nebular H$`\alpha `$ luminosity from Oey & Kennicutt (1997), and column 6 shows the adopted value from Paper I, of the inner radius of the gas distribution, as a fraction of the Strömgren radius $`R_\mathrm{S}`$. This gives some indication of the nebular morphology. Finally, column 7 indicates the presence of significant shock excitation: DEM L243 includes an embedded or superimposed supernova remnant (SNR), and DEM L301 also shows evidence of significant shock activity (Paper I). While shocks may be present in the other objects as well, their contribution to the nebular emission is unimportant.
The detailed presentation of the objects is given in Paper I, including narrow-band images in H$`\alpha `$, \[O III\], and \[S II\], and spectroscopic observations over the wavelength range 3500 – 9200 Å. For each object, we observed two to three stationary, spatially-resolved slit positions. For all of the objects except DEM L301, we also obtained spatially integrated observations by scanning the long slit across the nebulae. The scanned data should therefore resemble typical observations of such H ii regions at distances of 10 – 20 Mpc. As seen in Table 1, the sample spans a range in dominant stellar spectral types, from O7 to early Wolf-Rayet (WR). There is also variety in the morphology of the objects, ranging from near perfect Strömgren sphere (DEM L323), to extreme shell structure (DEM L301), to highly complex (DEM L199).
Paper I provides a detailed analysis of the spatially resolved emission-line ratios with respect to the sequence in $`T_{}`$ represented by these objects, and a comparison with photoionization models using the current generation of stellar atmosphere models for both O stars and early WR stars. In general we found a gratifyingly high level of agreement, largely supporting the CoStar energy distributions of Schaerer & de Koter (1997) and early WR atmospheres of Schmutz, Leitherer, & Gruenwald (1992) and Hamann & Koesterke (1998). In conjunction with data from Kennicutt et al. (2000), we provide a first, empirical calibration of nebular diagnostics for the dominant $`T_{}`$. In addition to the well-known $`\eta ^{}`$ radiation softness parameter of Vílchez & Pagel (1988), we introduce \[Ne III\]$`\lambda `$3869/H$`\beta `$ as a similar diagnostic, which is more robust to nebular conditions and is sensitive for a higher range of $`T_{}`$.
We also presented in Paper I the spatially-resolved behavior of the O abundance parameter (Pagel et al. 1979),
$$R23\frac{[\mathrm{O}\mathrm{II}]\lambda 3727+[\mathrm{O}\mathrm{III}]\lambda \lambda 4959,5007}{\mathrm{H}\beta };$$
(1)
and S abundance parameter (Vílchez & Esteban 1996; Christensen et al. 1997; Díaz et al. 1999),
$$S23\frac{[\mathrm{S}\mathrm{II}]\lambda 6724+[\mathrm{S}\mathrm{III}]\lambda \lambda 9069,9532}{\mathrm{H}\beta },$$
(2)
where we designate the \[S II\] lines $`\lambda 6716+\lambda 6732`$ as $`\lambda `$6724, analogous to \[O II\]$`\lambda `$3727. We confirmed that both observationally and theoretically, $`R23`$ remains spatially uniform across the nebulae. In contrast, for uniform abundances, models of $`S23`$ vary across the nebulae, showing lower values in central regions. The observations clearly reflect this pattern, which is caused by the missing ionization stage of S IV. We address this issue in detail below in this paper. As is well-known, $`R23`$ and $`S23`$ are also sensitive to $`T_{}`$, and these effects are also shown in Paper I for our objects.
In this paper, we present the abundance determinations for our sample, for both the spatially-resolved and scanned longslit observations. We examine conventional assumptions for the nebular electron temperature ($`T_e`$) structure and resulting ionic and total abundance determinations. We then explore the metallicity diagnostics $`R23`$ and $`S23`$ in more detail, and introduce another diagnostic, $`S234`$. As before, we use the photoionization code Mappings II (Sutherland & Dopita 1993) in conjunction with CoStar stellar atmosphere models (Schaerer & de Koter 1997) for O stars.
## 2. Direct abundance determinations
We derived abundances from the measured line fluxes in Paper I using standard techniques. For calculation of line emissivities for elements heavier than helium, we used the five-level atom code Five\_L as implemented in Stsdas version 1.3.5 (Shaw & Dufour 1995). We initially obtained an estimate of electron density from the \[S II\] $`\lambda `$6716/$`\lambda `$6731 line intensity ratio, assuming a default $`T_e=10^4`$ K. The densities are almost all $`100\mathrm{cm}^3`$.
### 2.1. Temperature structure
The assumed $`T_e`$ structure, and thereby, ionization structure of the H ii region, can generate substantial uncertainties in even “direct” abundance determinations from typical optical emission-line spectra. The $`T_e`$ structure is in turn a function of density structure and stellar atmosphere models. Peimbert’s (1967) $`t^2`$ formulation for temperature fluctuations is probably the best-known example of addressing this problem, which can be used for both small-scale and large-scale variations in $`T_e`$.
We consider here the large-scale temperature structure. As discussed below, our photoionization models tend to overestimate $`T_e`$, suggesting that small-scale fluctuations are not significant. Standard abundance determinations adopt either a single characteristic $`T_e`$ or a two-zone model for $`T_e`$, and these assumptions can also cause substantial errors in abundance determinations. Garnett (1992) provides one of the more thorough investigations of this issue, which is a well-known problem for elements like S that do not conform well to a two-zone model. However, if the adopted zone temperatures do not adequately characterize the respective regions, then the inferred abundance can be significantly in error even for elements like O that are well-described by a two-zone model. This can be a problem especially in cooler nebulae with $`T_e10,000`$ K, that have strong temperature gradients (Garnett 1992; Stasińska 1980).
We assumed here a two-zone temperature structure for our nebulae, such that a common $`T_e`$ was adopted for excitation of O III, Ne III, S III, and Ar III; while a separate $`T_e`$ was adopted for the excitation of N II, O II, and S II. We used the same electron density for both zones.
For the high ionization zone, we adopted $`T`$(O<sup>++</sup>), the volume-averaged $`T_e`$ for the O<sup>++</sup> population. This is obtained essentially directly from the observed $`T`$\[O III\], the temperature inferred from the \[O III\] $`\lambda `$4363/($`\lambda `$4959 + $`\lambda `$5007) line intensity ratio. A temperature can also be obtained in principle from the \[S III\] $`\lambda `$6312/($`\lambda `$9069 + $`\lambda `$9532) ratio. However, the auroral line flux in this case was often described by a large fractional uncertainty, which introduced correspondingly large random errors in the resulting abundances; there is also some possibility that the near-IR lines may be affected by telluric absorption (e.g., Stevenson 1994). We consequently chose to adopt $`T`$(O<sup>++</sup>)$`=T`$\[O III\] for the high-ionization zone, including the S<sup>++</sup> region. Garnett (1992) has suggested on the basis of nebular models that the relation between ion-weighted temperatures for S<sup>++</sup> and O<sup>++</sup> is linear, but with a slope differing from unity. For objects in the current study, $`T`$\[O III\] values are generally close to 10,000 K, at which point Garnett’s work suggests that the ion-weighted $`T`$(S<sup>++</sup>) and $`T`$(O<sup>++</sup>) should be similar; thermal effects of dust in the nebula may also be expected to reduce the contrast between the two temperatures (Shields & Kennicutt 1995), lending support to use of a common value.
For the low-ionization zone, $`T_e`$ can be obtained directly from the \[S II\]($`\lambda `$4069 + $`\lambda `$4076)/($`\lambda `$6716 + $`\lambda `$6731) ratio, but we found that this option frequently suffered from a low signal-to-noise ratio. It is often standard practice to derive the lower-ionization $`T`$(O<sup>+</sup>) from the higher-ionization $`T`$(O<sup>++</sup>) using an analytic relation (e.g., Campbell, Terlevich, & Melnick 1986; Pagel et al. 1992), whose accuracy clearly affects that of the abundance determination. We investigated several alternatives before adopting a prescription for $`T`$(O<sup>+</sup>). In deciding what relation to adopt, our principal criterion was that the abundances input to Mappings photoionization models for the individual objects (Paper I) should be consistent with those inferred from the output emission-line spectra. For example, when using the expression employed by Pagel et al. (1992, their equation 6), our resultant H ii region model spectra implied an O abundance that was $`0.2`$ dex lower than the assumed input, a substantial departure from self-consistency!
The relationship between $`T`$(O<sup>++</sup>) and $`T`$(O<sup>+</sup>) is model-dependent, and varies with ionization parameter and metallicity. Figure 1 shows $`T`$(O<sup>+</sup>) vs. $`T`$(O<sup>++</sup>) for Mappings photoionization models at $`Z=0.3\mathrm{Z}_{}`$, with an inner radius of 0.4$`R_\mathrm{S}`$. The different symbols correspond to different CoStar stellar atmospheres as indicated, with B2, C2, and E2 corresponding to spectral types O8–O9, O6–O7, and O3–O4, respectively. We show models with $`\mathrm{log}U=2,3,`$ and $`4`$. The dashed and dot-dashed lines show the relation between $`T`$(O<sup>+</sup>) and $`T`$(O<sup>++</sup>) from Campbell et al. (1986; see also Garnett 1992), and Pagel et al. (1992), respectively, while the dotted line delineates $`T`$(O<sup>+</sup>)$`=`$$`T`$(O<sup>++</sup>).
The Pagel et al. relation deviates significantly from the models at these temperatures, while the Campbell et al. relation shows reasonable agreement in slope for $`T`$(O<sup>++</sup>$`>10,000`$ K, falling close to the theoretical predictions for $`\mathrm{log}U=3`$. At lower temperatures, the models are more consistent with an isothermal nebula. We consequently adopted the formulation,
$$T(\mathrm{O}^+)=\{\begin{array}{cc}0.7T(\mathrm{O}^{++})+3000\mathrm{K},\hfill & T(\mathrm{O}^{++})>10,000\mathrm{K}\hfill \\ T(\mathrm{O}^{++}),\hfill & T(\mathrm{O}^{++})<10,000\mathrm{K}\hfill \end{array}$$
(3)
where $`T`$(O<sup>++</sup>)$`=T`$\[O III\] as described above. The relation for $`T`$(O<sup>++</sup>) $`>10,000`$ K is that given by Campbell et al. Equation 3 yields consistent input and output abundances to $`0.05`$ dex, except for Ne, whose modeled line ratios are persistently discrepant with the observations (Paper I; see below). We caution that Figure 1 and equation 3 are optimized in the parameter space relevant to our LMC targets. Different formulations may be more appropriate at other metallicities and ionization parameters; it is beyond the scope of this work to fully examine this issue. However, it is clear that some care is necessary in choosing a relation between $`T`$(O<sup>++</sup>) and $`T`$(O<sup>+</sup>).
While we ensured consistency between the input and output abundances to the photoionization models, one worrisome problem that remains unresolved is the overprediction of $`T_e`$ in comparison to the observations, as presented in Paper I. To briefly summarize, the predicted $`T`$\[O III\] is greater than that inferred from the observed \[O III\]$`\lambda `$4363/$`\lambda `$5007 by 850 K in DEM L323 and 1500 K in DEM L199. The problem may also exist in the other two objects, DEM L243 and DEM L301, but the fainter emission from $`\lambda `$4363 in these lower-excitation nebulae prevented any useful constraints. We also found that the discrepancy persisted when using Hummer & Mihalas (1970) stellar atmosphere models and the Cloudy (Ferland 1998) photoionization code. Since $`T`$(O<sup>++</sup>) and $`T`$\[O III\] agree to $`<1`$% in the models, the effect is not caused by non-collisional excitation of $`\lambda `$4363 in the models. As discussed in Paper I, the temperature discrepancy occurs in lines of sight across the entire nebulae, and is in the opposite sense of that expected from small-scale temperature fluctuations (Peimbert 1967). Our sample therefore shows no evidence that such $`T_e`$ fluctuations systematically bias our abundance determinations. Mathis (1995) discusses evidence for and against the general existence of significant $`T_e`$ fluctuations.
### 2.2. Ionic abundances
Determining ionic abundances relative to H requires calculation of the H$`\beta `$ emissivity, which is also temperature dependent, but originates in both the high- and low-ionization zones. We adopted an intermediate temperature for calculation of the H$`\beta `$ emissivity, representing an average of $`T`$(O<sup>++</sup>) and $`T`$(O<sup>+</sup>), weighted by the relative abundances of O<sup>++</sup> and O<sup>+</sup>. The same intermediate temperature was used for computing He emissivities, using expressions taken from Benjamin et al. (1999). This construction was used to obtain ionic abundances from the line measurements in Paper I, which were corrected for reddening and, where necessary, underlying Balmer absorption. The abundance results for different nebular lines of a single ion were combined, weighted by the variance of the line fluxes.
Total abundances were obtained from the ionic abundances following the standard practice of using ionization correction factors (ICFs) to allow for unobserved ionization stages. We employ the following relations:
$$\frac{\mathrm{O}}{\mathrm{H}}=\frac{\mathrm{O}^++\mathrm{O}^{++}}{\mathrm{H}^+},$$
(4)
where the lack of detectable He II$`\lambda `$4686 emission indicates no significant O<sup>+3</sup> population;
$$\frac{\mathrm{N}}{\mathrm{H}}=\frac{\mathrm{N}^+}{\mathrm{O}^+}\frac{\mathrm{O}}{\mathrm{H}}$$
(5)
(Peimbert & Costero 1969; Garnett 1990); and
$$\frac{\mathrm{S}}{\mathrm{H}}=\left[\frac{\mathrm{S}^++\mathrm{S}^{++}}{\mathrm{H}^+}\right]/\mathrm{ICF}(\mathrm{S}^{+3}),$$
(6)
where
$$\mathrm{ICF}(\mathrm{S}^{+3})=\left[1\left(1\frac{\mathrm{O}^+}{\mathrm{O}}\right)^\alpha \right]^{1/\alpha },$$
(7)
with $`\alpha =2.5`$, corrects for unobserved S<sup>+3</sup> ions (Garnett 1989; Stasińska 1978). For the noble gases, we adopt
$$\frac{\mathrm{Ne}}{\mathrm{H}}=\frac{\mathrm{Ne}^{++}}{\mathrm{O}^{++}}\frac{\mathrm{O}}{\mathrm{H}}$$
(8)
(Peimbert & Costero 1969; Simpson et al. 1995), and
$$\frac{\mathrm{Ar}}{\mathrm{H}}=\frac{\mathrm{Ar}^{++}}{\mathrm{S}^++\mathrm{S}^{++}}\frac{\mathrm{S}}{\mathrm{H}}$$
(9)
(Garnett et al. 1997). Most of our objects are ionized by early-type stars (Table 1), thereby fully ionizing He, but not exhibiting detectable He II. Thus our default relation for He is simply,
$$\frac{\mathrm{He}}{\mathrm{H}}=\frac{\mathrm{He}^+}{\mathrm{H}^+}.$$
(10)
For DEM L243, which is ionized by O7 stars, the He I $`\lambda `$5876/H$`\beta `$ line ratios suggest that He is not fully ionized throughout the nebula (Paper I; see below). For this object, we experimented with the expression for He/H given by Peimbert & Torres-Peimbert (1977; their equation 15), that includes an ICF for He<sup>0</sup>; however, this prescription yields He abundances that are at least 0.1 dex higher than the rest of the objects in the sample. We therefore give results for DEM L243 using equation 10 and note that these should be treated as lower limits to the true abundance. It is worth remarking that our investigation supports the results of Baldwin et al. (1991), who find that the Orion He abundance was slightly overestimated by Peimbert & Torres-Peimbert as a result of their ICF for He<sup>0</sup>.
The theoretical support for use of equations 49 derives mostly from models that consider the integrated properties of H ii regions (e.g., Garnett 1992; Mathis 1985; Stasińska 1978). Thus, observations that sample only a small “pencil-beam” through a nebula may not yield the correct abundances if such an analysis scheme is used; Gruenwald & Viegas (1992) have discussed this problem in detail. However, previous observational studies have found generally good agreement in elemental abundances derived at variable positions across individual nebulae, using the integrated-spectrum methods (e.g., Díaz et al. 1987; Masegosa, Moles, & del Olmo 1991; González-Delgado et al. 1994). Such agreement was found even when substantial excitation gradients were seen within the H ii region. In the present study, we can directly compare abundances obtained from the integrated spectrum of a nebula with those calculated by the same means from small-aperture measurements at a variety of radii. We show below in §2.3.1 that we find a high degree of consistency between the different aperture measurements for the same source, including the integrated spectrum. These results support the validity of using integrated-spectrum prescriptions when only a part of the H ii region is observed. Apparently the characteristic physical parameters determined for the two ionization zones are adequate to determine the ionic abundances to high accuracy in our objects.
### 2.3. Abundance results
We used the prescriptions above to compute abundances for both the spatially resolved and scanned, spatially integrated observations for each object. Table 2 presents results for the individual, spatially resolved observations, along with error estimates. These were obtained from the uncertainties in measured fluxes, using a Monte Carlo method similar to that described by Kobulnicky & Skillman (1996). Input flux values were modified randomly by addition of Gaussian noise with amplitude corresponding to 1-$`\sigma `$ measurement uncertainties, for a total of 5000 iterations per spectrum. A few points with low signal-to-noise ratios yielded unphysical solutions in the $`T_e`$ and abundance determinations, and these were discarded from the sample. The standard deviation of the resulting abundance distribution was adopted as the final uncertainty.
Table 3 presents the mean values of the spatially-resolved measurements, weighted inversely by the variances. We also list the corresponding formal uncertainty in the mean. Note that the weighting scheme and quoted errors assume that actual variations are negligible in the quantities listed in Table 3, and that the scatter results from a normal distribution of measurement errors. As described below, we do not find significant evidence for abundance fluctuations; however, it is possible that the measurement errors may not be a strictly normal distribution. The listed errors in Table 3 are therefore likely to somewhat underestimate the true uncertainties, nor do these include systematic errors.
Regarding the abundances of Ne, it is important to note that there is probably a substantial systematic error in the values obtained in Table 3, and also later in Table 4. As discussed in Paper I, the observed line intensities for Ne are systematically discrepant with the tailored models. Although earlier generation stellar atmospheres caused \[Ne III\] intensities to be underpredicted, we now find a modest overprediction. The discrepancy can be resolved by reducing log(Ne/H) by 0.2 – 0.3 dex. However, since we do not understand the cause of the discrepancy, we have chosen to list the Ne abundance derived from the standard relations. It appears likely that errors in the stellar atmospheres are responsible for much of the problem (Paper I). Peimbert (1993) also emphasizes the uncertainty in equation 8. The high ionization potential (40.96 eV) required for Ne III gives it outstanding potential for probing hot stellar ionizing sources, so it is highly desireable to resolve the uncertainties regarding its emission.
#### 2.3.1 Spatial uniformity
In Figure 2$`ae`$, we show the spatial distribution of elemental abundances across the sightlines for our stationary observations of DEM L199. Figure 2$`f`$ shows the corresponding distribution in $`T`$\[O III\] measurements. For DEM L243, DEM L301, and DEM L323, we show in Figures 35 the results for He, N, and O, as representative of the other elements. These are shown respectively in panels $`a,b`$, and $`c`$; and panels $`d`$ show corresponding measurements of $`T`$\[O III\]. The different symbols correspond to individual slit positions as designated in Paper I (Figures 6, 8, 12, and 13), to aid cross-referencing. Following our convention in that work, we simply show the slit positions superposed, therefore these figures will not necessarily show clean radial profiles across the nebulae, although they do approximate this reasonably well. This issue may be inspected in Paper I, along with actual slit positions. The light, horizontal lines indicate the spatial extent of each extracted aperture, while the vertical error bars indicate uncertainties in the derived measurements.
In Figure 3, DEM L243 shows much larger scatter and uncertainty among the individual apertures, which can be attributed to the large measurement uncertainties for $`T`$\[O III\] (Figure 3$`d`$). There is also an SNR either embedded or superimposed in the line of sight to this object, but we have not included the affected apertures in our abundance estimates or in Figure 3. Assuming an essentially constant metallicity in DEM L243, we recomputed the abundances by fixing $`T`$(O<sup>++</sup>) to be the mean $`T`$\[O III\], using the same weighting and omitting the same deviant values as before, in computing the mean abundances. Figure 6 and Table 3 show the results using the new $`T`$(O<sup>++</sup>)$`=`$9700 K. The dramatic reduction in scatter for the heavy elements suggests that the original scatter in Figure 3 may indeed be caused by poor measurements in $`T`$\[O III\]. However, we emphasize a point by Mathis, Chu, & Peterson (1985), that true abundance fluctuations will induce corresponding fluctuations in $`T_e`$, owing to the more efficient cooling accompanying higher O/H. The reduced scatter with a fixed $`T_e`$ is therefore not a strong demonstration of truly constant abundances. To further test consistency with constant abundance distributions, we computed the reduced-$`\chi ^2`$ statistics for the original distribution, yielding 1.4 and 1.8 for log(N/H) and log(O/H), respectively. The probabilities of obtaining these values for 12 degrees of freedom are 16% and 4%, respectively. These values therefore hint that the scatter may in part be caused by real variations, although the significance is low.
In the event that the abundances for DEM L243 are essentially constant, the mean values obtained with fixed $`T`$\[O III\] should give a better estimate than the original values (Table 3). However, we caution that there is systematic uncertainty introduced by the adopted value of $`T`$(O<sup>++</sup>); comparison to the original mean abundances in Table 3 suggests the uncertainty is roughly 0.1 dex in the derived metallicities. Since the recombination lines used to determine log(He/H) are less temperature-sensitive than the collisional metal lines, log(He/H) (Figure 6$`a`$) still shows larger scatter than exhibited in the other objects. However, note that the magnitude of the variation is only $`0.3`$ dex. As discussed in Paper I, the He I $`\lambda `$5876/H$`\beta `$ ratios suggest that He is not uniformly fully ionized in DEM L243, therefore causing log(He/H) to be underestimated in many apertures. Our value for the He abundance is therefore a lower limit in this object. The upper envelope to the distribution in Figure 6$`a`$ is around log(He/H) $`1.1`$, which may be more indicative of the true He abundance. This value is more consistent with those for the other objects (Table 3).
Thus, in Figures 25 the abundances appear to show no spatial variations within the measurement uncertainties, with standard deviations typically around 0.10 – 0.15 dex. The uniformity of the abundance derivations is reassuring, and suggests that our adopted ionic relations (§2.2) and description of the nebular temperature structure (equation 3) yield abundance estimates with high accuracy. The apparent success of these methods is consistent with the finding by Mathis (1985) that ICFs for model nebulae appear fairly robust between volume averaged and radially averaged regions. The uniformity of results for apertures tracing different parts of the same nebula further suggests that spurious “pencil-beam” effects resulting from projection of radial $`T_e`$ gradients (Gruenwald & Viegas 1992) are negligible for our objects. This is consistent with the fact that sources in our sample have $`T`$\[O III\] restricted to the approximate range 9500 – 12,000 K, for which the nebulae are expected to be relatively isothermal (equation 3; Figure 1). We emphasize that this propitious circumstance may not apply to other H ii regions, that may exhibit stronger $`T_e`$ gradients; as an example, Walter, Dufour, & Hester (1992) reported a significant $`T_e`$ gradient in the Orion nebula, although their spatial scale of 0.5 pc is almost an order of magnitude smaller than is relevant for our objects in the present study. As in §2.1 above, we again emphasize the importance of choosing the correct parameterization for the $`T_e`$ structure.
#### 2.3.2 Search for self-enrichment
It is especially interesting to examine DEM L199 for possible abundance variations that are introduced by the three WR stars. Two of these are binaries with a WN3 component (Breysacher 1981) and one is a WC4 + O6 V-III binary (Moffat et al. 1990). The exact location of these stars with respect to the nebular gas distribution and slit positions may be examined in Paper I. We emphasize that DEM L199 is not a WR ejecta shell, but is a large, luminous H ii complex whose dominant ionizing stars are early WR stars. Walsh & Roy (1989) and Kobulnicky et al. (1997) identified two regions in the starburst galaxy NGC 5253 that appear to show enhanced N, which is suggested to result from self-enrichment by WR stars. Kobulnicky et al. (1997) point out that accompanying He enrichment would be expected from the WR sources, and acknowledge that the lack of He excess in these objects is puzzling.
DEM L199 has three WR stars within a large central cavity in a luminous H ii region. It would seem likely that, if self-enrichment from WR stars can be seen, it should be detected in our observations. In Figure 2, we indicate six apertures that are closest to the stars Br 32 (WC4) and Br 33 (WN3), with solid dots. These apertures are identified in Paper I as: D199.205-15, 16, and 17; and D199.205N120-17, 18, and 19. The characteristic physical distances from the WR stars are about 5 – 15 pc. Figure 2 shows that these points do not show any sign of abundance enhancements. Results from the full set of aperture measurements for DEM L199 are consistent with uniform abundances at the 90% confidence level, as indicated by reduced-$`\chi ^2`$ values of 0.59 and 0.60, for log(N/H) and log(O/H) respectively, with 17 degrees of freedom. For He, in particular, the variation is constrained to be $`0.1`$ dex. Although abundance enhancements are seen in WR ejecta nebulae, our data suggest that self-enrichment by WR stars can be an extremely subtle phenomenon. It is possible that the processed material is heated to coronal temperatures within the superbubble and is thus optically undetectable. Alternatively, the WR phase in these stars may not yet have lasted long enough to produce significant enrichment in the surrounding environment.
We were also interested to see whether the SNR in DEM L243 showed evidence of abundance anomalies. Abundance estimates from SNRs are necessarily more uncertain, owing to the more complicated radiative transfer, but meaningful estimates have been made by, e.g., Russell & Dopita (1990). These are based on matching the emission-line spectra with a shock code. In Paper I, we were able to find excellent agreement between Mappings shock models and the observed emission from the shock-affected apertures, using the abundances derived from the uncontaminated regions of the nebula listed in Table 3. The SNR therefore shows no evidence of abundance anomalies, although strong emission from photoionized gas in the same line of sight prevents strong constraints.
Likewise, the superbubble DEM L301 shows strong evidence of a recent SNR impact (Chu & Mac Low 1990; Oey 1996b; Paper I), and thus could conceivably show enrichment by massive star winds and supernova ejecta. However, its elemental abundances are often lower than for the other objects in the sample (Table 3). We caution that shock-excitation can significantly affect the abundance determinations for this object (Paper I). Peimbert, Sarmiento, & Fierro (1991) showed that contamination by shock activity can cause abundances to be underestimated, especially for higher-ionization species. Our measurements for DEM L301 are consistent with this behavior, although in §2.3.3 we find that the metallicities for DEM L243 do not appear strongly affected by SNR contamination.
#### 2.3.3 Spatially integrated abundances <br>and LMC metallicities
We also derived abundances for the scanned, spatially integrated apertures using the same methods. These are presented in Table 4, along with mean LMC H ii region abundances from compilations by Dufour (1984). A more recent compilation by Garnett (1999) shows essentially the same values. Within the uncertainties, the mean abundances for each object from Table 3 agree with the determinations from the spatially integrated observations (Table 4), although the offsets appear to be systematic across all elements for each object. This again points to uncertainties in the $`T`$(O<sup>++</sup>) determination, which can result from simple measurement errors, or factors related to the spatial integration of the line emission. We regard the values obtained from the mean of the spatially resolved data (Table 3) to be more reliable than the single observations from the integrated data. However, we caution that the ionic relations in §2.2 are based on spatially integrated models and observations, and this could introduce systematic variations between the spatially resolved and scanned data. But taking the derived values and uncertainties at face value, the mean abundances of the stationary apertures should be somewhat more reliable.
We include in Table 4 the abundances for DEM L243 derived from both the total, scanned observation and the scanned observation with the SNR-contaminated region subtracted. Interestingly, there is no significant difference between these, although the data including the SNR do show the expected decrease in computed abundances (Peimbert et al. 1991). Thus, while DEM L301 showed suspiciously low metallicity measurements attributable to effects of shock emission, DEM L243 is an example where the SNR is not a significant factor. In Paper I, we also found that the two shock-affected objects exhibit different behavior in their line emission with respect to the photoionized regions, thereby demonstrating how shocks contribute in different ways to the spectra of host H ii regions, depending on shock velocity and environment.
We find a tendency for our measurements to be $`0.2`$ dex lower than the mean LMC metal abundances compiled by Dufour (1984). One of the probable causes is our adopted temperature structure (equation 3), which varies slightly from those used by others. For example, we find that our mean abundances for DEM L243 would increase by about 0.1 dex if we adopted the relation of Campbell et al. (1986) at all values of $`T`$(O<sup>++</sup>).
Our data are generally consistent with there being no abundance variations between the four different H ii regions. It is interesting to note that DEM L199 is close to the LMC bar, about 1 kpc from the center of the galaxy; and DEM L243 is situated in the northern outskirts of the LMC-4 supergiant shell, at a galactocentric radius of $``$3.5 kpc. Pagel et al. (1978) have suggested that the LMC H ii regions possibly exhibit a slight abundance gradient. This has not been further examined, nor has a gradient been detected in the cluster population (Olszewski et al. 1991). In our data, it is suggestive that DEM L199 and DEM L243 delineate the extremes of any interpreted variation among our four objects. The difference in metallicity is consistent with the small gradient suggested by Pagel et al. (1978).
## 3. Semi-empirical bright-line methods
We turn now to examining more indirect emission-line diagnostics of metal abundances. In situations where $`T_e`$ cannot be adequately constrained by observation, it is common practice to estimate the metal abundances using the semi-empirical, “bright-line” abundance parameters. Here we examine the performance of these parameters in light of our detailed nebular data and highly-constrained, tailored photoionization models from Paper I.
We also compute model tracks of the abundance parameters, using Mappings II with generalized nebular parameters. These incorporate the stellar energy distribution of CoStar model C2 (Schaerer & de Koter 1997), which corresponds to an O6 – O7 stellar effective temperature. We assume an inner radius to the gas distribution of 0.4$`R_\mathrm{S}`$, and gas density $`n=10\mathrm{cm}^3`$. In the figures that follow, the dashed, solid, and dotted lines correspond to the volume-averaged $`\mathrm{log}U=2,3,`$ and $`4`$, respectively, which is equivalent to changing the total ionizing photon emission rate or gas filling factor.
The grid of models is computed with $`Z=0.05,`$ 0.1, 0.3, 0.5, 1.0, and 2.0 times $`\mathrm{Z}_{}`$. We included the elements (He, C, N, O, Ne, Mg, Al, Si, S, Ar, Ca, and Fe) with $`\mathrm{Z}_{}=(1.01,3.44,3.95,3.07,3.91,4.42,`$ $`5.53,4.45,4.79,5.44,5.88`$, and –4.96), respectively. We largely follow McGaugh (1991) in scaling the abundances of individual elements with respect to O. For He and N, we use the relations given by McGaugh, but scaled to match the Anders & Grevesse (1989) values for $`\mathrm{Z}_{}`$:
$$\mathrm{He}/\mathrm{H}=0.0850+15(\mathrm{O}/\mathrm{H})$$
(11)
and
$$\mathrm{log}(\mathrm{N}/\mathrm{H})=1.5\mathrm{log}(\mathrm{O}/\mathrm{H})+0.66.$$
(12)
For C and Fe, we adopt McGaugh’s relations directly (his equations 10 and 11). The remainder of the elements are fixed in their proportion to O at $`\mathrm{Z}_{}`$, as given by Anders & Grevesse. Models for $`\mathrm{Z}_{}`$ and $`2\mathrm{Z}_{}`$ use the CoStar C2 atmosphere at solar metallicity, while the rest use the corresponding SMC metallicity model; we find that the stellar metallicity is unimportant for these $`U`$-tracks.
### 3.1. $`R23`$
The O abundance parameter $`R23`$ (Pagel et al. 1979; equation 1 above), has been extensively used and empirically calibrated several times, by McGaugh (1991), Skillman (1989), and Dopita & Evans (1986), among others. In Figure 7$`a`$, we show log(O/H) vs. $`\mathrm{log}R23`$, with the tracks showing results from generalized photoionization models described above. Our assumptions differ somewhat from those of previous studies, in particular with the assumption of a fairly hollow morphology and new stellar atmosphere models. Most authors (e.g., McGaugh 1991; Dopita & Evans 1986) also assume some anticorrelation between the characteristic $`T_{}`$ or $`U`$, and $`Z`$, in their adopted calibration at high abundance. Our tracks do not assume this anticorrelation. Despite these differences, our tracks remain similar to those of previous authors, although our models show a slight offset to higher log(O/H) (see Kobulnicky et al. 1999 and McGaugh 1991 for comparisons of $`R23`$ calibrations).
We also plot in Figure 7$`a`$ the spatially integrated data for our objects, using the abundances derived from the means of our resolved apertures (Table 3). The values of $`R23`$ are computed in Paper I and shown here in Table 5. We caution that the nebular fractional area included in the spatial scans varies among the four objects, and we refer the reader to Paper I for the precise details. The three spatial scans of the spherical object DEM L323 (triangles) should give an indication of the degree to which subsampling is representative of the total spatial scan (solid triangle). The three measurements of $`R23`$ are in excellent agreement, which is consistent with our finding in Paper I that this index is robust to spatial variations. For DEM L243, we show $`R23`$ derived from the spectrum with the SNR-contaminated region subtracted (solid diamond); and also that from the total integrated region including the SNR (open diamond). The square and cross show DEM L199 and DEM L301, respectively.
Our data points are generally well-behaved with respect to the model tracks in Figure 7$`a`$. While we found excellent agreement between the observed emission-line spectra and our tailored photoionization models in Paper I, we see in Figure 7$`a`$ that most of the objects fall in their expected location with respect to the more generalized model tracks. DEM L323, DEM L243, and DEM L199 fall between tracks of $`\mathrm{log}U=3`$ and $`4`$, with DEM L199 showing the highest value of $`U`$, as expected in this high-excitation object. The one anomalous point is DEM L301 (cross), which we found in Paper I to have an unusual combination of excitation mechanisms. We concluded that this object, which has an extreme shell morphology, is most likely ionized by a combination of density-bounded photoionization plus shocks. While our tailored model for this object reproduced the observed $`R23`$ well, it is apparent that its value is anomalously high with respect to the tracks in Figure 7$`a`$. Ironically, the offset is in the sense of a higher ionization parameter, although the object in fact has a much lower $`U`$ than the others in the sample. The larger value of $`R23`$ is probably caused by the enhanced emission contributed by the shock activity. DEM L301 and DEM L199 have similar values of $`R23`$, which, lacking any additional information for these objects, would imply similar abundances; we see that in fact this would overestimate log(O/H) for DEM L301 by about 0.3 dex, taking the measured log(O/H) at face value.
We also overplot with small plus signs in Figure 7$`a`$ the data compiled by Díaz & Pérez-Montero (2000). These represent data from the literature (their Table 2) for which abundances were derived from a direct measurement of $`T`$\[O III\]. Although the points for our data from Paper I are consistent with the model tracks, it is apparent that the photoionization models in Figure 7$`a`$ do not track well the locus of the larger dataset. It is important to note that adopting softer stellar atmospheres can improve the correspondence slightly, since this would offset the tracks to slightly lower log(O/H). We refer the reader to McGaugh (1991) to evaluate the consequences of the stellar effective temperature. The atmospheres adopted here (CoStar C2) correspond to O6 – O7 stars, which are already cooler than most of our LMC objects, and only relatively small changes result if O3 – O4 atmospheres (CoStar E2) are used instead. The discrepancy between models and data has always been a difficulty in the use of $`R23`$, and therefore empirical calibrations of this parameter have been crucial for its successful use.
### 3.2. $`S23`$
As mentioned earlier, a parameter similar to $`R23`$ has been introduced for S by Vílchez & Esteban (1996) and Christensen et al. (1997), which was further explored by Díaz & Pérez-Montero (2000, hereafter DPM). It is important to note that this parameter, $`S23`$ (equation 2), is not strictly analogous to $`R23`$. While O and S have homologous energy levels, the ionization potentials (IP) for their respective ions are different. In particular, it is important to note that the IP required to reach S<sup>+3</sup> (34.83 eV) is virtually identical to that necessary for O<sup>++</sup> (35.12 eV). Therefore, although S<sup>+</sup> and S<sup>++</sup> are indeed the dominant ions for S, for typical H ii regions, there is likely to be non-negligible S<sup>+3</sup>, which is ionized by the same radiation that produces O<sup>++</sup>. Although Christensen et al. (1997) pointed out that the ionization fraction of S<sup>+3</sup> is relatively small, typically $`0.2`$, we show below that it nevertheless significantly affects the ionization balance of S<sup>+</sup> and S<sup>++</sup>, and consequently, the value of $`S23`$.
Figure 7$`b`$ is similar to panel $`a`$, now showing $`\mathrm{log}`$ (S/H) vs $`\mathrm{log}S23`$ (Table 5). The model line types and data symbols are the same as before. It is immediately apparent that, contrary to earlier claims in the literature, $`S23`$ is more sensitive to the ionization parameter than $`R23`$. The change in $`\mathrm{log}S23`$ between the model tracks, varying $`\mathrm{log}U`$ from –2 to –4, is almost 0.5 dex, whereas it is less than 0.3 dex for $`R23`$. The greater $`U`$-sensitivity of $`S23`$ is caused by the “missing” contribution of S IV. Figure 7$`b`$ shows that the models with high $`U`$ show lower $`\mathrm{log}S23`$, the opposite pattern to $`R23`$. This is consistent with the ionization behavior of S, since a larger population of S<sup>+3</sup> is expected at higher $`U`$.
As suggested by DPM, the lower-metallicity branch of $`S23`$ does span a larger range in values than $`R23`$, for a given range of $`Z`$ (Figure 7). However, the location of the inflection at the maximum $`S23`$ is at only a slightly higher $`Z`$ than that for $`R23`$. Figure 7 shows that our models for $`\mathrm{log}R23`$ have a maximum close to $`Z=0.3\mathrm{Z}_{}`$, and those for $`\mathrm{log}S23`$ have a maximum around $`Z=0.5\mathrm{Z}_{}`$. Thus, there is only $`0.2`$ dex extension in the dynamic range of $`Z`$ in the use of $`S23`$. Nevertheless, since so many of the observed objects in the literature have abundances around $`0.30.5\mathrm{Z}_{}`$, this augmentation makes a substantial difference in evaluating abundances. As is dramatically shown in Figure 1 of DPM, $`S23`$ empirically shows an evidently monotonic increase as a function of $`Z`$, in contrast to $`R23`$, which shows a distinct double-valued structure in $`Z`$.
Our data points are again reasonably consistent with the models in Figure 7, although they now perhaps show a tendency to fall between the $`\mathrm{log}U=2`$ and $`3`$ models, rather than –3 to –4 for Figure 7$`a`$. The tailored photoionization models and spatially-resolved data in Paper I showed similar minor discrepancies. The data points for the shock-affected regions, DEM L301 (cross) and the SNR-contaminated observation for DEM L243 (open diamond) are offset to higher $`\mathrm{log}S23`$. This suggests that $`S23`$ is increased by the presence of shock excitation, similar to the behavior of $`R23`$ found in the previous section.
We also see in Figure 7$`b`$ that the photoionization models do track the data well for $`S23`$, and much better than for $`R23`$ in panel $`a`$. However, it is also apparent that the points with the highest values of $`S23`$ fall outside the model tracks. In formulating a calibration for $`S23`$, we would therefore recommend that these values be excluded, and that the empirical calibration derived by DPM should not be used for $`Z0.5\mathrm{Z}_{}`$.
### 3.3. $`S234`$
As discussed in the previous section, the population of S<sup>+3</sup>, which is ionized by the same radiation as that ionizing O<sup>++</sup>, is not sampled by the $`S23`$ parameter. While the ionization fraction of S<sup>+3</sup> is relatively small, we have seen in the previous section that it significantly compromises the utility of $`S23`$ as an abundance diagnostic. We therefore suggest that the parameter,
$$S234\frac{[\mathrm{S}\mathrm{II}]\lambda 6724+[\mathrm{S}\mathrm{III}]\lambda \lambda 9069,9532+[\mathrm{S}\mathrm{IV}]\lambda 10.5\mu }{\mathrm{H}\beta }$$
(13)
is a better abundance indicator than $`S23`$. In the same way that $`R23`$ samples all significant ions of O, $`S234`$ more completely samples the significant ions of S. Note that any population of S<sup>+4</sup> (IP 47.30 eV) will be an insignificant fraction of the total S ions, for massive star sources: S<sup>+4</sup>/S $`0.02`$ even for an object ionized by a WR star where $`\mathrm{He}^{++}/\mathrm{He}=0.4`$. The principal difficulty with $`S234`$ is the inclusion of the mid-IR line \[S IV\]$`\lambda 10.5\mu `$, which is not readily observable with standard ground-based instrumentation. However, since the IP necessary to produce S<sup>+3</sup> is virtually the same as that for O<sup>++</sup>, it is possible to estimate the abundance of S<sup>+3</sup> based on that for O<sup>++</sup>. This was demonstrated earlier by Mathis (1982) and Dennefeld & Stasińska (1983). We present this approach here.
Figure 8 presents $`\mathrm{log}`$ (S/H) vs. $`S234`$, on the same scale as that for $`S23`$ in Figure 7$`b`$. The line types and symbols are the same as before. We see that $`S234`$ is dramatically less sensitive to the ionization parameter, owing to the inclusion of the S IV indicator. Indeed, the models for $`S234`$ are even less sensitive to $`U`$ than is $`R23`$ (Figure 7$`a`$), for $`Z\mathrm{Z}_{}`$.
It would therefore be desireable to estimate the intensity of \[S IV\]$`\lambda 10.5\mu `$ from that of the O ionization indicator, \[O III\]/\[O II\]. Figure 9 shows log(\[S IV\]$`\lambda 10.5\mu `$/\[S III\]$`\lambda `$$`\lambda `$9069,9532) vs. log(\[O III\]$`\lambda `$$`\lambda `$4959,5007/\[O II\]$`\lambda `$3727) for models with an E2 CoStar atmosphere, and line types as before. This atmosphere corresponds to an O3 – O4 stellar type, and we prefer this model in examining the relation between S IV and other ions since it is more relevant in environments with harder ionizing fields. We see that for $`Z0.5\mathrm{Z}_{}`$ (solid points), the relation between these ratios is essentially a simple power law. For these points, we fit:
$$\mathrm{log}\frac{[\mathrm{S}\mathrm{IV}]\lambda 10.5\mu }{[\mathrm{S}\mathrm{III}]\lambda \lambda 9069,9532}=0.984+1.276\mathrm{log}\frac{[\mathrm{O}\mathrm{III}]\lambda \lambda 4959,5007}{[\mathrm{O}\mathrm{II}]\lambda 3727},$$
(14)
which is shown by the dot-dashed line in Figure 9.
With observations of \[S III\], this relation allows an estimate of the \[S IV\] intensity, which can then be used to compute $`S234`$. We note that the adoption of the cooler C2 CoStar atmospheres, used in our other photoionization models, would result in a difference of less than 0.02 and 0.01 in the fitted intercept and slope, respectively. Furthermore, the correction for \[S IV\] will only be significant for moderate to high $`U`$ and/or high $`Z`$. As a test of equation 14, we use mid-infrared and optical line observations of the Orion nebula by Lester, Dinerstein, & Rank (1979). For their measurements, equation 14 predicts a volume emissivity for \[S IV\]$`\lambda `$10.5$`\mu `$ of $`6.5\pm 2.7\times 10^{21}\mathrm{erg}\mathrm{cm}^6\mathrm{s}^1`$, which agrees within measurement uncertainties with the observed value of $`9.0\pm 2.7\times 10^{21}\mathrm{erg}\mathrm{cm}^6\mathrm{s}^1`$. Considering the extremely narrow, 10$`\mathrm{}`$ line of sight on the Orion nebula, and much higher density ($`10^4\mathrm{cm}^3`$) than considered for our purposes, this agreement is highly encouraging.
It is thus relatively simple to convert from $`S23`$ into $`S234`$ and thereby almost eliminate the sensitivity to $`U`$. We used this method of estimating \[S IV\] to compute $`S234`$ for our spatially-integrated observations (Table 5), which are plotted in Figure 8, using the same symbols as before. In the errors for $`S234`$, we include in quadrature an uncertainty of 25% for the uncertainty of \[S IV\] from equation 14. We again have excellent agreement with the models. It is clear that $`Z`$ can be estimated with greater confidence based on $`S234`$ than $`S23`$ at these metallicities, because the large spread in the models for $`S23`$ (Figure 7$`b`$) has been vastly reduced for $`S234`$.
Similarly, we show in Figure 10 that spatial variations are also reduced from $`S23`$ to $`S234`$. Figures 10$`a`$ and $`b`$ show our spatially resolved observations of DEM L199 for $`S23`$ and $`S234`$. The solid line indicates the tailored model for this object, using the early WR model of Schmutz et al. (1992; see Paper I), central hole radius of 0.5$`R_\mathrm{S}`$, and gas density $`n=100\mathrm{cm}^3`$. While Figure 10$`a`$ shows a large spatial variation of $`0.4`$ dex for $`S23`$, we see in Figure 10$`b`$ that the variation in $`S234`$ is reduced by about a factor of 2 in the logarithm. Figures 10$`ef`$ show the same behavior for DEM L323. The solid line again represents the corresponding tailored model, with an O3 – O4 stellar atmosphere (Costar E2), central hole radius of 0.4$`R_\mathrm{S}`$, and gas density $`n=10\mathrm{cm}^3`$. In Figure 10$`cd`$, we show the spatially resolved data for DEM L199 superimposed on the model tracks for $`S23`$ and $`S234`$, respectively. The reduced scatter in $`S234`$ against the models again demonstrates the improved constraints in estimating log(S/H), compared to $`S23`$.
### 3.4. Calibrations
In Figure 11$`a`$ and $`b`$ we show the models for $`\mathrm{log}(\mathrm{S}/\mathrm{H})`$ as a function of $`S23`$ and $`S234`$, overplotted with the Galactic and LMC data presented by Dennefeld & Stasińska (1983). Their S abundances are computed from measurements of $`T`$\[O III\] and observations of \[S III\]$`\lambda `$$`\lambda `$9069,9532. We compute $`S234`$ from this dataset with the aid of equation 14, as described above. Figure 11 shows that the data present a well-defined sequence in both $`S23`$ and $`S234`$. It is especially encouraging that the locus of the models is in excellent agreement with that of the data, in contrast with the situation for $`R23`$, as we saw above in Figure 7$`a`$. We replot the $`R23`$ models with the Dennefeld & Stasińska data in Figure 11$`c`$, again suggesting the same discrepancy seen earlier. It is apparent that for these data, the values of $`R23`$ are fairly insensitive to log(O/H) as we saw before, owing to the location of the inflection and spread in $`U`$.
To estimate a theoretical calibration for $`S23`$, we take the mean of the three models at each metallicity, up to $`0.5\mathrm{Z}_{}`$. A resulting power-law fit is shown by the lighter, straight, solid line in Figure 11$`a`$. For $`S23`$, we obtain:
$$\mathrm{log}(\mathrm{S}/\mathrm{H})=5.43+1.33\mathrm{log}S23.$$
(15)
The light dash-dot and dashed lines in Figure 11$`a`$ show the DPM and Christensen et al. (1997) calibrations, respectively. DPM calibrated a relation for log(O/H) vs. $`S23`$, so we used the solar S/O ratio to convert their relation to a calibration of log(S/H). It is apparent that all three calibrations are similar. The DPM relation shows the best correspondence to the data, as is expected since it is fitted to the largest dataset. It is especially encouraging that our theoretical relation is intermediate between the two empirical ones, confirming that the theoretical calibration is fully consistent with the available data. However, in using any of these $`S23`$ calibrations, it is important to bear in mind that the models predict a double-valued relation around $`\mathrm{log}S230.0`$.
We use the same procedure to fit a theoretical calibration for $`S234`$ and obtain:
$$\mathrm{log}(\mathrm{S}/\mathrm{H})=5.58+1.27\mathrm{log}\mathrm{S234}.$$
(16)
As in the case for $`S23`$, the data are in excellent agreement with this rough theoretical fit, shown by the light, solid line in Figure 11$`b`$.
We again emphasize that, although the data at $`Z>0.5\mathrm{Z}_{}`$ are consistent with the calibrations for both $`S23`$ and $`S234`$, they strongly diverge from the models in that regime, and that a power-law approximation is necessarily crude near these values. We therefore consider the calibrations reliable only for $`Z0.5\mathrm{Z}_{}`$, and extreme caution should be exercised in extrapolating at higher metallicity. It is also essential to note that the double-valued nature of all of these abundance parameters still remains an issue.
In Figure 11$`d`$, we show the measured log(O/H) vs. $`\mathrm{log}S234`$ for the Dennefeld & Stasińska sample, where log(O/H) are again derived from direct measurements of $`T`$\[O III\]. We see that the scatter is much larger than for log(S/H) vs. $`\mathrm{log}S234`$ (Figure 11$`b`$). Although Garnett (1989), among others, suggests that there is no systematic variation in S/O with O/H, Figure 11 shows that there is still significant variation in the S/O ratio among the different objects. Therefore, while $`S234`$ appears reasonably reliable for estimating the S abundance, it appears to be significantly less reliable for inferring the O abundance, and caution should be exercised accordingly.
## 4. Conclusion
We have carried out a detailed investigation of elemental abundance derivations using four H ii regions in the LMC. We use tailored photoioinzation models to examine standard abundance analyses based on measured values of $`T_e`$. Our data (Paper I) are derived from both spatially-resolved observations extracted from stationary long slit positions, and scanned, spatially-integrated slit observations. We also examine the bright-line abundance diagnostics for O and S, in light of the direct abundance determinations and photoionization models.
Our abundance determinations are based on measurements of $`T`$\[O III\], which we take to represent $`T`$(O<sup>++</sup>), and we assume a two-zone temperature structure for the nebulae, represented by $`T`$(O<sup>++</sup>) and $`T`$(O<sup>+</sup>). We use standard ionic abundance relations to then determine the total elemental abundances for He, N, O, Ne, S, and Ar, with respect to H. Comparison with tailored Mappings photoionization models highlights the importance of choosing a relation between $`T`$(O<sup>++</sup>) and $`T`$(O<sup>+</sup>) that adequately represents the nebular temperature structure. Failure to do so can result in metallicity estimates that are discrepant by at least 0.2 dex from values indicated by tailored photoionization models.
Abundance measurements for the stationary slit positions show high spatial uniformity, with no evidence of variations or gradients to within 0.1 – 0.15 dex. Thus it is unlikely that there are systematic biases resulting from the strong ionization gradients seen in these objects. The adopted two-zone $`T_e`$ structure therefore appears to be highly reliable for estimating ionic abundance estimates even through “pencil-beam” apertures that sample only a small nebular area, at least for our fairly isothermal H ii regions.
No areas of local enrichment were detected in DEM L199, in spite of the presence of two WN3 stars and one WC4 star. The stellar products may be hidden in hot, coronal gas within the central superbubble, or the stars may not have produced enough enriched material to be readily detectable. The results show that self-enrichment by WR stars is likely to be a complex phenomenon, empirically. DEM L243 and DEM L301, both showing evidence of recent SNR activity, also do not show local enrichments, although with poorer constraints.
Abundance measurements from the scanned, spatially integrated apertures are consistent with those obtained from the spatially resolved observations. Our results are $`0.2`$ dex lower than average LMC H ii region measurements (Dufour 1984; Garnett 1999), probably resulting in part from different descriptions for the $`T_e`$ structure. The spatially-integrated measurements are also consistent with there being no variation between the four H ii regions, although, interestingly, they are also consistent with the marginal abundance gradient suggested by Pagel et al. (1978). While the presence of the SNR in DEM L243 did not affect the resulting abundances from the spatially integrated observation, the derived abundances for DEM L301 are on the low end of the distribution, hinting at spurious effects caused by the shock activity in that superbubble.
We computed the $`R23`$ O abundance parameters (Pagel et al. 1979) for the spatially integrated data, and compared these with model tracks constructed with Mappings. The models assume the Costar C2 (Schaerer & de Koter 1997) stellar atmosphere corresponding to an O6 – O7 spectral type, and an inner nebular radius of 0.4 $`R_\mathrm{S}`$. As has historically been the case, the models do not agree well with the locus of observations in the literature, although our LMC data do agree well, coincidentally, with both.
Similarly, we examined the $`S23`$ abundance parameter for S (e.g., Christensen et al. 1997; Díaz & Pérez-Montero 2000). Our models reveal that, contrary to previous suggestions, $`S23`$ is more sensitive to the ionization parameter than is $`R23`$. S IV is produced by the same radiation that ionizes O III, and is a significant ion of S in many H ii regions, but it is not sampled by $`S23`$. Its omission therefore causes $`S23`$ to be much more sensitive to $`U`$ than $`R23`$. The spatially resolved observations confirm this by showing, in agreement with model predictions, lower values of $`S23`$ in the central nebular regions where S IV is important. As shown in Paper I, this spatial variation is not predicted or observed in $`R23`$.
Our models also suggest that the maximum in $`S23`$ occurs at only $`0.2`$ dex higher in $`Z`$ than in $`R23`$. Nevertheless, this appears to significantly alleviate the effect of the double-valued structure of log(S/H) vs $`S23`$ when inferring abundances, as shown by Díaz & Pérez-Montero (2000). It is highly encouraging that the data, both from our sample and from the literature, are in excellent agreement with the models, in contrast to the behavior of $`R23`$. We offer a theoretical calibration of $`S23`$ (equation 15) which appears to be fully compatible with the data in the literature thus far. However, we caution that the locus of the available data may well be deceptive in suggesting that a power-law relation can be used at $`Z0.5\mathrm{Z}_{}`$.
To overcome the limitations of $`S23`$ in $`U`$-sensitivity and spatial variation, we introduce a similar S abundance parameter, $`S234`$. This is the same as $`S23`$ with the added emission of \[S IV\]$`\lambda `$10.5$`\mu `$. Although this mid-IR line is not readily observable with most conventional ground-based spectrographs, it is straightforward to estimate its intensity from the simple correspondence between \[S IV\]/\[S III\] and \[O III\]/\[O II\] (equation 14). Our models show that $`S234`$ is less dependent on $`U`$ than is even $`R23`$. $`S234`$ for our objects and for the larger sample of Dennefeld & Stasińska (1983) are in excellent agreement with the models. Likewise, the spatial variations for both models and observations are dramatically reduced for $`S234`$ in contrast to $`S23`$. We provide a theoretical calibration for log(S/H) vs $`S234`$ at $`Z0.5\mathrm{Z}_{}`$ (equation 16).
Finally, we reiterate some caveats for the use of $`R23`$, $`S23`$, and $`S234`$. We find that the presence of shock excitation increases the value of these parameters; for our objects, the effect is about 0.1 dex in magnitude. Secondly, significant variations in the S/O ratio dictate caution in inferring O abundances using $`S234`$ and $`S23`$ (Figure 11). It is also important to bear in mind the double-valued structure for all three of these parameters. Lastly, we emphasize the deviation between the data and models above $`0.5\mathrm{Z}_{}`$, and we therefore consider the calibrations presented thus far for $`S23`$ and $`S234`$ to be reliable only for $`Z0.5\mathrm{Z}_{}`$. Further empirical investigation is needed to understand the behavior of these parameters at higher metallicity. Bearing in mind these caveats, the excellent correspondence between the modeled $`S234`$, $`S23`$, and the available data, together with the more highly monotonic behavior of these parameters, promises greater effectiveness as metallicity indicators than $`R23`$. With improving access to the \[S IV\]$`\lambda `$10.5$`\mu `$ line, it should be possible to confirm the behavior of $`S234`$ directly.
It is a pleasure to acknowledge discussions with Mike Dopita, Annette Ferguson, Don Garnett, Dick Shaw, Evan Skillman, Elena Terlevich, and Bob Williams. We are also grateful to Mike Dopita for access to the Mappings II photoionization code and to Angelez Díaz for access to her work in advance of publication. Finally, we are pleased to acknowledge the referee, Bernard Pagel. |
warning/0003/hep-th0003129.html | ar5iv | text | # Testing the AdS/CFT correspondence beyond large N
## 1 Introduction
We begin by briefly reviewing some relevant basic aspects of the AdS/CFT correspondence , see in particular .
Consider type IIB string theory with a number $`N`$ of D3 branes. There are open strings ending on these D3 branes and closed strings in the bulk. The effective low energy action consists of IIB supergravity describing the bulk closed strings, and $`𝒩=4`$ supersymmetric $`\mathrm{U}(N)`$ gauge theory (SYM) describing the open strings ending on the D3 branes. The gauge group of the latter actually is $`\mathrm{SU}(N)`$ since the $`\mathrm{U}(1)`$ decouples. Considering low energies at fixed $`\alpha ^{}`$ is equivalent to fixed energy and taking the $`\alpha ^{}0`$ limit. In this limit the gravitational coupling $`\kappa g_s\alpha ^2`$ vanishes and the interactions between the branes and the bulk can be neglected, as well as all higher derivative terms in the brane action. Only free bulk supergravity and pure $`𝒩=4`$ $`\mathrm{SU}(N)`$ SYM in $`d=4`$ dimensions remain. The latter is a conformal field theory (CFT).
There is a different way to describe the same physics. D3 branes may be viewed as certain solutions of the supergravity field equations, namely
$`\mathrm{d}s^2={\displaystyle \frac{1}{\sqrt{f}}}\left(\mathrm{d}t^2+\mathrm{d}x_1^2+\mathrm{d}x_2^2+\mathrm{d}x_3^2\right)`$
$`+\sqrt{f}\left(\mathrm{d}r^2+r^2\mathrm{d}\mathrm{\Omega }_5^2\right),`$
$`f=1+{\displaystyle \frac{R^4}{r^4}},R^4=4\pi g_s\alpha ^2N.`$ (1)
Here $`t,x_1,x_2,x_3`$ are the longitudinal coordinates (on the D3) while $`r`$ and $`\mathrm{\Omega }_5`$ describe the transverse space. $`N`$ is the number of (coincident) D3 branes. There is also a five-form field strength $`F`$ which is proportional to $`N`$. Again, one wants to study the low energy excitations in this description and compare with the first one. Due to the non-trivial function $`f(r)`$ in the metric there is a red-shift factor between energies measured at $`r`$ and energies measured at $`r=\mathrm{}`$:
$$E_{\mathrm{}}=f^{1/4}E_r=\left(1+\frac{R^4}{r^4}\right)^{1/4}E_r.$$
(2)
One sees that if $`r0`$, $`E_{\mathrm{}}0`$ for any finite $`E_r`$. So there are two types of low energy excitations: low energy at finite $`r`$ (bulk) or finite energy near the horizon ($`r=0`$), and the two types decouple yielding (free) bulk supergravity and near horizon supergravity. Concentrate on the near horizon limit: as $`r0`$ one has $`f\frac{R^4}{r^4}`$. Since also $`\alpha ^{}0`$, it is convenient to introduce the finite quantity $`u=r/\alpha ^{}`$ so that the near horizon metric becomes
$`{\displaystyle \frac{1}{\alpha ^{}}}\mathrm{d}s^2=\lambda ^{1/2}u^2\left(\mathrm{d}t^2+\mathrm{d}x_1^2+\mathrm{d}x_2^2+\mathrm{d}x_3^2\right)`$
$`+\lambda ^{1/2}{\displaystyle \frac{\mathrm{d}u^2}{u^2}}+\lambda ^{1/2}\mathrm{d}\mathrm{\Omega }_5^2,`$ (3)
where we introduced the finite quantity
$$\lambda =4\pi g_sN=\frac{R^4}{\alpha ^2}.$$
(4)
The metric (1) is the metric of $`AdS_5\times S^5`$.
Comparing both descriptions of the same physics one then is led to identify the conformally invariant, $`𝒩=4`$ $`\mathrm{SU}(N)`$ SYM theory in $`d=4`$ with the supergravity or small $`\alpha ^{}`$ limit of IIB string theory on $`AdS_5\times S^5`$. We will be more precise shortly.
How do the different parameters on the SYM side compare to those of the supergravity/string theory? In the former we have the coupling $`g_{\mathrm{YM}}`$ and the integer $`N`$ determining the gauge group $`\mathrm{SU}(N)`$. On the supergravity/string side we have $`\alpha ^{}`$ and $`R`$ (with only the dimensionless ratio $`R^2/\alpha ^{}`$ being a relevant parameter) and the coupling $`g_s`$. A first relation is already obtained in eq. (1), namely $`R^4=4\pi g_s\alpha ^2N`$ or equivalently eq. (4). A second relation is obtained from the D3 brane action from which one reads the YM coupling in terms of the string coupling. Thus
$$\frac{1}{g_{\mathrm{YM}}^2}=\frac{1}{4\pi g_s}\mathrm{and}N=\frac{1}{4\pi g_s}\frac{R^4}{\alpha ^2}$$
(5)
expresses the SYM parameters $`g_{\mathrm{YM}}`$ and $`N`$ in terms of the string/supergravity parameters $`g_s`$ and $`R^2/\alpha ^{}`$ and vice versa. In large $`N`$ gauge theories the relevant loop-counting parameter is the ’t Hooft coupling $`g_{\mathrm{YM}}^2N`$ rather than $`g_{\mathrm{YM}}^2`$. Combining both eqs (5) yields $`g_{\mathrm{YM}}^2N=\frac{R^4}{\alpha ^2}`$ which by eq (4) is just the quantity called $`\lambda `$. It is useful to rewrite the relations between the parameters of the two descriptions as
$$\lambda =\frac{R^4}{\alpha ^2},\frac{N}{\lambda }=\frac{1}{4\pi g_s}$$
(6)
with $`\lambda `$ now meaning the ’t Hooft coupling.
Let us first comment on the first relation: perturbative SYM theory is a good description if the ’t Hooft coupling $`\lambda `$ is small. Supergravity, rather than string theory, should be a good description if the radius of curvature of $`AdS_5`$ and $`S^5`$ is large, meaning $`R^2\alpha ^{}`$ or $`\lambda `$ large. The two regimes are opposite as is often the case with dualities. This of course avoids the obvious contradiction that both descriptions look so different.
At fixed ’t Hooft coupling $`\lambda `$, the second relation (6) tells us that $`\frac{1}{N^2}`$ corresponds to the string loop-counting parameter $`g_s^2`$, so that the large $`N`$ limit of SYM corresponds to classical string theory (or classical supergravity if also $`\lambda 1`$), and $`\frac{1}{N^2}`$ corrections should correspond to one-loop effects in string theory.
One now has various possible conjectures: 1.) The weakest one is: SYM is dual to $`AdS`$ supergravity only for $`\lambda \mathrm{}`$, but the full string theory is different. This version would not be very useful. 2.) The SYM theory is dual to string theory on $`AdS`$ for finite $`\lambda `$ but only as $`N\mathrm{}`$ or equivalently $`g_s0`$. This includes $`\alpha ^{}`$ corrections beyond the supergravity approximation, but no string loops. 3.) This is the strongest version, generally referred to as the Maldacena conjecture: the SYM theory is dual to string theory for all $`\lambda `$ and all $`N`$, i.e. all $`R^4/\alpha ^2`$ and $`g_s`$.
While there is now reasonable evidence for version 2.) of the conjecture (see e.g. ) the strong version 3.) is hard to test since standard string or supergravity loop computations on an $`AdS_5\times S^5`$ background are difficult, to say the least, if not unfeasible, with the present state of the art.
Many successful tests are group theoretic in nature. Some are not restricted to tree-level or even perturbation theory, but on the other hand they do not really provide any real test at one-loop. Examples are global symmetries (disregarding possible anomalies for the moment): $`AdS_5`$ space-time has an $`\mathrm{SO}(4,2)`$ symmetry which also is the conformal group of the $`𝒩=4`$ SYM theory. The “internal” symmetry is $`\mathrm{SO}(6)\mathrm{SU}(4)`$: this is the isometry of the $`S^5`$ as well as the R-symmetry of the SYM theory. The latter actually is anomalous which will be important for us. Both theories have the same amount of supersymmetry, the full supergroup being $`\mathrm{SU}(2,2|4)`$ $`\mathrm{SO}(4,2)\times \mathrm{SU}(4)_R`$. Also the duality symmetry $`\mathrm{SL}(2,𝐙)`$ is the same as it acts on
$$\tau =\frac{4\pi i}{g_{\mathrm{YM}}^2}+\frac{\theta }{2\pi }=\frac{i}{g_s}+\frac{\chi }{2\pi }.$$
(7)
A promising arena for performing tests beyond the large $`N`$ limit or string tree-level is to look at certain anomalies. On the SYM side anomaly coefficients are easily established one-loop effects in $`\lambda `$ that are protected against higher order corrections. Typically such an anomaly coefficient will depend on the number of fermion fields in the SYM theory, i.e. on $`N`$. The goal then is to reproduce the exact $`N`$-dependence from the string theory including subleading terms $`\frac{1}{N^2}`$ coming from string loops. If we want to have any chance to be able to do this calculation, the relevant quantity to compute in string theory should be of a topological nature, like a Chern-Simons term, so that the actual metric on $`AdS_5`$ is irrelevant.
## 2 The chiral $`\mathrm{SU}(4)_R`$ anomaly in the $`𝒩=4`$ SYM
The anomaly we will consider is the chiral $`\mathrm{SU}(4)_R`$ anomaly. As already mentioned, $`\mathrm{SU}(4)_R`$ is a (classical) global symmetry of the $`𝒩=4`$ SYM theory. Due to the presence of chiral fermions transforming in complex conjugate representations of $`\mathrm{SU}(4)_R`$ this symmetry is spoiled at one loop and there is an anomaly: the one-loop effective action in the presence of external $`\mathrm{SU}(4)_R`$ gauge fields is no longer invariant under $`\mathrm{SU}(4)_R`$ and the non-invariance is proportional to the number of fermions. Since they are also in the adjoint representation of the “true” gauge group $`\mathrm{SU}(N)`$ there are $`N^21`$ of them, and the anomaly is proportional to $`N^21`$. As we recall below, the leading term $`N^2`$ is accounted for by tree-level supergravity . It is the $`1`$ correction which should originate from a string/supergravity loop effect, and it indeed does as we showed in and explain in the remainder of this paper.
Before explaining the string/supergravity loop correction let us review how the leading $`N^2`$ term is obtained in the string/supergravity description. Here $`\mathrm{SU}(4)\mathrm{SO}(6)`$ acts as an isometry on the $`S^5`$. As a consequence, the $`AdS_5`$ supergravity is actually a gauged supergravity and there is an $`\mathrm{SU}(4)_R`$ gauge group with gauge fields $`\stackrel{~}{A}_\mu ^a(x,z)`$, $`\mu =0,1,\mathrm{}4`$ and $`a=1,\mathrm{}15=\mathrm{dim}\mathrm{SU}(4)`$. This gauge group is of course not to be confused with the $`\mathrm{SU}(N)`$ of the conformal SYM theory. Note also that in the latter, $`\mathrm{SU}(4)_R`$ is a global symmetry, hence there are $`\mathrm{SU}(4)_R`$ currents $`J_\mu ^a(x)`$, $`\mu =0,1,2,3`$, but no associated gauge fields. We can nevertheless couple these currents to external gauge fields $`A_\mu ^a(x)`$, $`\mu =0,1,2,3`$ which act as sources for these currents. Then by a standard argument, the non-invariance of the one-loop effective action $`\mathrm{\Gamma }[A_\mu ]`$ under gauge transformations of these external gauge fields is equivalent to the covariant non-conservation of the currents: let $`\delta _v`$ be such a gauge transformation with parameter $`v`$, then
$`\delta _v\mathrm{\Gamma }[A_\mu ]={\displaystyle \delta _vA_\mu ^a\frac{\delta \mathrm{\Gamma }}{\delta A_\mu ^a}}={\displaystyle \delta _vA_\mu ^aJ^{a,\mu }}`$
$`={\displaystyle (D_\mu v)^aJ^{a,\mu }}={\displaystyle v^a(D_\mu J^\mu )^a}`$ (8)
with
$$(D_\mu J^\mu )^a(N^21)d^{abc}ϵ^{\mu \nu \rho \sigma }_\mu A_\nu ^b_\rho A_\sigma ^c+\mathrm{}$$
(9)
the precise numerical coefficient being given below.
There is a standard prescription in the AdS/CFT correspondence how to compute correlation functions: we will give this prescription for the case of present interest. For any ($`\mathrm{SU}(N)`$ gauge-invariant) operator $`𝒪(x)`$ like the currents $`J_\mu ^a(x)`$ of the SYM theory, introduce a source $`\varphi _0(x)`$ like the $`A_\mu ^a(x)`$. Then the generating functional for correlators of $`J_\mu ^a`$ is just
$$\mathrm{e}^{\mathrm{\Gamma }[A]}\mathrm{e}^{{\scriptscriptstyle \mathrm{d}^4xA_\mu ^a(x)J^{a,\mu }(x)}}_{\mathrm{SYM}}.$$
(10)
In $`AdS_5`$ string theory there is a field $`\varphi (x,z)`$ such that at the boundary $`z=0`$ of $`AdS_5`$ (note that $`z=1/u`$) which is just the four-dimensional space of the SYM theory one has $`\varphi (x,z=0)=\varphi _0(x)`$. In our case this is just $`\stackrel{~}{A}_\mu ^a(x,z=0)=A_\mu ^a(x)`$ (for $`\mu =0,1,2,3`$ only) where the $`\stackrel{~}{A}_\mu ^a`$ are the gauge fields of the gauged supergravity. The prescription then is
$$\mathrm{e}^{\mathrm{\Gamma }[A]}=Z_{\mathrm{string}}|_{\stackrel{~}{A}_\mu ^a(x,z=0)=A_\mu ^a(x)},$$
(11)
meaning that the string partition function should be evaluated subject to the boundary condition $`\stackrel{~}{A}_\mu ^a(x,z=0)=A_\mu ^a(x)`$ for $`\mu =0,1,2,3`$. Writing
$$Z_{\mathrm{string}}=\mathrm{e}^{S_{\mathrm{string}}^{\mathrm{cl}}S_{\mathrm{string}}^{1\mathrm{loop}}\mathrm{}}\mathrm{e}^{S_{\mathrm{string}}^{\mathrm{eff}}}$$
(12)
eq (11) together with eq (2) implies that, if the AdS/CFT correspondence is correct, we should have
$`\delta _vS_{\mathrm{string}}^{\mathrm{eff}}|_{\stackrel{~}{A}_\mu ^a(x,z=0)=A_\mu ^a(x)}=\delta _v\mathrm{\Gamma }[A]`$
$`={\displaystyle v^a(D_\mu J^\mu )^a}`$ (13)
which is non-vanishing according to (9). Thus the $`\mathrm{SU}(4)_R`$ gauge variation of $`S_{\mathrm{string}}^{\mathrm{eff}}`$ should directly reproduce the SYM chiral $`\mathrm{SU}(4)_R`$ anomaly. Actually for the purpose of reproducing the leading $`N^2`$ part of the anomaly it is enough to consider the classical supergravity action .
Let us now determine the exact anomaly coefficient of the $`𝒩=4`$ SYM theory in 4 dimensions. This theory has four complex Weyl fermions $`\lambda `$ in the fundamental representation of $`\mathrm{SU}(4)_R`$ with the chirality part (0,1/2) in $`\mathrm{𝟒}`$ and (1/2,0) in $`\mathrm{𝟒}^{}`$ (see for example . Our conventions here are equivalent to those of .) Moreover, all fields are also in the adjoint representation of the “true” gauge group $`\mathrm{SU}(N)`$ which acts as a “flavour” group with respect to the $`\mathrm{SU}(4)_R`$. Thus there are actually $`N^21`$ complex Weyl fermions $`\lambda `$ in the $`\mathrm{𝟒}`$, resp. $`\mathrm{𝟒}^{}`$. The correctly normalised R-symmetry anomaly is given by
$$\delta _v\mathrm{\Gamma }[A]=(N^21)_{S^4}\omega _4^1(v,A).$$
(14)
The differential forms
$`\omega _4^1(v,A)={\displaystyle \frac{1}{24\pi ^2}}\mathrm{Tr}[vd(AdA+{\displaystyle \frac{1}{2}}A^3)],`$
$`\omega _5(A)={\displaystyle \frac{1}{24\pi ^2}}\mathrm{Tr}[A(dA)^2+{\displaystyle \frac{3}{2}}A^3dA+{\displaystyle \frac{3}{5}}A^5]`$
satisfy the descent equations $`d\omega _5`$ $`=`$ $`\frac{1}{24\pi ^2}\mathrm{Tr}F^3`$, and $`\delta _v\omega _5=d\omega _4^1`$ with $`F=dA+A^2`$, $`A=A^aT^a`$ and $`v=v^aT^a`$ as usual, the $`T^a`$ being the generators of $`\mathrm{SU}(4)`$ in the fundamental $`\mathrm{𝟒}`$ representation. For later use we note that for $`T^a`$ in a general representation $`𝐑`$ of $`\mathrm{SU}(4)`$, the corresponding quantities with the trace taken in $`𝐑`$ are
$$\omega _{2n}^{1}{}_{}{}^{𝐑}=A(𝐑)\omega _{2n}^1,\omega _{2n+1}^𝐑=A(𝐑)\omega _{2n+1},$$
(16)
where $`A(𝐑)`$ is the anomaly coefficient defined by the ratio of the $`d`$-symbols taken in the representation $`𝐑`$ and in the fundamental representation. In general $`2n`$ or $`2n+1`$ dimensions, since the $`d`$-symbol is given by a symmetrized trace of $`n+1`$ Lie algebra generators, it is easy to show that the complex conjugate representation $`𝐑^{}`$ has an anomaly coefficient
$$A(𝐑^{})=(1)^{n+1}A(𝐑).$$
(17)
Due to the connection of anomalies and Chern-Simons actions in one higher dimension, it is natural to expect that the four-dimensional field theory anomaly is dual to a Chern-Simons action in the gauged $`AdS_5`$ supergravity. This is indeed the case as was first pointed out in . The tree level supergravity action on $`AdS_5`$ contains the following terms
$$S_{\mathrm{cl}}[A]=\frac{1}{4g_{SG}^2}d^5x\sqrt{g}F_{\mu \nu }^aF^{\mu \nu a}+k_{AdS_5}\omega _5.$$
(18)
Note that here $`F`$ is the field strength associated with the five-dimensional gauge field $`\stackrel{~}{A}_\mu `$. The exact values of the coefficients $`\frac{1}{4g_{SG}^2}`$ and $`k`$ will be important for us. Their ratio is fixed by supersymmetry . They may be obtained by dimensional reduction of the ten-dimensional IIB supergravity on $`S^5`$ using the fact that the radius of $`S^5`$ is given by eq. (4) as $`R^4/\alpha ^2=4\pi g_sN`$. Then it is easy to determine the normalization of the gauge kinetic energy term and one finds
$$g_{SG}^2=\frac{16\pi ^2}{N^2},k=N^2.$$
(19)
Note that the action (18) with the values (19) has been used to compute the 2-point and 3-point correlators of the currents $`J_\mu ^a`$ in the SYM theory . To leading order in $`N`$ this gives the correct result.
In usual considerations of supergravity on $`AdS`$, one considers gauge configurations $`\stackrel{~}{A}_\mu `$ that vanish at the boundary and so the Chern-Simons term is gauge invariant since $`\delta \omega _5=d\omega _4^1`$ and the integral vanishes. For the considerations of the AdS/CFT correspondence however, we precisely want nonvanishing boundary values for $`\stackrel{~}{A}_\mu `$ as explained above, cf eq (11). Then under a gauge variation $`\delta _v\stackrel{~}{A}`$, the variation of the Chern-Simons term is a boundary term
$$\delta _vS_{cl}=k_{S^4}\omega _4^1(v,A).$$
(20)
(We take the SYM theory to be defined on compactified Euclidean space, i.e. on $`S^4`$.) Now by eq (2), approximating $`S_{\mathrm{string}}^{\mathrm{eff}}S_{\mathrm{cl}}`$ and using (20) one can read off the $`\mathrm{SU}(4)_R`$ anomaly obtained from the supergravity action (18). It is
$$\delta _v\mathrm{\Gamma }[A]=\delta _vS_{\mathrm{cl}}=N^2_{S^4}\omega _4^1(v,A),$$
(21)
which agrees with the gauge theory computation (14) to leading order in $`N`$.
We thus see that the IIB supergravity action contains a Chern-Simons term at tree level which can account for the chiral anomaly of the gauge theory to leading order in $`N`$. But there is also a mismatch of “-1” which is of order $`1/N^2`$ relative to the leading term. As discussed above, this should correspond to a 1-loop effect in IIB string theory. Thus we are lead to examine the string one-loop effective action.
## 3 One-loop induced Chern-Simons action
Loop effects in $`AdS_5`$ supergravity are technically very diffciult to compute due to the complicated propagators in $`AdS`$ geometry. Here however, this is possible due to the topological character of the Chern-Simons action.
Fermionic contributions
Consider a Dirac fermion $`\psi `$ in odd dimensions (flat) minimally coupled to vector bosons $`A_\mu `$ of a group $`G`$. At the quantum level, a regularization needs to be introduced to make sense of the theory and one cannot preserve both the gauge symmetry (small and large) and the parity at the same time . If one chooses to preserve the gauge symmetry by doing a Pauli-Villars regularization, then there will be an induced Chern-Simons term generated at one loop. The result is independent of the fermion mass. In our notation, the induced Chern-Simons term is
$$\mathrm{\Delta }\mathrm{\Gamma }=\pm \frac{1}{2}\omega _{2n+1}^𝐑=\pm \frac{1}{2}A(𝐑)\omega _{2n+1},$$
(22)
where $`𝐑`$ is the representation of the Dirac fermion. The $`\pm `$ sign depends on the regularization and can often be fixed within a specific context.
This result was originally obtained for fermions coupled to gauge fields in a flat spacetime and has been extended to full generality for arbitrary curved backgrounds and any odd dimension $`d=2n+1`$ . The induced parity violating terms are given (up to a normalization factor) by the secondary characteristic class $`Q(A,\omega )`$ satisfying
$$dQ(A,\omega )=\widehat{A}(R)ch(F)|_{2n+2},$$
(23)
where $`\omega `$ is the gravitational connection. Since $`\widehat{A}(R)=1+𝒪(R^2)`$ and $`\mathrm{Tr}F=0`$ for SU-groups, it is clear that for $`n=2`$ ($`d=5`$) there are no mixed gauge/gravitational terms. Also, there can be no purely gravitational term since it would correspond to a gravitational anomaly in four dimensions which is not possible. Hence for the present case of $`\mathrm{SU}(4)`$ with $`n=2`$, (23) simply reduces to $`dQ=ch(F)|_6`$ giving rise to the Chern-Simons action upon descent, which does not depend on the geometry of $`AdS_5`$ at all! Hence the result of (22) for a Dirac fermion in flat space(-time) remains valid on $`AdS_5`$.
Now we need the particle spectrum of the type IIB string theory on $`AdS_5\times S^5`$ . The only explicitly known states are the KK states coming from the compactification of the 10 dimensional IIB supergravity multiplet . So we will examine them first. We will argue in the discussion section that the other string states are not likely to modify the result.
Particles in $`AdS_5`$ are classified by unitary irreducible representations of $`\mathrm{SO}(2,4)`$. $`\mathrm{SO}(2,4)`$ has the maximal compact subgroup $`\mathrm{SO}(2)\times \mathrm{SU}(2)`$ $`\times \mathrm{SU}(2)`$ and so its irreducible representations are labelled by the quantum numbers $`(E_0,J_1,J_2)`$. The complete KK spectrum of the IIB supergravity on $`AdS_5\times S^5`$ was obtained in together with information on the representation content under $`\mathrm{SU}(4)_R`$. We reproduce these results in the table below. Actually, all fermions are symplectic Majorana, giving half the anomaly of a Dirac fermion. But there also is a mirror table with conjugate $`\mathrm{SU}(4)_R`$ representations and $`\mathrm{SU}(2)\times \mathrm{SU}(2)`$ quantum numbers exchanged (opposite chiralities). So these “mirror” fermions give the same anomaly as those in the table and the net effect is that we may restrict ourselves to the fermions of the table treating them as if they were Dirac fermions.
$$\begin{array}{ccccc}& \mathrm{SU}(2)\times \mathrm{SU}(2)& & \mathrm{SU}(4)_R\hfill & \\ \psi _\mu \hfill & (1,1/2)& & \mathrm{𝟒},\mathrm{𝟐𝟎},\mathrm{}\hfill & \hfill \\ & (1,1/2)& & \mathrm{𝟒}^{},\mathrm{𝟐𝟎}^{},\mathrm{}\hfill & \\ \lambda \hfill & (1/2,0)& & \mathbf{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.17em}20}^{},\mathrm{}\hfill & \hfill \\ & (1/2,0)& & \mathrm{𝟒},\mathrm{𝟐𝟎},\mathrm{}\hfill & \\ \lambda ^{}\hfill & (1/2,0)& & \mathrm{𝟒}^{},\mathrm{𝟐𝟎}^{},\mathrm{}\hfill & \hfill \\ & (1/2,0)& & \mathrm{𝟒},\mathrm{𝟐𝟎},\mathrm{}\hfill & \\ \lambda ^{\prime \prime }\hfill & (1/2,0)& & \mathrm{𝟑𝟔},\mathrm{𝟏𝟒𝟎},\mathrm{}\hfill & \\ & (1/2,0)& & \mathrm{𝟑𝟔}^{},\mathrm{𝟏𝟒𝟎}^{},\mathrm{}\hfill & \end{array}$$
(24)
Notice that the fermion towers always come in pairs with conjugate representation content, except for a missing $`\mathrm{𝟒}^{}`$ state in the first tower of $`\lambda `$. As a result , all contributions cancel two by two except for the contribution from the unpaired $`\mathrm{𝟒}`$ of the $`\lambda `$ tower. The net resulting induced Chern Simons action is
$$\mathrm{\Delta }\mathrm{\Gamma }=\frac{1}{2}_{AdS_5}\omega _5.$$
(25)
While this is almost what we want, it is only half of the desired result. However this is not the whole story.
Doubleton multiplet
There are similar “missing states” in the bosonic towers. Together they are identified with the doubleton multiplet of $`\mathrm{SU}(2,2|4)`$ which consists of a gauge potential, six scalars and four complex spinors. These are nonpropagating modes in the bulk of $`AdS_5`$ and can be gauged away completely , which is the reason why they don’t appear in the physical spectrum. These modes are exactly dual to the $`\mathrm{U}(1)`$ factor of the $`\mathrm{U}(N)`$ SYM living on the boundary . We will now show that the other half of the induced Chern-Simons action is due to the corresponding Faddeev-Popov ghosts.
Let us recall that the doubleton multiplet is absent because it has been gauged away by imposing the gravitino gauge fixing condition (see also for the gauging in the case of $`AdS_7\times S^4`$ case). The basic idea is that upon compactifying on $`S^5`$, the original supersymmetyries in 10 dimensions decompose into an infinite tower of (unwanted) supersymmetries according to the Fourier expansion. This can be fixed however by imposing a certain condition on the variation
$$\delta \psi _\alpha =D_\alpha ϵ+\frac{i}{2R}\gamma _\alpha ϵ$$
(26)
of the gravitino. Denote the local coordinates of $`AdS\times S^5`$ by $`(x^\mu ,y^\alpha )`$. A general spinor $`ϵ`$ has the decomposition
$$ϵ=ϵ^{I,\pm }(x)\mathrm{\Xi }^{I,\pm }(y)$$
(27)
where $`\mathrm{\Xi }^{I,\pm }(y)`$ are the spinor spherical harmonics on $`S^5`$ and satisfy ($`D/_y=\gamma ^\alpha D_\alpha ,\alpha =5,\mathrm{}9`$)
$$D/_y\mathrm{\Xi }^{I,\pm }=i(k+\frac{5}{2})\frac{1}{R}\mathrm{\Xi }^{I,\pm }$$
(28)
where $`k=I0`$ and $`\mathrm{\Xi }^{I,\pm }`$ can be written in terms of the killing spinors $`\eta ^{I,\pm }`$ on $`S^5`$. Subsituting (27) into (26) and using (28) we get
$$\delta (\gamma ^\alpha \psi _\alpha )=\frac{i}{R}\left[(k+\frac{5}{2})+\frac{5}{2}\right]ϵ^{I,\pm }(x)\mathrm{\Xi }^{I,\pm }(y).$$
(29)
So one finds that only the component corresponding to $`\mathrm{\Xi }^{0,+}`$ is gauge invariant while all other components of $`\gamma ^a\psi _\alpha `$ can be gauged away. Thus we arrive at the gravitino gauge fixing condition
$$\gamma ^\alpha \psi _\alpha (x,y)\chi ^{I_0}(x)\eta ^{I_0,+}(y)$$
(30)
where $`\chi ^{I_0}(x)`$ are some arbitrary spacetime spinor fields. We refer the reader to for more details. Therefore we see that (30) is the closest one can get to the gauge condition $`\gamma ^\alpha \psi _\alpha =0`$. One can also rewrite this condition as
$$\psi _\alpha =\psi _{(\alpha )}+\chi ^{I_0}(x)\gamma ^\alpha \eta ^{I_0,+}(y)$$
(31)
where the part $`\psi _{(\alpha )}`$ satisfies $`\gamma ^\alpha \psi _{(\alpha )}=0`$. The other Killing spinors $`\eta ^{}`$ have been gauged away. The coefficient of $`\eta ^{}`$ would be a field in the $`4^{}`$ of $`\mathrm{SU}(4)`$ and is precisely the doubleton spinors we are after. Now the constraint can be taken care of in the functional approach by introducing in the path integral the factor
$`{\displaystyle 𝑑b𝑑\overline{b}\frac{1}{\mathrm{det}M}e^{{\scriptscriptstyle d^5x\overline{b}Mb}}}`$ (32)
$`\delta (\gamma ^\alpha \psi _\alpha {\displaystyle \chi ^I\eta ^{I,+}}b^I(x)\eta ^{I,}(y))\delta (\text{h.c.})`$
where $`b(x)`$ is a complex fermionic field in the $`4^{}`$ of $`\mathrm{SU}(4)`$ and $`M=D/_x`$. Integrating over $`b,\overline{b}`$ results in the gauged fixed lagrangian. The factor $`(\mathrm{det}M)^1`$ can be handled by intoducing ghost fields $`c,\overline{c}`$, which are bosonic spinor fields on $`AdS_5`$ and are in the $`4^{}`$ of $`\mathrm{SU}(4)`$. Thus
$$\frac{1}{\mathrm{det}M}=𝑑c𝑑\overline{c}e^{{\scriptscriptstyle d^5x\overline{c}Mc}}.$$
(33)
and so give rises to another -1/2 contribution to the induced Chern-Simons action. So altogether we get a total induced Chern-Simons term of -1,
$$\mathrm{\Delta }\mathrm{\Gamma }=_{AdS_5}\omega _5,$$
(34)
which is exactly the desired result. Notice that the induced Chern-Simons action (coming with a constant integer coefficient) is independent of the radius $`R`$ and this is consistent with the AdS/CFT proposal since the anomaly and its corrections are independent of $`\lambda `$.
Bosonic contributions
There is another interesting effect related to the Chern-Simons action. It is known that in three dimensions, the gluons at one loop can modify the coefficient of the Chern-Simons action by an integer shift. It has been argued that there is no such shift for the present case. Therefore only spinor loops contribute to the induced Chern-Simons action and we find that at finite $`N`$, the coefficient $`k`$ is shifted by
$$kk1\text{or}N^2N^21$$
(35)
due to the quantum effects of the full set of Kaluza-Klein states.
A few comments about the absence of a shift due to gluon loops are in order. The bosonic shift in pure Chern-Simons theory was first computed in using a saddle point approximation. Later calculations trying to reproduce this results from the perturbative point of view revealed that the precise shift depends on the choice of regularization scheme <sup>1</sup><sup>1</sup>1 We thank R. Stora for a useful discussion about the issues of regularization. . In the present case of 5-dimensions, one may try to employ a regularization scheme and do a 1-loop perturbative calculation to determine the possible shift. However, like in the 3-dimensional case, it can be expected that the result will depend on the choice of regularization and a better way to determine the shift is called for. One possibility might be to do a string theory calculation by embedding the Chern-Simons action in a string setting and to determine the quantum loop effects from the string loop effects. Since string theory is free from divergences, no regularization related ambuigities should occur and a definite answer can be expected.
## 4 Discussion
We have reproduced the correct shift $`N^2`$ $`N^21`$ of the anomaly coefficient as a one-loop effect in IIB supergravity/string theory on $`AdS_5\times S^5`$. This shift is entirely due to the towers of Kaluza-Klein states. No massive string states need to be invoked. It is indeed likely that the latter play no role at all since anomalies are usually due to massless fields only. Note however that we need the full towers of Kaluza-Klein states to get the correct result. A truncation to five-dimensional $`AdS`$ supergravity alone would not give the desired result. Also the $`AdS_5`$ supergravity Chern-Simons term originates from compactifying the full IIB supergravity. This is another indication that string states beyond the Kaluza-Klein towers are unlikely to modify our result.
We have been able to obtain a non-trivial one-loop result within a particularly favourable case. In general, one-loop calculations in $`AdS_5`$ are very difficult - already tree computations are quite non-trivial! Of course, the anomaly coefficient $`N^21`$ should be exact and there cannot be any further higher-loop corrections $`N^2\frac{1}{N^4}`$. Again, since the induced Chern-Simons term in 5 dimensions is closely related to anomalies in 4 dimensions, we expect some sort of non-renormalisation theorem to be at work, although it would be nice to have a proof of this statement.
Finally, we would like to make some comments on more or less related situations. There are other dualities like those involving $`AdS_7\times S^4`$ where one can expect similar Chern-Simons terms and doubleton multiplets. The issue of the trace-anomaly in $`AdS_5`$ should be closely related to the present study. Already the leading $`N`$ behaviour of this conformal anomaly is non-trivial to establish and to explicitly obtain the subleading terms might well turn out to be more difficult than for the chiral anomaly studied here. Effects that are of lower order than $`N^2`$ have also been considered in which essentially studies situations where the leading effect corresponds to open strings at tree level and hence comes with just one power of $`N`$. A somewhat related discussion is .
It will also be interesting to investigate these anomaly issues within the non-commutative version of the AdS/CFT correspondence to see the origin of the higher derivative Chern-Simons terms on the supergravity side.
###### Acknowledgments.
This work was partially supported by the Swiss National Science Foundation, by the European Union under TMR contract ERBFMRX-CT96-0045. |
warning/0003/astro-ph0003061.html | ar5iv | text | # 1 Introduction
## 1 Introduction
The last five years have seen great progress in the detection of brown dwarfs in the Local Neighborhood, young Galactic Clusters and star formation regions, starting with the near simultaneous discovery of the first clearly confirmed brown dwarfs, Teide 1 in the Pleiades (Rebolo, Zapatero-Osorio & Martin 1995) and Gl229b in the Local Neighborhood (Nakajima et al.1995). Star formation regions offer the advantage that substellar objects are 3 orders of magnitude more luminous at an age of a few Myr than at an age of a few Gyr. Early photometric and spectroscopic work (Comeron et al.1993,1996; Williams et al.1995) indicated that brown dwarfs are probably very common in star formation regions. However, confirmation of substellar status is problematic in star formation regions, owing to the ubiquity of Lithium in young objects and the complicating effects of extinction on both photometry and spectroscopy.
Recently, high quality spectroscopy (Luhman & Rieke 1998; Luhman et al. 1998, Wilking, Greene & Meyer 1999) and the publication of theoretical evolutionary models for young substellar objects (Burrows 1997;D’Antona & Mazzitelli 1998, hereafter B97 and DM98) have provided convincing evidence that photometric identification of young brown dwarf candidates is reliable. In photometric studies, masses of candidate objects are derived by comparison of the observables (luminosity and temperature) with the evolutionary tracks. The isochrones of B97 and DM98 are in fairly good agreement in regard to the mass-luminosity relation at an age of about 1 Myr but there is some disagreement about the mass-temperature relation (HR diagrams are compared by Luhman & Rieke 1998). Even if the theoretical effective temperatures were without flaw, there is considerable uncertainty in the derivation of temperatures from photometry or spectroscopy, at the level of $`\pm 200K`$ in M and L dwarfs. Hence, we use luminosity, which is more easily measured, to derive masses for our sources.
In this paper we report the results of a deep infrared photometric survey of the Trapezium Cluster in Orion. A large population of substellar objects is discovered, including the first free-floating objects of planetary mass. We note that the IAC group (Bejar et al.1999, not yet published) has simultaneously reported a similar discovery of planetary mass objects in the adjacent $`\sigma `$ Orionis cluster. The Trapezium has been intensively studied for many years and we have been able to draw upon a large body of publications to aid in our work. We selected the Trapezium for several reasons. (1) Its very high stellar density allowed photometry of several hundred sources in a fairly small survey. (2) False positive detections are essentially eliminated because the dense backdrop of OMC-1 obscures all background stars even at K band, as shown by Hillenbrand & Hartmann (1998) through optical-infrared comparison of the cluster stellar density profile. (3) The extinction within the cluster is relatively low ($`0<A_V<15`$ for most sources), permitting reasonably precise dereddening. (4) Star formation is essentially complete in the cluster and the age range is thought to be 0.3-2 Myr, so the age-luminosity degeneracy is not large.
## 2 Observations
Deep Imaging of the Trapezium cluster was carried out at United Kingdom Infrared Telescope (UKIRT) on 14-16 December and 22-23 December 1998, using UFTI, the UKIRT Fast Track Imager. The observations of 22-23 Dec were made by observatory staff to compensate for time lost to poor weather and equipment failures on 15-16 Dec. UFTI is a high resolution camera constructed by the authors at Oxford University with assistance from several other UK institutions (Roche & Lucas 1998). It has a 1024$`\times `$1024 HAWAII array sensitive between 0.78 and 2.5 microns. The image scale is 0.091 arcsec/pixel, yielding a field of view of 92.6 arcsec. Observations of 15 contiguous fields were made in the $`I,J`$ and $`H`$ filters, with 900s exposures in each filter. Twilight flatfields were taken to reduce the data. Seeing conditions were typically 0.6 arcsec FWHM in all 3 filters, with only slightly poorer image quality at $`I`$ band. The fine pixel scale led to a high degree of over-sampling, which was very useful for distinguishing stars from small knots of nebulosity and in permitting reliable photometry of low signal to noise detections.
### 2.1 Filter Selection
Taking advantage of the short wavelength sensitivity of the HAWAII array UFTI contains $`I_U`$ (0.786-0.929 $`\mu `$m) and $`Z_U`$ (0.85-1.05 $`\mu `$m) band filters. These filters sample the steeply rising part of brown dwarf spectra from the optical to the near infrared flux peak. The $`I_U`$ filter was selected because the $`Z_U`$ band contains a very strong \[SIII\] emission line which might have contaminated the photospheric flux in “proplyd” sources and because the (I-J) colour is a more reliable temperature indicator than than the smaller (Z-J) colour given that fluxes have to be dereddened. The $`J`$ and $`H`$ filters were chosen to measure extinction using the temperature insensitive (J-H) colour. The $`K`$ filter was not selected because a significant fraction of the flux comes from hot circumstellar dust at 2.2 $`\mu `$m, making it unreliable for determining extinction or luminosity and because it contains fairly strong low-excitation emission lines.
The UFTI $`I_U`$ band calibration (equ.1) relative to Cousins $`I`$ was made by observation of 8 faint red standards of approximately solar metallicity taken from Leggett et al.(1998). The calibration has been confirmed by observatory staff, including observations of blue standards, and can be fit remarkably well by a linear equation. The $`I_U`$ bandpass is somewhat redder than $`I_C`$, which compensates for the low quantum efficiency of the HAWAII array at these wavelengths ($`23\%`$) when observing cool stars. The $`J`$ and $`H`$ filters are of the new type commissioned by the Mauna Kea Consortium. The transformation to the CIT colour system (equ.2,3) was determined by combining the transformations of S.Leggett for CIT to UKIRT(<sub>IRCAM</sub>) with that for UKIRT(<sub>IRCAM</sub>) to UKIRT(<sub>UFTI</sub>). We use the UFTI $`J`$ and $`H`$ magnitudes in this paper but use Cousins $`I`$ for ease of comparison with other studies. This is possible because fortuitously the net effect of extinction on equ.1 is much less than the measurement errors.
$$I_C=I_U+0.273(I_UJ_U)$$
(1)
$`(JH)_U=1.03(JH)_{CIT}`$ (2)
$`J_U=K_{CIT}+1.132(J_{CIT}K_{CIT})`$ (3)
### 2.2 Data Reduction and Photometry
The data were reduced using IRAF and the Starlink package CCDPACK for image mosaicing. Photometry was carried out using the DAOPHOT package in IRAF. Photometry in the core of the Trapezium cluster is complicated by the pervasive nebulosity, which has structure on all the observed spatial scales. This often leads to inaccurate measurement of sky background when doing automated photometry, and causes the DAOFIND algorithm to misidentify many small-scale nebular flux variations as stars. To overcome these problems, we cross correlated the stars found in each filter to remove most of the nebulous sources and also rejected all very blue sources ($`(JH)<0.2`$), which inspection indicated were all spurious. Every source was then visually inspected and photometry was performed manually in order to select the best approporiate sky annulus in each case. The results of manual photometry were generally used in preference to the results of automated crowded-field photometry, with a few exceptions where the ALLSTAR routine was needed for photometry of close binaries. The final photometric precision is $`5\%`$ in the outlying regions of the survey where the nebulosity is faint, limited by temporal variations in the image profile. Precision in the bright nebulous core (in which the majority of sources lie) is between 5% and 20% (for the worst cases), depending on the degree of spatial variation of surface brightness and the source magnitude. Since a large fraction of young stars are variable, more precise photometry at one epoch would not have had much greater value.
## 3 Results
The main results of the survey are presented graphically in Figure 1(a-b). 515 unsaturated point sources were detected in both the $`J`$ and $`H`$ bands, of which 313 were also detected at I band. An additional 48 sources were detected at greater than 5-$`\sigma `$ in the $`H`$ band alone, and appear as upper limits in Figure 1(b). An approximate colour-magnitude sequence for zero extinction is indicated by the dotted line. The value of the nearly temperature independent intrinsic (J-H) colour over the range $``$2200-4000 K is clear, since nearly all low mass stars and substellar sources will deredden to a colour near (J-H) = 0.6. However, we prefer to use a two-colour sequence for formal dereddening where possible (see Section 3.2). The empty region to the upper right of the diagram represents the saturation limit near m$`{}_{H}{}^{}=11.7`$.
A large fraction ($`32\%`$) of the $`JH`$ sources are brown dwarf candidates, lying below the reddening track for a 0.08 M star at an age of 1 Myr, as calculated from the B97 isochrones. Approximately 13 sources appear to lie below the 1 Myr track for an object with the minimum mass to burn Deuterium ($``$ 0.013 M). Following the definition suggested by Burrows (1997) it is convenient to call such objects free-floating planets, even though they are likely to have formed by cloud core fragmention in the same manner as stars and brown dwarfs. A definition by mass has the advantage that it can be applied before all the formation mechanisms are known. An interesting feature of Figure 1(a) is the paucity of very faint blue sources with m$`{}_{J}{}^{}>19`$, given that several fairly red sources are seen below this limit. This observation is based on a small number of objects but it is supported by the upper limits in Figure 1(b), which show that many very faint red sources are detected at H band only, but none with (J-H) $`<1.3`$.This appears to indicate a sharp drop in the cluster Luminosity Function, at a level corresponding to about 8 M<sub>Jup</sub>. The significance of this is discussed in Section 4.2.
$`IJH`$ photometry of the observed point sources is presented in Table 1<sup>1</sup><sup>1</sup>1 Table 1 is available electronically by FTP to star.herts.ac.uk, in pub/Lucas/Orion. together with astrometry, dereddened magnitudes, luminosities, derived masses (using both the B97 and DM98 isochrones) and temperatures. We have surveyed the inner regions of the Orion Nebula cluster, in which most star formation is thought to have occurred over the period between 0.3 and 2 Myrs ago (Ali & Depoy 1995, Hillenbrand 1997), which leads to a small uncertainty in the derived masses, which we have indicated on Figure 1(a) by plotting the brown dwarf threshold for three different ages. A small proportion of younger sources is undoubtedly present, whose masses will be less than we have estimated, but we have excluded the 18 sources which exhibited extended (non-stellar) profiles from our photometry and performed a further colour selection against proplyds (see Section 3.2) so only a few extremely young sources should remain in Table 1.
The $`I`$ vs.(I-J) data in Figure 2 show that nearly $`90\%`$ of the sources follow a well defined sequence which is almost parallel to the reddening line. However, 41/313 of the sources ($`13\%`$) have much bluer (I-J) colours and lie to the left of the arbitrary line, parallel to the reddening line, which is plotted in Figure 2. An initial suspicion that this might be due to poor photometry was disproved by observing that the same two sequences are followed in a subset of 92 sources located more than $`2`$ arcminutes away from the cluster core, where photometry is not compromised by bright nebulosity (not shown). Since we have selected filters with no strong low-excitation emission lines the obvious interpretation is that the blue colours are due to scattered light from those objects which are very young or are viewed close to the plane of the accretion disk. This interpretation is confirmed by comparison with the list of proplyds detected via emission line imaging with the Hubble Space Telescope (HST) in O’Dell & Wong (1996). Only 11/41 were detected in their relatively shallow optical surveys, which suffer from greater extinction and do not overlap precisely with our survey. However, 10/11 are listed as proplyds, and 1/11 is listed as a star. The star is presumed to illuminate circumstellar matter which was not detected by HST because it does not receive sufficient UV radiation from the central O-type stars, or because the proplyd “tail” lies too close to the line of sight. These 10 proplyds are marked with crosses in Figure 2.
The 41 blue sources are widely distributed throughout the survey region and do not display obviously unusual (J-H) colours, only a weak tendency to be bluer than the rest. The $`I`$ vs. I-J diagram appears to be an efficient way of detecting proplyds at large distances from the photoionising O stars. However, the effect of circumstellar matter on infrared colours is not obvious, since it depends on the orientation of the system and distribution of matter in a complicated way (eg. Kenyon et al.1993). The observed (J-H) colours may well be different from the photospheric colours in these systems so we have excluded all the blue sources from our dereddening analysis. It is likely that some sources in the red group also have slightly modified colours due to anomalous extinction (i.e. both absorption and scattering) but this is not expected to be significant for most sources of age $`1`$ Myr, since the spectral energy distributions of T Tauri stars are usually well fitted by a Planckian (photospheric) function at wavelengths between the visible and the thermal infrared (eg. Rydgren et al. 1976, Wilking et al. 1989). The 202 faint sources detected only at $`J`$ and $`H`$ cannot be probed for anomalous colours in this way and probably include some sources which are therefore inaccurately dereddened. However, the proportion of anomalous sources will be lower than $`13\%`$ in this group because the blue sources are easier to detect at I band. Those sources in Figure 1(a) with m$`{}_{J}{}^{}>16.2`$ which are detected in all 3 filters are almost exclusively members of the blue group. Hence, blue sources detected only at $`J`$ and $`H`$ will be very few except below the $`I`$ band detection limit which corresponds to $`m_J18`$. A likely example of one such faint source at lower left in Figure 1(a-b) is Orion 131-047 (adopting the O’Dell & Wen (1994) naming convention), for which m$`{}_{J}{}^{}=18.22`$ and (J-H) = 0.39. This is an unrealistically blue colour for such a low luminosity source (which has accurately measured fluxes) so it is likely that the (J-H) colour has been reduced by at least 0.2 magnitudes by scattering effects.
### 3.1 Extinction Law
Many observers have found evidence for an anomalous extinction law in the Trapezium at optical and infrared wavelengths (eg. Davis et al.1986; Cardelli, Clayton and Mathis 1989, hereafter CCM). We adopt the R$`{}_{V}{}^{}=5.5`$ extinction law of CCM, calculated for $`\theta _1`$C Ori using data at similar wavelengths to those observed here. The infrared extinctions for this law are A(I<sub>C</sub>)=0.643, A(I<sub>U</sub>)=0.583, A(J)=0.334, A(H)=0.214 for A(V)=1. The effect of the unusually high R<sub>V</sub> is to increase the slope of the reddening lines, resulting in higher derived luminosities and masses for sources with significant extinction. However, the change in near infrared reddening is quite small because, as showed by CCM, all extinction laws converge at wavelengths $`\stackrel{>}{}0.9\mu `$m, such that A($`\lambda `$)/A(I) is similar in all known clusters. Only the optical extinction is greatly modified and we are not concerned with this. Hence, we use $`A(J)=2.783E(JH)`$, which compares with $`A(J)=2.364E(JH)`$ for the R=3.05 interstellar extinction law derived from Whittet (1990) and $`A(J)=2.543E(JH)`$ which follows from the oft-quoted $`\lambda ^{1.8}`$ law for near infrared interstellar extinction. This convergence of extinction laws in the near infrared is fortunate, since we doubt whether a common extinction law will apply to all the stars in a given star formation region, especially in Orion where the nebulosity has such complex spatial structure.
### 3.2 Dereddening Procedure
For the $`IJH`$ detections, we have used the unusual procedure of dereddening to an empirically derived curve in the (I-J) vs.(J-H) two-colour diagram, rather than to a theoretical curve in a colour-magnitude diagram. This was for two reasons: firstly, theoretical models are at an early stage of development for such young substellar sources and appropriate colour predictions have yet to be published at the time of writing; secondly, theoretical colour predictions are subject to the significant uncertainty in the absolute temperature-colour calibration which we referred to in Section 1. The empirical curve (Figure 3(a)) was derived from a fit to observations of M and L dwarfs of near solar metallicity from Leggett et al.(1998) and UKIRT stellar standards of unknown metallicity for (I-J) $`<1`$, where there is no known metallicity dependence. A possible flaw in this approach is that young stars and brown dwarfs have lower surface gravities than main sequence objects of the same effective temperature. However, the colour predictions by Baraffe et al.(1998) for young low mass stars indicate that much larger changes in $`log(g)`$ ($`>1`$ dex) have a negligible effect on these colours. The B97 models also indicate that $`log(g)`$ at 1 Myr does not approach red giant values for the masses considered here, so the empirical curve is not likely to be far in error. A cubic polynomial was used, with a Gaussian addition to fit the peak at (I-J) $`1.0`$. The form of the relation is a good match to the plots of Bessell & Brett (1988) and Leggett (1992).
The results for the 272 $`IJH`$ sources which are not anomalously blue in (I-J) are plotted in Figure 3(b). 95 sources have double valued solutions but only 1 solution has a plausible colour and flux in nearly every case. The handful of uncertain choices are low mass stars near the bump in the curve at $`(IJ)1.0`$, where the two solutions lie close together in (J-H) and derived luminosity but differ substantially in (I-J) and hence derived temperature. These ambiguous sources are listed as such in Table 1. We are confident that the dereddened (J-H) colours are accurate to $`\pm `$ 0.1 mag in nearly every case, given the weak temperature dependence of this colour; a standard error of 0.05 mag is estimated, due to measurement error and uncertainties in the empirical curve and dereddening law. The $`J`$ band extinction correction should therefore have typical uncertainties of $`\pm 0.14`$ mag, which is small enough to produce a useful Luminosity Function. The dereddened (I-J) colours are more sensitive to any errors in the process, such that the standard error is approximately 0.25 mag. However, the very strong temperature dependence of the (I-J) colour means that this leads to only a modest uncertainty in derived values of $`T_{eff}`$ (q.v Section 4.3).
The 202 $`JH`$-only detections were dereddened to the theoretical track shown in the colour-magnitude diagrams (Figure 1(a-b), which is a simple linear fit to an $`L`$-$`T_{eff}`$ relation (taking the average of the DM98 and B97 predictions at 1 Myr), and a $`T_{eff}`$-(J-H) relation (taking the average of the Wilking et al (1999) and Baraffe et al.(1998) relations for main sequence stars, which agree to 0.02 mag). Only the B97 and Baraffe predictions extend to the faintest magnitudes and lowest temperatures ((I-J)$`>3.3`$) . In this region the (J-H) colour changes more rapidly with T<sub>eff</sub> and the uncertainties increase.
## 4 Interpretation
### 4.1 Luminosity Function and IMF
The Luminosity Function (LF) is plotted for all sources detected at $`H`$ band with m$`{}_{H}{}^{}>12.25`$ in Figure 4(a). The function declines from a strong peak at m$`{}_{H}{}^{}12.5`$, which is not well measured here due to saturation. Zinnecker, McCaughrean & Wilking (1993) and Ali & Depoy (1995) observed the equivalent K band function and found a peak at m<sub>K</sub>=11-12, the function declining to fainter magnitudes but flattening off and possibly rising beyond m<sub>K</sub>=14. In our data a strong peak at $`H`$=16.5 is apparent, which probably has physical significance given that the function is based on more than 500 sources, and any real features are blurred by extinction of typically 1 magnitude at H band. The peak exists independently of the magnitude binning and is seen in Figure 1(b) as a clump of sources with 16$`<m_H<`$17. A corresponding feature exists in the J band LF at J=17.5. The completeness falls gradually, due to the variable nebular surface brightness but is estimated at $`>90\%`$ to m<sub>H</sub>=18.0, as evidenced by the small secondary peak there.
The observed function is converted to the absolute Luminosity Function shown in Figure 4(b), including only the dereddened sources (i.e. excluding H band only detections and blue $`IJH`$ sources. $`M_{bol}`$ is determined for each source using $`J`$ band magnitudes and bolometric corrections (these being well established at $`J`$) and a distance modulus of 8.22. The bias due to necessarily retaining some faint, anomalously blue sources due to lack of $`I`$ band data is only significant below m<sub>J</sub>=18, which is approximately the completeness limit (the nebulosity reduces sensitivity more in the $`I`$ and $`J`$ bands than at $`H`$.) We adopted the following bolometric corrections, derived from a simple fit to the relations between T<sub>eff</sub>, J-H and BC<sub>J</sub> of Wilking et al.(1999) (and Baraffe et al.(1998) for T$`{}_{eff}{}^{}>3500K`$) and using the DM98 1 Myr isochrone to connect T<sub>eff</sub> to Luminosity:
$`BC_J=1.955;J_{dr}>13.77`$ (4)
$`BC_J=0.1583J_{dr}0.2248;J_{dr}<13.77`$ (5)
where $`J_{dr}`$ is the dereddened $`J`$ magnitude and the BC<sub>J</sub> refers to the UFTI filter, which has smaller bolometric corrections than the CIT filter.
Figure 4(b) shows a primary peak at M$`{}_{bol}{}^{}=6`$ and a small secondary peak at M$`{}_{bol}{}^{}=10.5`$. Incompleteness is significant for M$`{}_{bol}{}^{}<6.2`$, due to saturation of bright sources, and for M$`{}_{bol}{}^{}>11.75`$, which corresponds to $`J>18`$. The secondary peak corresponds roughly to the peak at $`H=16.5`$ in Figure 4(a). We convert the LF to the IMF using the tracks of B97 and DM98. To remove bias due to non-detection of highly reddened sources we include only sources with $`(JH)_{obs}<1.5`$. The results in Figure 5 therefore represent an unbiased sample of the IMF complete to $`log(M/M_{})=1.5`$. In a log-log plot, both IMF’s show a fairly flat stellar function which has a tendency to fall slightly into the brown dwarf regime. Both IMFs also show some indication of a rise beyond the completeness limit but deeper observations will be needed to quantify this.
The discovery of a large population of brown dwarf candidates contrasts with previous surveys (eg. Hillenbrand & Hartmann 1998) which have concluded that few substellar objects exist. This conclusion appears to have been based on the well-established decline in the LF beyond the principal peak at m$`{}_{K}{}^{}11.5`$. This survey is the first to go deep enough at infrared wavelengths to detect the secondary peak in the LF.
### 4.2 A Cut-off in the Luminosity Function ?
The absence of faint blue sources (see Section 3.1) is well established by the inclusion of the $`H`$ band upper limits. This may indicate a sharp turn-down in the LF, and perhaps a cut-off, at a level corresponding to about 8 Jupiter masses. However, such sources would be close to the survey sensitivity limit (which we believe to lie just below 5 M<sub>Jup</sub>) particularly if their intrinsic (J-H) colours are redder than we expect. Moreover, the B97 mass-luminosity relation becomes steeper below about 8 M<sub>Jup</sub>, so a turn down in the LF would occur even for a flat IMF. Hence a deeper survey will be needed to confirm the reality of the fall in the IMF. The least massive detection with good photometry is Orion 023-115, which has J=19.38, (J-H) =0.97 and a derived mass of $`8.4_{2.7}^{+1.4}`$ M<sub>Jup</sub> ($`8.0_{2.6}^{+1.3}\times 10^3M_{}`$) for an age of 1 Myr, using the B97 tracks. The quoted $`+/`$ uncertainties refer to alternative ages of 2 Myr and 0.3 Myr respectively. None of the faint sources in Figure 1(a) with highly uncertain $`J`$ band fluxes have a lower derived mass. If real, the turn-down might be attributed to a minimum Jeans mass for gravitational cloud core collapse, below which star formation cannot occur. Alternatively, it may be due to the ending of star formation in the cluster before extremely low mass cloud cores had time to collapse, since this process is believed to take longer in less massive cores. This may be attributable to dispersal of dense molecular gas by the photoionising O-type stars at the centre of the cluster. The IAC group apparently find no sign of the turn down in the LF at planetary masses in the neighboring $`\sigma `$ Orionis cluster, so we favour the second explanation. If correct, this explanation may still be significant with regard to any galactic population of free-floating planets: most star formation is believed to occur in high mass star formation regions like the Trapezium or M16, with the consequence that free-floating planets may be relatively rare. However, the effectiveness of this mechanism for reducing planet formation efficiency will vary depending on local conditions.
### 4.3 Effective Temperatures
The dereddened I-J colours are converted to the effective temperatures in Table 1 using the models of Baraffe et al.(1998). The derived values of T<sub>eff</sub> are in reasonable agreement with the predictions of B97 and DM98, in which $`2600<T_{eff}<2900K`$ for objects of brown dwarf mass at 1 Myr. However a few sources in this mass range are detected with dereddened $`(IJ)3`$, which implies $`T_{eff}2500`$ K, at least for main sequence objects. We estimate that our derived temperatures are accurate to $`\pm 200K`$, leaving aside the uncertainty in the absolute T<sub>eff</sub> vs. I-J relation and assuming this is not altered significantly by the youth of the sources. In any case the I-J colours should provide a useful guide to relative temperatures.
### 4.4 Potential Problems
We have carefully avoided several potential problems in studies of young clusters, such as cluster membership, infrared excess and line emission but some serious issues remain. We consider each in turn. (1) Foreground contamination is minimal in the tiny area surveyed, introducing perhaps 5 red dwarfs into the sample over a range of about 3 magnitudes, using the stellar space densities of Tinney (1993). (2) Background contamination is removed by the dark backdrop of OMC-1 (see Hillenbrand & Hartmann 1998). Our avoidance of the K band filter removes any chance of seeing through OMC-1 at the faint limits. (3) Infrared excess due to hot dust is not believed to be significant at H band, since the spectra of T Tauri stars are well fitted by black bodies at this wavelength. (4) Line emission was minimised by the choice of filters. (5) Variation in the infrared extinction law will probably be small (see Section 3.4) and the effect on the derived IMFs is minimised by excluding highly reddened sources. (6) Scattering is a potentially serious problem. Even sources without anomalous colours in the $`I`$ vs.(I-J) diagram may have distorted (J-H) colours and fluxes. As noted in Section 3 however, photospheric flux dominates the spectral energy distributions of most T Tauri stars for 0.6 $`\mu `$m $`<\lambda <`$$`\mu `$m) so significant distortion of broad band colours and fluxes is unlikely to be common. This should be investigated in future by searching for the polarisation signature of scattered light. (7) The evolutionary tracks for substellar objects are in an early stage of development, which leads to a significant uncertainty in derived masses. However the fairly close similarity of the B97 and DM 98 tracks, for luminosity and $`T_{eff}`$, is encouraging.
## 5 Conclusions
A large population of brown dwarf candidates is detected in the Trapezium Cluster and a small population of objects with planetary masses. We have confidence that these are true cluster members and, though many uncertainties exist in deriving the masses, they are not likely to be large enough to cause misclassification of low mass stars as low mass brown dwarfs or free-floating planets. The derived IMF is fairly flat on a log-log plot at low stellar masses but declines slightly at brown dwarf masses, indicating that brown dwarfs are a little less numerous than stars. There is a possible small peak near $`0.02`$M, which is below the completeness limit. Approximately 13 planetary mass objects are detected but none with M$`<8\times 10^3`$M. We suggest that this is due to dispersal of the star-forming cloud by the photoionising O-stars before such objects had time to form.
Approximately 13% of sources have anomalously blue (I-J) colours, which we attribute to scattering from circumstellar material. These blue excesses are strongly correlated with detection as ’proplyds’ by HST , so this colour selection may prove to be a powerful new tool for detecting sources with circumstellar envelopes. The effect of scattering on the colours of the general population should be investigated via polarimetry and spectroscopy.
Acknowledgements
We wish to thank the staff of UKIRT, which is operated by the Joint Astronomy Centre on behalf of the UK Particle Physics and Astronomy Research Council (PPARC). Particular thanks are due to Sandy Leggett and Andy Adamson for providing information on the colours of cool stars and for carrying out reactive observations for us. We also thank the Panel for the Allocation of Telescope Time for providing us with reactively rescheduled observing time. We thank Juliette White for helping us with the dereddening procedure and we are grateful to the referee, Richard Jameson, for useful and timely comments. PWL is grateful for support by PPARC via a Post Doctoral Fellowship at the University of Hertfordshire.
References
Ali B., & Depoy D.L. 1995, AJ, 109,709
Baraffe I., Chabrier G., Allard F., & Hauschildt P.H. 1998, A&A, 337,403
Bessell M.S., & Brett J.M., PASP, 100,1134
Burrows A. 1997, ApJ, 491,856
Cardelli J.A., Clayton G.C., & Mathis J.S. 1989, ApJ, 345,245
Comeron F., Rieke G.H., Burrows A., Rieke M.J. 1993, ApJ 416,185
Comeron F., Rieke G.H., & Rieke M.J., ApJ 473,294
D’Antona F., & Mazzitelli 1997, MmSAI, 68,607
Davis D.S., Larson H.P., & Hofmann R., 1986, ApJ, 304,481
Hillenbrand L.A. 1997, AJ 113,1733
Hillenbrand L.A., & Hartmann L.W. 1998, ApJ, 492,540
Kenyon S.J., Whitney B.A., Gomez M., & Hartmann, L., 1993, ApJ 414,773
Leggett S.K., Allard F., & Hauschildt P.H. 1998, ApJ 509,836
Luhman K.L., & Rieke G.H., 1998, ApJ 497,354
Luhman K.L., Rieke G.H., Lada C.J., & Lada E.J., 1998, ApJ 508,347
Nakajima T., Oppenheimer B.R., Kulkarni S.R., Golimowski D.A., Mathews K., & Durrance S.T., 1995, Nature 378,463
O’Dell C.R., & Wong K. 1996, AJ 111,846
Rebolo R, Zapatero-Osorio M.R., & Martin E.L., 1995, Nature 377,129
Roche P.F., & Lucas P.W., 1998, on-line, www-astro.physics.ox.ac.uk/ pwl/camera.html
Rydgren A.E., Strom S.E., & Strom K.M., ApJS 30,307
Tinney C.G. 1993, ApJ, 414, 279
Whittet D.C.B., 1992, Dust in the Galactic Environment. Institute of Physics, Bristol.
Wilking B.A., Greene T.P., & Meyer M.R. 1999, AJ 117,469
Wilking B.A., Lada C.J., & Young E.T. 1989, ApJ 340,823
Williams D.M., Comeron F., Rieke G.H., & Rieke M.J., 1995, ApJ 454,144
Zinnecker H., McCaughrean M.J., & Wilking B.A. 1993, in “Protostars and Planets III”, p429, edited by E.H.Levy & J.I.Lunine, pub. Tucson: University of Arizona Press.
Figure 1: (a) J vs. (J-H) plot. Open circles are highly uncertain data points. The dotted line is an approximate zero reddening track (see text). The solid lines are parallel to the A(V)=7 reddening vector and divide the population into stars, brown dwarfs and planets, using the B97 prediction and an age of 1 Myr. The dashed lines correspond to the 0.3 Myr and 2 Myr predictions, indicating the effect of the age spread on the classification. The effect is similar at the planetary boundary.
(b) H vs (J-H) plot. This includes upper limits for sources with $`J>20`$, which confirms the paucity of faint blue sources.
Figure 2: I vs (I-J) plot. Anomalously blue sources lie to the left of the arbitrary line parallel to the reddening track. Of the 11 blue sources detected by HST, 10 are proplyds, plotted as crosses.
Figure 3: (a) Empirical 2-colour curve fitted to the plotted observations. Two example dereddening tracks are also shown, indicating the possibility of double-valued solutions.
(b) results of dereddening Orion data.
Figure 4: (a) Observed H band luminosity Function. The equivalent J band function is overplotted as a dashed line. Both functions are complete to approximately magnitude 18.
(b) Dereddened Luminosity Function, complete between magnitudes 6.2 and 11.75.
Figure 5: IMFs for Burrows 1997 and DM98 tracks, complete to log(M/M$`{}_{}{}^{})=1.5`$. Errorbars are plotted assuming Poisson statistics. The IMF appears to fall slowly into the Brown Dwarf regime, but rises again below the completeness limit. |
warning/0003/hep-th0003227.html | ar5iv | text | # RU-NHETC-2000-10 hep-th/0003227 Noncommutative Gauge Theory, Divergences and Closed Strings
## 1 Introduction
Noncommutative field theories have been the focus of much interest recently. Following their appearance in Matrix theory , and in string theory , there has been renewed interest in their perturbative study. In scalar field theories have been studied perturbatively. Surprisingly, some remnants of stringy behaviour are visible even in the perturbation theory. Subsequent discussions of the scalar field theories include classical solutions , and finite temperature effects .
In the Wilsonian effective action of non-commutative scalar theories was discussed. The non-commutativity parameter acts as an effective ultra-violet cut-off, suppressing all ultraviolet divergences in non-planar diagrams. Instead, one encounters curious infrared divergences in the 1PI effective action. These divergences come from high momentum integration. Therefore they are incorrectly cut-off in the Wilsonian approach.
In order to repair the Wilsonian approach, a general procedure was suggetsted in . The Wilsonian effective action includes extra light (non-propagating) modes, which have the required couplings to correct the infrared behaviour of the Wilsonian effective action. Those modes were interpreted as closed string modes.
We are interested here in examining the issues raised in in a context which is more closely related to string theory. To this effect we study the perturation theory of supersymmetric non-abelian gauge theory with matter<sup>1</sup><sup>1</sup>1Noncommuatative gauge theories have been studied perturbatively in .. One can then discuss noncommutative theories that have both the above mentioned infrared divergences (unlike the $`𝒩`$=4 theory), and also a string theory realization (unlike the scalar field theories).
We find that similar infrared effects arise also in those theories. In particular the procedure of adding closed string modes to the effective action works in the present context as well. Studying the stringy realization of the gauge theories reveals the origin of the extra modes. They are indeed closed string modes of the underlying string theory.
The paper is organized as follows:
In the next section we introduce the classical action for the noncommutative theories we study. These theories include an arbitrary product gauge group, with unitary factors, coupled to matter in the fundamental and in the adjoint representations. A somewhat surprising result is the existence of gauge invariant local operators if one includes matter in the fundamental representation. This is unlike the case of the pure gauge theory.
We then turn to studying the renormalization properties of noncommutative theories. After reviewing the results in about the renormalization of scalar theories, we calculate similar results in the gauge theory case. We concentrate on the $`\beta `$ functions of each of the gauge factors, and on the associated IR divergences.
As was the case in , we find that the UV does not decouple from the IR physics. When forming a Wilsonian effective action, there is a need to add some ”closed string” modes to account for infrared divergences. We use the procedure outlined in in the present context to write explicitly the required modes and their couplings.
In the rest of the paper we study a string realization of such theories. A general class of $`𝒩`$=1 supersymmetric theories, the so called quiver theories, can be realized as the worldvolume theories of branes transverse to an orbifold singularity. We review the construction of the quiver gauge theories and specify their matter content. The limit considered by Seiberg and Witten should then yield a noncommutative version of the quiver gauge theories.
In the last section we discuss the stringy realization of the closed string modes required to fix the Wilsonian effective action. We conclude that the twist fields are indeed of the right form to be these closed string modes. Their inclusion in the effective action summarizes the effect of the high momentum gauge theory modes that have been integrated out. They have the correct couplings by virtue of a relation between the $`\beta `$ function coefficients and twist field tadpoles, studied in . We identify all the required modes in the large $`N_c`$ limit of the gauge theories, and point out a universal discrepency to do with the overall $`U(1)`$ factor.
We conclude by discussing open questions regarding the (absence of) massive closed string contributions to the infrared divergences, and quadratic divergences in the orbifold realization of non-supersymmetric quiver gauge theories.
The relation between the perturbative calculation of and string theory has been also discussed in .
## 2 Noncommutative Gauge Theories with Matter
In a ordinary non-abelian gauge theory matter fields transform by a matrix representation of the gauge group. For a non-commutative gauge theory there can be two types of representations: left modules and right modules. This simply asserts that the gauge group acts on the field from the left or from the right. Gauge invariance restricts possible couplings of such matter fields as described below.
Suppose $`A_\mu `$ is a non-commutative gauge field, transforming in the adjoint representation of $`G=U(N)`$. The gauge transformation of $`A_\mu `$ is:
$$\delta A_\mu =_\mu ϵ+iϵA_\mu iA_\mu ϵ$$
(1)
Where we suppress the fundamental $`U(N)`$ indices $`i,j=1,\mathrm{},N`$ in the gauge field $`A_\mu `$ and in the gauge parameter $`ϵ`$. With respect to the global part of the gauge transformation, the gauge field transforms as a bi-module: $`G`$ acts simoultaneously from the left and from the right. The field strength that transforms covariantly is defined as:
$$F_{\mu \nu }=_\mu A_\nu _\nu A_\mu +iA_\mu A_\nu iA_\nu A_\mu $$
(2)
Then one can write the standard action for the gauge fields:
$$I=\frac{1}{4g^2}d^4xTr\left[F_{\mu \nu }F^{\mu \nu }\right]$$
(3)
Raising and lowering of spacetime indices is done with flat space metric. The trace is in the fundamental representation of $`U(N)`$. The kinetic action for several $`U(N)`$ gauge factors is simply the sum of this action for each of the gauge factors.
We are now ready to discuss matter couplings (matter couplings are discussed in ). The gauge transformations for the fundamental left or right modules are:
$`\delta \mathrm{\Phi }_L=iϵ\mathrm{\Phi }_L`$
$`\delta \mathrm{\Phi }_R=i\mathrm{\Phi }_Rϵ`$ (4)
One can define covariant derivatives which transform similarly, as follows:
$`D_\mu \mathrm{\Phi }_L=_\mu \mathrm{\Phi }_L+iA_\mu \mathrm{\Phi }_L`$
$`D_\mu \mathrm{\Phi }_R=_\mu \mathrm{\Phi }_Ri\mathrm{\Phi }_RA_\mu `$ (5)
In the commutative limit, the left and right modules go over to fields in the fundamental and anti-fundamental respectively. It is natural to define Hermitian conjugation which pairs up left and right modules. In the commutative limit this notion of Hermitian conjugation goes over to the usual one.
A gauge invariant action can be written for a field and its Hermitian conjugate, which transform in the fundamental left and right modules, respectively. The gauge invariant kinetic term is:
$$I=d^4xTr\left[D_\mu \overline{\mathrm{\Phi }}_RD^\mu \mathrm{\Phi }_L\right]$$
(6)
We note that the kinetic term is gauge invariant before integration. The Lagrangian density provides therefore a gauge invariant local operator. One can easily consruct other such operators. Similarly one can use the fundamental representation to construct Wilson lines, which are gauge invariant for any particular path chosen. This is in contrast to the pure gauge theory case, where no such objects exist.
In the following we are interested in quiver gauge theories. These are product gauge theories with gauge group factors $`U(N)`$. The matter fields transform in bi-fundamental representation. In the non-commutative case this means that one factor of the gauge group acts from the left, and another from the right. We denote such a field schematically by $`\mathrm{\Phi }_{LR}`$, and its hermitian conjugate by $`\overline{\mathrm{\Phi }}_{RL}`$. The covariant derivatives of the fields $`\mathrm{\Phi }_{LR},\overline{\mathrm{\Phi }}_{RL}`$ are:
$`D_\mu \mathrm{\Phi }=\mu \mathrm{\Phi }+iA_\mu ^{(1)}\mathrm{\Phi }i\mathrm{\Phi }A_\mu ^{(2)}`$
$`D_\mu \overline{\mathrm{\Phi }}=\mu \overline{\mathrm{\Phi }}+iA_\mu ^{(2)}\overline{\mathrm{\Phi }}i\overline{\mathrm{\Phi }}A_\mu ^{(1)}`$ (7)
A gauge invariant kinetic term is then:
$$I=d^4xTr\left[D_\mu \overline{\mathrm{\Phi }}D^\mu \mathrm{\Phi }\right]$$
(8)
Note that the Lagrangian density now is gauge invariant with respect to one of the gauge factors. With respct to the other gauge factor acting on $`\mathrm{\Phi }_{LR}`$, it is gauge invariant only up to total derivative. Therefore, in a quiver gauge theories it is still difficult to construct simple gauge invariant local operators.
In addition we note that there is no longer a decoupled $`U(1)`$ in this case, as is the case of the single $`U(N)`$ gauge theory.
The quiver gauge theories appear naturally in string theory, as reviewed below. In the next section we study The Wilsonian effective action of these gauge theories. We find appearance of closed string modes similar to .
## 3 IR Divergences in Scalar Field Theory
We first review IR divergences in noncommutative $`\varphi ^4`$ theory. The action is
$$S=d^4x\left[\frac{1}{2}(_\mu \varphi )^2+\frac{1}{2}m^2\varphi ^2+\frac{1}{4!}g^2\varphi \varphi \varphi \varphi \right]$$
(9)
This 2 point function is calculated using the diagram above, which yields:
$$\mathrm{\Gamma }^{(2)}=\frac{g^2}{3(2\pi )^4}\frac{d^4k}{k^2+m^2}cos(\frac{k.p}{2})$$
(10)
Rewrite the integrals in terms of Schwinger parameters using
$$\frac{1}{k^2+m^2}=𝑑\alpha e^{\alpha (k^2+m^2)}$$
(11)
The integrals are regulated by multiplying the integrands by $`e^{\frac{1}{\alpha \mathrm{\Lambda }^2}}`$. Then
$`\mathrm{\Gamma }^{(2)}=\mathrm{\Gamma }_{planar}^{(2)}+\mathrm{\Gamma }_{nonplanar}^{(2)}`$ (12)
$`\mathrm{\Gamma }_{planar}^{(2)}={\displaystyle \frac{g^2}{3(2\pi )^4}}{\displaystyle 𝑑\alpha d^4ke^{\alpha (k^2+m^2)\frac{1}{\alpha \mathrm{\Lambda }^2}}}`$ (13)
$`\mathrm{\Gamma }_{nonplanar}^{(2)}={\displaystyle \frac{g^2}{6(2\pi )^4}}{\displaystyle 𝑑\alpha d^4ke^{\alpha (k^2+m^2)\frac{1}{\alpha \mathrm{\Lambda }^2+ikp}}}`$ (14)
These can evaluated to give
$`\mathrm{\Gamma }_{planar}^{(2)}={\displaystyle \frac{g^2}{3(2\pi )^4}}\left(\mathrm{\Lambda }^2m^2ln\left({\displaystyle \frac{\mathrm{\Lambda }^2}{m^2}}\right)+\mathrm{}\right)`$
$`\mathrm{\Gamma }_{nonplanar}^{(2)}={\displaystyle \frac{g^2}{6(2\pi )^4}}\left(\mathrm{\Lambda }_{eff}^2m^2ln\left({\displaystyle \frac{\mathrm{\Lambda }_{eff}^2}{m^2}}\right)+\mathrm{}\right)`$ (15)
where
$`\mathrm{\Lambda }_{eff}^2={\displaystyle \frac{1}{\frac{1}{\mathrm{\Lambda }^2}+\stackrel{~}{p}^2}}`$
$`\stackrel{~}{p}_j=p^i(\mathrm{\Theta })_{ij}`$ (16)
The 1PI effective action is then
$`S={\displaystyle d^4p\frac{1}{2}(p^2+m^2)}+{\displaystyle \frac{g^2}{96\pi ^2(\frac{1}{\mathrm{\Lambda }^2}+\stackrel{~}{p}^2)}}{\displaystyle \frac{g^2}{96\pi ^2}}ln\left({\displaystyle \frac{1}{M^2(\frac{1}{\mathrm{\Lambda }^2}+\stackrel{~}{p}^2)}}\right)`$ (17)
In , the authors showed that the first new term in the above 1PI action could be obtained from a Wilsonian action with an extra $`\chi `$ field coupled linearly to $`\varphi `$.
$`S_{eff}={\displaystyle d^4x\frac{1}{2}(_\mu \varphi )^2}+{\displaystyle \frac{1}{2}}m^2\varphi ^2+{\displaystyle \frac{1}{4!}}g^2\varphi \varphi \varphi \varphi `$
$`+{\displaystyle d^4x\frac{1}{2}(\chi )o(\chi )}+{\displaystyle \frac{1}{2}}\mathrm{\Lambda }^2(o\chi )^2+{\displaystyle \frac{i}{\sqrt{96\pi ^2}}}g\chi \varphi `$ (18)
Integrating out $`\chi `$ correctly reproduces the first correction to the 1PI action.
Similarly the logarithmic term can be obtained by adding a second field $`\chi _2`$ with a coupling $`d^4xg\chi _2\varphi `$ and a logarithmic propagator
$$\chi _2(p)\chi _2(p)=2ln\left(\frac{\frac{1}{\mathrm{\Lambda }^2}+\stackrel{~}{p}^2}{\stackrel{~}{p}^2}\right)$$
(19)
The inclusion of fields with logarithmic propagators seems arbitrary, but showed that there was a natural interpretation of these fields as coming from closed string fields living in 2 extra dimensions. The 3+1 dimensional theory where the $`\varphi `$\- quanta live is taken to be a 3-brane living in 5+1 dimensions. The $`\chi _2`$ fields live in all 5+1 dimensions, but couple to the $`\varphi `$ fields at the brane location. The $`\chi _1`$ fields live on the brane only.
The $`\chi `$ fields have the closed string metric, which is $`g^{\mu \nu }=\frac{1}{\alpha ^{}_{}{}^{}2}(\mathrm{\Theta }^2)^{\mu \nu }`$ in the brane directions, and $`\delta ^{\mu \nu }`$ in the transverse directions. Furthermore, there is a cutoff $`\frac{1}{\alpha ^{}\mathrm{\Lambda }}`$ on the transverse momenta of the $`\chi `$ fields.
Then the effective 4-dimensional propagator of the $`\chi `$ fields is
$`\chi _2(p)\chi _2(p)={\displaystyle ^{\frac{1}{\alpha ^{}\mathrm{\Lambda }}}}{\displaystyle \frac{d^2q}{(2\pi )^2}}{\displaystyle \frac{1}{\frac{\stackrel{~}{p}^2}{\alpha ^{}_{}{}^{}2}+q^2}}`$ (20)
$`={\displaystyle \frac{1}{4\pi }}ln\left({\displaystyle \frac{\frac{1}{\mathrm{\Lambda }^2}+\stackrel{~}{p}^2}{\stackrel{~}{p}^2}}\right)`$ (21)
as required.
## 4 Gauge Theories
We start with the case of $`𝒩`$=1 $`U(N)`$ noncommutative gauge theory.
The $`U(N)`$ gauge field can be written as:
$`A_\mu =A_\mu ^AT^A={\displaystyle \frac{1}{\sqrt{N}}}A_\mu ^0\mathrm{𝟏}+A_\mu ^at^a`$ (22)
where $`t^a`$ are $`SU(N)`$ matrices.
The standard action for the gauge fields is:
$$I=\frac{1}{4g^2}d^4xTr\left[F_{\mu \nu }F^{\mu \nu }\right]$$
(23)
with
$$F_{\mu \nu }=_\mu A_\nu _\nu A_\mu +iA_\mu A_\nu iA_\nu A_\mu $$
(24)
In momentum space one can write:
$`F_{\mu \nu }=p_\mu A_\nu ^AT_Ap_\nu A_\mu ^AT_A+ig(e^{i\stackrel{~}{p}^{(1)}p^{(2)}}A_\mu ^A(p^{(1)})T_AA_\nu ^B(p^{(2)})T_B`$
$`e^{i\stackrel{~}{p}^{(2)}p^{(1)}}A_\nu ^B(p^{(2)})T_BA_\mu ^A(p^{(1)})T_A)`$
$`=p_\mu A_\nu ^AT_Ap_\nu A_\mu ^AT_A+igA_\mu ^A(p^{(1)})A_\nu ^B(p^{(2)})(cos(\stackrel{~}{p}^{(1)}p^{(2)})[T_A,T_B]`$
$`+isin(\stackrel{~}{p}^{(1)}p^{(2)})\{T_A,T_B\})`$ (25)
The interaction terms are then
$`A_\mu ^A(p^{(1)})A_\nu ^B(p^{(2)})A^{\nu C}(p^{(3)})(p_\mu ^{(1)}cos(\stackrel{~}{p}^{(2)}p^{(3)})tr(T_A[T_B,T_C]+`$
$`ip_\mu ^{(1)}sin(\stackrel{~}{p}^{(1)}p^{(2)})T_A\{T_B,T_C\})`$
and
$`A_\mu ^A(p^{(1)})A_\nu ^B(p^{(2)})A^{\mu C}(p^{(3)})A^{\nu D}(p^{(4)})`$
$`(cos(\stackrel{~}{p}^{(1)}p^{(2)})[T_A,T_B]+isin(\stackrel{~}{p}^{(1)}p^{(2)})\{T_A,T_B\})`$
$`(cos(\stackrel{~}{p}^{(3)}p^{(4)})[T_C,T_D]+isin(\stackrel{~}{p}^{(3)}p^{(4)})\{T_C,T_D\})`$ (26)
We wish to compute the 1PI two point function $`FF`$ which is obtained from the diagrams in Fig. 2.
Note that every interaction involving an anticommutator is down by a factor $`\sqrt{N}`$ due to the normalization of $`A^0`$. We start by calculating the terms which are leading order in $`N`$. To this order, the vertices are identical to the commutative $`SU(N)`$ gauge theory with the replacement
$`f^{abc}f^{abc}cos(\stackrel{~}{p}^{(1)}p^{(2)})`$ (27)
The diagrams are each of the form
$`{\displaystyle \frac{d^4p}{(2\pi )^4}\frac{i}{p^2}\frac{i}{(p+q)^2}g^2C_2(G)\delta ^{ab}cos^2(\stackrel{~}{p}q)N^{\mu \nu }}`$ (28)
where
$`N_{(1)}^{\mu \nu }={\displaystyle \frac{1}{2}}(g^{\mu \rho }(qp)^\sigma +g^{\rho \sigma }(2p+q)^\mu +g^{\sigma \mu }(p2q)^\rho )`$
$`(\delta _\rho ^\nu (pq)_\sigma +g_{\rho \sigma }(2pq)^\nu +\delta _\sigma ^\nu (p+2q)_\rho )`$
$`N_{(2)}^{\mu \nu }=3(p+q)^2g^{\mu \nu }`$
$`N_{(3)}^{\mu \nu }=(p+q)^\mu p^\nu `$
$`N_{(4)}^{\mu \nu }=tr[\gamma ^\mu (i/[k])\gamma ^\nu (i/[k+q])]`$ (29)
We can combine the denominators by
$`{\displaystyle \frac{1}{p^2(p+q)^2}}={\displaystyle _0^1}𝑑x{\displaystyle \frac{1}{((1x)p^2+x(p+q)^2)^2}}={\displaystyle _0^1}𝑑x{\displaystyle \frac{1}{(P^2M^2)^2}}`$ (30)
where $`P=p+xq`$ and $`M^2=x(1x)q^2`$. We then write the $`N_{(i)}^{\mu \nu }`$ in terms of $`P,q`$. We can drop terms linear in $`P`$ by symmetry.
The diagrams are then of the form
$`{\displaystyle _0^1}𝑑x{\displaystyle \frac{d^4P}{(2\pi )^4}\frac{1}{(P^2M^2)^2}g^2C_2(G)\delta ^{ab}cos^2(\stackrel{~}{P}q)\overline{N}^{\mu \nu }}`$ (31)
with
$`\overline{N}_{(1)}^{\mu \nu }={\displaystyle \frac{1}{2}}(2g^{\mu \nu }P^210P^\mu P^\nu g^{\mu \nu }q^2((2x)^2+(1+x)^2)`$
$`+q^\mu q^\nu (2(12x)^2+2(1+x)(2x))`$
$`\overline{N}_{(2)}^{\mu \nu }=3g^{\mu \nu }(P^2+(1x)q^2)`$
$`\overline{N}_{(3)}^{\mu \nu }=P^\mu P^\nu q^\mu q^\nu x(1x)`$
$`\overline{N}_{(4)}^{\mu \nu }=4P^\mu P^\nu 2g^{\mu \nu }P^2+2g^{\mu \nu }q^2x(1x)2q^\mu q^\nu x(1x)`$ (32)
The terms linear in $`P^2`$ in the above expressions give quadratic divergences which are cancelled in the usual commutative case, but give unpleasant IR divergences in the noncommutative case . However, in the supersymmetric case, these divergences cancel.
The terms quadratic in $`q`$ are then summable to give the final answer
$`{\displaystyle _0^1}𝑑x{\displaystyle \frac{d^4P}{(2\pi )^4}\frac{1}{(P^2M^2)^2}g^2C_2(G)\delta ^{ab}cos^2(\stackrel{~}{P}q)(q^\mu q^\nu g^{\mu \nu }q^2)}`$ (33)
which is manifestly gauge invariant as in the commutative case.
The noncommutative case therefore differs from the commutative answer only through the replacement
$`{\displaystyle \frac{d^4P}{(2\pi )^4}\frac{1}{(P^2M^2)^2}}{\displaystyle \frac{d^4P}{(2\pi )^4}\frac{1}{(P^2M^2)^2}cos^2(\stackrel{~}{P}q)}`$ (34)
and this is always the case if the quadratic divergences cancel in the integrand. As shown in , this is always the case in supersymmetric theories. The LHS of the equation above produces logarithmic UV divergences (in the commutative theory). The RHS (in the noncommutative theory) produces logarithmic IR divergences. The above analysis says that we can obtain the logarithmic IR divergences of the noncommutative theory by replacing $`ln\mathrm{\Lambda }`$ of the commutative theory by $`ln\mathrm{\Lambda }_{eff}`$. This is identical to the results in for the scalar field theory case.
Now the effective action of the usual Yang-Mills theory is of the form
$`S={\displaystyle TrF^2}+\beta (g)ln({\displaystyle \frac{p^2}{\mathrm{\Lambda }^2}})TrF^2+\mathrm{}`$ (35)
where the second term is from the running coupling.
The noncommutative effective action is then of the form
$`S={\displaystyle TrF^2}+\beta (g)ln(\stackrel{~}{p}^2)TrF^2+\mathrm{}`$ (36)
Thus, the coefficient of the IR divergence is proportional to the beta function coefficient, to the leading order in $`N`$. This is also the case for the matter diagrams. Hence, the beta functions for the quiver gauge theories are accompanied by IR divergences with the same coefficient. This will be important in the relation to closed strings.
For completeness, we calculate the subleading $`\frac{1}{N}`$ corrections to the amplitude. We will keep the external legs in the nonabelian part of the theory.
The diagrams are then the same as in the commutative case, with the replacement
$`f^{abc}sin(\stackrel{~}{p}^{(1)}p^{(2)})(\delta ^{ab}\delta ^{c0}+\delta ^{ac}\delta ^{b0}+\delta ^{bc}\delta ^{a0})`$ (37)
The final answer is then obtained from the commutative theory by the replacement
$`{\displaystyle \frac{d^4P}{(2\pi )^4}\frac{1}{(P^2M^2)^2}C_2(G)}{\displaystyle \frac{d^4P}{(2\pi )^4}\frac{1}{(P^2M^2)^2}\frac{1}{N}sin^2(\stackrel{~}{P}q)}`$ (38)
This leads to further IR divergences, which are however supressed in the large $`N`$ expansion.
## 5 Orbifold Constructions
Constructions of SUSY gauge theories as the worldvolume theories of Dirichlet branes transverse to an orbifold singularity was pioneered in . The orbifold action is accompanied by an action of the discrete point group $`\mathrm{\Gamma }`$ on the Chan-paton matrices, via a finite dimensional matrix representation. In the regular representation was utilized. This was further developed by the introduction of fractional branes , corresponding to Chan-Paton factors in arbitrary representations of $`\mathrm{\Gamma }`$.
This construction allows for a construction of a general class of gauge theories, with a prescribed matter content and interactions. As the methods involved are well known, we refer the reader to for a more detailed derivation of results used below.
We are interested in putting D3 branes transeverse to an orbifold. We choose for simplicity two types of orbifolds:
First, we can consider orbifolds of the form $`C^2/Z_m`$, when interested in $`𝒩`$=2 SUSY theories. The single generator of the orbifold acts on two complex coordinates $`X_1,X_2`$ as follows:
$`X_1e^{2\pi i/m}X_1`$
$`X_2e^{2\pi i/m}X_2`$ (39)
One obtaines a product gauge theory $`U(N_1)\times \mathrm{}\times U(N_m)`$, with a bi-fundamental hypermultiplet, $`(N_r,\overline{N}_{r+1})`$, for each neighboring gauge groups (which are cyclically ordered). The case where all the factors $`N_r`$ are identical is the case studied in . In this case the theory turns out to be conformal. This is reflected in the orbifold model having no twisted sector tadpoles . For other choices of integers $`N_r`$, one can have logarithmic divergences, for example a non-vanishing $`\beta `$ function. This class of gauge theories has $`𝒩`$ =2 supersymmetry, and are therefore non-chiral.
Furthermore, we can consider orbifolds of the form $`C^3/(Z_m\times Z_n)`$, when interested in $`𝒩`$=1 supersymmetric theories. The $`𝒩`$ =2 example is a special case of this class of orbifolds. The two generators of $`\mathrm{\Gamma }`$, $`\alpha ,\beta `$, act as above on the complex planes spanned by $`(X_1,X_2)`$ (for $`\alpha `$), and on the plane spanned by $`(X_2,X_3)`$(for $`\beta `$). The action of a group element $`\alpha ^k\beta ^k^{}`$ on the Chan-Paton factors is given by a matrix $`\gamma _{(k,k^{})}`$.
The matter content can be summarized by the brane box rules . The gauge group is a product of unitary gauge group, one for each number $`N_{rs}`$, where $`r=1,\mathrm{}m`$ and $`s=1,\mathrm{},n`$. There are also chiral multiplets in bifundamental representations of neighbouring gauge groups. If we cyclically order $`r,s`$, there are the following chiral multiplets for each $`r,s`$:
$`(N_{r,s},\overline{N}_{r1,s}),(N_{r,s},\overline{N}_{r,s+1}),(N_{r,s},\overline{N}_{r+1,s1})`$ $`\text{in fundamental of}U(N_{r,s})`$
$`(N_{r+1,s},\overline{N}_{r,s}),(N_{r,s1},\overline{N}_{r,s}),(N_{r1,s+1},\overline{N}_{r,s})`$ in the anti-fundamental
This orbifold theory can be chiral, and care has to be taken to obtain anomaly free gauge theories. This is done by cancelling a certain class of tadpoles . These are dubbed dimension zero tadpoles in . They are tadpoles of (unphysical) twisted sector fields which are allowed to propagate in a dimension zero plane (a point) in $`C^3`$. Cancellation of such tadpoles is a consistency condition which has to be imposed in orbifold theories . This means that the quiver gauge theory is consistent if and only if the complete string theory is consistent in the corresponding background.
Still, fairly general gauge theories can be obtained by this construction, consistent with gauge anomaly cancellation. The logarithmic divergences of those theories result from tadpoles for closed string fields which were dubbed ”partially twisted” in . Those are allowed tadpoles for twisted sector fields which propagate in a (real) dimension 2 plane in $`C^3`$. We call those twisted sector fields dimension 2 fields in what follows.
To describe the relation more precisely, denote the beta function coefficients of each non-abelian gauge group factor by $`\beta _{r,s}`$. These coefficients are given by:
$$\beta _{rs}=3N_{r,s}\frac{1}{2}\left(N_{r1,s}+N_{r,s+1}+N_{r+1,s1}+N_{r+1,s}+N_{r,s1}+N_{r1,s+1}\right)$$
(41)
In the orbifold description, there are $`mn`$ twisted sector scalar fields, denoted by $`\chi _{k,k^{}}`$. These are fields twisted by the generator $`\alpha ^k\beta ^k^{}`$ of the orbifold group. Some of those twisted sector scalars, for example those which are twisted only by one of the factors in $`Z_m\times Z_n`$, propagate in dimension two plane in $`C^3`$. The tadpoles for those fields encode the beta function coefficients . We review this correspondence and compare it to the noncommutative case in the next section.
It is now straightforward to construct the noncommutative version of the quiver gauge theories. In all of the models described above , the spectrum includes an untwisted NS-NS two form. Therefore one can use the Seiberg-Witten construction , and obtain non-commutative quiver gauge theories with the matter content described above. We study this realization of the gauge theories below.
## 6 Closed String Modes
Having broken the $`𝒩`$=4 supersymmetry, which exists on D3-branes in flat space, we expect to discover the phenomena discussed in . In particular, the logarithmic UV divergences in SUSY gauge theories are now accompanied by logarithmic IR singularities. As reviewed above, one then discovers closed string modes when trying to reproduce the correct logarithmic singularity within a Wilsonian effective action.
In the present context, having obtained the gauge theory from a string theory, one should be able to account explicitly for the required closed string modes. A closed string mode $`\chi `$ is introduced for every case there is a logarithmic UV divergence in the commutative limit. The field $`\chi `$ couples linearly to a relevant or marginal operator in the gauge theory, and is allowed to propagate in two dimensions transverse to the brane.
In particular, for the quiver $`𝒩`$ =1 gauge theories, there are $`mn`$ independent $`\beta `$-function coefficients. As shown above, the Wilsonian effective action is then forced to have additional ”closed string“ fields $`\chi _{rs}`$, one for each gauge factor $`U(N_{rs})`$. They propagate in two dimensions transverse to the brane, and couple linearly to the operator $`Tr(F^2)`$ in each gauge factor. Note that the latter coupling is not gauge invariant, as the pure gauge theory has no local gauge invariant operators. Presumably, the linear coupling is a part of an infinite series of terms that couple $`\chi _{r,s}`$ to the gauge theory. Only the leading order term in such series contributes to the logarithmic divergence<sup>2</sup><sup>2</sup>2Similarly, the divergence is not sensitive to the difference between regular product and \*-product, when multiplying $`\chi `$ with the corresponding operator..
In order to gain intuition about the fields $`\chi _{r,s}`$ and their couplings we consider again the commutative quiver theories. Consider calculating the $`\beta `$-functions in open string theory. This can be extracted from the two point function of the gauge fields, $`A^{(1)}A^{(2)}`$, on the annulus. Indeed, in the limit when the annulus degenerates to a loop of open string modes, the annulus reduces to the standard gauge field self-energy diagram. The existence of a non-vanishing beta function manifests itself in a logarithmic divergence in the integration over the Schwinger parameter, which is the modulus of the annulus. The relevant part of the diagram is then:
$$A=v_4\frac{dl}{l}\underset{r,s}{}\beta _{r,s}Tr(F_{rs}^2)$$
(42)
where $`v_4`$ is the volume of the non-compact directions in string units, and $`F_{r,s}`$ is the field strength of the gauge group $`U(N_{r,s})`$. The Schwinger parameter represented by the modulus of the annulus is denoted by $`l`$
Now, one can evaluate the annulus diagram in the closed string channel, where it reduces to an exchange of closed string modes. The only contribution to a logarithmic modular divergence was shown to arise from the dimension 2 twisted sector fields. Concentrating on the contribution of those fields to the annulus diagram reproduces the logarithmically divergent part of the self-energy diagram, equation (42). On the other hand it can be written as:
$$A=\underset{k,k^{}}{}Tr(\gamma _{(k,k^{})})Tr(\gamma _{(k,k^{})}\lambda _a^{rs}\lambda _b^{r^{}s^{}})F_{rs}^aF_{r^{}s^{}}^b\frac{dt}{t}$$
(43)
Here $`\lambda _a^{r,s},\lambda _b^{r^{},s^{}}`$ are the Chan-Paton matrices of the two gauge fields. The closed string modulus is $`t=\frac{1}{2l}`$. The sum is constrained to include only dimension 2, or partially twisted, sector fields.
The amplitude therefore factorizes:
$$A=A^{(1)}A^{(2)}\chi _{(k,k^{})}\frac{dt}{t}\chi _{(k,k\mathrm{`})}$$
(44)
The logarithmic divergence $`\frac{dt}{t}`$ comes from a massless closed string field propagating in two transverse dimensions. We note that this is the correct factorization of the diagram where both open string vertex operators are on the same boundary component of the annulus. The diagram which has the vertex operators on different boundaries factorizes differently, but does not contributes to the self-energy of the non-abelian gauge bosons.
The argument above gives the following relations between the linear couplings of the fields $`\chi _{(k,k^{})}`$ and the $`\beta `$-function coefficients:
$$\underset{k,k^{}}{}Tr(\gamma _{(k,k^{})})Tr(\gamma _{(k,k^{})}\lambda _a^{rs}\lambda _b^{r^{}s^{}})=\beta _{r,s}\delta ^{ab}$$
(45)
where the sum is again over the partially twisted sectors only. The first factor on the left hand side is the tadpole of $`\chi _{(k,k^{})}`$ and the second factor is the coupling of $`\chi _{(k,k^{})}`$ to the operator $`Tr(F_{rs}^aF_{r^{}s^{}}^b)`$ in the gauge theory.
We see that the non-abelian $`\beta `$-functions translate in the closed string channel to the existence of linear couplings between closed string fields $`\chi _{(k,k^{})}`$ and the operators $`Tr(F^2)`$ and $`\mathcal{1}`$ of the gauge theory. This is very similar to the couplings needed in the non-commutative case . In the commutative case the contribution of those closed string modes to open string scattering vanishes in the decoupling limit, as explained below.
We now turn to the non-commuatative case and discuss the effects of the fields $`\chi _{(k,k^{})}`$. The Wilsonian effective action of the quiver gauge theory is forced to have some fields $`\chi `$ which couple to operators in the gauge theory. We see that in the orbifold models there are natural candidates for the fields $`\chi `$. As they are predicted to propagate only in two extra dimensions they must be the partially twisted sector fields discussed above. Furthermore, we saw that those closed string fields have linear couplings to operators in the gauge theory.
To compare explicitely to the prescription of , we consider the couplings of the fields $`\chi _{(k,k^{})}`$. First, their kinetic terms live in 6 dimensions, and couple to the closed string metric. Therefore their bulk action is:
$$I_{bulk}=d^6x\left[_\mu \chi _\nu \chi g^{\mu \nu }+_a\chi _b\chi g^{ab}\right]$$
(46)
where $`\mu ,\nu `$ are the commutative directions, on and off the brane, and $`a,b`$ are the non-commuting coordinates. The closed string metric in the non commuting directions is given in the Seiberg-Witten limit as:
$$G_{ab}=\frac{\alpha _{}^{}{}_{}{}^{2}}{(\mathrm{\Theta }^2)^{ab}}$$
(47)
There are also linear couplings to the operators $`Tr(F_{rs}^2)`$ and $`1`$ in the gauge theory, as discussed above<sup>3</sup><sup>3</sup>3The linear couplings can be calculated from disc diagrams in string theory. The presence of a $`B`$-field merely changes the commmutative gauge fields to non-commutative ones.
In we are instructed to include the closed string modes $`\chi `$ up to a certain momentum scale, $`\frac{1}{\alpha ^{}\mathrm{\Lambda }}`$, in the transverse directions. This scale is clear if we consider open-closed channel duality. The effect of integrating out open strings of momenta higher than $`\mathrm{\Lambda }`$ can be summarized by inclusion of closed strings up to momenta $`\frac{1}{\alpha ^{}\mathrm{\Lambda }}`$. We comment on this cutoff further below.
Finally, the relation (45) is exactly the correct relation between the linear couplings of the fields $`\chi _{(k,k^{})}`$ and the IR divergences. Such relation guarentees that closed string mode exchange diagrams indeed reproduce all the IR divergences, to the leading order in the ’tHooft large $`N_c`$ expansion.
The crucial difference between the commmutative and noncommutative cases lies in the relation between the open and closed string metrics, giving a different scaling of the closed string metric with $`\alpha ^{}`$. The Seiberg-Witten decoupling limit sends $`\alpha ^{}`$ to zero, while keeping open string quantities fixed. This includes the UV cut-off scale $`\mathrm{\Lambda }`$.
In the commutative case the effect of the fields $`\chi `$ vanishes in the decoupling limit $`\alpha ^{}0`$, up to counterterms which renormalize the Wilsonian effective action. This fits with the interpretation of these effects as summarizing high momentum gauge theory effects.
In the noncommutative case, as shown in , the effects of the closed string modes are non-vanishing in the decoupling limit. Integrating out the gauge theory high momentum modes does not decouple them in the usual way. Rather, they have some IR effects which are nicely summarized by inclusion of some light closed string modes.
## 7 Conclusions
To summarize, we find that closed string modes, conjectured by to be a necessary ingredient of the Wilsonian effective action, do indeed exist in a concrete example. In the example studied, working with a finite cutoff couples the open string back to the closed string in a controllable manner. Using the closed string fields we were able to reproduce the infrared behaviour in the leading order in the large $`N_c`$ limit.
The discrepancy has to do with the overall $`U(1)`$ contribution in the loop, and is independent of the details of the gauge group or the matter content. This suggests an additional ”singleton“ $`\chi `$ field to account for this divergence. It would be interesting to discover that closed string directly.
Another puzzling aspect of the analysis is the origin of the cutoff on the closed string momenta. The effect of integrating out open strings of momenta higher than $`\mathrm{\Lambda }`$ is summarized by inclusion of the massless closed strings, which have momenta up to $`\frac{1}{\alpha ^{}\mathrm{\Lambda }}`$. However, this momentum scale is higher than the string scale, and therefore it is not clear why massive closed string modes should not be included<sup>4</sup><sup>4</sup>4We thank M. Berkooz for conversations on the subject.. This might have to do with supersymmetry, as suggested in . We note that any such contribution is surprising, and it is conceivable that only a small subset of closed string mode can produce the correct behaviour. For example, massless closed string states that propagate in more than two dimensions do not contribute to the infrared divergences. We do not, however, have a clear understanding why massive twisted sector mode apparently make no contribution.
Finally, in the context of orbifolds, the quadratic divergences present a puzzle. The results in the supersymmetric cases discussed above strongly suggest that quadratic divergences can be accounted for by dimension zero tadpoles: tadpoles for twisted sector scalars that are forced to a point in the transverse space. However, consistency condition for string on orbifolds require vanishing of all such tadpoles . These conditions stem from an inconsistent equation of motion for such fields, and naively have nothing to do with supersymmetry.
## 8 Acknowledgements
We thank O. Aharony, M. Berkooz, M. Douglas, H. Liu, and G. Moore for useful conversations.
The research was supported in part by DOE grant DE-FG02-96ER40959. |
warning/0003/hep-ph0003256.html | ar5iv | text | # Effect of flavor mixing on the time delay of massive supernova neutrinos
## Abstract
The neutrinos and antineutrinos of all the three flavors released from a galactic supernova will be detected in the water Cerenkov detectors. We show that even though the neutral current interaction is flavor blind, and hence neutrino flavor mixing cannot alter the total neutral current signal in the detector, it can have a non-trivial impact on the delay of massive neutrinos and alters the neutral current event rate as a function of time. We have suggested various variables of the neutral and charged current events that can be used to study this effect. In particular the ratio of charged to neutral current events can be used at early times while the ratio of the energy moments for the charged to the neutral current events can form useful diagnostic tools even at late times to study neutrino mass and mixing.
PACS: 14.60.Pq, 97.60.Bw, 95.55.Vj
Keywords: massive neutrinos, supernovae, neutrino mixing
The detection of the SN1987A neutrinos by the water Cerenkov detectors at Kamioka and IMB settled many important issues in the subject of type II supernova theory. The observation of neutrinos from any future galactic supernova event will answer the remaining questions regarding the understanding of the supernova mechanisms. A galactic supernova event will also bring in a lot of information on neutrino mass, which of late, has been an issue of much discussion. Among the various problems which demand non-zero neutrino mass are the atmospheric neutrinos anomaly , the solar neutrino problem and the LSND experiment in Los Alamos . While all the three above mentioned experiments give information on the mass squared differences, the supernova neutrinos can be used to place direct limits on the $`\nu _\mu /\nu _\tau `$ masses and at the same time can also constrain the neutrino mixing parameters which will be useful in the understanding of the above three experiments.
About $`10^{58}`$ neutrinos, in all three flavors carrying a few times $`10^{53}`$ ergs of energy are released in a type II supernova. These neutrinos for a galactic supernova events can be detected by the current water Cerenkov detectors, the Super-Kamiokande (SK) and the Sudbury Neutrino Observatory (SNO). The effect of neutrino mass can show up in the observed neutrino signal in these detectors in two ways,
* by causing delay in the time of flight measurements
* by modifying the neutrino spectra through neutrino flavor mixing
Massive neutrinos travel with speed less than the speed of light and for typical galactic supernova distances $``$ 10 kpc, even a small mass results in a measurable delay in the arrival time of the neutrino. Many different analyses have been performed before to give bounds on the neutrino mass ( and references therein). Neutrino oscillations on the other hand convert the more energetic $`\nu _\mu /\nu _\tau (\overline{\nu }_\mu /\overline{\nu }_\tau )`$ into $`\nu _e(\overline{\nu }_e)`$ thereby hardening the resultant $`\nu _e(\overline{\nu }_e)`$ energy spectra and hence enhancing their signal at the detector . In a previous work we studied quantitatively the effects of neutrino flavor oscillations on the supernova neutrino spectrum and the number of charged current events at the detector using a realistic supernova model. In this work we study the neutral current signal as a function of time in the water Cerenkov detectors, for a mass range of the neutrinos where both the phenomenon of delay and flavor conversion are operative. That the time response of the event rate in the detector is modified if the neutrinos have mass alone and hence delay is a well known feature . In this letter we stress the point that since neutrino flavor conversions change the energy spectra of the neutrinos, and since the time delay of the massive neutrinos is energy dependent, the time dependence of the event rate at the detector is altered appreciably in the presence of mixing. We suggest various variables which act as tools for measuring this change in the time response curve of the neutral current events and in differentiating the cases of (a) massless neutrinos (b) neutrinos with mass but no mixing and (c) neutrinos with mass as well as mixing. In particular we study the ratio of the charged current to neutral current ratio R(t), as a function of time in the SNO detector and show that the change in the value and the shape of R(t) due to flavor mixing cannot be emulated by uncertainties. We also study other variables like the normalized $`n`$-th energy moments of the neutral current events and the ratio of charged to the neutral current $`n`$-th moments as important diagnostic tools in filtering out the effects of neutrino mass and mixing.
The differential number of neutrino events at the detector for a given reaction process is
$$\frac{d^2S}{dEdt}=\underset{i}{}\frac{n}{4\pi D^2}N_{\nu _i}(t)f_{\nu _i}(E)\sigma (E)ϵ(E)$$
(1)
where $`i`$ runs over the neutrino species concerned, $`N_{\nu _i}(t)=L_{\nu _i}(t)/E_{\nu _i}(t)`$, are the number of neutrinos produced at the source where $`L_{\nu _i}(t)`$ is the neutrino luminosity and $`E_{\nu _i}(t)`$ is the average energy, $`\sigma (E)`$ is the reaction cross-section for the neutrino with the target particle, $`D`$ is the distance of the neutrino source from the detector (taken as 10kpc), $`n`$ is the number of detector particles for the reaction considered and $`f_{\nu _i}(E)`$ is the energy spectrum for the neutrino species involved, while $`ϵ(E)`$ is the detector efficiency as a function of the neutrino energy. By integrating out the energy from eq.(1) we get the time dependence of the various reactions at the detector. To get the total numbers both integrations over energy and time has to be done.
For the neutrino luminosities and average energies, though it is best to use a numerical supernova model, but for simplicity, we will here use a profile of the neutrino luminosities and temperatures which have general agreement with most supernova models. We take the total supernova energy radiated in neutrinos to be 3 $`\times 10^{53}`$ ergs. This luminosity, which is almost the same for all the neutrino species, has a fast rise over a period of 0.1 sec followed by a slow fall over several seconds in most supernova models. We use a luminosity that has a rise of 0.1 sec using one side of the Gaussian with $`\sigma `$ = 0.03 and then an exponential decay with time constant $`\tau `$ = 3 sec for all the flavors .
The average energies associated with the $`\nu _e,\overline{\nu }_e\mathrm{and}\nu _\mu `$ (the $`\nu _\mu ,\overline{\nu }_\mu ,\nu _\tau \mathrm{and}\overline{\nu }_\tau `$ have the same energy spectra) are 11 MeV, 16 MeV and 25 MeV respectively in most numerical models. We take these average energies and consider them to be constant in time. We have also checked our calculations with time dependent average energies and estimated its effect. The neutrino spectrum is taken to be a pure Fermi-Dirac distribution characterized by the neutrino temperature alone.
We will here be concerned with two important water Cerenkov detectors, the SK and the SNO. The column 1 of Table 1 lists all the important reactions in SK and SNO. In column 2 of Table 1 we report the calculated number of expected events for the various reactions in SNO, when neutrinos are assumed to be massless. The corresponding values for SK can be obtained by scaling the number of events in $`\mathrm{H}_2\mathrm{O}`$ to its fiducial volume of 32 kton. The detector efficiency is taken to be 1 and the energy threshold is taken to be 5 MeV for both SK and SNO . For the cross-section of the $`(\nu _ed),(\overline{\nu }_ed),(\nu _id)`$ and $`(\overline{\nu }_ep)`$ reactions we refer to . The cross-section of the $`(\nu _e(\overline{\nu }_e)e^{})`$ and $`(\nu _ie^{})`$ scattering has been taken from while the neutral current $`(\nu _i^{16}O)`$ scattering cross-section is taken from . For the $`{}_{}{}^{16}O(\nu _ee^{})^{16}F`$ and $`{}_{}{}^{16}(\overline{\nu }_e,e^+)_{}^{16}N`$ reactions we refer to and use it’s cross-sections for the detector with perfect efficiency. The expected number of events that we get agree quite well with the one reported in , where the results of a numerical supernova model was used.
If the neutrinos are massless then the time response of their signal at the detector reflect just the time dependence of their luminosity function at the source, which is the same for all the three flavors and hence the same for the charged current and neutral current reactions. If neutrinos have mass $`eV`$ then they pick up a measurable delay during their course of flight from the supernova to the earth. For a neutrino of mass m (in eV) and energy E (in MeV), the delay (in sec) in traveling a distance D (in 10 kpc) is
$$\mathrm{\Delta }t(E)=0.515(m/E)^2D$$
(2)
where we have neglected all the small higher order terms. The time response curve then has contributions from both the luminosity and the mass. We will now consider a scheme of neutrino masses such that $`\mathrm{\Delta }m_{12}^210^6eV^2`$ consistent with the solar neutrino problem and $`\mathrm{\Delta }m_{13}^2\mathrm{\Delta }m_{23}^2110^4eV^2`$. The neutrino mass model considered here is one of several, given for the purpose of illustration only. In this scheme the atmospheric neutrino anomaly will have to be explained by the $`\nu _\mu \nu _{sterile}`$ oscillation mode . The mass range for the neutrinos as the hot component of hot plus cold dark matter scenario in cosmology is a few $`eV`$ only , which will conflict with the higher values in the range of $`m_{\nu _3}=1100eV`$ that we consider here if $`\nu _3`$ is stable. Hence, we assume that the $`\nu _3`$ state is unstable but with a large enough life time so that it is does not conflict with the observations of SN 1987A (even though SN1987A observations did not correspond to any $`\nu _\tau `$ event, one can put limits on the $`\nu _3/\overline{\nu _3}`$ lifetime as the $`\nu _e/\overline{\nu }_e`$ state is a mixture of all the three mass eigenstates) and is also consistent with Big Bang Nucleosynthesis. In fact, from the ref. we know that using the time delay technique, the SK and SNO can be used to probe neutrino masses down to $`50eV`$ and $`30eV`$ respectively. Hence we have presented all our results for a particular representative value of $`m_{\nu _3}=40eV`$. There have been proposals in the past for an unstable neutrino with mass $`30eV`$ and lifetime $`10^{23}sec`$ . Since direct kinematical measurements give $`m_{\nu _e}<5eV`$ , we have taken the $`\nu _e`$ to be massless and the charged current events experience no change. But since the $`\nu _\tau (\overline{\nu }_\tau )`$ pick up a detectable time delay (for the mass spectrum of the neutrinos that we consider here, the $`\nu _\mu (\overline{\nu }_\mu )`$ do not have measurable time delay), the expression for the neutral current events gets modified to,
$`{\displaystyle \frac{dS_{nc}^d}{dt}}`$ $`=`$ $`{\displaystyle \frac{n}{4\pi D^2}}{\displaystyle }dE\sigma (E)\{N_{\nu _e}(t)f_{\nu _e}(E)+N_{\overline{\nu }_e}(t)f_{\overline{\nu }_e}(E)+N_{\nu _\mu }(t)f_{\nu _\mu }(E)+`$ (3)
$`+`$ $`N_{\overline{\nu }_\mu }(t)f_{\overline{\nu }_\mu }(E)+N_{\nu _\tau }(t\mathrm{\Delta }t(E))f_{\nu _\tau }(E)+N_{\overline{\nu }_\tau }(t\mathrm{\Delta }t(E))f_{\overline{\nu }_\tau }(E)\}`$ (4)
where $`dS_{nc}^d/dt`$ denotes the neutral current $`(nc)`$ event rate with delay $`(d)`$. Delay therefore distorts the neutral current event rate vs. time curve. By doing a $`\chi ^2`$ analysis of this shape distortion one can put limits on the $`\nu _\tau `$ mass .
We next consider the neutrinos to have flavor mixing as well. The mixing angle $`\mathrm{sin}^2\theta _{12}`$ can be constrained from the solar neutrino data ($`\mathrm{sin}^2\theta _{12}10^3`$) while for $`\mathrm{sin}^2\theta _{13}`$ there is no experimental data to fall back upon, but from r-process considerations in the “hot bubble” of the supernova, one can restrict $`\mathrm{sin}^2\theta _{13}10^6`$ . In this scenario there will be first a matter enhanced $`\nu _e\nu _\tau `$ resonance in the mantle of the supernova followed by a $`\nu _e\nu _\mu `$ resonance in the envelope. The MSW mechanism in the supernova for the neutrino mass scheme that we consider here is discussed in details in ref. . As the average energy of the $`\nu _\mu /\nu _\tau `$ is greater than the average energy of the $`\nu _e`$, neutrino flavor mixing modifies their energy spectrum. Hence as pointed out in , the $`\nu _e`$ flux though depleted in number, gets enriched in high energy neutrinos and since the detection cross-sections are strongly energy dependent, this results in the enhancement of the charged current signal. The total number of events in SNO, integrated over time in this scenario with complete flavor conversion ($`P_{\nu _e\nu _e}=0`$) are given in column 2 of Table 1. Of course since the $`\overline{\nu }_e`$ do not have any conversion here, the $`\overline{\nu }_e`$ signal remains unaltered. Also as the neutral current reactions are flavor blind, the total neutral current signal remains unchanged. But whether the time response curve of the neutral current signal remains unchanged in presence of mixing, in addition to delay, is an interesting question.
If the neutrinos have mass as well as mixing, then the neutrinos are produced in their flavor eigenstate, but they travel in their mass eigenstate. The neutrino mass eigenstates will travel with different speeds depending on their mass and will arrive at the detector at different times. For the scenario that we are considering only $`\nu _3`$ and $`\overline{\nu }_3`$ will be delayed. Hence to take this delay in arrival time into account, the eq.(4) has to be rewritten in terms of the mass eigenstates. It can be shown that expression for the neutral current event rate in terms of the mass eigenstates is,
$`{\displaystyle \frac{dS_{nc}^{do}}{dt}}`$ $`=`$ $`{\displaystyle \frac{n}{4\pi D^2}}{\displaystyle }dE\sigma (E)\{N_{\nu _1}(t)f_{\nu _1}(E)+N_{\overline{\nu }_1}(t)f_{\overline{\nu }_1}(E)+N_{\nu _2}(t)f_{\nu _2}(E)`$ (5)
$`+`$ $`N_{\overline{\nu }_2}(t)f_{\overline{\nu }_2}(E)+N_{\nu _3}(t\mathrm{\Delta }t(E))f_{\nu _3}(E)+N_{\overline{\nu }_3}(t\mathrm{\Delta }t(E))f_{\overline{\nu }_3}(E)\}`$ (6)
where $`N_{\nu _i}`$ is the $`\nu _i`$ flux at the source. If the neutrinos are produced at densities much higher than their resonance densities, all the mixings in matter are highly suppressed, and the neutrinos are produced almost entirely in their mass eigenstates. For the three generation case that we are considering, $`\nu _e\nu _3`$, $`\nu _\mu \nu _1`$ and $`\nu _\tau \nu _2`$. For the antineutrinos on the other hand, at the point of production in the supernova $`\overline{\nu }_e\overline{\nu }_1`$, $`\overline{\nu }_\mu \overline{\nu }_2`$ and $`\overline{\nu }_\tau \overline{\nu }_3`$. Hence the above expression for the neutral current event rate in the presence of delay and mixing can be written as,
$`{\displaystyle \frac{dS_{nc}^{do}}{dt}}`$ $`=`$ $`{\displaystyle \frac{n}{4\pi D^2}}{\displaystyle }dE\sigma (E)\{N_{\nu _\mu }(t)f_{\nu _\mu }(E)+N_{\overline{\nu }_e}(t)f_{\overline{\nu }_e}(E)+N_{\nu _\tau }(t)f_{\nu _\tau }(E)+N_{\nu _\tau }(t)f_{\nu _\tau }(E)`$ (7)
$`+`$ $`N_{\overline{\nu }_\mu }(t)f_{\overline{\nu }_\mu }(E)+N_{\overline{\nu }_\mu }(t)f_{\overline{\nu }_\mu }(E)+N_{\nu _e}(t\mathrm{\Delta }t(E))f_{\nu _e}(E)+N_{\overline{\nu }_\tau }(t\mathrm{\Delta }t(E))f_{\overline{\nu }_\tau }(E)\}`$ (8)
Note that the above expression does not depend on the neutrino conversion probability as the neutral current interaction is flavor blind.
In fig. 1 we have plotted the neutral current event rate for the reaction ($`\nu _i+dn+p+\nu _i`$, where $`\nu _i`$ stands for all the 6 neutrino species) as a function of time for massless neutrinos along with the cases for mass but no mixing (eq.(4)) and mass along with mixing (eq.(8)). The figure looks similar for the other neutral current reactions as well, apart from a constant normalization factor depending on the total number of events for the process concerned. The curves corresponding to the massive neutrinos have been given for $`m_{\nu _\tau }=40eV`$. As expected, the shape of the neutral current event rate changes due to the delay of massive $`\nu _\tau `$. Since the delay given by eq.(2) depends quadratically on the neutrino mass, the distortion is more for larger masses . But the noteworthy point is that the presence of mixing further distorts the rate vs. time curve. The reason for this distortion can be traced to the fact that the time delay $`1/E^2`$. As the energy spectrum of the neutrinos change due to flavor mixing, the resultant delay is also modified and this in turn alters the neutral current event rate as a function of time. In fact the flavor conversion in the supernova results in de-energising the $`\nu _\mu /\nu _\tau `$ spectrum and hence the delay given by eq.(2) should increase. As larger delay caused by larger mass results in further lowering of the neutral current event rate vs. time curve for early times, one would normally expect that the enhanced delay as a result of neutrino flavor conversion would have a similar effect. But the fig. 1 shows that during the first second, the curve corresponding to delay with mixing is higher than the one with only time delay. This at first sight seems unexpected. But then one realizes that while the flavor conversion reduces the average energy of the massive $`\nu _\tau `$ increasing its delay and hence depleting its signal at early times, it energizes the massless and hence undelayed $`\nu _e`$ beam, which is detected with full strength. Therefore, while for no mixing the $`\nu _\tau `$ gave the larger fraction of the signal, for the case with mixing it is the $`\nu _e`$ that assume the more dominant role, and so even though the $`\nu _\tau `$ arrive more delayed compared to the case without mixing, the delay effect is diluted due to the enhancement of the $`\nu _e`$ fraction and the depletion of the $`\nu _\tau `$ fraction of the neutral current events. We have also checked that although it may seem that the curve with delay and mixing can be simulated by another curve with delay alone but with smaller mass, the actual shape of the two curves would still be different. This difference in shape though may not be statistically significant and hence one may not be able to see the effect of mixing in the time delay of the neutrinos just by looking at the time response of the neutral current event rate in the present water Cerenkov detectors. We therefore look for various other variables which can be studied to compliment this.
One such variable which carries information about both the neutrino mass and their mixing is R(t), the ratio of charged to neutral current event rate as a function of time. In fig. 2 we give the ratio R(t) of the total charged current to the neutral current event rate in $`\mathrm{D}_2\mathrm{O}`$ in SNO as a function of time. Plotted are the ratios (i) without mass, (ii) with only mixing, (iii) with delay but zero mixing and (iv) with delay and flavor mixing. The differences in the behavior of R(t) for the four different cases are clearly visible. For no mass R(t)=0.3 and since the time dependence of both the charged current and neutral current reaction rates are the same, their ratio is constant in time. As the presence of mixing enhances the charged current signal keeping the neutral current events unaltered, R(t) goes up to 0.61 for only mixing, remaining constant in time, again due to the same reason. With the introduction of delay the ratio becomes a function of time as the neutral current reaction now has an extra time dependence coming from the mass. At early times as the $`\nu _\tau `$ get delayed the neutral current event rate drops increasing R(t). These delayed $`\nu _\tau `$s arrive later and hence R(t) falls at large times. This feature can be seen for both the curves with and without mixing. The curve for only delay starts at R(t)=0.52 at t=0 sec and falls to about R(t)=0.26 at t=10 sec. For the delay with mixing case the corresponding values of R(t) are 0.83 and 0.51 at t=0 and 10 sec respectively. The important point is that the curves with and without mixing are clearly distinguishable and should allow one to differentiate between the two cases of only delay and delay with neutrino flavor conversion.
In order to substantiate our claim that the two scenarios of only delay and delay with mixing are distinguishable in SNO, we divide the time into bins of size 1 second. The number of events in each bin is then used to estimate the $`\pm 1\sigma `$ statistical error in the ratio R(t) in each bin and these are then plotted in fig. 2 for the typical time bin numbers 1, 4 and 7. From the figure we see that the two cases of delay, with and without mixing, are certainly statistically distinguishable in SNO for the first 6 seconds.
We next focus our attention on $`M_n^{nc}(t)`$, the neutral current $`n`$-th moments of the neutrino energy distributions observed at the detector, defined as
$$M_n^{nc}(t)=\frac{d^2S}{dEdt}E^n𝑑E$$
(9)
while the corresponding normalized moments are given by
$$\overline{M}_n^{nc}(t)=\frac{M_n^{nc}(t)}{M_0^{nc}(t)}$$
(10)
We have shown the behavior of the 1st normalized moment $`\overline{M}_1^{nc}(t)`$ in fig. 3 as a function of time in SNO. For massless neutrinos, the $`\overline{M}_1^{nc}`$ has a value 40.97, constant in time, as this is again a ratio and hence the time dependence gets canceled out as in the case of R(t). For the case where the $`\nu _\tau `$ is massive and hence delayed, it assumes a time dependence. Since the delay $`1/E^2`$ and since the neutrinos are produced at the source with an energy distribution, hence at each instant the lower energy $`\nu _\tau `$ will be delayed more than the higher energy $`\nu _\tau `$. Therefore $`\overline{M}_1^{nc}(t)`$, which gives the energy centroid of the neutral current event distribution in $`\mathrm{D}_2\mathrm{O}`$, starts from a low value 38.76 at t=0 sec as all the $`\nu _\tau `$ are delayed, rises sharply as the higher energy neutrinos arrive first and then falls slowly as the lower energy delayed $`\nu _\tau `$ start arriving. If the $`\nu _\tau `$ are allowed to mix with the $`\nu _e`$, then they are de-energized and the above mentioned effect is further enhanced. To make an estimate of whether SNO would be able to distinguish the three cases discussed above, we compute the $`\pm 1\sigma `$ statistical errors in the $`1^{st}`$ normalized moment for the two scenarios of delay, with and without mixing, and show them for the $`1^{st}`$, $`6^{th}`$ and $`11^{th}`$ bins. We see that the errors involved are large enough to completely wash out the differences between the energy moments with and without neutrino mass and mixing. Hence the normalized energy moments fail to probe neutrino mass and mixing as at early times we don’t see much difference between the different cases considered, while at late times the number of events become very small so that the error in $`M_0^{nc}(t)`$ becomes huge, increasing the error in $`\overline{M}_1^{nc}(t)`$.
The variable that can be a useful probe for differentiating the case for delay with mixing from the case for delay without mixing is the ratio of the unnormalized moment of the charged to neutral current events
$$r_n(t)=\frac{M_n^{cc}(t)}{M_n^{nc}(t)}$$
(11)
We present in fig. 4, for SNO, the $`r_n(t)`$ vs. time plots (for n=1) for the cases of (a) massless neutrinos (b) with mixing but no delay (c) with delay but no mixing and (d) with delay as well as mixing. Since this is a ratio, the supernova flux uncertainties get canceled out to a large extent and since the unnormalized moments have smaller statistical errors, this is a better variable than the normalized moments to observe the signatures of neutrino mixing. In the figure we have shown the $`\pm 1\sigma `$ statistical errors in $`r_1(t)`$ for the two cases of delay alone and delay with mixing, for the $`1^{st}`$, $`8^{th}`$ and $`15^{th}`$ bins in time, and the two cases are clearly distinguishable in SNO for early as well as late times. Note that $`r_1(t)`$ is different from the ratio R(t) as it gives information about the ratio of the energy centroids of the charged current and neutral current distributions as a function of time, while the latter gives only the ratio of the number of events as a function of time.
The advantage of using ratios is that, they are not only sensitive to the mass and mixing parameters but are also almost insensitive to the details of supernova models. Since they are a ratio they are almost independent of the luminosity and depend only on some function of the ratio of neutrino temperatures. All the calculations presented so far are for fixed neutrino temperatures. In order to show that the time dependence of the neutrino temperatures does not alter our conclusions much, we present our analysis with time dependent neutrino temperatures. We take
$$T_{\nu _e}=0.16\mathrm{log}t+3.58,T_{\overline{\nu }_e}=1.63\mathrm{log}t+5.15,T_{\nu _\mu }=2.24\mathrm{log}t+6.93$$
(12)
These forms for the neutrino temperatures follow from fits to the results of the numerical supernova model given in Totani et al. . In fig. 5 we compare the ratio R(t) for the cases of delay and delay with mixing for the two cases of fixed temperatures and the time dependent temperatures. It is clear from the figure that that the time dependence of the neutrino temperatures does not have much effect on the time dependence of the ratio of the charged current to neutral current rates. In fact the two curves corresponding to fixed and time dependent temperatures, fall within $`\pm 1\sigma `$ statistical errorbars for both the cases of only delay and delay with mixing.
In conclusion, we have shown that even though neutrino flavor mixing cannot alter the total neutral current signal in the detector - the neutral current interaction being flavor blind, it can have a non-trivial impact on the delay of massive neutrinos, which alters the neutral current event rate as a function of time. The neutral current event rate though does not depend on the neutrino conversion probability. In order to study the effect of neutrino mass and mixing we have suggested various variables. Of the different variables that we have presented here, the ratio of the charged to neutral current event rate R(t), can show the effect of mixing during the first few seconds, while the charged to neutral current ratio of the energy moments are useful diagnostic tools for all times. These variables are not just sensitive to flavor mixing and time delay, they are also insensitive to supernova model uncertainties and hence are excellent tools to study the effect of flavor mixing on the time delay of massive supernova neutrinos.
In this letter we have considered a mass spectrum for the neutrinos where only the $`\nu _\tau `$ have a measurable delay. The model considered is one of many, but one can easily extend the above formalism to include more general classes of neutrino models . In addition to the energy moments that we have presented here, the $`n`$-th order moments of the arrival time of the neutrinos as a function of energy can also be analyzed to study the effect of neutrino mass and mixing, and we plan to present them in a future work .
The authors thank S. Goswami, P.B. Pal, S. Mohanty and D. Majumdar for discussions.
Table 1
The expected number of neutrino events in SNO. To get the number of events in SK, one has to scale the number of events in $`\mathrm{H}_2\mathrm{O}`$ given here to its fiducial mass of 32 kton. The column A corresponds to massless neutrinos, column B to neutrinos with complete flavor conversion The $`\nu _i`$ here refers to all the six neutrino species.
reactions in 1 kton $`\mathrm{D}_2\mathrm{O}`$ A B $`\nu _e+dp+p+e^{}`$ 75 239 $`\overline{\nu }_e+dn+n+e^+`$ 91 91 $`\nu _i+dn+p+\nu _i`$ 544 544 $`\nu _e+e^{}\nu _e+e^{}`$ 4 6 $`\overline{\nu }_e+e^{}\overline{\nu }_e+e^{}`$ 1 1 $`\nu _{\mu ,\tau }(\overline{\nu }_{\mu ,\tau })+e^{}\nu _{\mu ,\tau }(\overline{\nu }_{\mu ,\tau })+e^{}`$ 4 3 $`\nu _e+^{16}Oe^{}+^{16}F`$ 1 55 $`\overline{\nu }_e+^{16}Oe^++^{16}N`$ 4 4 $`\nu _i+^{16}O\nu _i+\gamma +X`$ 21 21 reactions in 1.4 kton $`\mathrm{H}_2\mathrm{O}`$ $`\overline{\nu }_e+pn+e^+`$ 357 357 $`\nu _e+e^{}\nu _e+e^{}`$ 6 9 $`\overline{\nu }_e+e^{}\overline{\nu }_e+e^{}`$ 2 2 $`\nu _{\mu ,\tau }(\overline{\nu }_{\mu ,\tau })+e^{}\nu _{\mu ,\tau }(\overline{\nu }_{\mu ,\tau })+e^{}`$ 6 5 $`\nu _e+^{16}Oe^{}+^{16}F`$ 2 86 $`\overline{\nu }_e+^{16}Oe^++^{16}N`$ 6 6 $`\nu _i+^{16}O\nu _i+\gamma +X`$ 33 33
Figure Captions
Fig.1 The neutral current event rate as a function of time in $`\mathrm{D}_2\mathrm{O}`$ in SNO. The solid line corresponds to the case of massless neutrinos, the long dashed line to neutrinos with only mass but no mixing, while the short dashed line gives the event rate for neutrinos with mass as well as flavor mixing.
Fig.2 The ratio R(t) of the total charged current to neutral current event rate in SNO versus time. The solid line is for massless neutrinos, the short dashed line for neutrinos with complete flavor conversion but no delay, the long dashed line for neutrinos with only delay and no flavor conversion and the dotted line is for neutrinos with both delay and complete flavor conversion. Also shown are the $`\pm 1\sigma `$ statistical errors for delay with and without mixing in the $`1^{st}`$, $`4^{th}`$ and the $`7^{th}`$ time bins.
Fig. 3 The 1st normalized energy moment of the neutral current events in SNO $`\overline{M}_1^{nc}(t)`$ versus time. The solid line corresponds to the case of massless neutrinos, the long dashed line to neutrinos with only mass but no mixing, while the short dashed line gives the event rate for neutrinos with mass as well as flavor conversion. Also shown are the $`\pm 1\sigma `$ statistical errors for delay with and without mixing in the $`1^{st}`$, $`6^{th}`$ and the $`11^{th}`$ time bins.
Fig. 4 The variation of $`r_1(t)`$ with time in SNO. The solid line is for massless neutrinos, the short dashed line for neutrinos with complete flavor conversion but no delay, the long dashed line for neutrinos with only delay and no flavor conversion and the dotted line is for neutrinos with both delay and complete flavor conversion. Also shown are the $`\pm 1\sigma `$ statistical errors for delay with and without mixing in the $`1^{st}`$, $`8^{th}`$ and the $`15^{th}`$ time bins.
Fig. 5 The ratio R(t) in SNO for the two cases of fixed and time dependent neutrino temperatures. The solid line and the long dashed line give R(t) for the cases of fixed temperatures and varying temperatures respectively for only delay, while the short dashed line and the dotted line give the corresponding R(t) for delay with mixing. We have also given the $`\pm 1\sigma `$ statistical errors in the $`1^{st}`$ and the $`4^{th}`$ time bin, for the both the curves for fixed and time dependent temperatures. |
warning/0003/cond-mat0003301.html | ar5iv | text | # Effect of incoherent scattering on shot noise correlations in the quantum Hall regime
## I Introduction
The effect of incoherent scattering on transport and noise in mesoscopic structures has been of interest in a number of previous works (for review see ). In this paper we are interested in the effect of inelastic and quasi-elastic scattering on the correlations of the current in a structure submitted to a strong magnetic field so that the current is carried by edge states propagating along the boundary of the sample (see figure 1).
To demonstrate the reality of edge states (despite their small contribution to the overall density of states) the possibility of creating a non-equilibrium population has been crucial. Several experiments have investigated the equilibration of edge states selectively populated with the help of various contacts by transport measurements . A natural problem is to investigate the effect of inter-edge state scattering on the noise on such structures.
Recently, Hanbury Brown and Twiss (HBT) experiments studying current correlations using partially degenerate stream of fermions were performed , starting from the idea of using the edge states in a conductor submitted to a strong magnetic field. A Y-structure is discussed in Ref. and a HBT experiment without a magnetic field has also been realized . Here we are interested in the experimental arrangement of Oberholzer et al. which is depicted in figure 1, with the only difference that the magnetic field, in this experiment, was adjusted in such a way that only one spin-degenerate edge state (filling factor $`\nu =2`$) carries the current, and not two, as it is represented on the figure. Contact 1, which is at a potential $`eV`$ above the potentials of the two other contacts, acts as a carrier source and contacts 2 and 3 as detectors for the beams of electrons. The incident beam is splitted at two quantum point contacts (QPC) which can be tuned by applied voltages. In the following we will denote by $`T_1`$ and $`R_1=1T_1`$ the transmission and reflection coefficients at QPC at contact 1, and introduce $`T_3`$ and $`R_3`$ for QPC at contact 3. The quantity of interest in this experiment, revealing statistical properties of the carriers, is the correlation $`S_{23}=\mathrm{\Delta }I_2\mathrm{\Delta }I_3`$ between the current at the two contacts. Compared to the initial experiment by Henny et al. the experiment by Oberholzer et al. introduces fluctuations in the incident beam and thus the current correlations are not determined already from current conservation alone. In the following we will be interested in the case where the distance between the QPC’s is long so that incoherent processes may occur along this edge. If only one edge state is populated, phase breaking processes are of little interest since $`S_{23}`$ measures quantum-statistical properties of the carriers revealed by the separation of the beam at QPC 3, and what happens before this separation is of little importance. We can indeed check that the introduction of incoherent processes within the framework that will be used in the following does not affect the auto-correlations (noise) and the correlations if only one edge state is present.
In contrast, when several edge states are populated, inter-edge state scattering along one boundary may cause a redistribution of the carriers between the edge states and thus modify the noise properties of each channel. This is the situation that will interest us in this paper. Inter-edge state scattering centers were investigated experimentally very recently . We will not take into account the spin degeneracy which only would multiply the conductances and the noise by a factor 2.
The inelastic and quasi-elastic scattering are modeled by introducing an additional probe at the edge along which incoherent scattering occurs (see figure 2). We have to impose that the current through this probe is zero at any time. This approach, followed in a number of works (see for a review), has the advantage to reduce the problem to the study of coherent scattering in a conductor with one additional contact. The method will be recalled in more detail in the following.
In the first section we are interested in the situation where only coherent scattering is present. In the following section we describe the influence of incoherent scattering along the long edge of the sample.
## II Coherent Scattering
We first discuss the situation when only elastic scattering is present in the system. We consider the situation depicted in figure 1 where the magnetic field is chosen so that two edge states are populated. The edge state associated with the lowest Landau level (LL) is perfectly transmitted at the two QPC’s whereas the second edge state associated with the second LL is only partially transmitted at the QPC’s. We are interested in the current correlations between contacts 2 and 3.
The noise spectrum is defined as $`S_{\alpha \beta }(\omega )2\pi \delta (\omega +\omega ^{})=\mathrm{\Delta }\widehat{I}_\alpha (\omega )\mathrm{\Delta }\widehat{I}_\beta (\omega ^{})+\mathrm{\Delta }\widehat{I}_\beta (\omega ^{})\mathrm{\Delta }\widehat{I}_\alpha (\omega )`$ with $`\mathrm{\Delta }\widehat{I}_\alpha (\omega )=\widehat{I}_\alpha (\omega )\widehat{I}_\alpha (\omega )`$, where $`\widehat{I}_\alpha (\omega )`$ is the Fourier transform of the current operator at contact $`\alpha `$. (For a recent presentation of the formalism and notations, see ). The zero frequency limit will be denoted: $`S_{\alpha \beta }=S_{\alpha \beta }(\omega =0)`$. Following the scattering approach, we construct the scattering matrix for the system to calculate the noise :
$$S_{\alpha \beta }=\frac{2e^2}{h}dE\underset{\gamma ,\lambda }{}Tr\left\{A_{\gamma \lambda }^\alpha A_{\lambda \gamma }^\beta \right\}f_\gamma (1f_\lambda ),$$
(1)
where the matrices $`A`$’s are related to the on-shell $`S`$-matrix: $`A_{\gamma \lambda }^\alpha =\delta _{\alpha \gamma }\delta _{\alpha \lambda }s_{\alpha \gamma }^{}(E)s_{\alpha \lambda }(E)`$, $`f_\alpha `$ being the Fermi-Dirac distribution at the contact $`\alpha `$. We consider the case of zero temperature and will discuss at the end the effect of temperature on some of the results.
Using this expression and the $`S`$-matrix we can compute the correlations between the currents at contacts $`2`$ and $`3`$, which is eventually found to be :
$$S_{23}=\frac{2e^2}{h}|eV|R_3T_3T_1^2,$$
(2)
where $`R_1`$, $`T_1`$ and $`R_3`$, $`T_3`$ are the reflection and transmission coefficients for the edge state corresponding to the second LL at QPC 1 and 3, respectively. Since the first edge state is totally transmitted through QPC’s, it carries a noiseless current and does not contribute to the noise or to the correlations. The $`R_3T_3`$ factor is the usual partition factor due to the separation of the electron beam at QPC 3. The $`T_1^2`$ term is due to the fact that the processes leading to correlations between contacts 2 and 3 involve two electrons transmitted through the QPC 1, which happens with probability $`T_1^2`$.
Before discussing the effect of inelastic and quasi-elastic scattering, we may consider first the effect of coherent scattering between the two edge states along the upper edge. This question might seem academic but it will be important to have this result in mind, to appreciate the difference in the correlations $`S_{23}`$ when the two edge states exchange carriers coherently or incoherently. Let us recall that in the presence of coherent scattering, it was proven that correlations between two contacts are always negative as a consequence of the fermionic nature of the carriers<sup>*</sup><sup>*</sup>* The proof applies only to conductors which are part of a zero-impedance external circuit. . To describe coherent scattering between edge states we introduce the probability $`\epsilon `$ that an electron, starting in one of the two channels, is scattered into the other edge state when it travels between the two QPC’s. Here we do not need to enter into more details about the scattering however let us mention that it has been studied within a microscopic approach in . After having constructed the $`S`$-matrix, formula (1) gives the correlations:
$`S_{23}^{\mathrm{coh}}={\displaystyle \frac{2e^2}{h}}|eV|R_3[(1\epsilon )^2T_3T_1^2+\epsilon ^2T_3`$ (3)
$`+\epsilon (1\epsilon )(12R_3T_1+T_1^2)].`$ (4)
Let us now discuss a few limiting cases. If $`\epsilon =0`$ (no inter-edge state elastic scattering) we fall back to the previous situation (2). If $`\epsilon =1`$, the two edge states are perfectly exchanged between the QPC’s and the current separated by QPC 3 issues from the first edge state, perfectly transmitted at QPC 1 and we have simply $`S_{23}^{\mathrm{coh}}=\frac{2e^2}{h}|eV|R_3T_3`$. Note that the outer edge state in contact 3 is not noiseless but these fluctuations are not correlated with fluctuations at contact 2 and then do not contribute to $`S_{23}`$ since this edge state is not splitted at QPC 3.
If $`T_1=0`$ the correlation is $`S_{23}^{\mathrm{coh}}=\frac{2e^2}{h}|eV|[R_3T_3\epsilon ^2+\epsilon (1\epsilon )R_3]`$. The first term is the partition noise for the beam separated at QPC 3; this term is proportional to the probability $`\epsilon ^2`$ that two electrons are transmitted from the first edge state to the second. The second term is the partition noise due to the separation of the beam between QPC’s. In this process, after the separation of the beam, one electron is transmitted with probability $`1`$ in the first edge state towards contact 3 and the other is reflected with probability $`R_3`$ towards contact 2.
Another interesting case occurs for $`T_3=0`$. Then the two edge states are perfectly separated at QPC 3. The first edge state flows towards contact 3 whereas the second goes to contact 2. This situation is particularly interesting since it allows us to study separately the noise spectrum of each edge state after having travelled along the upper edge of the conductor. In the absence of scattering between edge states, the currents remain uncorrelated, whereas if redistribution of charges occurs between edge states, it may induce correlations. For coherent scattering, we find: $`S_{23}^{\mathrm{coh}}=\frac{2e^2}{h}|eV|\epsilon (1\epsilon )R_1^2`$. The only separation of the beam that can cause correlations occurs along the edge with usual partition factor $`\epsilon (1\epsilon )`$; the factor $`R_1^2`$ ensures that, if the currents in the two channels are noiseless before exchanging carriers along the upper edge, they remain noiseless and uncorrelated. Let us also give the currents auto-correlations: $`S_{22}^{\mathrm{coh}}=\frac{2e^2}{h}|eV|[R_1T_1(1\epsilon )^2+\epsilon (1\epsilon )R_1]`$ and $`S_{33}^{\mathrm{coh}}=\frac{2e^2}{h}|eV|[R_1T_1\epsilon ^2+\epsilon (1\epsilon )R_1]`$. The noise spectral densities have contributions from the splitting of the beam at the first QPC and from the redistribution of the charges between the QPC’s. Let us finally mention the value of the noise when $`\epsilon =1/2`$, i.e. when the average currents in the two edge states are equally equilibrated due to scattering, then
$$S_{23}^{\mathrm{coh}}=\frac{e^2}{h}|eV|\frac{R_1^2}{2}\text{ for }T_3=0,\epsilon =1/2$$
(5)
and
$$S_{22}^{\mathrm{coh}}=S_{33}^{\mathrm{coh}}=\frac{e^2}{h}|eV|\frac{1+T_1}{2}R_1.$$
(6)
Next we proceed to treat incoherent scattering.
## III Introduction of incoherent scattering
As we have already evoked, the presence of inelastic or quasi-elastic scattering will be treated by adding a fictitious contact at the edge along which incoherent scattering is expected to occur . Let us remark that the only place where it can have some influence on the correlations is the edge between the two QPC’s where we have introduced it (see figure 2). Since the current is obviously conserved along the edge, we have to impose that the current $`I_4`$ through this contact is zero at any time. The consequence is that the potential $`\mu _4`$ at this probe is fluctuating.
The average currents are related to the potentials by the conductance: $`I_\alpha =\frac{1}{e}dE_\beta G_{\alpha \beta }(E)\overline{f}_\beta (E)`$, where we recall that $`G_{\alpha \beta }=\frac{e^2}{h}(N_\alpha \delta _{\alpha \beta }Tr\{s_{\alpha \beta }^{}s_{\alpha \beta }\})`$, $`N_\alpha `$ being the number of open channels at contact $`\alpha `$. We first discuss the inelastic case for which we require that the current is zero on average $`I_4=0`$ and that the average distribution function at the fictitious contact is an equilibrium Fermi distribution. This leads to the following expression for the average chemical potential at contact $`4`$:
$$\overline{\mu }_4=\mu _0+\frac{1+T_1}{2}eV.$$
(7)
The fact that $`I_4`$ is zero on average determines the average distribution function at contact $`4`$. We have also to ensure that the fluctuating part of $`I_4`$ remains zero. At contact $`\alpha `$ the current may be written as $`I_\alpha =\frac{1}{e}dE_\beta G_{\alpha \beta }f_\beta +\delta I_\alpha `$, where $`\delta I_\alpha `$ is the intrinsic part of the fluctuations, whose spectrum is given by (1). We now write the currents as $`I_\alpha =I_\alpha +\mathrm{\Delta }I_\alpha `$. Imposing that the fluctuating current is zero $`\mathrm{\Delta }I_4=0`$ leads to the relation $`\delta I_4=\frac{1}{e}G_{44}(\mu _4\overline{\mu }_4)`$ which gives the expression of the fluctuating part of the current:
$$\mathrm{\Delta }I_\alpha =\delta I_\alpha \frac{G_{\alpha 4}}{G_{44}}\delta I_4,$$
(8)
the first term corresponds to intrinsic fluctuations of the current and the second term to the fluctuations due to the existence of a fluctuating potential at contact $`4`$.
In the quasi-elastic case we impose that not only the total current $`I_4=dEj_4(E)`$ is zero on average but the contribution to the current of the states of energy $`E`$ is zero $`j_4(E)=0`$, which gives the averaged distribution function at contact $`4`$:
$$\overline{f}_4(E)=\frac{1+T_1}{2}f_1(E)+\frac{R_1}{2}f_2(E).$$
(9)
The distribution function at the fictitious probe is plotted in the two cases in figure 3. Since the integral of these two distribution functions are equal, the currents in the two edge states are equally equilibrated by these two kinds of incoherent scattering. Indeed, if we compute the average currents for $`T_3=0`$, when the two edge states are directed each to a different contact, we find in the two cases: $`I_2=I_3=\frac{e^2}{h}\frac{1+T_1}{2}V`$.
In the quasi-elastic case, the distribution function $`f_4`$ is itself a fluctuating quantity. Imposing that the contribution of the states of energy $`E`$ to the fluctuating part of the current at the fictitious contact is zero, $`\mathrm{\Delta }j_4(E)=0`$, leads to a relation between the fluctuating part of the distribution $`f_4\overline{f}_4`$ and the contribution of those states to the intrinsic noise $`\delta j_4(E)`$. This relation is of the same form as (8): $`\mathrm{\Delta }j_\alpha (E)=\delta j_\alpha (E)\frac{G_{\alpha 4}}{G_{44}}\delta j_4(E)`$.
Finally we find for the two different kinds of incoherent scattering the current correlations $`S_{\alpha \beta }^{\mathrm{in},\mathrm{qe}}=\mathrm{\Delta }I_\alpha \mathrm{\Delta }I_\beta `$:
$$S_{\alpha \beta }^{\mathrm{in},\mathrm{qe}}=S_{\alpha \beta }\frac{G_{\alpha 4}}{G_{44}}S_{\beta 4}\frac{G_{\beta 4}}{G_{44}}S_{\alpha 4}+\frac{G_{\alpha 4}G_{\beta 4}}{G_{44}^2}S_{44}.$$
(10)
They involve the conductances $`G_{\alpha \beta }`$ and the intrinsic correlations $`S_{\alpha \beta }=\delta I_\alpha \delta I_\beta `$ for the four-terminal conductor of figure 2, calculated with formula (1) where the distribution function at contact 4 is either a step like Fermi function for the potential $`\overline{\mu }_4`$ in the inelastic case (7), or the average distribution $`\overline{f}_4`$ given by (9) in the quasi-elastic case (see figure 3).
Let us now come to the results. First, for the quasi-elastic case, we find:
$$S_{23}^{\mathrm{qe}}=\frac{e^2}{h}|eV|\frac{R_3}{4}\left[2T_3(1+T_1)^2+(1+T_3)R_1^2\right].$$
(11)
In particular, in the interesting limit $`T_3=0`$ where the two edge states are perfectly separated at QPC 3, we get: $`S_{23}^{\mathrm{qe}}=\frac{e^2}{h}|eV|\frac{R_1^2}{4}`$, i.e. half of the result (5) obtained when scattering between edge states is coherent. The auto-correlations take the values: $`S_{22}^{\mathrm{qe}}=S_{33}^{\mathrm{qe}}=\frac{e^2}{h}|eV|\frac{R_1}{2}\frac{1+3T_1}{2}`$.
More surprising is the result obtained for inelastic scattering:
$$S_{23}^{\mathrm{in}}=\frac{e^2}{h}|eV|\frac{R_3}{2}\left[2T_3(1+T_1)(1+T_3)R_1T_1\right]$$
(12)
leading to the possibility of positive correlations, as figure 4 shows (let us recall again that correlations in the presence of coherent scattering only are always negative due to the fermionic nature of carriers ). If $`T_3=0`$, we get $`S_{23}^{\mathrm{in}}=+\frac{e^2}{h}|eV|\frac{R_1T_1}{2}`$. This result means that this modelization of inelastic scattering gives the possibility for a fluctuation of the potential $`\mu _4`$ to inject in a correlated way in the two channels some current. Let us also mention the result for the noise in this limit: $`S_{22}^{\mathrm{in}}=S_{33}^{\mathrm{in}}=\frac{e^2}{h}|eV|\frac{R_1T_1}{2}`$.
If $`T_1=T_3=0`$, only one edge state is transmitted at QPC $`1`$, the currents in the two channels are equilibrated by inelastic processes between QPC’s, and finally the two edge states are separated by QPC 3. We remark that in this case the correlations and the noise vanish. This shows that this modelization of inelastic process does not introduce noise when it distributes the current of a noiseless channel into two channels. (This is not the case if coherent or quasi-elastic scattering occurs between edge states: $`S_{23}^{\mathrm{coh}}=\frac{e^2}{h}|eV|\frac{1}{2}`$ and $`S_{23}^{\mathrm{qe}}=\frac{e^2}{h}|eV|\frac{1}{4}`$).
Note that if more than two edge states carry the current, still with only one partially transmitted and all others being perfectly transmitted, the region in the paramater space where the correlations are positive diminishes.
Finally we would like to discuss the effect of a finite temperature on the result (12) for $`T_3=0`$. We will show that for a high enough temperature $`S_{23}^{\mathrm{in}}`$ is negative. For a finite temperature $`T`$, the Fermi distributions at the three contacts are smoothed as well as the average distribution $`\overline{f}_4`$ at the fictitious contact. For $`T_3=0`$, equation (1) gives:
$`S_{23}^{\mathrm{in}}={\displaystyle \frac{e^2}{h}}[k_BT(2+R_1+R_1T_1)`$ (13)
$`+{\displaystyle \frac{R_1T_1}{2}}eV\mathrm{coth}{\displaystyle \frac{eV}{2k_BT}}].`$ (14)
If $`T=0`$ we indeed recover the positive result $`S_{23}^{\mathrm{in}}=+\frac{e^2}{h}|eV|\frac{R_1T_1}{2}`$ for the shot noise, and if $`V=0`$ we find $`S_{23}^{\mathrm{in}}=\frac{2e^2}{h}k_BT\left(1+\frac{R_1}{2}\right)`$, which is the result of the fluctuation-dissipation theorem: $`S_{23}^{\mathrm{in}}=2k_BT(G_{23}^{\mathrm{in}}+G_{32}^{\mathrm{in}})`$ where $`G_{\alpha \beta }^{\mathrm{in}}=G_{\alpha \beta }\frac{G_{\alpha 4}G_{4\beta }}{G_{44}}`$ is the conductance of the three-terminal conductor in the presence of incoherent scattering.
Let us define $`T_c`$, the critical temperature above which the correlations $`S_{23}^{\mathrm{in}}`$ are negative. For small transmission $`T_11`$ we find: $`k_BT_c|eV|\frac{T_1}{6}`$, and for large transmission $`R_11`$: $`k_BT_c|eV|\frac{R_1}{4}`$. The transmission that maximizes the critical temperature is $`T_1=3\sqrt{6}0.55`$. In this case we have: $`k_BT_c^{\mathrm{max}}|eV|\frac{5\sqrt{6}12}{2(6\sqrt{6}12)}\frac{|eV|}{21.8}`$.
## IV Summary
To summarize we have shown that incoherent scattering can have a strong effect on the correlations in a HBT situation. In our modelization, inelastic scattering is responsible for positive correlations in a certain parameter range. We emphasize that the investigation of the correlations with the conductor of figure 1 for $`T_3=0`$ provides direct information about inter-edge state scattering: in the absence of inter-edge state scattering the correlation $`S_{23}`$ vanishes, for elastic scattering and for quasi-elastic scattering it is negative and for inelastic scattering it is positive. In principle it is possible to discriminate also between elastic inter-edge state scattering and quasi-elastic scattering: in the case of elastic scattering a single parameter ($`\epsilon `$) determines the noise spectra and the conductances if $`T_1`$ and $`T_3`$ are known.
We have constructed here only the limiting case of a fictitious contact. Partially transmitting contacts would allow to interpolate between fully coherent inter-edge state scattering and fully quasi-elastic or fully inelastic inter-edge state scattering.
## Acknowlegments
Interest in the topic of this work was stimulated by a question of Bart J. van Wees. We acknowledge Yaroslav M. Blanter and Andrew M. Martin for very interesting discussions. This work was supported by the Swiss National Science Foundation and by the TMR Network Dynamics of Nanostructures. |
warning/0003/astro-ph0003299.html | ar5iv | text | # I Introduction
## I Introduction
The shape and the intensity of the intergalactic UV background radiation are crucial factors in determining the ionization balance of the intergalactic medium and therefore influence the structure formation in the universe. Knowledge of this radiation field is thus necessary for the understanding of the early universe. AGNs are believed to be the major contributors to this background, though a significant contribution from star forming galaxies can not be ruled out. Detailed calculations of the propagation of AGN like ionizing radiation through intergalactic space, taking into account the absorption and reradiation by the galactic and intergalactic material, have been carried out by Haardt and Madau (1996, hereafter HM96). They have determined the frequency and redshift dependence of the background. Observationally, the intensity of the radiation has been determined in recent years, by studying the proximity effect in the Lyman alpha forest of the absorption lines in the spectra of QSOs (Bajtlik, Duncan & Ostriker 1988). This analysis is insensitive to the shape of the radiation (Bechtold, 1994 and Das and Khare, 1997). Values of the intensity of the background at the Lyman limit, J$`_{\nu _{\mathrm{LL}}}`$, obtained by Bechtold (1994) and Cooke et al (1996) are considerably higher than the value expected from the distribution of visible QSOs. Several sources of uncertainty in the value of the flux obtained by the proximity effect analysis have been considered by various authors (Bechtold, 1994, 1995, Srianand and Khare, 1995, Das and Khare, 1997). It has also been suggested (Fall & Pei 1993) that the actual number of QSOs may be larger than their observed number and that several QSOs may be rendered invisible due to dust extinction in the intervening absorbers. It is possible that the background radiation gets a significant contribution from star forming galaxies (Madau & Shull 1996, Giroux & Shapiro 1996, Khare and Ikeuchi, 1998). Here we try to obtain an independent estimate of the background flux by studying the ionization state of the QSO absorption systems for which an estimate of the particle density is available from the observations of the fine structure excited lines of C II. Where ever possible, we also try to estimate the contribution of the galactic flux to the total ionizing flux for the systems. In section II we present our analysis, the results are discussed in section III.
## II Data and Analysis
Several absorption systems reported in the literature, have the absorption lines of C II, C II\*, C IV, H I, Si II, Si IV etc. Particle density in the absorber can be obtained from the column densities of C II and C II\*. The column density ratio of C II to C IV is a good indicator of the ionization parameter as it changes rapidly with change in the ionization parameter, $`\mathrm{\Gamma }=\frac{\mathrm{\Phi }}{\mathrm{c}\mathrm{n}_\mathrm{H}}`$, where $`\mathrm{\Phi }`$ is the flux of the ionizing radiation i.e. the number of photons cm<sup>-2</sup> s<sup>-1</sup>, $`\mathrm{n}_\mathrm{H}`$ is the particle density and c is the velocity of light. This ratio is insensitive to the particle density and the abundance of carbon, but is very sensitive to the neutral hydrogen column density, N$`_{\mathrm{H}\mathrm{I}}`$, in the absorber, specially for N$`{}_{\mathrm{H}\mathrm{I}}{}^{}>10^{17}`$ cm<sup>-2</sup> (Bergeron and Stasinska 1986 ). This particular column density ratio is also very sensitive to the shape of the ionizing radiation. This is shown in Fig.1 which shows the ratios for neutral hydrogen column densities of 10<sup>17</sup> cm<sup>-2</sup> and 10<sup>19</sup> cm<sup>-2</sup> for (i) spectral shape as given by HM96 for a redshift of 2.5, (ii) galactic spectra as taken from Bruzal (1983) and (III) a combination of both the above with the AGN flux taken to be same as the actual value given by HM96 for z=2.5. These results have been obtained from the code ’CLOUDY’, kindly supplied to us by Prof. Ferland. The galactic spectra produces much smaller values of the ratio compared to the AGN or power law spectra. This is due to the fact that the galactic spectra has much smaller number of photons having sufficient energy to produce C IV. Thus it is necessary to know the shape of the ionizing radiation to determine the ionization parameter from the C II to C IV ratio. In order to get information about the intensity and the shape of the radiation field we can make use of the column density ratio of other ions. Si II to Si IV ratio is an useful ratio for this purpose. In Fig. 2 we have plotted the ratio of column densities of Si II to Si IV for the three cases listed above. This ratio is not very sensitive to the shape and can be used with the C II to C IV ratio to constrain the shape as well as the intensity of the ionizing radiation. Fe II to Fe III and Al II to Al III ratios can also be used for this purpose. These ratios are also plotted in Fig. 2. We have, therefore, selected from the literature, absorption systems for which the column densities of H I, C II and C II\*, or limits on their values, have been determined and for which additional ions like the C IV, Si II, Si IV, Fe II, Fe III etc are also available. The details are given in Table 1.
The fine structure excited level of C II is primarily populated by collisions with electrons for the absorption systems considered here, which are either the Lyman limit or the damped Lyman alpha systems, while the collisions with H I and also the collisional deexcitations by electrons can be ignored (Bahcall and Wolfe, 1968, Morris et al 1986). As we are considering the absorption systems at high redshifts, the excitation of the finestructure level by the absorption of the cosmic microwave background photons may be important. In order to check this we calculated the excitation temperatures of C II\* from the observed column densities. These are much higher than the corresponding temperatures of the microwave background at the redshifts of the absorbers as can be seen from Table 2. We have therefore, ignored the excitation by the microwave background photons and have assumed the electron densities (assumed to be equal to the proton density) to be given by n<sub>e</sub> = 21 \[$`\frac{\mathrm{N}_{\mathrm{C}\mathrm{II}}}{\mathrm{N}_{\mathrm{C}\mathrm{II}}}`$\] (Morris et al 1986). The hydrogen density obtained by adding the neutral hydrogen density to the electron density, corrected for the electrons coming from He II and He III, for each system is also given in Table 2. For a few of the systems, only an upper limit on C II\*/C II was available. For these systems only an upper limit on the particle density could be obtained. For these systems we have assumed a lower limit on particle density to be 0.045 cm<sup>-2</sup>, which is smaller than all the lower limits to the particle densities obtained for the systems considered here and is also considerably lower than the mean interstellar particle density. For each absorption system we have constructed a number of photoionization models for the observed neutral hydrogen column density and different spectral shapes. The AGN spectral shape at the redshift of the absorption system is taken from HM96. For some of the systems only the total neutral hydrogen column density is available, while C II\* has been observed in some particular velocity component of the absorption line. For such systems we have assumed the neutral hydrogen column density in individual components to be in the same ratio as the Si II column densities, as the ionization potential of Si II is close to that of H I. Details of the models and comparison of their predictions with the observations are discussed below for individual absorption systems. The results are given in Table 2. Note that all the absorption systems are sufficiently far away from the respective QSOs (relative velocity is greater than 15000 km s<sup>-1</sup>) and can be considered to be intervening (however, see Richards et al, 1999) so that the radiation of the parent QSO can be ignored. In the following analysis we have only used the ratios of the column densities of different ions of the same element. Our conclusions are, therefore, independent of the assumed values of chemical abundances.
### A z=1.7765 system towards Q1331+170
This QSO has been observed by Kulkarni et al (1996). However, for this system C IV column density has not been reported and so we could not constrain the spectral shape. Errors on column densities have not been reported by the authors and we assumed errors of 25$`\%`$ in the column densities. Neutral hydrogen column density in the component showing C II\* has been obtained from the total H I column density (Green et al 1995) by assuming the H I column densities to be in the same ratio as the Si II column densities. Analysis of the Si lines assuming HM96 spectral shape for z=2 yields -2.4 $`>\mathrm{\Gamma }>2.6`$ giving 3.2 $`\times 10^7\mathrm{\Phi }1.4\times 10^8`$. This is considerably higher than the value of HM96 flux at the redshift of the absorption system. It is, however, possible that the excess flux comes from the galaxies. We explored this possibility by constructing photoionization models with the shape as well as the intensity of the background as given by HM96 at the redshift of the absorber, the rest coming from the galaxies. For these models the limits become -2.7 $`>\mathrm{\Gamma }>2.9`$ giving 1.5 $`\times 10^7\mathrm{\Phi }_\mathrm{G}6.9\times 10^7`$, $`\mathrm{\Phi }_\mathrm{G}`$ being the galactic flux in cm<sup>-2</sup> s<sup>-1</sup>.
### B z=2.279 system towards Q2348-14
This QSO has been observed by Pettini et al (1994). An upper limit on the column density ratio of Si II to Si IV gives, for HM96 spectral shape for z=2.5, a lower limit of -2.2 for log $`\mathrm{\Gamma }`$. However, as only an upper limit on the particle density could be obtained, this can not be converted to a limit on the flux. Assuming the minimum value of the particle density to be 0.045 cm<sup>-3</sup>, which is smaller than the observed upper limit by 1.2 dex, requires the flux to be larger than 8.5 $`\times 10^6`$. The column density ratio of Al II to Al III gives -1.3 $`>`$ log $`\mathrm{\Gamma }>1.4`$, giving $`5.3\times 10^7<\mathrm{\Phi }<1.2\times 10^9`$. Taking the actual value of the HM96 flux and assuming the rest of the contribution from the galactic sources requires -1.4 $`>`$ log $`\mathrm{\Gamma }>1.5`$, giving the galactic flux to be between 4.1 $`\times 10^7`$ and 9.7 $`\times 10^8`$.
### C z=2.638 system towards PKS 2126-158
This system has been observed by Giallongo et al (1993) and has 7 components spread over a velocity width of 270 km s<sup>-1</sup>. A neutral hydrogen column density for the whole system has been obtained by Young et al (1979) to be 1.1 $`\times 10^{19}`$ cm<sup>-2</sup>. We have assumed the neutral hydrogen column density in individual components to be in the same ratio as the Si II column densities. C II\* is observed in two of the components. These are considered below.
(i) z=2.6376: We obtain log N<sub>HI</sub> = 17.92. As Si IV lines are not observed we take a 3 $`\sigma `$ upper limit on equivalent width, which translates to an upper limit of 10<sup>13.23</sup> cm<sup>-2</sup> on the column density of Si IV, assuming a velocity dispersion parameter of 24.2 km s<sup>-1</sup>, same as that for C IV. Assuming AGN shape, we get -3.1 $`>`$ log $`\mathrm{\Gamma }`$ $`>`$ -3.3, giving 1.6 $`\times 10^6\mathrm{\Phi }1.4\times 10^8`$. Taking the AGN flux to be that given by HM96 for z=2.5, and additional flux from the galaxies, the observed column density ratios can not be explained and a minimum flux of 9.4 $`\times 10^5`$ from the AGN is required, needing a total of -2.9 $`>`$ log $`\mathrm{\Gamma }`$ $`>`$ -3.1, giving 1.6 $`\times 10^6\mathrm{\Phi }_\mathrm{G}2.3\times 10^8`$. Note that the Al II to Al III ratio can not, however be explained by the same range of $`\mathrm{\Gamma }`$ for any shape and requires log $`\mathrm{\Gamma }3.3`$.
(ii) z=2.6364. We obtain log N<sub>HI</sub> = 16.4. As Si II and Si IV lines have not been detected for this system, we could not obtain any constraints on the relative contribution from galaxy to the radiation flux. Taking all of the radiation to be AGN type, we obtained limits on the ionization parameter to be -2.7 $``$ log $`\mathrm{\Gamma }3.3`$ so that the flux lies between 1.05$`\times 10^6`$ and 6.97$`\times 10^8`$. Taking the actual value of the AGN flux (HM96 value at z=2.5), the rest coming from galaxy, requires log $`\mathrm{\Gamma }1.0`$, giving $`\mathrm{\Phi }_\mathrm{G}1.2\times 10^{10}`$.
### D z=2.6522 system towards Q2231-00
This Lyman limit system has been analysed in details by Prochaska (1999). We have reanalysed this system taking the shape of the background to be that given by HM96 for z=2.5. We determined the column density ratios for various ions for log N<sub>HI</sub> of 19.12. The Fe II to Fe III and Si II to Si IV column density ratios constrain the ionization parameter to between -2.4 to -2.55 resulting in a background flux between 3.9$`\times 10^6`$ and 8.5$`\times 10^6`$. This is considerably higher than the value of 8.59$`\times 10^5`$ for HM96 for z=2.5. Models with the shape as well as the intensity of the background as given by HM96 for z=2.5, the rest of the flux coming from galaxy fail to yield a result as the column density ratios of Si and Fe can not simultaneously be produced by a single value (range) of ionization parameter. By gradually increasing the value of the flux contributed by the AGN background above the HM96 value, we find that a minimum AGN flux of 3.3 $`\times 10^6`$ was needed to explain the ion ratios, for which we get -2.4 $`>`$ log $`\mathrm{\Gamma }`$ $`>`$ -2.6, requiring $`\mathrm{\Phi }_\mathrm{G}`$ to be between 6.7 $`\times 10^5`$and 5.2 $`\times 10^6`$. Thus the minimum of AGN type flux required is more than a factor of 3.8 larger than that obtained by HM96.
### E z=2.844 system towards HS1946+7658
This system has been observed and analysed by Fan and Tytler (1994). Cloudy models with HM96 spectral shape at z=3 give -2.3 $`>`$ log $`\mathrm{\Gamma }>`$ -2.8, giving $`10^8\mathrm{\Phi }10^9`$. A minimum flux of 8.4$`\times 10^7`$ of the HM96 type is needed, with -1.9 $``$ log $`\mathrm{\Gamma }2.6`$, giving 1.9 $`\times 10^8\mathrm{\Phi }_\mathrm{G}3.9\times 10^9`$. Thus almost all of the flux is being contributed by galaxy. Note that the flux for z=3 of HM96 is 7.8$`\times 10^5`$.
### F z=3.025 system towards Q0347-38
This QSO has been observed by Prochaska and Wolfe (1999). Only a lower limit is available for the C II to C IV ratio, so that the galactic fraction of the flux could not be constrained. AGN shape for z=3.0 for the radiation gives -2.6 $`>`$ log $`\mathrm{\Gamma }>2.8`$, giving 2.1 $`\times 10^6\mathrm{\Phi }9.8\times 10^7`$, for the assumed minimum value of the particle density. Taking the actual value of the HM96 flux for z=3 and assuming the rest of the contribution from the galactic sources requires -2.5 $``$ log $`\mathrm{\Gamma }2.7`$, giving 1.9 $`\times 10^6\mathrm{\Phi }_\mathrm{G}1.2\times 10^8`$, for the assumed minimum value of n<sub>H</sub>.
### G z=3.6617 system towards Q2212-1626
This QSO has been observed by Lu et al (1996). Only a lower limit is available for the C II to C IV ratio, so that the galactic fraction of the flux could not be constrained. AGN shape for z=3.5 for the radiation gives -2.5 $`>`$ log $`\mathrm{\Gamma }>2.6`$, giving $`3.4\times 10^6\mathrm{\Phi }1.1\times 10^8`$ for the assumed minimum value of n<sub>H</sub>. Taking the actual value of the HM96 flux at z=3.5 and assuming the rest of the contribution from the galactic sources requires -2.4 $`>`$ log $`\mathrm{\Gamma }>2.5`$, giving $`3.8\times 10^6\mathrm{\Phi }_\mathrm{G}1.4\times 10^8`$, for the assumed minimum value of n<sub>H</sub>. Note that the value of flux for HM96 for z=3.5 is 5.0 $`\times 10^5`$.
### H z=4.0803 system towards Q2237-0608
This QSO has been observed by Lu et al (1996). Only a lower limit is available for the C II to C IV ratio, so that the galactic fraction of the flux could not be constrained. AGN shape for z=4 for the radiation gives -2.6 $`>`$ log $`\mathrm{\Gamma }>2.7`$, giving $`2.6\times 10^6\mathrm{\Phi }2.1\times 10^7`$ for the assumed minimum value of n<sub>H</sub>. Taking the actual value of the HM96 flux at z=4 and assuming the rest of the contribution from the galactic sources requires -2.4 $``$ log $`\mathrm{\Gamma }>2.5`$, giving $`4.0\times 10^6\mathrm{\Phi }_\mathrm{G}3.3\times 10^7`$ , for the assumed minimum value of n<sub>H</sub>. Note that the value of flux for HM96 for z=4 is 2 $`\times 10^5`$
## III Discussion
For 5 of the systems we could derive the range of flux values assuming the radiation to be AGN type. All of these are higher than the corresponding HM96 values by minimum factors ranging from 1.2 to 158. Note that we have taken into account the uncertainties in the column densities of all the ions which is the reason for obtaining large ranges for the flux values. For three of these systems we could obtain a minimum value for the flux of the AGN background. These values are 1.1, 3.8 and 98.8 times higher than the HM96 values at the appropriate redshifts. For these systems a large flux is needed from galaxies. For 4 other systems a lower limit to the flux could only be obtained with an assumption of the lower limit on the particle density to be 0.045 cm<sup>-3</sup>, which is about half of the mean interstellar value of the particle density and which indicates that the actual C II column densities are higher than the observed lower limits by 0.64 to 1.37 dex. This is a reasonable lower limit as the systems being considered are Lyman limit or damped Lyman alpha systems and also as this value is considerably lower than the range of density values for systems for which the values could be obtained from the observations. For these systems, the required values of flux are higher than the HM96 values by minimum factors of 2.7 to 62. On the other hand, assuming the AGN flux to be that given by HM96, and assuming the rest of the required flux to be of local, galactic origin, very high galactic flux is required. For most of the systems, this high flux requires the absorption systems to be present with in 100 parsecs of typical O stars. The typical radius of the Stromgren spheres of these stars is of the same order, indicating that the absorption systems are inside the H II regions. Such conclusions have earlier been rejected on the basis of statistical arguments about the properties of the absorption systems (Srianand and Khare, 1994). The flux could come from QSOs which happen to lie close to the lines of sight at redshifts similar to the redshifts of the absorption systems. We have searched the catalogues for presence of any such QSOs near the line of sight to Q2231-00. However, no QSO is found to lie closer than 1000 Mpc to the line of sight within the required redshift range. The high values of the flux indicated by our analysis for almost all the systems, may be interpreted to indicate the presence of an unseen population of dust extinct QSOs.
Note that in all our analysis we have assumed that all the ions producing absorption in a given velocity range in an absorption system are physically located in the same region (cloud). This may not be always valid. Kirkman and Tytler (1999) and Churchill and Charlton (1999) have found evidence for ions with the same velocity structure in their absorption lines belonging to a given redshift system, arising in physically different gaseous components. If C IV ions are from a more widely distributed component, then, the C II/C IV column density ratio in the region of interest will be smaller and may require lower values of the flux.
## acknowledgement
This work was partially supported by a grant (No. SP/S2/013/93) by the Department of Science and Technology, Government of India. A.S. is supported by a C.S.I.R. fellowship.
Figure Captions
Figure 1. Column density ratios of C II to C IV as a function of the ionization parameter, for different shapes of the background radiation spectrum as explained in the text. The solid lines are for N<sub>HI</sub>=10<sup>17</sup> cm<sup>-2</sup> and dashed lines are for N<sub>HI</sub>=10<sup>19</sup> cm<sup>-2</sup>.
Figure 2. Column density ratios of Si II to Si IV, Al II to Al III and Fe II to Fe III. for the three shapes of the background radiation spectrum as explained in the text, for N<sub>HI</sub>=10<sup>18</sup> cm<sup>-2</sup>. Solid, dotted and dashed lines are for galactic spectra, AGN (HM96) spectra and AGN together with the galactic spectra respectively. |
warning/0003/hep-th0003300.html | ar5iv | text | # Gravity of higher-dimensional global defects
## I Introduction
In recent years there has been renewed interest in ‘brane-world’ models in which the universe is represented by a $`(3+1)`$-dimensional subspace (3-brane) embedded in a higher-dimensional (bulk) spacetime . In such models, all the familiar matter fields are constrained to live on the brane, while gravity is free to propagate in the extra dimensions. Initially it was thought that realistic models required compact extra dimensions, but it has been shown by Randall and Sundrum that it is possible to have infinite extra dimensions and still have gravity effectively localized on the brane. This is achieved by introducing a negative cosmological constant which has the effect of ‘warping’ the extra-dimensional space, so that most of the physical volume is concentrated near the brane.
In most of the recent work, including that of Randall and Sundrum, the brane is pictured as a domain wall propagating in a 5-dimensional bulk spacetime. The case of two extra dimensions has also been considered, when the branes are similar to strings and the bulk has 6 dimensions. For a gauge string, the metric outside the string core is flat with a conical deficit angle, and Sundrum suggested that the extra dimensions can be compactified by introducing a sufficient number of branes, so that the total deficit angle is equal to $`2\pi `$. This was generalized by Chodos and Poppitz to include a positive cosmological constant. Cohen and Kaplan considered the case of a global sting which has a curvature singularity at a finite distance from the string core. They argued that the singularity can provide an effective compactification of the extra dimensions. Gregory has shown that a non-singular global string solution exists in the presence of a negative cosmological constant. In this solution the extra dimensions are infinite and strongly warped, as in the Randall-Sundrum model. Garriga and Sasaki discussed a Euclidean continuation of the Randall-Sundrum spacetime and interpreted the resulting instanton as describing quantum nucleation of a 5-dimensional brane-world from nothing.
In this paper, we shall explore a more general case of a brane carrying a global charge in a higher-dimensional spacetime with a nonzero cosmological constant $`\mathrm{\Lambda }`$. We shall consider a $`(p1)`$-dimensional brane in a bulk spacetime of $`D=p+n`$ dimensions. The physically interesting case is $`p=4`$, but we shall allow an arbitrary $`p`$ for greater generality. For $`n=3`$ and $`\mathrm{\Lambda }=0`$, one can expect to recover the global monopole metric with a solid deficit angle , but for $`n>3`$ the defects do not have 3-dimensional analogues.
The paper is organized as follows. In the next Section we introduce the scalar field and metric ansatz and present the corresponding Einstein’s equations. Our solutions for $`\mathrm{\Lambda }0`$ and $`\mathrm{\Lambda }<0`$ are given, respectively, in Sections III and IV. Euclidean instanton solutions are discussed in Section V, and our conclusions are summarized in Section VI.
## II Einstein’s equations
We shall use the notation $`\{x^\mu \}`$ with $`\mu =0,\mathrm{},p1`$ for the coordinates on the brane worldsheet, $`\{\xi ^a\}`$ with $`a=1,\mathrm{},n`$ for coordinates in the extra dimensions, and $`\{X^A\}`$ with $`A=0,\mathrm{},D1`$ for general coordinates in the $`D`$-dimensional spacetime.
A global defect in $`n`$ extra dimensions is described by a multiplet of $`n`$ scalar fields $`\varphi ^a`$ with a Lagrangian
$$L=\frac{1}{2}_A\varphi ^a^A\varphi ^aV(\varphi ),$$
(1)
where the potential $`V(\varphi )`$ has its minimum on the $`n`$-sphere $`\varphi ^a\varphi ^a=\eta ^2`$. One can use, for example,
$$V(\varphi )=\frac{\lambda }{4}(\varphi ^a\varphi ^a\eta ^2)^2.$$
(2)
The defect solution should have $`\varphi =0`$ at the center of the defect and approach the radial ‘hedgehog’ configuration outside the core,
$$\varphi ^a(\xi )=\eta \frac{\xi ^a}{\xi }$$
(3)
with $`\xi ^2\xi ^a\xi ^a`$. We shall be interested only in the exterior solutions, where $`V(\varphi )0`$ and $`\varphi (\xi )`$ is accurately approximated by (3).
We shall adopt the following ansatz for the metric:
$$ds^2=A(\xi )^2d\xi ^2+\xi ^2d\mathrm{\Omega }_{n1}^2+B(\xi )^2\widehat{g}_{\mu \nu }dx^\mu dx^\nu ,$$
(4)
where $`d\mathrm{\Omega }_m^2`$ stands for the metric on a unit $`m`$-sphere, and the spherical coordinates in the extra dimensions are defined by the usual relations, $`\xi ^a=\{\xi \mathrm{cos}\theta _1,\xi \mathrm{sin}\theta _1\mathrm{cos}\theta _2,\mathrm{}\}`$. (A different ansatz will be considered in sections III.C and IV.C). The energy-momentum tensor for the field configuration (3) is then given by
$`T_\xi ^\xi ={\displaystyle \frac{1}{2}}(n1){\displaystyle \frac{\eta ^2}{\xi ^2}},`$ (5)
$`T_{\theta _b}^{\theta _a}={\displaystyle \frac{1}{2}}(n3){\displaystyle \frac{\eta ^2}{\xi ^2}}\delta _b^a,`$ (6)
$`T_\mu ^\nu ={\displaystyle \frac{1}{2}}(n1){\displaystyle \frac{\eta ^2}{\xi ^2}}\delta _\mu ^\nu .`$ (7)
(8)
Our goal will be to solve Einstein’s equations in $`D`$ dimensions,
$$R_{AB}\frac{1}{2}G_{AB}R=\kappa ^2T_{AB}\mathrm{\Lambda }G_{AB},$$
(9)
where $`G_{AB}`$ is the $`D`$-dimensional metric, $`\mathrm{\Lambda }`$ is the cosmological constant and $`T_{AB}`$ is from Eq.(8). The line element (4) is a special case of the more general class of metrics,
$$ds^2=d\stackrel{~}{s}_n^2+B(\xi ^a)^2d\widehat{s}^2,$$
(10)
where $`d\stackrel{~}{s}_n`$ depends only on the transverse coordinates $`\{\xi ^a\}`$ and $`d\widehat{s}^2`$ only on those on the brane $`\{x^\mu \}`$. For such metrics, the Ricci tensor splits in the following way:
$`R_{mn}`$ $`=`$ $`\stackrel{~}{R}_{mn}p{\displaystyle \frac{B_{;mn}}{B}},`$ (11)
$`R_{\mu \nu }`$ $`=`$ $`\widehat{R}_{\mu \nu }\widehat{g}_{\mu \nu }\left[B\stackrel{~}{}^2B+(p1)(\stackrel{~}{}B)^2\right].`$ (12)
Since $`T_{\mu \nu }\widehat{g}_{\mu \nu }`$, through Einstein’s equations we have that $`R_{\mu \nu }\widehat{g}_{\mu \nu }`$ and finally, from equation (12) above, that $`\widehat{R}_{\mu \nu }\widehat{g}_{\mu \nu }`$. That is, $`\widehat{R}`$, the curvature associated with the metric $`\widehat{g}_{\mu \nu }`$, must be constant. Einstein’s field equations then reduce to
$`{\displaystyle \frac{1}{A^2}}\left[p(p1)\left({\displaystyle \frac{B^{}}{B}}\right)^2+2p{\displaystyle \frac{n1}{\xi }}{\displaystyle \frac{B^{}}{B}}+{\displaystyle \frac{(n1)(n2)}{\xi ^2}}\right]`$ (13)
$`{\displaystyle \frac{n1}{\xi ^2}}(n2\kappa ^2\eta ^2)+2\mathrm{\Lambda }{\displaystyle \frac{\widehat{R}}{B^2}}`$ $`=0`$ , (14)
$`{\displaystyle \frac{1}{A^2}}[2p({\displaystyle \frac{B^{\prime \prime }}{B}}{\displaystyle \frac{A^{}}{A}}{\displaystyle \frac{B^{}}{B}})+p(p1)\left({\displaystyle \frac{B^{}}{B}}\right)^2+{\displaystyle \frac{2(n2)}{\xi }}(p{\displaystyle \frac{B^{}}{B}}{\displaystyle \frac{A^{}}{A}})`$ (15)
$`+{\displaystyle \frac{(n3)(n2)}{\xi ^2}}]{\displaystyle \frac{n3}{\xi ^2}}(n2\kappa ^2\eta ^2)+2\mathrm{\Lambda }{\displaystyle \frac{\widehat{R}}{B^2}}`$ $`=0,`$ (16)
$`{\displaystyle \frac{1}{A^2}}[2(p1)({\displaystyle \frac{B^{\prime \prime }}{B}}{\displaystyle \frac{A^{}}{A}}{\displaystyle \frac{B^{}}{B}})+{\displaystyle \frac{2(n1)}{\xi }}((p1){\displaystyle \frac{B^{}}{B}}{\displaystyle \frac{A^{}}{A}})+(p1)(p2)\left({\displaystyle \frac{B^{}}{B}}\right)^2`$ (17)
$`+{\displaystyle \frac{(n1)(n2)}{\xi ^2}}]{\displaystyle \frac{n1}{\xi ^2}}(n2\kappa ^2\eta ^2)+2\mathrm{\Lambda }+{\displaystyle \frac{2p}{p}}{\displaystyle \frac{\widehat{R}}{B^2}}`$ $`=`$ $`0,`$ (18)
supplemented by the equation for the metric on the $`(p1)`$-brane,
$$\widehat{R}_{\mu \nu }=\frac{\widehat{R}}{p}\widehat{g}_{\mu \nu }.$$
(19)
It can be shown that only two of the three equations (14)-(18) are independent.
We have been able to find several classes of solutions to this set of equations. We shall discuss separately the cases of positive and negative $`\mathrm{\Lambda }`$ and consider both Lorentzian and Euclidean versions of the metric.
## III Solutions with $`\mathrm{\Lambda }0`$
### A Class I
The first class of solutions is obtained with the ansatz $`A(\xi )=B(\xi )^1`$, which is the same as the one used for a global monopole in . With this ansatz, Einstein’s equations are considerably simplified. The two independent equations can be written as
$$(p+1)\frac{A^{}}{A^3}\xi \frac{n2}{A^2}+(n2)\kappa ^2\eta ^2\frac{2\mathrm{\Lambda }}{n+p2}\xi ^2=0,$$
(20)
$$\frac{A^{\prime \prime }}{A^3}+(p+2)\left(\frac{A^{}}{A^2}\right)^2\frac{(n1)}{\xi }\frac{A^{}}{A^3}\frac{\widehat{R}}{p}A^2+\frac{2\mathrm{\Lambda }}{n+p2}=0,$$
(21)
and we find the following solution:
$`A^2(\xi )=B^2(\xi )=1{\displaystyle \frac{\kappa ^2\eta ^2}{n2}}{\displaystyle \frac{2\mathrm{\Lambda }}{(n+p2)(n+p1)}}\xi ^2`$ (22)
$`\widehat{R}={\displaystyle \frac{2\mathrm{\Lambda }p(p1)}{(n+p1)(n+p2)}}(1{\displaystyle \frac{\kappa ^2\eta ^2}{n2}})`$ (23)
The solution (22) is only valid for $`n>2`$. For $`n=3`$ and $`\mathrm{\Lambda }=0`$, the transverse to the brane part of the solution coincides with the metric of a global monopole. In higher dimensions, $`n>3`$, the form of the metric is quite similar: the defect introduces a solid angle deficit in extra dimensions. This is remarkable, considering the fact that defect solutions are quite different for $`n=1,2`$.
With an appropriate rescaling of the coordinates, the $`\mathrm{\Lambda }=0,n3`$ solution can be written as
$$ds^2=d\xi ^2+\left(1\frac{\kappa ^2\eta ^2}{n2}\right)\xi ^2d\mathrm{\Omega }_{n1}^2+\eta _{\mu \nu }dx^\mu dx^\nu ,$$
(24)
where $`\eta _{\mu \nu }`$ is the Minkowski metric. As the symmetry breaking scale $`\eta `$ is increased, the solid angle deficit grows and eventually consumes the entire solid angle at the critical value
$$\eta _c=(n2)^{1/2}\kappa ^1.$$
(25)
One expects that the transverse dimensions in this case have the geometry of an infinite cylinder whose cross-sections are $`(n1)`$-spheres of a fixed radius. This expectation will be verified in Section III.C.
We next consider solutions with $`\mathrm{\Lambda }>0`$. Requiring that the right-hand side of (22) is positive, we should have $`\kappa ^2\eta ^2/(n2)<1`$, so that $`\widehat{R}>0`$, and $`\xi `$ should be constrained to the interval $`0<\xi <\xi _m`$ where $`\xi _m`$, defined by the condition $`B(\xi _m)=0`$, is
$$\xi _m^2=\frac{(n+p2)(n+p1)}{2\mathrm{\Lambda }}\left(1\frac{\kappa ^2\eta ^2}{n2}\right).$$
(26)
Another form of the metric can be obtained using the transformation:
$$\xi =\xi _m\mathrm{sin}\chi ,$$
(27)
which gives
$$ds^2=K\left[d\chi ^2+\alpha ^2\mathrm{sin}^2\chi d\mathrm{\Omega }_{n1}^2+\mathrm{cos}^2\chi d\widehat{s}_+^2\right],$$
(28)
$$\widehat{R}=p(p1),$$
(29)
where
$$K=(n+p1)(n+p2)/2|\mathrm{\Lambda }|,$$
(30)
$$\alpha ^2=\left|1\frac{\kappa ^2\eta ^2}{n2}\right|,$$
(31)
and $`\chi `$ varies in the interval $`0<\chi <\pi `$. The absolute value signs on the right-hand sides of Eqs.(30),(31) are introduced for later use.
The positive-curvature metric $`d\widehat{s}_+^2`$ can be given by any solution of Eq.(19) with $`\widehat{R}`$ from Eq.(29). In this paper we shall assume it to be the $`p`$-dimensional de Sitter space,
$$d\widehat{s}_+^2=(dt^2+\mathrm{cosh}^2td\mathrm{\Omega }_{p1}^2).$$
(32)
This is the case of highest symmetry, when all points on the brane worldsheet are equivalent.
The transverse part of the solution (28) describes an $`n`$-sphere with a solid angle deficit. This may create an impression that the extra dimensions are compactified with a fixed compactification radius $`\sqrt{K}`$. However, this impression is misleading. In the limit $`\eta 0`$, the deficit angle vanishes, and the metric (28),(32) becomes that of $`(p+n)`$-dimensional de Sitter space written in somewhat unfamiliar coordinates . With a more familiar form of de Sitter metric,
$$ds^2=K(dt^2+\mathrm{cosh}^2td\mathrm{\Omega }_{p+n1}^2),$$
(33)
it is clear that all dimensions are equally large and expanding. Spatial sections of the universe are $`(p+n1)`$-spheres, and spatial sections of the brane are $`(p1)`$-spheres of the same radius. So the brane is wrapped around the universe along one of the ‘big circles’. Both the brane and the universe expand exponentially with time. A nonzero $`\eta `$ introduces a deficit angle, but does not change the qualitative character of the spacetime.
We note that de Sitter space also appears to be static in the coordinates
$$ds^2=K(\mathrm{cos}^2\psi dt^2+d\psi ^2+\mathrm{sin}^2\psi d\mathrm{\Omega }_{p+n2}^2).$$
(34)
The reason for this is well known: this coordinate system does not cover the whole spacetime; it covers only the interior of a sphere of radius equal to the de Sitter horizon. Our solution (28),(32) uses a mixed representation in which the metric has a static form like (34) in the transverse dimensions and an expanding form like (33) on the brane. The coordinate system in (28) covers the region from $`\chi =0`$ to the horizon surface $`\chi =\pi /2`$ where the determinant of the metric vanishes, indicating a coordinate singularity.
The metric (28) is somewhat similar to the dilatonic string solution found by Dando and Gregory . They interpreted their solution as describing a string-antistring pair in a universe with compact static transverse dimensions. Our interpretation of (28) is quite different, and we believe a similar interpretation should also apply to the Dando - Gregory solution.
The induced metric on the brane is $`ds_p^2=Kd\widehat{s}^2`$, and the curvature of the brane worldsheet is
$$R_p=\frac{2\mathrm{\Lambda }p(p1)}{(n+p1)(n+p2)}.$$
(35)
This shows that the curvature of the brane is determined only by the cosmological constant $`\mathrm{\Lambda }`$, while the symmetry breaking scale $`\eta `$ affects only the deficit angle in the extra dimensions.
The solution (28) has curvature singularities at $`\chi =0,\pi `$ (since $`T_A^B`$ is singular there), but these singularities are rather mild, and the metric coefficients are non-singular. One should remember that Eq.(28) gives a solution only in the exterior region outside the defect core. But since the metric is well-behaved at $`\chi 0,\pi `$, one can expect that it gives a reasonably accurate representation of the full spacetime in the limit when the defect thickness $`\delta `$ can be neglected, $`\delta \xi _m`$.
### B Class II
We find another solution to the equations (14)-(18) by considering a different ansatz: $`B(\xi )=\xi `$. Again the equations simplify considerably and it is possible to find an analytic solution:
$`A^2(\xi )={\displaystyle \frac{n2\kappa ^2\eta ^2}{n+p2}}{\displaystyle \frac{2\mathrm{\Lambda }}{(n+p2)(n+p1)}}\xi ^2,B(\xi )=\xi `$ (36)
$`\widehat{R}=p(n2\kappa ^2\eta ^2)`$ (37)
As in the previous case, for $`\mathrm{\Lambda }>0`$ we have from the condition $`A^2>0`$ that $`n2\kappa ^2\eta ^2>0`$, and thus $`\widehat{R}>0`$ and $`\xi `$ is constrained to the interval $`0<\xi <\xi _m`$ with
$$\xi _m^2=\frac{(n+p1)(n2\kappa ^2\eta ^2)}{2\mathrm{\Lambda }}$$
(38)
As before, we redefine the radial coordinate as
$$\xi =\xi _m\mathrm{sin}\chi ,$$
(39)
and the metric takes the form
$$ds^2=K\left[d\chi ^2+\stackrel{~}{\alpha }^2\mathrm{sin}^2\chi (d\mathrm{\Omega }_{n1}^2+d\widehat{s}^2)\right]$$
(40)
where $`K`$ is given by Eq.(30),
$$\stackrel{~}{\alpha }^2=\left|\frac{n2\kappa ^2\eta ^2}{n+p2}\right|,$$
(41)
$`d\widehat{s}^2`$ stands for a $`p`$-dimensional spacetime of constant curvature $`\widehat{R}=p(p1)`$, and $`\chi `$ takes values in the interval $`0<\chi <\pi `$.
An unphysical feature of the solution (40) is that the deficit angle does not vanish even for $`\eta =0`$, that is, in the absence of a defect. We have verified that the curvature invariant $`R^{\mu \nu \sigma \tau }R_{\mu \nu \sigma \tau }`$ diverges at $`\chi =0,\pi `$ for $`\eta =0`$. These singularities appear to be unrelated to the defect, and we dismiss class-II solutions as unphysical.
### C Class III
As we mentioned in Section III.A, the ‘conical’ geometry of the extra dimensions is expected to degenerate into a cylinder at some critical value of the symmetry breaking scale $`\eta `$. In order to verify this expectation, we introduce the following ansatz:
$$ds^2=d\xi ^2+C^2d\mathrm{\Omega }_{n1}^2+B(\xi )^2d\widehat{s}^2,$$
(42)
where $`C`$ is a constant radius of the $`(n1)`$-spheres. This is again of the form (10), so Eqs.(11), (12) can be used, and we obtain
$`p{\displaystyle \frac{B^{\prime \prime }}{B}}{\displaystyle \frac{2\mathrm{\Lambda }}{n+p2}}`$ $`=`$ $`0,`$ (43)
$`{\displaystyle \frac{n2\kappa ^2\eta ^2}{C^2}}{\displaystyle \frac{2\mathrm{\Lambda }}{n+p2}}`$ $`=`$ $`0,`$ (44)
$`{\displaystyle \frac{1}{p}}{\displaystyle \frac{\widehat{R}}{B^2}}{\displaystyle \frac{B^{\prime \prime }}{B}}(p1)\left({\displaystyle \frac{B^{}}{B}}\right)^2{\displaystyle \frac{2\mathrm{\Lambda }}{n+p2}}`$ $`=`$ $`0.`$ (45)
For $`\mathrm{\Lambda }=0`$, Eq.(44) gives $`\eta =(n2)^{1/2}\kappa ^1`$, which agrees with the critical value (25). From Eq.(43), $`B^{}=\mathrm{const}`$, and it follows from (45) that the worldsheet curvature $`\widehat{R}`$ can be either positive or zero. For $`\widehat{R}=0`$, $`B=\mathrm{const}`$, and the solution is
$$ds^2=d\xi ^2+C^2d\mathrm{\Omega }_{n1}^2+\eta _{\mu \nu }dx^\mu dx^\nu .$$
(46)
The radius of the cylinder $`C`$ is arbitrary; we expect it to be determined by matching to an appropriate interior solution in the defect core, with the complete geometry being that of a ‘cigar’.
For $`\widehat{R}>0`$ and with a suitable normalization of the radial coordinate, the solution can be written as
$$ds^2=C^2d\mathrm{\Omega }_{n1}^2+d\chi ^2+\chi ^2d\widehat{s}_+^2$$
(47)
with $`d\widehat{s}_+^2`$ from Eq.(32). It can be shown that the last two terms in the metric (47) describe a $`(p+1)`$-dimensional Minkowski space in unfamiliar coordinates . This metric is therefore equivalent to (46).
For $`\mathrm{\Lambda }>0`$, Eq.(44) gives
$$C^2=(n+p2)(n2\kappa ^2\eta ^2)/2\mathrm{\Lambda },$$
(48)
and we find a solution of the form
$$ds^2=C^2d\mathrm{\Omega }_{n1}^2+\omega ^2(d\chi ^2+\mathrm{sin}^2\chi d\widehat{s}_+^2),$$
(49)
where
$$\omega =\sqrt{\frac{2\mathrm{\Lambda }}{p(n+p2)}}.$$
(50)
The last two terms in the metric (49) describe a $`(p+1)`$-dimensional de Sitter space. Note that, in contrast to the $`\mathrm{\Lambda }=0`$ case, solutions now exist for all values of $`\eta <\eta _c`$, while for $`\eta =\eta _c`$ the solution becomes singular, with $`C=0`$. This shows that the flat cylindrical solution (46) with $`\eta =\eta _c`$ is unstable with respect to the introduction of an arbitrarily small cosmological constant $`\mathrm{\Lambda }`$.
## IV Solutions with $`\mathrm{\Lambda }<0`$
The solutions (22) and (36) given in the previous section also allow for negative values of $`\mathrm{\Lambda }`$. There are actually three different possibilities, since now $`n2\kappa ^2\eta ^2`$ can be either positive, negative, or zero.
### A Class I
For $`\mathrm{\Lambda }<0`$ and depending on the sign of $`n2\kappa ^2\eta ^2`$, we can define a new radial coordinate $`\chi `$ as
$$\xi =\sqrt{K}\alpha \mathrm{sinh}\chi ,\sqrt{K}e^\chi ,\sqrt{K}\alpha \mathrm{cosh}\chi $$
(51)
for $`n2\kappa ^2\eta ^2`$ less, equal and greater than zero, respectively. The range for the new coordinate is $`0\chi <\mathrm{}`$ in the first case and $`\mathrm{}<\chi <\mathrm{}`$ in the other two. Then we can write the metric as
$`\widehat{R}<0:`$ $`ds^2=K[d\chi ^2+\alpha ^2\mathrm{sinh}^2\chi d\mathrm{\Omega }_{n1}^2`$ (53)
$`+\mathrm{cosh}^2\chi d\widehat{s}_{}^2],`$
$`\widehat{R}=0:`$ $`ds^2=K\left[d\chi ^2+e^{2\chi }(d\mathrm{\Omega }_{n1}^2+d\widehat{s}_0^2)\right],`$ (54)
$`\widehat{R}>0:`$ $`ds^2=K\left[d\chi ^2+\alpha ^2\mathrm{cosh}^2\chi d\mathrm{\Omega }_{n1}^2+\mathrm{sinh}^2\chi d\widehat{s}_+^2\right].`$ (55)
Here, $`d\widehat{s}_\pm ^2`$ is the metric on a space of constant curvature satisfying (19) with $`\widehat{R}=\pm p(p1)`$, and $`d\widehat{s}_0^2`$ is a Ricci-flat metric. In the case of negative curvature, we can choose, for example, the anti-de Sitter space
$$d\widehat{s}_{}^2=(dt^2+\mathrm{sin}^2t(d\psi ^2+\mathrm{sinh}^2\psi d\mathrm{\Omega }_{p2}^2)).$$
(56)
Flat space metric can be used for $`d\widehat{s}_0^2`$, and the de Sitter metrics (32) can be used for the constant positive curvature space $`d\widehat{s}_+^2`$.
For $`\widehat{R}<0`$, the defect is located at $`\chi =0`$. For $`\widehat{R}=0`$ it is removed to $`\chi =\mathrm{}`$, and for $`\widehat{R}>0`$ there is no defect at all. In the latter case, there is a minimum radius for the $`(n1)`$-spheres in the extra dimensions, $`r_{min}=K\alpha `$. We thus have a wormhole connecting a monopole configuration at $`\chi >0`$ with an antimonopole configuration at $`\chi <0`$.
### B Class II
For the solutions defined by expressions (36) we find a similar situation. With a new coordinate $`\chi `$ defined as in (51), but with $`\alpha `$ replaced by $`\stackrel{~}{\alpha }`$, we have
$`\widehat{R}>0:`$ $`ds^2=K\left[d\chi ^2+\stackrel{~}{\alpha }^2\mathrm{sinh}^2\chi (d\mathrm{\Omega }_{n1}^2+d\widehat{s}_+^2)\right]`$ (57)
$`\widehat{R}=0:`$ $`ds^2=K\left[d\chi ^2+e^{2\chi }(d\mathrm{\Omega }_{n1}^2+d\widehat{s}_0^2)\right]`$ (58)
$`\widehat{R}<0:`$ $`ds^2=K\left[d\chi ^2+\stackrel{~}{\alpha }^2\mathrm{cosh}^2\chi (d\mathrm{\Omega }_{n1}^2+d\widehat{s}_{}^2)\right].`$ (59)
Once again, the metric (57) is singular at $`\chi =0`$ even in the absence of a defect $`(\eta =0)`$, and we dismiss this solution as unphysical.
### C Class III
We finally consider the cylindrical metric ansatz (42). The solutions of Eqs.(43)-(45) for $`\mathrm{\Lambda }<0`$ have the form
$`\widehat{R}>0:`$ $`ds^2=C^2d\mathrm{\Omega }_{n1}^2+\omega ^2(d\chi ^2+\mathrm{sinh}^2\chi d\widehat{s}_+^2),`$ (60)
$`\widehat{R}<0:`$ $`ds^2=C^2d\mathrm{\Omega }_{n1}^2+\omega ^2(d\chi ^2+\mathrm{cosh}^2\chi d\widehat{s}_{}^2),`$ (61)
$`\widehat{R}=0:`$ $`ds^2=C^2d\mathrm{\Omega }_{n1}^2+d\chi ^2+e^{\pm 2\omega \chi }d\widehat{s}_0^2,`$ (62)
where
$`\omega =\sqrt{{\displaystyle \frac{2\mathrm{\Lambda }}{p(n+p2)}}},`$ (63)
$`C^2=(n+p2)(\kappa ^2\eta ^2(n2))/2\mathrm{\Lambda }.`$ (64)
Of greatest interest are the flat brane solutions (62) which generalize the solutions considered by Gregory in the $`n=2`$ case. The geometry of the extra dimensions in the metric (62) is that of a cylinder with a cross-section being an $`(n1)`$-sphere of a fixed radius $`C`$. It would be interesting if this solution could be matched to an appropriate interior solution, so that the complete geometry is that of a ‘cigar’. Gregory has argued that this is possible for $`n=2`$, but her analysis does not directly apply to $`n3`$.
Cigar-like defect solutions with an exponential warp factor would be of interest, since they would have features similar to those of the Randall-Sundrum geometry. If the brane is located at $`\chi =0`$ and the asymptotic metric is given by (62) with a negative sign in the exponential, then the volume of the extra dimensions would be finite, despite their infinite extent in the $`\chi `$-direction. As in the Randall-Sundrum case, most of the volume would be concentrated near the brane, and one can expect that gravitons would be effectively confined to the brane.
The right-hand side of (64) should be positive, so we must have $`\kappa ^2\eta ^2(n2)>0`$. While this does not give any additional information for $`n=2`$, this condition requires a super-Planckian symmetry breaking scale, $`\eta >\kappa ^1`$, for the defects when $`n>2`$.
## V Instanton solutions
Euclidean continuations of brane-world solutions are of interest, since they can be interpreted as gravitational instantons describing quantum nucleation of a brane-world. The nucleation probability is given by
$$𝒫e^{\pm |S|},$$
(65)
where $`S`$ is the instanton action. The choice of sign in the exponential is determined by the choice of boundary conditions for the wave function of the universe. The lower sign is chosen for the tunneling and Linde boundary conditions, and the upper sign for the Hartle-Hawking boundary condition . For definiteness we shall adopt the tunneling boundary condition below.
For the instantons to give a nonvanishing contribution to the nucleation probability, they must have a finite action, with instantons of the smallest absolute value of the action giving the dominant contribution. The action is typically extremized for solutions of the highest symmetry, so we shall consider instantons with $`d\widehat{s}_+^2`$, $`d\widehat{s}_{}^2`$ and $`d\widehat{s}_0^2`$ being maximally symmetric spaces of positive, negative and zero curvature, that is, Euclidean de Sitter, anti-de Sitter, and flat spaces, respectively.
The Euclidean action for our model is given by
$$S=\frac{1}{2\kappa ^2}d^{(n+p)}x\sqrt{g}[R2\mathrm{\Lambda }2\kappa ^2L(\varphi )],$$
(66)
where $`R`$ is the $`D`$-dimensional scalar curvature and $`L(\varphi )`$ is the scalar field Lagrangian. We can eliminate $`R`$ by making use of Einstein’s equations to obtain
$$R=2\kappa ^2L(\varphi )+\frac{2(n+p)}{n+p2}\mathrm{\Lambda }$$
(67)
and
$$S=\frac{\mathrm{\Lambda }}{\kappa ^2(n+p2)}d^{n+p}x\sqrt{g}$$
(68)
For class-I and class-II solutions with $`\mathrm{\Lambda }<0`$, the volume of the transverse space is infinite, and $`|S|=\mathrm{}`$. If cigar-like class-III solutions exist, they may have a finite transverse volume, but the action is still infinite due to the divergence of the $`p`$-dimensional volume of the flat brane worldsheet. Hence, we only need to consider solutions with $`\mathrm{\Lambda }>0`$. In this case the curvature of the brane must be positive, $`\widehat{R}>0`$, and thus the metric $`d\widehat{s}^2`$ should be that of a Euclidean de Sitter space, that is, a $`p`$-sphere:
$$ds_E^2=K\left[d\chi ^2+\alpha ^2\mathrm{sin}^2\chi d\mathrm{\Omega }_{n1}^2+\mathrm{cos}^2\chi (d\psi ^2+\mathrm{sin}\psi ^2d\mathrm{\Omega }_{p1}^2)\right].$$
(69)
One can model the nucleation of a closed universe with a brane by allowing $`\psi `$ to vary in the interval $`[0,\pi /2]`$ in the Euclidean region and then continuing it in the imaginary direction in the Lorentzian region, $`\psi =\pi /2+it`$. This turns (69) into the metric
$$ds^2=K\left[d\chi ^2+\alpha ^2\mathrm{sin}^2\chi d\mathrm{\Omega }_{n1}^2+\mathrm{cos}^2\chi (dt^2+\mathrm{cosh}^2td\mathrm{\Omega }_{p1}^2)\right]$$
(70)
describing an expanding braneworld.
We can easily calculate the action for the instanton solution (69):
$`S`$ $`=`$ $`{\displaystyle \frac{1}{2\kappa ^2}}{\displaystyle \frac{V_p}{2}}V_{(n1)}K^{(n+p)/2}\alpha ^{n1}{\displaystyle _0^\pi }𝑑\chi |\mathrm{cos}\chi |^p(\mathrm{sin}\chi )^{(n1)}{\displaystyle \frac{4\mathrm{\Lambda }}{n+p2}}`$ (71)
$`=`$ $`{\displaystyle \frac{4}{\kappa ^2}}{\displaystyle \frac{\sqrt{\pi }^{(n+p+1)}}{\mathrm{\Gamma }[(n+p1)/2]}}K^{(n+p2)/2}\alpha ^{n1}.`$ (72)
where $`V_k`$ stands for the volume of a k-sphere of unit radius, that is, $`V_k=2\pi ^{(k+1)/2}/\mathrm{\Gamma }[(k+1)/2]`$.
Apart from nucleation of the entire brane-world, the instanton (69) can also describe nucleation of spherical branes in an inflating $`(n+p)`$-dimensional de Sitter space. The situation here is very similar to the nucleation of circular loops of string and of spherical domain walls in a $`(3+1)`$-dimensional de Sitter space, as discussed by Basu et. al. . The nucleation rate is given by
$$\mathrm{\Gamma }e^B$$
(73)
with
$$B=SS_0,$$
(74)
where $`S`$ is the instanton action and $`S_0`$ is the action for the Euclidean de Sitter space without a brane. From Eq.(72) we have
$$B=\frac{4}{\kappa ^2}\frac{\pi ^{(n+p+1)/2}}{\mathrm{\Gamma }[(n+p1)/2]}K^{(n+p2)/2}(1\alpha ^{n1}).$$
(75)
The initial radius of the brane is $`r=\sqrt{K}`$. After nucleation, it is stretched by the exponential expansion of the universe.
## VI Conclusions
In this paper we have found a number of solutions describing global defects in a higher-dimensional space. We assumed that the core of the defect is centered on a $`(p1)`$-dimensional brane and concentrated on the case when the number of extra dimensions is $`n3`$.
In the absence of a cosmological constant, we found that for all $`n3`$ the defect solution is very similar to that for a global monopole . The brane worldsheet is flat, and there is a solid angle deficit in the extra dimensions. This is rather surprising, considering the fact that solutions are very different for $`n=1`$ and $`n=2`$. The maximal solid angle deficit is reached at the critical value $`\eta _c=(n2)^{1/2}\kappa ^1`$, when the transverse metric becomes that of a cylinder.
For a positive cosmological constant, $`\mathrm{\Lambda }>0`$, our solutions describe spherical branes in an inflating higher-dimensional universe. In the limit $`\eta 0`$, when the gravitational effect of the defect can be neglected, the universe can be pictured as an expanding $`(p+n1)`$-dimensional sphere with a brane wrapped around it in the form of a sphere of lower dimensionality $`(p1)`$. A nonzero $`\eta `$ introduces a deficit angle in the dimensions orthogonal to the brane worldsheet. It is interesting that the expansion rate of the universe (and of the brane) is independent of the symmetry breaking scale $`\eta `$ and is determined only by $`\mathrm{\Lambda }`$, while the deficit angle is determined by $`\eta `$ and independent of $`\mathrm{\Lambda }`$. Gravitational instantons obtained by a Euclidean continuation of this class of solutions have the geometry of a $`(p+n)`$-sphere with the brane represented by a maximal $`p`$-sphere and with a deficit solid angle in the dimensions transverse to the brane. These instantons can be interpreted as describing quantum nucleation either of the entire brane-world, or of a spherical brane in an inflating $`(p+n1)`$-dimensional universe.
Another class of solutions has curvature singularities even in the absence of a defect $`(\eta =0)`$, and we have dismissed such solutions as unphysical.
The third class of solutions has the geometry of a $`(p+1)`$-dimensional de Sitter space, with the remaining $`(n1)`$ dimensions having the geometry of a cylinder.
We have also found 3 classes of solutions for $`\mathrm{\Lambda }<0`$. The first two are essentially analytic continuations of the positive-$`\mathrm{\Lambda }`$ solutions. The third class is similar to Randall-Sundrum $`(n=1)`$ and Gregory $`(n=2)`$ solutions, exhibiting an exponential warp factor. If solutions of the third class can be matched to appropriate interior solutions in the defect core, one may be able to use them as a basis for realistic brane-world models.
## VII Acknowlegements
We are grateful to Gia Dvali, Jaume Garriga and Ruth Gregory for useful discussions and comments on the manuscript. This work was supported in part by the Basque Government under fellowship number BFI.99.89 (I.O.) and by the National Science Foundation (A.V.). |
warning/0003/cond-mat0003200.html | ar5iv | text | # Bond-order and charge-density waves in the isotropic interacting two-dimensional quarter-filled band and the insulating state proximate to organic superconductivity
## I Introduction
Theoretical discussions of spatial broken symmetries in strongly correlated electron systems have largely focused on the 1/2-filled band Mott-Hubbard semiconductor. The one-dimensional (1D) case has been widely discussed in the context of polyacetylene . Here it is known that Coulomb electron-electron (e-e) interactions can strongly enhance the 2k<sub>F</sub> (k<sub>F</sub> = one-electron Fermi wavevector) bond-alternation expected within the Peierls purely electron-phonon (e-ph) coupled model, giving rise to a periodic modulation of the bond-order, a bond-order wave (BOW). In the limit of very strong on-site Coulomb interaction, the BOW instability is usually referred to as the spin-Peierls (SP) instability. In the presence of intersite Coulomb interactions, and for certain relative values of the on-site and intersite interaction parameters, a charge-density wave (CDW), periodic modulation of the site charge density, can be the dominant instability . The BOW and the CDW occur in largely nonoverlapping regions of the parameter space and compete against each other . True antiferromagnetism (AFM)—ie, a long-range order (LRO) 2k<sub>F</sub> spin-density wave (SDW)—is absent in for spin-rotationally invariant models in 1D, and the ground state is dominated by singlet spin coupling, which favors the BOW over the SDW. Two-dimensionality is thus essential for the SDW.
The 1/2-filled isotropic two-dimensional (2D) case has been investigated in great detail in recent years (mostly for the case of large intrasite Hubbard interaction but zero intersite interaction) , as this limiting case is known to describe the parent semiconductor compounds of copper-oxide based high temperature superconductors. The BOW instability that characterises the 1D chain is destabilized in 2D by Coulomb interaction , and the dominant broken symmetry here is the 2k<sub>F</sub> SDW, with periodic modulation of the spin density. Most recently, it has been demonstrated that this SDW state appears for the smallest nonzero interchain hopping in weakly coupled 1/2-filled band chains , in agreement with previous renormalization group calculations . As in 1D , there is no CDW-SDW coexistence in 2D . The absence of coexistence between the BOW and SDW for the 1/2-filled band in both 1D and 2D can be readily understood intuitively: the BOW requires spin-singlet coupling between alternate nearest neighbor spins, which clearly has to disappear in the SDW. An alternate way of viewing this is to observe that the probability of charge-transfer to the left and to the right in the AFM are exactly equal, and therefore the SDW cannot coexist with the BOW. On the other hand, both the BOW and the SDW require that the site-occupancies by electrons are strictly uniform, and thus neither the 1D BOW nor the 2D SDW will coexist with the CDW.
Coupled 1/2-filled band chains have also been discussed within the context of the so-called ladder systems . Whether or not a given n-leg ladder system, for small n, exhibits the BOW now depends on whether n is odd or even. This feature of the ladder systems could have been anticipated from the physics of the odd versus even S Heisenberg chains . Thus at least for the simplest monatomic lattices, ground states of the 1/2-filled band are known: the BOW, CDW and SDW phases compete against one another and do not coexist, and 2D behavior emerges for the smallest 2D coupling.
In contrast to the 1/2-filled band, broken symmetries in non-1/2-filled bands with strong e-e interactions have been investigated primarily in 1D limit or at most in the quasi-1D regime of weak interchain coupling. This emphasis likely arises from the theoretical preconception that finite one-electron hopping between chains destroys the nesting feature that characterizes the 1D limit, leading necessarily to the restoration of the metallic phase . A recent work has examined coupled chains in the limit of weak e-e interactions . The weak-coupling approximation employed in reference reproduces the loss of nesting predicted within band theory. While the continuum renormalization group calculations predicted CDW-SDW coexistence for incommensurate bandfillings, early quantum Monte Carlo calculations for the 1/4-filled band failed to find this coexistence . Many more recent numerical simulations on discrete finite systems assume the absence of coexistence between the 2k<sub>F</sub> BOW, the 2k<sub>F</sub> CDW and the 2k<sub>F</sub> SDW that characterizes that 1/2-filled band also applies to the non-1/2-filled bands. Indeed, it is often assumed that the CDW is driven by the e-ph interactions and the SDW by e-e interactions and that their effects are competing. This assumption is made despite the result mentioned above that already in the simplest case of the 1D 1/2-filled band, e-e and e-ph interaction effects are known not to be competing but to act in a co-operative way to give the enhanced 2k<sub>F</sub> BOW .
Recently, we have begun a systematic study of the nature of the broken symmetry ground states in the 2D 1/4-filled band on an anisotropic rectangular lattice with both e-ph and e-e interactions. Earlier work by us had already established the cooperative coexistence between the BOW and the period 4 “2k<sub>F</sub>” CDW in the
1D 1/4-filled band, with each broken symmetry enhancing the other, for both noninteracting and interacting electrons. The latter results have been subsequently confirmed by Riera and Poilblanc. In the more recent work we have demonstrated an apparently unique feature of the 1/4-filled band: namely, the coexistence of the BOW-CDW with the period 4 “2k<sub>F</sub>” SDW, giving rise to a coupled Bond-Charge-Spin density wave (BCSDW) that appears for weak interchain electron transfer between chains.
In the present paper, we extend our calculations to the full range of anisotropies, from uncoupled chains to an isotropic 2D lattice. We include both the SSH intersite phonons that drive a BOW and the Holstein phonons that drive a CDW . We list three primary motivations for this extension. First, the cooperative coexistence between the BOW and the 2k<sub>F</sub> SDW found in the 1/4-filled band for weak interchain transfer is exactly opposite to the competition between the 2k<sub>F</sub> BOW and the 2k<sub>F</sub> SDW (with the latter dominating for nonzero interchain transfer) in the 1/2-filled band. It is then immediately natural to ask what the nature of the ground state is for strong interchain hopping of electrons in the 1/4-filled band. Second, from a more general theoretical perspective, whether or not the vanishing of density waves that is predicted by one-electron nesting ideas remains true for strongly correlated electrons is of considerable general interest. Finally, our results are likely to have relevance to experimental observations in the organic CTS, including those that exhibit superconductivity .
Our investigations yield the surprising result that the coexisting Bond-Charge density wave (BCDW) persists as the ground state of the strongly correlated 1/4-filled band in 2D for all values of the interchain electron transfer, including the isotropic limit. We show that this result can be understood physically as a consequence of interchain confinement arising from strong intrachain Coulomb interactions . The SDW component of the BCSDW, on the other hand, attains a maximum amplitude at some intermediate interchain transfer, after which it typically vanishes at a critical value of the transfer.
In order to discuss applications of results to real materials, including the 2:1 cationic CTS, we need to clarify an important aspect of our approach vis-a-vis most previous work on models of these materials. In our above discussion of band-filling, “1/4-filled” is defined in the usual manner: namely, in the absence of the BCDW, the lattice is uniform in at least one direction, and the average density of electrons per site is 1/2. In real materials, crystal structure effects often cause a lattice dimerization that is unrelated to any underlying electronic or magnetic instability (see below) . As shown in Fig. 1(a), this dimerization leads to a gap in the single electron spectrum at $`k_F=\pi /2a`$, and consequently suggests using an effective 1/2-filled band model that focuses on the upper subband. In real space terms, this approximation amounts to considering the system as a set of (tightly bound) dimers (i.e., a diatomic lattice) with one electron per dimer site, as shown in Figure. 1(b). This approach has been widely applied, particularly with considerable success in the context of the magnetic field-induced spin density wave (FISDW) in 2:1 salts of TMTSF . As we show below, a further dimerization of the dimer lattice is unconditional in both 1D (the well-known spin-Peierls transition) and 2D (a surprising new result), and that this dimerization of the dimer lattice leads spontaneously to different electronic populations on the sites within a dimer, i.e., to the same 2k<sub>F</sub> CDW that occurs in the 1/4-filled (monoatomic) band (see Fig. 1(c)). For small interchain electron transfer, the BCSDW will therefore have nearly the same structure as the original 1/4-filled band. This is a third new result, perhaps also surprising, and shows that the number of electrons per site within a unit cell is a more fundamental parameter than the bandfilling: the latter is strictly a one-electron concept of limited use in the interacting electron picture.
We expect our results to be relevant for the 1D semiconductors (TMTTF)<sub>2</sub>X, the so-called “quasi-1D” organic superconductors (TMTSF)<sub>2</sub>X, as well as the 2D organic superconductors (BEDT-TTF)<sub>2</sub>X and the more recently synthesized (BETS)<sub>2</sub>X . In reference we showed that the highly unusual “mixed CDW-SDW state” found in (TMTTF)<sub>2</sub>Br, (TMTSF)<sub>2</sub>PF<sub>6</sub> and (TMTSF)<sub>2</sub>AsF<sub>6</sub> can be explained naturally as the BCSDW state within the strongly correlated 1/4-filled band scenario. Our current work shows that dimerization of the dimer lattice leads to the same results, and hence the weak high temperature dimerization along the stack axis is effectively irrelevant: starting from either the 1/4-filled model or the effective 1/2-filled scenario, the final outcome is the same .
With these comments complete, we can describe the organization of the remainder of the paper. In Section II we introduce our model Hamiltonian, as well as that of the dimerized dimer model. In Section III, we present physical, intuitive arguments, based on a configuration space picture of broken symmetry that predict both the BCSDW for weak interchain electron transfer and the persistent BCDW state in the isotropic limit. In Section IV, we present the results of extensive numerical studies, exploring behavior in both the strict 1D limit and for the full range of anisotropies in the quasi-2D case. These studies, in confirmation of the qualitative predictions of Section III: (i) establish the persistence of the BCDW up to the isotropic limit; (ii) suggest the occurrence of two quantum critical transition as an SDW first appears for weak transverse hopping and then disappears for the nearly isotropic case; and (iii) prove the equivalence of the 1/2-filled dimerized dimer and 1/4-filled monatomic lattices. For clarity, in Section V we summarize our theoretical conclusions; readers not interested in the underlying physical arguments or numerical details can skip directly to this summary in Section V. In Section VI, we examine in some detail several recent experiments that indicate the applicability of our theory to the insulating states that are observed to be proximate to the superconducting states in the organic CTS. Finally, in Section VII, we indicate possible future directions for our research, focusing on commensurability defects in the BCDW state and their possible role in the proximate superconducting phases. We point out several intriguing similarities between this potential microscopic mechanism for superconductivity and other recent phenomenological models. We conclude the article with three appendices, which deal with various more technical arguments and details of the numerical methods.
## II Models and Observables
We consider two different extended Peierls-Hubbard Hamiltonians on a rectangular lattice 2D with (in general) anisotropic electron hopping. The first model describes a monatomic 1/4-filled band and is defined by the Hamiltonian
$$H=H_0+H_{ee}+H_{inter}$$
$`(1a)`$
$$H_0=\underset{j,M,\sigma }{}[t\alpha (\mathrm{\Delta }_{j,M})]B_{j,j+1,M,M,\sigma }+\beta \underset{j,M}{}v_{j,M}n_{j,M}$$
$$+K_1/2\underset{j,M}{}(\mathrm{\Delta }_{j,M})^2+K_2/2\underset{j,M}{}v_{j,M}^2$$
$`(1b)`$
$$H_{ee}=U\underset{j,M}{}n_{j,M,}n_{j,M,}+V\underset{j,M}{}n_{j,M}n_{j+1,M}$$
$`(1c)`$
$$H_{inter}=t_{}\underset{j,M,\sigma }{}B_{j,j,M,M+1,\sigma }$$
$`(1d)`$
In the above, $`j`$ is a site index, $`M`$ is a chain index, $`\sigma `$ is spin, and we assume a rectangular lattice . As $`t_{}`$ varies from 0 to $`t`$, the electronic properties vary from 1D to 2D. An implicit parameter in the above Hamiltonian is the bandfilling, or more precisely $`\rho `$. We shall focus on the 1/4-filled case, for which $`\rho `$ = 1/2. In applications to the organic CTS, each site is occupied by a single organic molecule, the displacement of which from equilibrium is described by $`u_{j,M}`$ (with $`\mathrm{\Delta }_{j,M}=(u_{j+1,M}u_{j,M})`$); $`v_{j,M}`$ is an intra-molecular vibration, $`n_{j,M,\sigma }=c_{j,M,\sigma }^{}c_{j,M,\sigma }`$, $`n_{j,M}=_\sigma n_{j,M,\sigma }`$, and $`B_{j,k,L,M,\sigma }[c_{j,L,\sigma }^{}c_{k,M,\sigma }+h.c.]`$, where $`c_{j,L,\sigma }^{}`$ is a Fermion operator. We treat the phonons in the adiabatic approximation and are interested in unconditional broken symmetry solutions that occur for e-ph couplings $`(\alpha ,\beta )0^+`$. All energies such as $`U`$, $`V`$, and $`t_{}`$ will be given in units of the undistorted intra-chain hopping integral $`t`$.
The second model describes a diatomic/dimer lattice, with one electron per dimer. The Hamiltonian for this case is similar to that above, with identical $`H_{ee}`$ and $`H_{inter}`$, but with modified intrachain one-electron term $`H_0^{}`$,
$`H_0^{}`$ $`=`$ $`t_1{\displaystyle \underset{j,M,\sigma }{}}B_{2j1,2j,M,M,\sigma }`$ (3)
$`{\displaystyle \underset{j,M,\sigma }{}}[t_2\alpha \mathrm{\Delta }_{j,M}]B_{2j,2j+1,M,M,\sigma }+{\displaystyle \frac{K}{2}}{\displaystyle \underset{j,M}{}}(\mathrm{\Delta }_{j,M})^2`$
In the above each pair of sites (2j–1,M) and (2j,M) forms a dimer with fixed hopping $`t_1>t`$ between them, $`\mathrm{\Delta }_{j,M}=(u_{2j+1,M}u_{2j,M})`$, with $`u_{2j1,M}=u_{2j,M}`$; this means that there is no modulation of the intradimer bond length, and the dimer unit is displaced as a whole. As written, the model assumes an “in-phase” 2D arrangement of the dimer units (i.e., dimers on different chains lie directly above one another), which we have determined to be the lower energy configuration for both zero and nonzero $`\mathrm{\Delta }_{j,M}`$. Notice that $`H_0^{}`$ does not contain the Holstein on-site e-ph coupling. Nevertheless, we will show that a site-diagonal CDW is a consequence of the BOW here.
The broken symmetries we are interested in are (i) the BOW, with periodic modulations of the intrachain nearest neighbor bond order $`_\sigma B_{j,j+1,M,M,\sigma }`$; (ii) the CDW, with periodic modulations of the site charge-density $`n_{j,M}`$; and (iii) the SDW, with periodic modulations of the site spin-density $`n_{j,M,}n_{j,M,}`$. Note that in case of the dimer lattice (Eq. (2)) we are interested in both intra- and interdimer charge and spin modulations, although bond modulations can occur only between dimers. Furthermore, in the CDW and the SDW the modulations of the site-based densities occur along both longitudinal and transverse directions (though not necessarily with the same periodicities, see below). In case of the BOW, a complete description would require the determination of the phase difference between consecutive chains.
## III Configuration space picture of spatial broken symmetry
The physical arguments presented in this section provide crucial insights that allow us to anticipate the apparently counterintuitive results of this paper. The need to develop such arguments arises from the limitations inherent in all true many-body numerical simulations of strong correlated electron systems: namely, one can study only systems of limited size and distinguishing finite-size artifacts from true results requires physical understanding. In turn, true many-body numerical methods are essential here because of the intermediate magnitude of the e-e interactions (comparable to the bandwidths) in the organic CTS, which renders both mean field and perturbation theoretic approaches questionable. For instance, even in the strictly 1D limit, where well-established RG and bosonization techniques have existed for decades, for the intermediate coupling regime, there have recently been some surprising discoveries in the phase diagram of the extended Hubbard model. In 2D, developing a clear physical intuition is still more crucial, as numerically tractable lattices are even farther from the thermodynamic limit, and the competition among broken symmetries is likely to be more subtle. Brief presentations of these physical ideas for $`t_{}`$ = 0 and t$`{}_{}{}^{}<<t`$ have been made previously. Here we discuss these ideas for the complete range 0 $``$ t$`{}_{}{}^{}t`$, focusing on (i) the transition from 1D to 2D, and (ii) the difference from the 1/2-filled band monatomic lattice.
A physical picture of spatial broken symmetry in strongly correlated electron systems must necessarily be based on configuration space ideas, as one-electron bands have simply ceased to exist for strong e-e interaction. Within the configuration space picture of broken symmetry , each broken symmetry state, independent of band-filling, can be associated with a small number of equivalent configurations that are related by the symmetry operator in question. For commensurate $`\rho `$, these configurations are easily determined by inspection. The relevant configurations consist of repeat units which themselves possess the same periodicity as the density wave. For illustration, we choose the 1D 1/2-filled band. In this case, each broken symmetry has two extreme configurations, the pairs corresponding to the SDW, BOW and CDW being, respectively: the two Néel states $`\mathrm{}\mathrm{}`$ and $`\mathrm{}\mathrm{}`$ (SDW); the two nearest neighbor valence bond diagrams (1,2)(3,4)(5,6)….(N – 1,N) and (N,1)(2,3)(4,5)….(N – 2, N – 1) (where (i,j) is a spin singlet bond between sites i and j and N is the number of sites) (BOW); and the configurations …202020… and …020202…(where the numbers denote site occupancies) (CDW). N applications of the one-electron hopping term in Eq. (1) on any one extreme configuration (corresponding to a given broken symmetry) generates the other extreme configuration, but for N $`\mathrm{}`$ this mixing of configurations is small, and the ground state resembles one or the other of the extreme configurations qualitatively, with reduced spin moment, bond order or charge-density difference due to quantum fluctuations .
The key insight of the configuration space heuristics is that the qualitative effects of many-body Coulomb interactions, as well as additional one-electron terms, can be deduced from their effects on any one of the extreme configurations . As a trivial example of this, a repulsive Hubbard $`U`$ destroys the CDW in the 1/2-filled band, simply because double occupancies in the extreme configuration …202020… “cost” prohibitively high energy. Significantly, in the 1/2-filled band, the extreme configurations favoring the SDW, the BOW and the CDW are different, and there is a complete lack of overlap between them. This essentially guarantees the absence of coexistence among these broken symmetries in both 1D and 2D.
To apply these ideas to the 1D 1/4-filled band, we begin by considering the on-site charge configurations. A 2k<sub>F</sub> (4k<sub>F</sub>) density wave here has period 4 (2) in configuration space. As discussed above, the extreme configurations of interest must also have period 4 or 2, and there are then only three distinct sets of extreme charge configurations. These contain the repeat units …2000…, …1100…, and …1010…, respectively, where the numbers again denote site occupancies. There are four distinct configurations for sets 1 (…2000…) and 2 (….1100….), whereas there are only 2 for set 3 (….1010….). By analogy with the 1/2-filled band (see above), we now introduce spins and note that configurations belonging to sets 2 and 3 can again have spin singlet bonds between pairs of nearest neighbor singly occupied sites, or the spins of the occupied sites can alternate as in the 1/2-filled band Néel configurations. Let us now show, by considering the different cases separately, how e-e interactions affect these configurations and how an understanding of these effects suggests (correctly!) the broken symmetries to be studied.
### A 1/4-filled band, $`t_{}`$ = 0, $`U=V`$ = 0
The non-interacting case provides a simple example to introduce some of the important differences between the 1/4-filled and 1/2-filled bands. Actual calculation indicates that within the 1D Holstein model the charge densities $`\rho _j`$ on the sites have the functional form
$$\rho _j=0.5+\rho _0\mathrm{cos}(2k_Fja)=0.5+\rho _0\mathrm{cos}(\pi j/2)$$
(4)
This charge density pattern could have been anticipated by focusing on the extreme configuration …2000…, which also predicts three different charge densities (large, intermediate, small and intermediate), since each ‘0’ that is immediately next to a ‘2’ is different from the other pair of sites labeled ‘0’ that are further away from the ‘2’. occupancy scheme …2000…, the probabilities of charge-transfer between a ‘2’ and the two neighboring ‘0’s are larger than that between the two neighboring ‘0’s themselves. For nonzero $`\alpha `$ in Eq. (1), this difference in charge-transfers leads to lattice distortion of the form
$$u_j=u_0\mathrm{cos}(2k_Fja)=u_0\mathrm{cos}(\pi j/2),$$
(5)
with bonding pattern “SSWW” (for strong, strong, weak, weak), where a strong (weak) bond has hopping $`t_S>t`$ ($`t_W<t`$). This then is one very important difference from the 1/2-filled band: whereas in the 1/2-filled band differences in bond-orders arise from spin-effects only (the probability of charge-transfer is greater between nearest neighbor singlet-coupled sites than between nearest neighbor non-bonded sites ), in non-1/2-filled bands this difference can also originate from site occupancies. Precisely because the BOW and the CDW here are both derived from the same extreme configuration, they coexist in the noninteracting 1/4-filled band .
### B 1/4-filled band, $`t_{}`$ = 0, $`U,V>`$ 0
For nonzero (positive) $`U`$ and $`V`$, the interplay among the various possible broken symmetries becomes both more subtle and more interesting. Since double occupancies “cost” energy, the extreme configuration …2000… is suppressed even at a relatively small $`U`$ . For the strongly correlated ($`U\mathrm{}`$) 1D 1/4-filled band with convex long range interactions, Hubbard showed that there exist two different Wigner crystals, with occupancy schemes …1100… and …1010…. At first glance, the extreme configuration …1010…, corresponding to a period 2 “4k<sub>F</sub>”CDW , appears to be strongly preferred, but in fact more careful analysis shows that it dominates the ground state only for fairly substantial $`V`$ . This can be seen rigorously for $`U\mathrm{}`$, where the 1/4-filled spinful band can be mapped rigorously to the 1/2-filled spinless band , which in turn can be mapped (via a Jordan-Wigner transformation) to an anisotropic Heisenberg spin 1/2 chain . Using this approach, one finds that the period 2 “4k<sub>F</sub>” CDW becomes the ground state only for $`V>V_c=2`$ (in units of $`|t|`$) . For finite $`U`$, numerical results show that $`V_c`$ is slightly larger than $`2`$. Given the estimated values of $`V`$ in the organic CTS, it seems unlikely that they will exhibit this (…1010…) intrachain ordering. This expectation is strongly supported by the result that the …1010… CDW cannot coexist with the BOW , whereas the (TMTTF)<sub>2</sub>X are known to exhibit a low-temperature transition to a SP-BOW ground state .
For $`V<V_c`$ the extended 1D Hubbard model at 1/4-filling is a Luttinger liquid that is also susceptible to a 2k<sub>F</sub> bond and charge distortion, and it is this distortion that can be described by any one of the four equivalent configurations …1100…. The 2k<sub>F</sub> CDW compatible with the …1100… configuration has the form
$`\rho _c(j)`$ $`=`$ $`0.5+\rho _0\mathrm{cos}(2k_Fja3\pi /4)`$ (6)
$`=`$ $`0.5+\rho _0\mathrm{cos}(\pi j/23\pi /4),`$ (7)
This particular CDW also coexists with a BOW, since the charge-transfer across a ‘1 – 1’ bond is different from that across a ‘1 – 0’ (or ‘0 – 1’) bond, which again is different from the charge-transfer across a ‘0 – 0’ bond. It is a subtle but crucial fact, confirmed by earlier numerical studies, that this same CDW can now promote two different BOWs, each with three different bond strengths. In each of these the ‘0 - 0’ bond is the weakest, but depending upon the strength of the Coulomb interaction, the ‘1 - 1’ bond can be stronger than a ‘1 – 0’ (or ‘0 – 1’) bond (since charge-transfer in the former can occur in both directions), but it can also be weaker (since charge-transfer in the former leads to double occupancy, while no double occupancy is created in the charge transfer between a ‘1’ and a ‘0’). Consistent with this and the numerical results, we shall refer to the first bonding pattern as “SUWU” (for a strong ‘1 – 1’ bond, undistorted ‘1 –0’ bond, weak ‘0 – 0’ bond, followed by an undistorted ‘0 – 1’ bond), where a strong bond has $`t_S>t`$, an undistorted bond has $`t_U=t`$, and a weak bond has $`t_W<t`$. This BOW has pure period 4 “2k<sub>F</sub>” periodicity and is accompanied by lattice distortion
$$u_j=u_0\mathrm{cos}(2k_Fja\pi /4)=u_0\mathrm{cos}(\pi j/2\pi /4).$$
(8)
Again consistent with the numerical results, we call the second bonding pattern “WSWS” (for a stronger weak ‘1 – 1’ bond, strong ‘1 – 0’ bond, weak ‘0 – 0’ bond and strong ‘0 – 1’ bond, with $`t_S>t>t_W^{}>t_W`$). Interestingly, the WSWS bonding pattern is a superposition of the pure 2k<sub>F</sub> period 4 SUWU structure and the pure 4k<sub>F</sub> period 2 SWSW structure and is accompanied by lattice distortion
$`u_j`$ $`=`$ $`u_0[r_{2k_F}\mathrm{cos}(2k_Fja\pi /4)+r_{4k_F}\mathrm{cos}(4k_Fja)]`$ (9)
$`=`$ $`u_0[r_{2k_F}\mathrm{cos}(\pi j/2\pi /4)+r_{4k_F}\mathrm{cos}(\pi j)],`$ (10)
where $`r_{2k_F}`$ and $`r_{4k_F}`$ are the relative weights of the 2k<sub>F</sub> and 4k<sub>F</sub> bond distortions, respectively . These results were established numerically in reference , where from comparisons to available experimental data in the 1:2 anionic TCNQ systems it was also shown that the phase relationship between the coexisting 2k<sub>F</sub> CDW and the WSWS BOW (the W bond connects sites with greater charge densities than the W bond) is precisely in agreement with theory.
Very importantly, we show below that the dimerization of the dimer lattice with one electron per dimer also leads to a WSWS bonding pattern (see Fig. 1(c)), which in its turn promotes the site occupancy scheme …1100…. This coexistence will therefore occur in either the full 1/4-filled band model or the effective 1/2-filled, dimerized dimer approach.
### C 1/4-filled band, $`t_{}<<t`$, $`U,V`$ 0.
The above two BOW-CDWs describe the ground state of the interacting 1/4-filled band in the limit of $`t_{}`$ = 0, where the ‘1 - 1’ bond is a singlet. As in the 1/2-filled band though, singlets are expected to give way to SDW order for $`t_{}0`$. Thus we must understand the role of the spin degrees of freedom. Once specific spins are assigned to the sites labeled ‘1’ in the …1100… configuration, the sites labeled ‘0’ become distinguishable, as a given ‘0’ site is now closer to one particular ‘1’ (up or down) than the other . In this case the ‘0’ site is expected to acquire the spin characteristic of its neighboring ‘1’. The charge and spin along a chain can now thus be denoted as $`,`$,$`,`$, where the sizes of the arrows are schematic measures of the charge and spin densities on the sites. Note that this represents the SDW of the form
$`\rho _s(j)`$ $``$ $`c_{j,M,}^{}c_{j,M,}c_{j,M,}^{}c_{j,M,}`$ (11)
$`=`$ $`\rho _{s2k_F}\mathrm{cos}(2k_Fja\pi /4)+\rho _{s4k_F}\mathrm{cos}(4k_Fja\pi ),`$ (12)
which coexists with the BOW and CDW.
Commensurability effects imply that the possible phase shifts between adjacent chains in the anisotropic 2D system are 0, $`\pi `$/2 and $`\pi `$, and we have performed
explicit numerical calculations to determine that the lowest energy state is obtained with a phase shift of $`\pi `$. The intrachain bond orders, determined by the probabilities of nearest neighbor charge-transfers, continue to be different for the different pairs of neighboring sites. This is the major difference between the possible broken symmetries in the 1/2-filled and 1/4-filled band. While in the 1/2-filled band there is no overlap between the extreme configurations favoring the BOW, CDW and SDW, in the weakly 2D 1/4-filled band the same extreme configuration supports all three broken symmetries . For small nonzero $`t_{}`$, we therefore expect a strong cooperative coexistence between the BOW, the CDW and the SDW. Furthermore, since the same CDW coexists with both the SUWU BOW and the WSWS BOW, this coexistence is independent of which particular BOW dominates. This has been explicitly demonstrated in reference , where it was shown that the overall ground state for small $`t_{}`$ is one of the two BCSDW states shown in Fig. 2, with overall 2D periodicity of (2k<sub>F</sub>, $`\pi `$).
### D 1/4-filled band, $`t_{}t`$, $`U,V`$ 0.
What happens as $`t_{}`$ is further increased? Within k-space single-particle theory, increasing $`t_{}`$ should destroy the nesting of the Fermi surface. But as we have indicated above, our real space analysis predicts, and our numerical results will establish, that this destruction does not occur. To argue this convincingly, we must first show how this destruction of the nesting, which certainly does occur for non-interacting electrons, can be correctly described within our configuration space picture of the broken symmetry. Recall that the one-electron hopping term in Eq. (1) introduces “paths” between the extreme configurations, where each step in a given path connects two configurations related by a single hop . Nonzero $`t_{}`$ introduces many additional paths connecting the extreme configurations that are the 2D equivalents of …1100… (with a $`\pi `$-phase shift between consecutive chains). For $`U=V`$ = 0, there is no inhibition of these paths, and it therefore becomes easier to reach one extreme configuration from another, leading to enhanced configuration mixing (relative to 1D), which in its turn destroys the “nesting” and the broken symmetry.
The situation described above changes, however, for nonzero Coulomb interaction. Interchain hopping $`t_{}`$ leads to partial double occupancy on a single site ($``$ $``$) with an energy barrier that, while less than the bare $`U`$, is a $`U_{eff}`$ that increases with $`U`$. The energy barrier to interchain hopping leads to “confinement” of the electrons to single chains, a concept that has been widely debated recently, in the context of high $`T_c`$ superconductors. For large enough $`U_{eff}`$, the confinement can be strong enough that the broken symmetry state can persist up to the isotropic limit $`t_{}t`$.
More precisely, the bond and charge components of the BCSDW can persist up to the isotropic limit $`t_{}t`$, leading to the BCDW state we have previously introduced. The evolution of the spin structure is different from and more subtle than the bond and charge components. From the cartoons in Fig. 2, we see that for the SDW to exist it is essential that the ‘0’s have a spin “direction”. In the small $`t_{}`$ case, the sign of the spin on a ‘0’ is necessarily that of the nearest intrachain ‘1’. Note, however, that each ‘0’ also has two interchain ‘1’s as neighbors and that for a stable SDW, the spin densities of the ‘1’s that are neighbors of a specific ‘0’ must be opposite (as shown in the Figure). Therefore, with increasing $`t_{}`$, competing effects occur. On the one hand, the magnitude of the interchain exchange coupling $`J_{}`$ $`t_{}^2/U_{eff}`$ increases. On the other hand, the spin density on a site labeled ‘0’ decreases because of the canceling effects of the intra\- and inter-chain neighboring ‘1’s. We thus expect the SDW of the 2D lattice to vanish at a $`t_{}^c`$ that will depend on the magnitudes of the bare U and V.
This description of the evolution of the SDW applies to the true 1/4-filled band. In lattices that are dimerized initially, further dimerization leads to the occupancy ‘10’
or ‘01’ on each dimer. If the original dimerization is very strong, the spin on a given ‘0’ will continue to be strongly influenced by the spin on its partner in the dimer, and $`t_{}^c`$ at which the SDW vanishes in this case will be larger.
The robustness of the BCSDW and the BCDW relative to the uniform metallic state can be understood from the cartoon occupancy schemes in Fig. 2. It is instructive to discuss the BCDW state in terms of the two large $`U`$ Wigner crystal structures discussed by Hubbard. We refer to the …1100… electron arrangement as that of a “paired electron crystal”, and the ..1010… as the “monatomic Wigner crystal.” For the 3D low density electron gas, Moulopoulos and Ashcroft showed that there exists an intermediate density range where the paired electron crystal has lower energy than the monatomic Wigner crystal, and the region $`0<V<V_c`$ in our discrete lattice case can be thought of as intermediate between the $`V=0`$ and $`V>V_c`$. A striking feature of the BCSDW and the BCDW occupancy scheme is that it is a paired electron crystal along the chains (…1100…, periodicity 2k<sub>F</sub>), a monatomic Wigner crystal transverse to the chains (…1010…., periodicity 4k<sub>F</sub>), as well as a paired electron crystal along both diagonals (…1100…, periodicity 2k<sub>F</sub>). It is thus possible to predict that even in the presence of interactions not explicitly included in Eq. (1), the BCDW continues to persist. For instance, by enhancing the 4k<sub>F</sub> charge ordering along the transverse direction, the nearest neighbor interchain Coulomb interaction $`V_{}`$ will further enhance the stability of the BCDW. Similarly, the diagonal …1100… charge ordering implies that even the additions of hopping $`t_{diag}`$ and Coulomb repulsion $`V_{diag}`$ along the diagonals will not destroy the BCDW state for realistic parameters: in particular, $`V_{diag}`$ stabilizes the BCDW relative to the other Wigner crystal (…1010…) along both $`x`$ and $`y`$ directions.
In the above our goal has been to predict a novel semiconducting state that is more stable than the metallic state. Even if this semiconducting state is assumed, however, there is an additional surprise in our claim, viz., the dominance of the singlet BOW over the SDW for strong two-dimensionality in the interacting quarter-filled band. This is exactly opposite to what is observed in the 1/2-filled band. While in the half-filled band a single singlet-to-antiferromagnet transition occurs with increasing $`t_{}`$, for the 1/4-filled band, a second antiferromagnet-to-singlet transition is predicted at large $`t_{}`$. Since a full discussion of this second transition at this junction would interrupt the flow of the narrative, we defer it to Appendix 1, which presents arguments based on variational concepts and valence bond theory to motivate this result.
## IV Numerical results
### A Results for 1D lattices
Computational limitations will compel us to use fairly small lattices in 2D and will prevent us from studying dynamical phonons (even at a classical, self-consistent level). As a consequence, we will have to work with explicitly distorted lattices, rather than allowing the distortions to arise naturally, as they would in larger lattices calculated with dynamical phonons. To provide justification for this approach, in this section we (a) extend our previous 1D results obtained with nonzero $`\alpha `$ and $`\beta `$ to zero e-ph couplings, to demonstrate that these bond and charge distortions are unconditional, and (b) show that the dimerization of the dimer lattice (see Eq. (2)) leads to the same CDW as the monatomic 1/4-filled band.
It is known that in a sufficiently long open chain the bond orders and the charge densities at the center of the chain show the behavior in the long chain limit, even in the absence of the e-ph coupling. In Fig. 3(a) we show the exact nearest neighbor bond orders and charge densities at the center of an open undistorted chain of 16 atoms with all hopping integrals equal, for $`U=6`$, $`V=1`$. Note that both the BOW and the CDW show the 2k<sub>F</sub> modulations discussed in section III, and appear in spite of uniform hopping integrals.
Second, we recall that in a purely 1D system, a LRO SDW can occur only if an external staggered magnetic field is applied. We therefore incorporate an additional (external field-like) term
$$H_{SDW}=\underset{j}{}ϵ[n_{j,}cos(2k_Fj)+n_{j,}cos(2k_Fj+\pi /2)]$$
(13)
and consider $`H+H_{SDW}`$ for the 1/4-filled band with amplitude $`ϵ=0.1`$. In reference the same Hamiltonian was investigated for the case of finite bond distortion. Figs. 3(b) and (c) show the bond orders and CDW for a periodic ring (zero e-ph coupling and undistorted hopping integrals) with the SDW $``$$``$ superimposed on it. Note that because of the periodicity, the bond orders are uniform for the finite ring for $`ϵ`$ = 0. For $`ϵ`$ = 0.05 (Fig. 3(b)) and 0.1 (Fig. 3(c)), the externally imposed SDW creates spontaneous BOWs with r$`_{4k_F}`$ = 0 and r$`{}_{4k_F}{}^{}`$ 0, respectively.
In Fig. 3(d) we show the charge densities on a periodic ring of 16 sites, now for the dimerized dimer lattice (the hopping integrals here are 1.2, 0.9, 1.2 and 0.7). The charge modulations (which appear entirely due to modulations of the interdimer bond orders) on the sites are exactly as in Figs. 3(a)–(c), with the larger charges occurring on the sites connected by the stronger weak bond (the W bond, with $`t_W^{}`$ = 0.9). In discussions of the spin-Peierls transition within the effective 1/2-filled band (corresponding to the dimer lattice), it is usually assumed that the electronic populations within each dimer cell remains uniform in the spin-Peierls state. Fig. 3(d) clearly shows that this is not true.
### B Results for 2D lattices
To confirm the expectations based on the qualitative arguments of Section III, we use exact diagonalization and Constrained Path quantum Monte Carlo (CPMC) numerical techniques to calculate for representative finite 2D lattices: (i) the electronic energy gained upon bond distortion,
$$\mathrm{\Delta }EE(0)E(u_{j,M}),$$
(14)
where $`E(u_{j,M})`$ is the electronic energy per site with fixed distortion $`u_{j,M}`$ along the chains; (ii) the site charge densities $`\rho _{j,M}`$ for the bond-distorted lattices; due to the coexistence of the BOW and the CDW, measuring the CDW amplitude that results as a consequence of the external modulation of the hopping integrals is exactly equivalent to the measurement of the bond order differences in the charge-modulated lattices; and (iii) the z-z component of the spin-spin correlations, for a range of $`U`$, $`V`$ and $`t_{}`$. We consider three distinct distorted lattices, two of which correspond to those shown in Figs. 2(a) and (b), where we have indicated the hopping integrals along the chain (the uniform lattice has a hopping integral of 1.0 corresponding to all intrachain bonds). The third distorted lattice we consider is the dimerized dimer lattice, the hopping integrals for which will be discussed later.
Ideally, calculations that aim to demonstrate persistence of a spatial broken symmetry should do fully self-consistent calculations of the total energy, which is a sum of the the electronic energy gain $`\mathrm{\Delta }E`$ (including effects of both e-e and e-ph interactions) and the loss in lattice distortion energy. Unfortunately, in true many-body simulations (such as exact diagonalizations or CPMC) of the very large 2D lattices we investigate (see below), such self-consistent calculations are not possible. A well-tested alternate approach is to calculate only the electronic energy gain for fixed lattice distortion and compare the calculated $`\mathrm{\Delta }E`$ against a known reference configuration, where the distortion is known to occur. This approach works because for a fixed distortion, the contribution of the elastic energy to the total energy is constant, independent of the other parameters; therefore the gain in electronic energy, relative to that for the reference configuration, is a direct measure of the tendency to distortion. An example of a previous successful application of this approach is the enhancement by e-e interactions of the bond alternation in the 1D 1/2-filled band; here, the reference configuration corresponds to the limit of zero e-e interaction (SSH model), where the Peierls bond alternation is known to occur . For nonzero e-e interaction, the electronic energy gain for fixed bond alternation can be larger (see Figs. 2.26 and 2.31 in reference ), indicating the enhancement of the bond alternation by e-e interaction, a theoretical result that has been confirmed by all subsequent studies. Similarly, in the 2D 1/2-filled band, calculations of the electronic energy gain for fixed bond distortion have been used to prove the decrease in the tendency to Peierls bond alternation upon the inclusion of e-e interaction (see Fig. 10 in reference ), a result that is in agreement with other studies as well as the determination of long range AFM in this case . Thus the approach has been shown to work in two cases in which exactly opposite outcomes, – in one case, an increase in dimerization, in the other case, a decrease, occurred, indicating its robustness.
At first glance, it appears that there exist two different reference configurations in the present case. First, for given $`t_{}`$, one could study $`\mathrm{\Delta }E`$ as a function of $`U`$ and $`V`$: in essence, this amounts to comparing uncorrelated and correlated lattices for each $`t_{}`$. Second, for given $`U`$ and $`V`$, one could calculate $`\mathrm{\Delta }E`$ as a function of $`t_{}`$. In fact, the first approach does not yield correct results for two reasons: (i) the uncorrelated 2D lattices are undistorted, so there is no obvious $`\mathrm{\Delta }E`$ with which to compare the correlated results; and (ii) magnitude of $`\mathrm{\Delta }E`$ decreases with $`U`$ and $`V`$ even in the 1D limit, where we know that the bond and charge distortions are unconditional (see references , as well as the immediately previous subsection on 1D numerical results). Thus to determine properly the tendency to distortion in 2D, our reference configuration should be the single chain. We therefore normalize the energy gained for coupled chains ($`\mathrm{\Delta }E`$) against that for the single chain ($`\mathrm{\Delta }E_0`$) with the same $`U`$ and $`V`$. A decreasing $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ as a function of $`t_{}`$ signals the destruction of the distortion by increasing two-dimensionality, while a constant or increasing $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ indicates a persistent distortion . Since the BOW and the CDW are coupled cooperatively, the behavior of the charge ordering gives a second measure for the tendency to bond distortion. Decreasing charge ordering for fixed bond distortion, as a function of $`t_{}`$ (as occurs for noninteracting electrons), indicates the tendency to decreasing bond distortion, while constant or increasing charge ordering indicates persistent bond distortion. The expected (and calculated, see below) charge ordering pattern is the same for all bond distortion patterns and is the same as in 1D (with, however, a $`\pi `$-phase shift between consecutive chains).
As mentioned above, our numerical calculations involve both exact diagonalization and the CPMC technique. Because of the sign errors that plague quantum Monte Carlo calculations in 2D, it is critical to obtain a precise idea about the accuracy of the numerical results. This is especially so because CPMC calculations that have been reported so far are only for the simple Hubbard Hamiltonian and did not include the nearest neighbor interaction $`V`$. In Appendix 2 we discuss our methodology and give detailed comparisons of energies and correlation functions obtained for finite lattices within the CPMC and exact diagonalization procedures. As shown there, although the CPMC technique is not variational, the accuracies in both energy and correlation functions are sufficient for our purposes.
For numerical results obtained from finite-size calculations to be relevant in the thermodynamic limit, it is essential to choose proper boundary conditions. In the present case, we choose lattices and boundary conditions based on the physical requirement that for noninteracting electrons any nonzero $`t_{}`$ must destabilize the BCDW on that particular finite lattice. Details of the analysis that guided our choice of 2D lattices are also presented in Appendix 2. There we show N $`\times `$ M lattices
(with N the number of sites per chain and M the number of chains) that obey the above physical requirement are restricted to those for which N = 8n, where n is an integer. On the other hand, there is no restriction on M, except that M be even to avoid even/odd effects. In our calculations below, we have chosen M = 4n + 2, for reasons that are also discussed in Appendix 2.
We make one final point before presenting the 2D numerical data. The restriction to N = 8n sites coupled with the 1/4-filling introduces a potential subtlety into the numerical computations of $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ for nonzero $`U`$ and $`V`$. Finite 4n-electron non-1/2-filled 1D undistorted periodic rings have their ground state in the total spin S = 1 subspace, and even the distorted system’s ground state can be in the S = 1 subspace for the smallest 4n-electron rings. We have confirmed from exact
diagonalizations of the 8 $`\times `$ 2 lattice that the ground state is in the S = 0 state for the smallest nonzero $`t_{}`$. Thus while $`\mathrm{\Delta }E_0`$ can correspond to the energy gained upon distortion in the S = 1 subspace, $`\mathrm{\Delta }E`$ necessarily corresponds to the energy gained upon distortion in the S = 0 subspace. As this important but subtle point requires extensive discussion that would interrupt the presentation here, we present the details in Appendix 3, where we show that despite this subtlety, the behavior of $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ nevertheless is a proper measure of the stability of the distorted state for nonzero $`t_{}`$.
#### 1 Exact diagonalization and CPMC calculations, r$`_{4k_F}`$ = 0
In Fig. 4 we show the behavior of $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ for the non-interacting and interacting ($`U=6`$, $`V=1`$) cases for three different lattices satisfying our boundary condition constraints. In all cases we measure the electronic energy gained upon 2k<sub>F</sub> SUWU bond distortion (corresponding to nearest neighbor hopping integrals $`t_S`$ = 1.14, $`t_U`$ = 1.0, and $`t_W`$ = 0.86), relative to that of the undistorted state with equal hopping integrals. For the 8$`\times `$2 lattice the calculations involved both exact diagonalization and the CPMC technique. The 8$`\times `$2 results, taken together, then provide an estimate of the precision of the CPMC calculation. The exact diagonalization studies also confirm that the system is in the total spin state S = 0 for $`t_{}`$ as small as 0.01 (see Appendix 3).
The large scatter in the normalized $`\mathrm{\Delta }E`$ at very large and very small $`t_{}`$ may be due to the degeneracies in the non-interacting system at $`t_{}0`$ and $`t_{}1`$. Furthermore, as pointed out in Appendix 2 (subsection A), the absolute values of $`\mathrm{\Delta }E`$ are rather small, especially for the pure 2k<sub>F</sub> (r$`_{4k_F}`$ = 0) distortion. The systematic errors due to the CPMC approximation are therefore large in these two regions. Nevertheless, except for the $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ value at $`t_{}=0.1`$ for the 8$`\times `$6 lattice, at all other $`t_{}`$ the $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ values are above 1 for all three lattices, and far above the normalized non-interacting values. As seen in Fig. 4, while for the non-interacting cases the $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ decreases rapidly with $`t_{}`$, for the interacting cases the $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ either remains unchanged or is enhanced by $`t_{}`$. Because of the strong degeneracies in the one-electron occupancy scheme at the Fermi level at $`t_{}`$ = 1, a single well-defined one-electron wavefunction is missing here. The CPMC calculations therefore could not be done for $`t_{}`$ = 1.0. It is, however, highly unlikely that the BCDW persists for $`t_{}=0.9`$ but vanishes at $`t_{}=1`$; this expectation is corroborated by the results of the exact diagonalization studies for the 8 $`\times `$ 2 lattice, which were performed for the full range of $`t_{}`$, including $`t_{}=1`$ and showed enhanced distortion throughout the whole region. In the following sections we also show $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ for the
2k<sub>F</sub>+4k<sub>F</sub> (r$`_{4k_F}`$ $``$ 0) and dimerized dimer lattice. In both of these cases, the magnitude of $`\mathrm{\Delta }E`$ is larger and hence easier to compute, but degeneracies restrict CPMC simulations to smaller $`t_{}`$. In both cases, $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ is close to or above 1 for all $`t_{}`$ we have studied.
As discussed in the above, the bond-distorted lattices (both $`r_{4k_F}`$ = 0 and $`r_{4k_F}`$ 0) have a synergetic coexistence with the CDW. Thus the amplitude of the CDW, defined as $`\mathrm{\Delta }\rho _c=\rho _{cl}\rho _{cs}`$, where $`\rho _{cl}`$ and $`\rho _{cs}`$ are the larger and smaller charge densities on the …1100… 2k<sub>F</sub> CDW, is an alternate measure of the stability of the BOW. If the nonzero $`t_{}`$ destabilized the bond-distortion, then even with fixed 2k<sub>F</sub> distorted hopping integrals the amplitude of the BOW (measured as the differences in the bond orders) would decrease, and the diminished strength of the BOW in turn would decrease $`\mathrm{\Delta }\rho _c`$. This is easily confirmed for the noninteracting Hamiltonian, where the amplitude of the CDW decreases with increasing $`t_{}`$. In Fig. 5(a) we show the charge densities on a single chain for a bond-distorted 8$`\times `$6 lattice (because of periodicity, all chains are equivalent) for $`U=6`$, $`V=1`$, and $`t_{}=0.2`$. In Fig. 5(b) we have shown the behavior of $`\mathrm{\Delta }\rho _c`$ for all the three lattices we have studied, now as a function of $`t_{}`$. Degeneracies in the one-electron energy levels in the 16$`\times `$6 lattice for $`t_{}>0.6`$ even with finite bond-distortion cause the CPMC ground states in this region to be S = 1. Exact calculations in the 1D limit show that the amplitude of the CDW in S = 1 is less than that in S = 0. Thus the weak decrease in the $`\mathrm{\Delta }\rho _c`$ values with $`t_{}`$ in the 16$`\times `$6 lattice is a spin effect: the bond distorted state is S = 0 at small $`t_{}`$ and S = 1 at large $`t_{}`$. The $`\mathrm{\Delta }\rho _c`$ values at large $`t_{}`$ for the 16 $`\times `$ 6 lattice should therefore be considered as lower limits (the $`\mathrm{\Delta }\rho _c`$ values of the 16 $`\times `$ 6 lattice are considerably larger than that of the S = 1 single chain of 16 sites). In agreement with the behavior of the $`\mathrm{\Delta }E`$ in the interacting case (see Fig. 4), the CDW amplitude now increases or remains constant with increasing $`t_{}`$ for all the lattices studied, indicating a greater tendency to bond and charge distortion with increasing $`t_{}`$. Taken together, the results of Figs. 4 and provide quantitative proof of our qualitative arguments establishing that the BCDW is a robust broken symmetry state for the interacting 2D $`\frac{1}{4}`$-filled band.
In Fig. 6 we show the inter-chain spin-spin correlations between sites 1 and 2 on the first chain, and sites j = 1 – 8 on the second chain, for the 2k<sub>F</sub> bond-distorted 8$`\times `$6 lattice for several values of $`t_{}`$. The SDW profile is somewhat different from what is expected from a pure …1100… charge modulation along the chains because the wavefunction of this finite lattice also has contributions from the …1010… type intrachain charge modulation. The small …1010… contribution to the wavefunction affects the charge density, $`n_{j,M,}+n_{j,M,}`$ only weakly, but the spin density, being the difference $`n_{j,M,}n_{j,M,}`$ is a smaller quantity and is affected relatively more strongly. It is useful here to recall however that within the rectangular lattice, …1010… charge orderings along both longitudinal and transverse directions give triangular lattice of occupied sites, and thus a pure …1010… cannot give the SDW profiles of Fig. 6 (see also below) .
Qualitatively, at $`t_{}=0.1`$ the SDW behavior is the same as in Ref. , where these calculations were done for the 12$`\times `$4 lattice: the amplitude of the interchain spin-spin correlation is independent of the distance between the sites, indicating long-range order. The qualitative behavior of the spin-spin correlations is the same for $`t_{}`$ = 0.4, where, however, the amplitude of the SDW is larger. At still larger $`t_{}(=`$ 0.6), the inter-chain correlations are very strongly antiferromagnetic at short distances (j = 1,2 on chain 2), but the antiferromagnetic correlations have disappeared at larger distances. This can be seen from comparisons of the spin-spin correlations corresponding to values of j lying near the center of the second chain (j = 5), which are farthest from the spins occupying sites 1 and 2 on the first chain. While the spin-spin correlations near j=5 increase from $`t_{}`$ = 0.1 to 0.4, they decrease as $`t_{}`$ is further increased to 0.6. Similarly, focusing on site 8 of the second chain, we see that the spin-spin correlation with site 1 on the first chain has actually changed sign upon increasing $`t_{}`$ to 0.6 from 0.4 (due to the very strong short-range antiferromagnetic correlations), and the magnitude of the positive spin-spin correlation with site 2 on the first chain has decreased. All of these results indicate the absence of long-range spin order for large $`t_{}`$ 0.6 in the 8$`\times `$6 lattice. The loss of the long-range spin-order is most clear at $`t_{}`$ = 0.9, where spin-spin correlations are nonzero only for the nearest interchain neighbors.
Fig. 7 shows the inter-chain spin-spin correlations between sites 2 and 3 on the first chain and sites j=1$`\mathrm{}`$16 on the second chain for the 16$`\times `$6 lattice. The admixture of the intrachain …1010… CDW is weaker in this larger system: this is because the “tunneling’‘ between the extreme configurations …1100… and, say, …0110…, decreases with size, and as consequence, $`V_c`$ increases with size in finite systems. This can be seen by simply comparing the figures on the left and right panels for $`t_{}`$ = 0.1 and 0.2. If the intrachain CDW were a pure …1010…,
the signs of the spin-spin correlations for each j would be the same for both i = 2 and i = 3. Different signs for these correlations are signatures of the …1100… CDW (see Fig. 2). As in the 8$`\times `$6 system, long-range SDW behavior is seen for $`t_{}=0.1`$. Focusing on sites j=7 – 12 on the second chain, the amplitude of the SDW increases from $`t_{}=0.1`$ to $`t_{}=0.2`$, but further increasing $`t_{}`$ to 0.3 destroys the long-range order, as evidenced again by very large AFM correlations at short distances and vanishing correlations at large distances (sites j=7$`\mathrm{}`$12 on the second chain). The vanishing of the SDW is seen most clearly at very large $`t_{}`$ ($`t_{}=0.9`$ in Fig. 7). We observe this same behavior of the SDW on 8 $`\times `$ 2 lattice. In all cases, the SDW amplitude initially increases, exhibits a maximum, and then vanishes at a $`t_{}^c`$ which decreases with the size of the system. As discussed in section III.D, this behavior is to be expected from the nature of the BCSDW in Fig. 2. The initial increase of the SDW amplitude indicates that $`t_{}^c`$ is nonzero, a conclusion that is also in agreement with the experimental observation of the BCSDW state in the weakly 2D organic CTS (see below). Based on the calculations for 16 $`\times `$ 6 lattice, we estimate $`0.1<t_{}^c<0.3`$ for the strictly rectangular lattice for $`U=6`$, $`V=1`$.
#### 2 Persistent distortions with r$`_{4k_F}`$ $``$ 0
The bond modulation pattern in the 1/4-filled band given in Eq. (7) has in general both 2k<sub>F</sub> and 4$`k_F`$ components. Figs. 4 and 5 show persistent distortion at large inter-chain couplings for $`r_{4k_F}`$ = 0 (purely 2k<sub>F</sub> bond distortion). The persistent BCDW is expected also for $`r_{4k_F}`$ 0.
Physically, the reason for this persistence is the coexisting site CDW, whose nature is independent of $`r_{4k_F}`$ . We show in Fig. 8 the calculated $`\mathrm{\Delta }`$E/$`\mathrm{\Delta }`$E<sub>0</sub> for $`r_{4k_F}`$ = $`r_{2k_F}`$ (equal admixtures of 2k<sub>F</sub> and 4k<sub>F</sub> bond distortions), for the 8 $`\times `$ 2 and 8 $`\times `$ 6 lattices for $`U=6`$ and $`V=1`$. The hopping integrals corresponding to the distorted lattice here are 1.089, 0.974, 1.089 and 0.848, and the energy gained is being measured against the uniform lattice. Starting from $`t_{}`$ = 0.5, the one-electron $`\mathrm{\Delta }`$E is highly discontinuous. This is because distortions with $`r_{4k_F}0`$ do not correspond to a natural periodicity for the noninteracting system. As a consequence the noninteracting wavefunctions are not suitable trial wavefunctions for the CPMC calculation. For the same reason the 16$`\times `$6 calculation could not be performed here. The similarities between the results for the 8 $`\times `$ 2 and the 8 $`\times `$ 6 lattices are obvious. The ratio $`\mathrm{\Delta }`$E/$`\mathrm{\Delta }`$E<sub>0</sub> is independent of $`t_{}`$ over a broad range of $`t_{}`$ and increases slightly for large $`t_{}`$, indicating once again a stable 2D BCDW. Although only limited data could be obtained for this case, the dimerized dimer lattice is very similar in character to $`r_{4k_F}`$ 0 (see Fig. 2(b)). In the following we show convincing evidence for persistent double-dimerization in 2D.
#### 3 The dimerized dimer lattice
We have previously noted that Fig. 2(b) suggests that an alternate way to view the BCDW/BCSDW states is as a dimer lattice with additional structure within each of the dimer cells; the dotted box in Fig. 2(b) represents one dimer. Each dimer has one electron, leading to an “effective half-filled” dimer band . Bond dimerization in the 1D 1/2-filled band is unconditional for all $`U>2V`$ , and thus this dimer lattice itself distorts in a period 2 dimerization pattern in 1D. In this section we show the additional result that the (anisotropic) 2D dimer lattice is unconditionally unstable to a second dimerization for all $`t_{}`$.
We choose the hopping integrals between the two sites within the dimer cell to be 1.2 in our calculations. The two inter-dimer hopping integrals for the uniform dimer lattice were taken to be 0.8, while for the distorted (”dimerized”) dimer lattice these were taken to be 0.7 and 0.9, respectively (i.e., the dimerized dimer lattice has hopping integrals 1.2, 0.7, 1.2, 0.9 along each chain). Exact diagonalizations show that a $`\pi `$-phase shift between the chains (i.e., dimer cells lying directly above each other, but a strong inter-dimer bond on one chain facing a weak inter-dimer bond on the next chain) gives the lowest total energy. Again we define $`\mathrm{\Delta }`$E and $`\mathrm{\Delta }`$E<sub>0</sub> as the electronic energies gained per site upon interdimer bond distortion by the 2D and 1D lattices. Fig. 9 shows the $`\mathrm{\Delta }`$E/$`\mathrm{\Delta }`$E<sub>0</sub> behavior for the 8$`\times `$2 lattice over the complete range of $`t_{}`$ and for the 8$`\times `$6 and 16$`\times `$6 lattices for several different $`t_{}`$ for $`U=6`$ and $`V=1`$. The 8$`\times `$6 and 16$`\times `$6 lattices, taken together, cover nearly the full range of $`t_{}`$, and the $`\mathrm{\Delta }`$E/$`\mathrm{\Delta }`$E<sub>0</sub> behavior for these lattices closely follow the curve for the 8$`\times `$2 lattice. As before, $`\mathrm{\Delta }`$E/$`\mathrm{\Delta }`$E<sub>0</sub> is significantly greater than 1 for the complete range $`0<t_{}<1`$, indicating the persistence of the dimerization of the dimer lattice in the interacting case, whereas for the non-interacting case, the dimerization vanishes, as expected.
Fig. 10 shows the interchain spin-spin correlations between sites 2 and 3 on one chain and sites j = 1 – 16 on a neighboring chain, for a 16$`\times `$6 dimerized dimer system. Notice the far smaller contribution by the …1010… intrachain charge ordering here. This is because of the large difference between the hopping integrals even in the “uniform” lattice with interdimer hopping integrals of 0.8 here. Such a large bond dimerization diminishes the intrachain …1010… contribution. The spin-spin correlation amplitudes cannot be directly compared to Fig. 7 because of the different distortion amplitudes, but Fig. 10 shows that the SDW amplitude is significantly greater in the intermediate $`t_{}`$ regime ($`t_{}`$ = 0.37 in the Figure) compared to the small $`t_{}`$ regime unlike the results in Fig. 7. Our calculations indicate that the larger the difference between the intra-dimer and the inter-dimer hopping integrals, the greater the range of the $`t_{}`$ over which the SDW is stable. Thus with hopping integrals of 1.2, 0.9, 1.2 and 0.7 along each chain, the SDW in the 8$`\times `$6 lattice persists even at $`t_{}=0.6`$ (in contrast to the 2k<sub>F</sub> bond-distorted lattice of Fig. 2), but vanishes at still larger $`t_{}`$. This is expected from our discussion of the behavior of the SDW in Section III.D. Recall that the smaller spin densities on the sites labeled ‘0’ are influenced by both the intrachain nearest neighbor as well as the interchain nearest neighbor with opposite spin, and this competition creates a disordering effect. The larger the hopping integral between the ‘0’ and the nearest intrachain ‘1’, the larger the $`t_{}`$ necessary to create the disordering of the spin, hence the greater stability of the SDW. We shall later argue that this same phenomenon is related to the very large magnetic moments of the $`\kappa `$-(BEDT-TTF) salts.
#### 4 Effects of additional Coulomb interactions
Fig. 2 clearly suggests that interchain nearest neighbor Coulomb interaction $`V_{}`$ stabilizes the BCDW further. We have confirmed this by exact numerical calculations for the 8 $`\times `$ 2 lattice, as shown in Fig. 11 below, where we have plotted $`\mathrm{\Delta }`$E/$`\mathrm{\Delta }`$E<sub>0</sub> for three different values of $`V_{}`$: 0, 0.5 and 1. Nonzero $`V_{}`$ increases $`\mathrm{\Delta }`$E further. Similar calculations were done also with variable $`V_{}`$ but fixed $`V_{}/t_{}`$. An even larger increase in $`\mathrm{\Delta }`$E is found in this case. Implementing $`V_{}`$ over and above $`V`$ is difficult within the CPMC, and therefore these calculations could not be performed for larger lattices. However, based on
the similarities between the $`\mathrm{\Delta }`$E behavior of the three lattices studied in Figs. 4 and 9, no difference in the larger lattices is expected.
## V Summary of theoretical results
We have performed detailed numerical calculations of various broken symmetries for the 2D 1/4-filled band within Eq. (1) and for the effective 1/2-filled band of dimer lattice within Eq. (2), for $`U`$ = 6, $`V`$ = 1. Regarding these parameter values, the broken symmetries we have found will occur for all intermediate to strong $`U`$ but require $`V`$ to be less than a critical $`V_c2t`$ .
We have discovered three distinct new results in 1D. First, we have confirmed that the BCDW state occurs spontaneously even for zero e-ph couplings (see Fig. 3(a)). The bond distortion pattern in the center of a long open chain corresponds to a pure 2k<sub>F</sub> distortion, and coexists with the 2k<sub>F</sub> …1100… type charge modulation. Second, we have shown that a BOW appears spontaneously in a uniform periodic ring when the SDW $`,`$,$`,`$ is superimposed, confirming the synergetic cooperation between e-e and e-ph interactions. The BOW pattern corresponds to r$`_{4k_F}`$ = 0 (see Eq. (7)) when the amplitude of the superimposed SDW is relatively weak (Fig. 3(b)), but switches over to r$`{}_{4k_F}{}^{}0`$ when the SDW amplitude is large (Fig. 3(c)). Our earlier demonstrations of the BOW-SDW coexistence were only for the bond distorted periodic systems. Finally, from exact calculations for a periodic dimerized dimer ring, we have established the new result that the BOW here also coexists with the …1100… 2k<sub>F</sub> CDW, with the large (small) charges occupying the sites connected by the stronger (weaker) interdimer W (W) bond (see Fig. 3(d)). Our earlier work had claimed that a 1/4-filled description was essential to obtain the BCDW and the BCSDW states. As shown in Fig. 3(d), the same result is obtained, however, even for the dimer lattice, provided the second dimerization is allowed to occur.
Three different bond distortion patterns were investigated in 2D. These correspond to r$`_{4k_F}`$ = 0 (Fig. 2(a)), r$`_{4k_F}`$ = r$`_{2k_F}`$ (Fig. 2(b)), and the dimerized dimer lattice. In all cases a $`\pi `$-phase shift in the bond distortion between consecutive chains gives the lowest energy. From calculations of energy gained upon bond distortion, we conclude that 2D bond distorted lattices with r$`_{4k_F}`$ = 0 and r$`_{4k_F}`$ = r$`_{2k_F}`$ are both more stable than the uniform lattice (see numerical results in Figs. 4 and 8). Similarly, the dimerization of the dimer lattice is also unconditional (see numerical results in Fig. 9). The persistence of the distortions is a novel effect of e-e interactions and is in contradiction to what is expected within one-electron nesting concepts. The ground state of the strongly correlated 1/4-filled band is therefore a novel insulating BCDW state for all $`t_{}`$.
The persistence of the BCDW for all anisotropies is also evident from the charge density calculations. In Fig. 5, we have shown the amplitude of the CDW that accompanies the r$`_{4k_F}`$ = 0 BOW as a function of $`t_{}`$. In the absence of e-e interaction, the CDW amplitude decreases rapidly with $`t_{}`$ even with nonuniform hopping integrals. One interesting aspect of these calculations is that the CDW pattern is the same for all bond distortion patterns. Our computer capabilities do not allow us to determine self-consistently which of the three BOW patterns dominate within Eqs.(1) and (2) for a given $`U`$, $`V`$, $`t_{}`$, $`\alpha `$ and $`\beta `$. This is, however, largely irrelevant, because the charge ordering is the same with all the bond distortion patterns.
The SDW behavior is different from those of the BOW and the CDW. As seen from our numerical calculations of interchain spin-spin correlations in Figs. 6 and 7, the SDW amplitude of the novel BCSDW state is initially enhanced by $`t_{}`$, but with further increase in $`t_{}`$ the SDW vanishes, indicating a singlet BCDW state again in the large $`t_{}`$ region. The range of $`t_{}`$ within which a stable SDW is found depends on the BOW pattern, and within the dimerized dimer lattice (see Fig. 10) the SDW can be stable over a wider range of $`t_{}`$.
## VI Comparison to Experiments on the Insulating States in 2:1 organic CTS
Experimentally, the organic cationic CTS, with cation:anion ratio of 2:1, span the range $`t_{}0.1`$ in (TMTTF)<sub>2</sub>X to $`t_{}`$ $``$ $`1`$ in certain (BEDT-TTF)<sub>2</sub>X. Hence these materials provide a critical testing ground for our theoretical results. In reference , we compared our theoretical predictions regarding the BCSDW state to the mixed CDW-SDW found experimentally in (TMTTF)<sub>2</sub>Br, (TMTSF)<sub>2</sub>PF<sub>6</sub> and $`\alpha `$-(BEDT-TTF)<sub>2</sub>KHg(SCN)<sub>4</sub>. Here we make additional, more detailed comparisons, distinguishing between 1D TMTTF and weakly 2D TMTSF-based compounds, and also emphasizing the similarities and differences between the salts of BEDT-TTF and BETS with different crystal structures. In the case of the TMTTF and TMTSF band structure calculations of hopping integrals have been been summarized by Yamaji . In both cases the lattice is anisotropic triangular in nature, which would correspond to our rectangular lattice with one additional diagonal hop $`t_{diag}`$ beyond the usual $`t_{}`$. Both $`t_{}`$ and $`t_{diag}`$ are small in the 1D TMTTF, while they are comparable in TMTSF and about 0.1$`|t|`$ in magnitude. As discussed in section III.D, the paired electron crystal ordering even along the diagonal directions in the configurations shown in Figs. 2(a) and (b) indicate that the BCDW and the BCSDW states continue to be stable for nonzero $`t_{diag}`$ and there is thus no loss of generality in considering a rectangular lattice. Several crystal structures occur in the BEDT-TTF systems, and more subtle and individual analyses for the different cases are required. Our aim is to show that a variety of recent experiments indicate that the BCSDW and the BCDW are appropriate descriptions of the insulating states of this entire class of 2:1 cationic CTS, and conversely, the very nature of the insulating ground state in certain cases provides direct verification for some of our more surprising theoretical results. We discuss below each class of material individually.
### A (TMTTF)<sub>2</sub>X
The (TMTTF)<sub>2</sub>X compounds are nearly 1D semiconducting materials with weak to moderate dimerization along the stacks at high temperature. Because of this dimerization, they have often been described within the effective 1/2-filled band picture . Further dimerization of the dimerization occurs below the SP transition temperature T<sub>SP</sub> ($``$ 15 K). Existing theories of the SP transition in these systems do not discuss the simultaneous appearance of the 2k<sub>F</sub> CDW and assume that the site populations continue to be uniform below T<sub>SP</sub>. As depicted in Fig. 1(c), and as confirmed in Fig. 3(d), independent of whether these systems are considered as 1/4-filled or effective 1/2-filled with a dimer lattice, the appearance of this 2k<sub>F</sub> CDW is unconditional and the site populations are therefore not uniform. In a recent NMR study of <sup>13</sup>C spin-labeled (TMTTF)<sub>2</sub>PF<sub>6</sub> and (TMTTF)<sub>2</sub>AsF<sub>6</sub> charge-ordered states have been found . Although such a charge-ordering suggests agreement with the theory presented here, one problem is that the initial appearance of the charge-ordered phase (at $``$ 70 K in (TMTTF)<sub>2</sub>PF<sub>6</sub>) occurs considerably above T<sub>SP</sub> (15 K) . There are two possible reasons why the charge-ordering might appear at a temperature T$`{}_{CO}{}^{}>`$ T<sub>SP</sub>. First, this might be due to fluctuation effects associated with the 1D nature of the crystals. As has been shown by Schulz , fluctuation effects associated with the SP transition may be seen at temperatures as high as 4T<sub>SP</sub>, in which case signatures of charge ordering would also become visible at these high temperatures. The observation of diffuse X-ray scattering at 2k<sub>F</sub> in this material already at $``$ 60 K seems to support this possibility. A second possibility is that the charge-ordering is driven primarily by the Holstein e-ph coupling $`\beta `$ in Hamiltonian (1), and the SSH coupling $`\alpha `$ is small, such that actual lattice displacement and spin singlet formation takes place at lower temperature. Independent of which mechanism dominates to give T$`{}_{CO}{}^{}>`$ T<sub>SP</sub>, it is important to keep in mind that (a) no charge-ordering is expected at all within conventional theories of SP transition, and (b) as discussed extensively in section III, charge ordering of the type …1010…, as has sometimes been suggested (see below and footnote ), promotes equal intrachain bonds, and therefore the SP transition could not occur if the …1010…charge-ordering had taken place. Finally as has been pointed out by us previously , charge-ordering of the type …1100… also occurs in the SP phase of the anionic 1:2 TCNQ solids.
Although most (TMTTF)<sub>2</sub>X exhibit the SP transition, the material (TMTTF)<sub>2</sub>Br exhibits a transition to a SDW , like the (TMTSF)<sub>2</sub>X. Also like the (TMTSF)<sub>2</sub>X, this material can become superconducting, although at a relatively high pressure of 26 kbar. Within the structural classification scheme described by Jerome , this difference is due to the larger $`t_{}`$ in (TMTTF)<sub>2</sub>Br (relative to the other TMTTF). We therefore discuss this material along with the (TMTSF)<sub>2</sub>X.
### B (TMTTF)<sub>2</sub>Br and (TMTSF)<sub>2</sub>X
X-ray scattering studies by Ravy and Pouget have shown that in both (TMTTF)<sub>2</sub>Br and the prototype TMTSF system, (TMTSF)<sub>2</sub>PF<sub>6</sub>, CDW distortions occur below the SDW transition temperature T<sub>SDW</sub>. Similar conclusions have been reached also by Kagoshima et al. . In (TMTTF)<sub>2</sub>Br evidence for a 4k<sub>F</sub> lattice instability was found , clearly suggesting that the insulating state here is the BCSDW of Fig. 2(b). In (TMTSF)<sub>2</sub>PF<sub>6</sub> the authors claim a “purely electronic CDW”, which would indicate the dominance of the 2k<sub>F</sub> CDW over the BOW. Since, however, in both the 1/4-filled band and the effective 1/2-filled band, the 2k<sub>F</sub> CDW necessarily coexists with a BOW, the experimental work merely indicates that the transition to the BCSDW state is driven mainly by the Holstein e-ph coupling in Eq. (1) rather than the SSH coupling (i.e., $`\alpha `$ is small), so that the actual modulations of the intermolecular distances are small . This would agree with one of the two possible reasonings given by us for T<sub>CO</sub> being larger than T<sub>SP</sub> in (TMTTF)<sub>2</sub>PF<sub>6</sub> and (TMTTF)<sub>2</sub>AsF<sub>6</sub>, as discussed above.
One additional comment appears to be necessary. Fröhlich mode sliding conductivity has been seen in (TMTSF)<sub>2</sub>X . While this indicates a weak incommensurability of the density wave (see below), an equally important point is that the sliding conductivity in the past has been ascribed to a SDW: the SDW collective transport is viewed as that of two CDWs, one for each spin subband. The actual displacement of the charge density is difficult to visualize in configuration space within this picture. We believe that the experimental demonstration of the coexisting CDW and the present theoretical work, taken together, suggest the more coherent viewpoint that the sliding mode conductivity is that of a BCSDW.
### C $`\alpha `$-(BEDT-TTF)<sub>2</sub>MHg(SCN)<sub>4</sub>
This class of materials, with M = K, Rb, Tl and NH<sub>4</sub> has been of considerable interest recently. M = NH<sub>4</sub> is a superconductor, but M = K, Rb, Tl are non-superconducting. Early magnetic susceptibility studies in the M = K material had indicated anisotropic susceptibility below the so-called “kink” transition that occurs at 10 K, indicating a SDW; here the kink refers to the change in slope that occurs in the temperature dependence of the resistivity and the Hall coefficient. On the other hand, analysis of the angle-dependent magnetoresistance oscillations by Sasaki and Toyota led these authors to conclude already in 1995, prior to the experiments by Pouget and Ravy in the (TMTSF)<sub>2</sub>PF<sub>6</sub>, that the dominant broken symmetry in $`\alpha `$-(BEDT-TTF)<sub>2</sub>MHg(SCN)<sub>4</sub> is a CDW . Since, however, a CDW would not explain the anisotropic susceptibility, Sasaki and Toyota concluded that the broken symmetry here is a “mysterious” state that is a “SDW accompanied by a CDW” or a “CDW accompanied by a SDW”. Muon spin resonance studies indicate very small magnetic moment per BEDT-TTF molecule here, $``$ 0.003 $`\mu _B`$ (to be compared against 0.08 $`\mu _B`$ in (TMTSF)<sub>2</sub>X and 0.4 – 1 $`\mu _B`$ per BEDT-TTF dimer in $`\kappa `$-(BEDT-TTF)<sub>2</sub>Cu(CN)<sub>2</sub>Cl , see below). More recent <sup>13</sup>C-NMR studies in the M = Rb indicate even smaller magnetic moment (if it exists at all) $``$ 1 $`\times `$ 10<sup>-4</sup> $`\mu _B`$ . Recent theoretical and experimental investigations conclude either that the dominant broken symmetry here is a CDW or that it is not a conventional SDW .
We point out here that a mixed state with very small magnetic moments is exactly what is expected within our theory. In Fig. 12 we have given a schematic view of the structure of the donor plane in $`\alpha `$-(BEDT-TTF)<sub>2</sub>MHg(SCN)<sub>4</sub>. The one-electron hopping integrals (called “t<sub>p</sub>” and “t<sub>c</sub>” in the figure) have been calculated using approximate one-electron techniques by Mori et al. and Ducasse and Fritsch . Here the t<sub>p</sub> correspond to the interstack hopping and the t<sub>c</sub> to the intrastack hopping. Four slightly different p-type integrals and three slightly different c-type integrals are obtained by these authors. We ignore the small differences within each type of hopping integrals, as a more important effect is the periodic modulation that appears with the BCDW. We believe that what is relevant in the present context is that $`t_p>t_c`$. The $`\alpha `$-BEDT-TTF lattice is then simply a rotated (by approximately 45<sup>o</sup>) version of our rectangular lattice with both $`t`$ and $`t_{}`$ = $`t_p`$ and $`t_{diag}=t_c`$. Our calculations (see Figs. 4, 5, 8 and 9) show that even at $`t_{}1`$ the correlated 1/4-filled band (or the dimerized dimer lattice) remains bond and charge-distorted, while based on the ..1100… ordering along the diagonals we have argued that $`t_{diag}`$ does not destroy this order (see section III.D). Furthermore, while $`t_{}>t_{}^c`$ destroys the SDW order (leaving the BCDW intact) by disordering the spins on the sites labeled ‘0’ (see section III), a small $`t_{diag}`$ will have a tendency to restore it, since now each small spin has two neighbors with spins of the same sign and one spin with opposite sign. Thus, the experimentally observed strong BCDW and a weak nearly vanishing SDW is exactly what we expect within our theory. Further evidence for a partial gap has been found in the <sup>13</sup>C-NMR studies of $`\alpha `$-(BEDT-TTF)<sub>2</sub>KHg(SCN)<sub>4</sub> in high magnetic fields, in a region where the system was previously thought to be a metal. In Fig. 12 we give a schematic of the spin arrangement in the $`\alpha `$-BEDT-TTF lattice; note that the underlying $`xy`$ symmetry in the isotropic 2D limit implies that there are two degenerate orthogonal 2D BCDW states here.
Since in $`\alpha `$-(BEDT-TTF)<sub>2</sub>MHg(SCN)<sub>4</sub> charge-ordering has also been discussed by Kino and Fukuyama , and more recently, by Seo , we should point out that the charge-ordering proposed by these authors is different from that in Fig. 12. Our charge-ordering in Fig. 12 is a rotated version of Fig. 2, where the occupancy scheme is …1100… along the x-direction and along the diagonals. The charge-ordering found by Kino and Fukuyama, and by Seo, assumes that the …1010… order dominates over the …1100… order. The ordering determined by Kino and Fukuyama is within a Hartree-Fock solution to the simple Hubbard model (zero intersite Coulomb interaction and zero e-ph coupling) and consists of a stripe structure with stack occupancies (c-direction in Fig. 12) alternating, i.e., stacks are either completely filled or completely devoid of holes). More recently, Seo has repeated these calculations by incorporating nearest neighbor Coulomb interaction $`V`$, but by treating $`U`$ within the Hartree-Fock approximation and the $`V`$ within the Hartree approximation. Different stripe structures, including that of Fukuyama and Kino, are found now, but once again, these are derived fundamentally from the occupancy scheme …1010… As has, however, been pointed out by previous authors , the …1010… charge ordering for the case of $`V`$ = 0 is an artifact of the Hartree-Fock approximation. Similarly, the Hartree approximation for $`V`$ also exaggerates the …1010… order while the Hartree-Fock treatment of the Hubbard term exaggerates the SDW order . This is precisely why these authors find very large magnetic moments in the $`\alpha `$-phase materials, in disagreement with experiments.
### D $`\kappa `$-(BEDT-TTF)<sub>2</sub>X
The deviation from the rectangular lattice is much stronger here . Crystal structure effects are very strong, and as a consequence the lattice is strongly dimerized, with the dimer sites forming an effective triangular lattice . The strong deviation from the rectangular lattice precludes direct comparisons against our theory. A more elaborate discussion of the spin arrangement will be given elsewhere. Here we only point out that (a) our calculations with the dimerized dimer lattice indicate that very large spin moments are possible when the intra-dimer hopping integrals are large compared to the inter-dimer hopping (see Fig. 10), in qualitative agreement with the observed very large magnetic moment in $`\kappa `$-(BEDT-TTF)<sub>2</sub>Cu(CN)<sub>2</sub>Cl , and (b) each dimer of BEDT-TTF molecules has the cartoon occupancy of 10 or 01 and the …1100… ordering along one direction and …1010… ordering along another (see Fig. 2), thereby reducing the spin frustration among the dimer sites forming the triangular lattice. In the absence of this population difference within each dimer cell (and the population difference is a consequence only of dimerization of the dimer lattice) the frustration within the triangular lattice would have severely reduced magnetic moments. We further point out that a pseudo-gap in the spectrum of magnetic excitations has been observed in the SDW phase of $`\kappa `$-(BEDT-TTF)<sub>2</sub>Cu\[N(CN)<sub>2</sub>\]Br and $`\kappa `$-(BEDT-TTF)<sub>2</sub>Cu(NCS)<sub>2</sub> ; this is in agreement with the dimerization of the dimer lattice, since without the second dimerization there should be no spin gap within the 2D antiferromagnet.
The material $`\kappa `$-(BEDT-TTF)<sub>2</sub>Cu<sub>2</sub>(CN)<sub>3</sub> merits separate discussion. This material is not antiferromagnetic, and measurement of spin susceptibility due to the BEDT-TTF components exhibits a steep drop below 10 K, suggesting SP-like behavior . This behavior is very similar to that in the BETS-based materials, which we discuss below, where we point out that for $`\rho `$ = 1/2, this behavior is expected for the case of large $`t_{}`$ ($`>t_{}^c`$).
### E $`\lambda `$-(BETS)<sub>2</sub>GaBr<sub>z</sub>Cl<sub>4-z</sub> (BETS = BEDT-TSF)
These materials, discovered only recently , are superconducting for $`0<z0.8`$ and semiconducting for $`0.8<z<2.0`$. Thus the proximity between a semiconducting and a superconducting state that characterizes the TMTSF and the BEDT-TTF is also a characteristic feature of the $`\lambda `$-BETS. In contrast to the TMTSF and the BEDT-TTF systems, however, the semiconducting state in the BETS is nonmagnetic and possesses a spin gap . Magnetic susceptibility studies indicate absence of anisotropy in the susceptibility, and no spin-flop transition (signature of antiferromagnetism) was found down to 10 K, which is close to the maximum superconducting critical temperature T<sub>c</sub> (onset 7.5 K, and even higher in certain samples) . The absence of the SDW is particularly perplexing here in view of the strong two-dimensionality predicted within extended Hückel band calculations .
The lattice structures of the $`\lambda `$-(BETS)<sub>2</sub>GaBr<sub>z</sub>Cl<sub>4-z</sub> are known . The stacking of the organic donor molecules is very similar to the $`\beta `$-BEDT-TTF systems, i.e., a nearly rectangular lattice with strong intrastack coupling, weaker transverse coupling, and very weak coupling along one diagonal. The nearly rectangular lattice permits comparison with our theory. One interesting feature of the lattice structure is that the intrastack bonds have strengths that are WSWS, exactly the structure expected for the $`r_{4k_F}0`$ lattice in Fig. 2(b) as well as the dimerized dimer lattice. We believe that while the difference between the strong and weak bonds is a crystal structure effect, the further dimerization of the dimer lattice is a consequence of the BCDW instability discussed here.
Hartree-Fock calculations by Seo and Fukuyama within an anisotropic Hubbard Hamiltonian gave an antiferromagnetic ground state instead of the nonmagnetic state. Since Hartree-Fock calculations overestimate antiferromagnetism, these authors then chose the $`U\mathrm{}`$ limit of Hubbard model to arrive at a dimerized, anisotropic 2D Heisenberg spin Hamiltonian, each lattice site of which corresponds to one dimer of the original BETS lattice. The antiferromagnetic-SP boundary within the 2D dimerized Heisenberg spin Hamiltonian has been investigated by Katoh and Imada using QMC simulations . For the longitudinal and transverse exchange integrals derived by Seo and Fukuyama, the QMC calculations still predict the antiferromagnetic structure . Seo and Fukuyama explain the spin gap in $`\lambda `$-BETS by claiming that the second dimerization of the dimer lattice (i.e., intermolecular distances WSWS, instead of WSWS) takes these systems to the 1D side of the 1D-2D antiferromagnetic-SP boundary, exactly as (TMTTF)<sub>2</sub>PF<sub>6</sub>, even though the actual transverse hopping integrals are large.
We believe that the problem faced by these authors arises entirely from their effective 1/2-filled band approximation. As seen in Fig. 10, the dimerization of the dimer lattice enhances the SDW in the region of small to intermediate $`t_{}`$ and therefore cannot be the origin of the spin gap or supposedly 1D behavior. Recall also that (TMTTF)<sub>2</sub>PF<sub>6</sub>, which is certainly on the 1D side of the 1D-2D boundary, is nonsuperconducting. In contrast, $`\lambda `$-BETS does become superconducting and that too at a $`T_c`$ that is considerably higher than that in the (TMTSF)<sub>2</sub>X, indicating what we believe to be strongly 2D character . We believe that the solution to this puzzle lies in recognizing the $`\rho `$ =1/2 character of the (BETS)<sub>2</sub>X. An essential difference between the effective 1/2-filled band model of Seo and Fukuyama and ours is that within the former, there are only two regions, nearly 1D and 2D, with the spin states as singlet and antiferromagnetic, respectively. Our work indicates that there are three distinct regions, singlet, antiferromagnet, and singlet again, as a function of increasing $`t_{}`$, independent of whether one assumes a 1/4-filled band or an effective 1/2-filled band. We therefore believe that a more natural explanation of the spin gap phase is obtained within our theory, with the singlet ground state in semiconducting BETS not being due to $`t_{}`$ that is too small, but due to a $`t_{}`$ that is too large ($`>t_c`$) to give SDW. This would be in agreement with the strong two-dimensionality of these systems . We believe that the same explanation also applies to the $`\kappa `$-(BEDT-TTF)<sub>2</sub>Cu<sub>2</sub>(CN)<sub>3</sub>, discussed in the above. We predict that experiments that can probe charge ordering will find two kinds of BETS molecules with different electronic populations, with greater charge densities on the two BETS molecules that are linked by the W bond.
## VII Possible Implications for Organic Superconductivity
What might be the implications of our BCSDW and BCDW states to organic superconductivity, the mechanism for which remains unclear despite two decades of research? We present here several partial responses to this challenging question.
First, given the the robustness of the BCDW/BCSDW in the exactly 1/4-filled band, we believe that the superconductivity must be the result of weak incommensurability in the actual materials. Specifically, we suggest and discuss in more detail below, that superconductivity arises from the pairing of commensurability defects in the background BCDW/BCSDW. That such weak incommensurability exists is strongly indicated by (i) the observation of a zero-energy mode in the optical conductivity of (TMTSF)<sub>2</sub>PF<sub>6</sub> and (TMTSF)<sub>2</sub>ClO<sub>4</sub>; (ii) the observation of Fröhlich mode sliding transport in the same materials ; and (iii) the observation of a “partially gapped Fermi surface” in the metallic region of $`\alpha `$-(BEDT-TTF)<sub>2</sub>KHg(SCN)<sub>4</sub>. Extremely interesting results in this context were reported by Komatsu et. al. , who showed that the superconductivity in $`\kappa `$-(BEDT-TTF)<sub>2</sub>Cu<sub>2</sub>(CN)<sub>3</sub> was due to a subtle change in the valence state of the Cu. The pure $`\kappa `$-phase material is a semiconductor with the Cu valence of +1. According to the authors of Ref. , the superconducting phase corresponds to a different material ($`\kappa ^{}`$ in the authors’ notation) in which some of the Cu (several hundred ppm) have acquired valency 2+. This was confirmed from ESR studies. The increase in Cu valency decreases the overall negative charge on the anion, and therefore the overall positive charge on the cation, providing a weak incommensurability that appears to be essential for superconductivity . This result lends credence to our suggestion that organic superconductivity arises from the pairing of commensurability defects within the BCDW/BCSDW background.
Second, the similarities between the organic and high temperature oxide superconductors have been pointed out in recent years by several research groups . One obvious apparent similarity between these two classes of superconductors is the proximity of the SDW to superconductivity. Our studies suggest that superconductivity in the organics is actually occurring at the interface of a Coulomb-induced BCDW that for a range of $`t_{}`$ coexists with the SDW. It therefore seems more likely that the pair binding is actually driven by the BCDW, and not the SDW, although it is probable that the symmetry of the pairing state may depend on the SDW (see below). As noted above, the experimental observation of superconductivity in the $`\lambda `$-(BETS)<sub>2</sub>GaCl<sub>4</sub> (where no proximate SDW is observed ) supports this view. An important implication of this perspective is that it casts doubt on recent spin-fluctuation theories of organic superconductivity within the effective 1/2-filled correlated electron model . The consequences of this conclusion from the organics for the high $`T_c`$ materials are unclear, but it is perhaps not irrelevant in this context to point out that evidence for superconductivity within the 2D nearly 1/2-filled Hubbard model, which for large $`U`$ has strongly AFM behavior, has remained elusive, despite more than a decade of intense research.
Third, there are striking similarities between this “doped” BCSDW/BCDW scenario and several other theoretical suggestions of superconductivity induced by doping of exotic “paired” semiconductors. As we have noted previously, the BCSDW and the BCDW states are very similar to the “paired electron crystal” (as opposed to the monatomic Wigner crystal) found by Moulopoulos and Ashcroft for the intermediate density electron gas . Superconductivity near the “melting” transition of the paired electron crystal has been conjectured by a number of authors in the past , even before the discovery of organic or high T<sub>c</sub> superconductivity. The commensurate BCDW is also qualitatively similar to a “negative U - positive V” effective 1/2-filled extended Hubbard model, with the effective lattice sites sites consisting of (a) the “occupied” pair (‘1–1’) of nearest neighbor sites, and (b) the “unoccupied” pair (‘0–0’) of nearest neighbor sites, in Fig. 1(c). Within this scenario, there is an effective attraction between the carriers on the “occupied” pair of dimer sites, but an effective repulsion between two pairs of occupied dimers. For models of this type, it is known that diagonal and off-diagonal long-range order can in principle coexist slightly away from commensurate filling . Further, Imada has studied a 2D spin-Peierls state (not possible in the monatomic 1/2-filled band) in which each composite site is again a dimer, with the dimer sites now having occupancies ‘10’ and ‘01’ (see Fig. 2(b) and note that the bonds between a ‘10’ and ‘01’ and between a ‘01’ and a ‘10’ are different, giving rise to a spin-Peierls-like behavior). His numerical simulations find evidence for superconductivity in the hypothetical doped 2D spin-Peierls state . Finally, Emery, Fradkin, and Kivelson have recently suggested that superconductivity can exist for incommensurate fillings in models that support stripe phases and in which a spin gap is present. Since the analysis in Ref. does not make direct contact with an initial microscopic Hamiltonian, but rather posits the form of the effective Hamiltonian in the vicinity of an unpinned stripe phase, it is not possible immediately to make detailed comparisons with our results. We can, however, make two comments. First, Ref. reflects the widespread belief that models within which a spin gap persists in the doped state are strong candidates for a microscopic theory of correlated superconductivity. Our preliminary numerical evidence suggests that both the BCDW and the BCSDW will continue to have a spin gap when doped; further work is in progress to confirm this. Second, regarding the attractive possibility that our BCSDW/BCDW state provides the background charge order within which commensurability defects may pair to form a superconducting state, we note that the occupancy schemes in Fig. 1(c) and Fig. 2(a) and (b) resemble intersecting stripes, where each stripe is obtained by connecting the ‘1–1’ bonds along the x- and x+y (–x+y) directions.
The possible BCSDW/BCDW to superconductor transition in the organic CTS clearly requires further study. We close our present discussion of this topic with comments on three important open issues: (i) the possible mechanism for superconducting pairing; (ii) the problem of phase separation; and (iii) the symmetry of the order parameter.
First, the possible mechanism for pairing of commensurability defects within the 2D BCDW can be visualized most simply in the rigid bond limit, where nearest-neighbor bonds retain their individual distortions independent of the occupancies of the sites linked by these bonds. The commensurate BCDW in this limit can be viewed as consisting of “quasimolecules”, where each quasimolecule is a “1-1” dimer. If two holes are now removed from the system, it is energetically preferable to destroy one “quasichemical bond,” thereby creating an intersite (small) bipolaron, as opposed to destroying two bonds and creating two polarons. Thus, within the WSWS structure ($`t_S>t_W^{}>t_W`$), each W bond acts as a “negative-U” center in the rigid bond limit. It is of course highly unlikely that superconductivity can be obtained, at least at the experimental T<sub>c</sub>, due to condensation of small bipolarons , so this might appear to present a serious problem for this proposed mechanism. In fact, when one goes beyond the oversimplified rigid bond limit to the full model that correctly reflects the cooperation between e-e and e-ph interactions in the 1/4-filled band, one finds that the actual commensurability defects are more like the extended, “resonant” (and therefore mobile) bipolarons that are indeed candidates for explaining superconductivity in strongly correlated systems . To understand this in detail, consider again the weakly incommensurate BCDW, starting from the 1D limit, but now with the e-ph interactions included. Below the 4k<sub>F</sub> transition temperature T$`_{4k_F}`$, but above the 2k<sub>F</sub> transition temperature T$`_{2k_F}`$, incommensurability leads to fractionally charged solitons with charge e/2, and each vacancy creates two such defects . Previous work has assumed that the soliton charge remains e/2 even below the 2k<sub>F</sub> transition, which implies that two vacancies create four such defects . However, Ref. assumes that the site charge density remains uniform even below the 1D 2k<sub>F</sub> SP (dimerization of the dimer lattice) transition, which is precisely what we have shown here not to be the case. Indeed, as a consequence of this spatial charge inhomogeneity (charge ordering), the “solitons” now acquire integer charge (i.e., two fractionally charged solitons bind to give a single soliton with charge +e), as we have shown explicitly elsewhere . A pair of added vacancies within the 1D BCDW below T$`_{2k_F}`$ therefore creates (only) two solitons. In the strictly 1D limit, these do not bind, but with increasing $`t_{}`$, one expects binding to a large bipolaron. The source of this binding is precisely the same as the source of soliton confinement in coupled chains of polyacetylene : in the region between the two defect centers the phase relationships between the BCDW’s on neighboring chains is different from the preferred one (viz., $`\pi `$), and therefore a large separation between the defect centers would increase the energy (linearly with increasing separation). There exists therefore a space-dependent interaction between the polarons, which is repulsive at short range but attractive at some ($`t_{}`$-dependent) intermediate range. The bipolaron size, as well as its dimensionality, depends on $`t_{}`$ (as well as on $`U`$ and $`V`$). There is currently limited analysis of 2D large bipolarons in the strongly correlated limit, although some results suggest that these can indeed be mobile . Within this scenario, superconductivity occurs due to the condensation of these large bipolarons, which is not precluded by the theoretical analysis of Ref. . Resolving the question of whether static distortion is sufficient, or whether dynamical phonons will have to be included, will require further work.
Second, in many existing models of superconducting pairing involving correlated electrons, the interactions that bind two particles also lead to phase separation, since the attraction producing pairing does not saturate. Perhaps the best known example of this is the t-J model away from 1/2-filling. In contrast, within any “negative U” model there does exist a saturation in this attraction (since a single site can at most have two electrons), and the analogy between our BCDW model and the effective 1/2-filled “negative U – positive V” case suggests that phase separation will also not occur here. Further, the immediately previous discussion of the proposed binding mechanism makes clear that with small but macroscopic (say, 1%) concentration of commensurability defects, there is no particular energetic advantage in creating additional polarons or bipolarons proximate to the original bipolaron (in contrast to, say, the t-J model, where there is such an energetic advantage).
Third, what symmetry do we expect for the superconducting order parameter in our model? This is clearly a challenging issue, particularly since even with the same BCDW background the pairing symmetry in the highly anisotropic TMTSF might be different from that in the more two-dimensional BEDT-TTF and BETS. Several recent experiments have presented evidence consistent with nodes in the superconducting gap function in the BEDT-TTF . This is reminiscent of d-wave symmetry of the superconducting order parameter in the high temperature copper oxide based superconductors. On the other hand, Lee at al. have recently presented evidence suggesting that a spin triplet p-wave pairing is necessary to explain data in (TMTSF)<sub>2</sub>PF<sub>6</sub>, where the upper critical field H<sub>c2</sub> shows no saturation with the field in the plane of the organic molecules and exceeds the Pauli paramagnetic (Clogston) limit expected to hold for singlet superconductors and the temperature dependent Knight shift measurements of <sup>77</sup>Se show that the spin susceptibility remains unaltered through the superconducting T<sub>c</sub>. Within the continuum RG theories triplet superconductivity does indeed occur proximate to the SDW. However, within the discrete extended Hubbard model, triplet superconductivity occurs within a very narrow region of the the “positive $`U`$ – negative $`V`$” sector of the $`UV`$ phase diagram, bounded by the SDW phase and a phase segregated phase . Triplet pairing thus will not only require a change in sign of the nearest neighbor Coulomb interaction within our original Hamiltonian of Eq. (1), but will also occur for a very narrow critical range of this parameter. But to resolve definitively the issue of the symmetry of the order parameter within our model will be a non-trivial task, as the consequences of the interplay between e-e and e-ph interactions, as well as the effects of anisotropy, must be properly understood.
## VIII Acknowledgments
We thank Jim Gubernatis and Shiwei Zhang for discussions regarding the CPMC method, Eduardo Fradkin and Philip Phillips for discussions of their recent theoretical results, and Stuart Brown, Paul Chaikin, and Andrew Schwarz for discussions of their experiments. We also acknowledge stimulating discussions with Zlatko Tesanovic. Work at the University of Illinois was supported in part by the grants NSF-DMR-97-12765 and NSF-GER-93-54978 and by an allocation of supercomputer time through the NRAC program of the NCSA.
## Appendix 1: The AFM-singlet transition for weak anisotropy
Our goal here is to understand the second AFM-to-singlet transition that should occur in the quarter-filled band for large $`t_{}`$ from a perspective that is different from the one presented in section III. Specifically, we refer to the antiferromagnetic dimer lattice of Fig. 1(b) with weak intrachain interdimer links, and the frozen valence bond state of Fig. 1(c), in which one of the interdimer links (W in the notation of section III) is now stronger than the other (W in the notation of section III), and is a singlet bond. We aim to give variational arguments at the simplest level that (a) point out the difference between $`\rho `$ = 1 and $`\rho `$ = 1/2, and (b) indicate that the frozen valence bond state of Fig. 1(c) dominates over the antiferromagnetic dimer lattice of Fig. 1(b) for large $`t_{}`$ and therefore the dimerization of the dimer lattice is unconditional. The argument given below is not to be considered as a proof, but rather, it provides convincing physical motivation for the numerical work discussed in section IV.
Note that our discussion here is limited to the relative stabilities of two insulating states, and not the competition with any metallic state. We consider only the simple Hubbard Hamiltonian with $`V`$ = 0 (since for $`\rho `$ = 1/2 the periodicity of the CDW is the same for all $`V<V_c`$ and while for $`\rho `$ = 1 the $`V`$ merely reduces the effective on-site correlation) for $`t_{}`$ = 1. For completeness we begin by repeating the variational argument for the dominance of the SDW over the BOW in $`\rho `$ = 1. Consider the Heisenberg antiferromagnetic spin Hamiltonian
$$H=J\underset{ij}{}S_i.S_j$$
(15)
Consider also the singlet variational state (1,2)(3,4)….(N-1,N), with singlet bonds between nearest neighbors in 1D and the Néel state $`\mathrm{}\mathrm{}`$. The energy of an isolated singlet bond is –(3/4)J while that of a 2-site Néel state is –(1/4)J. The overall variational energy of the singlet state in 1D is –(3/8)NJ and that of the Néel state –(1/4)NJ, so that the singlet dominates over the Néel state in 1D. In the 2D isotropic N $`\times `$ N lattice, we compare (1) the frozen valence bond state in which each chain still has the same spin couplings as in 1D (note that at the level of our approximation the relative phases between consecutive chains make no difference), and (2) the 2D Néel state. The variational energy of the frozen valence bond state is –(3/8)N<sup>2</sup>J, but now because of the larger number of nearest neighbors the energy of the Néel state has a lower value –(1/2)N<sup>2</sup>J, which therefore dominates over the frozen valence bond state. Thus for $`\rho `$ = 1 in 1D the SP state dominates, while in 2D the SDW wins over the SP state. While this argument may appear simplistic, it nevertheless predicts the dominance of the antiferromagnet over the singlet in $`\rho `$ = 1.
Consider now the isotropic 2D dimerized $`\rho `$ = 1/2 lattice with moderately strong dimerization (Fig. 1(b)). The effective 1/2-filled band is clearly a SDW, with the dimerization pattern being necessarily “in-phase” between consecutive chains, as shown in Fig. 1(b) (to prevent confusion in what follows we have not shown the bonds in Fig. 1(b), but a strong bond between the two sites within the parentheses and weaker interdimer bonds have been assumed) The individual site populations are equal in this state and each is exactly 1/2. Our contention is that this state has a higher variational energy than that reached by further dimerization of the dimer lattice, which gives the $`\rho `$ = 1/2 frozen valence bond state shown in Fig. 1(c), where there occur interdimer singlet bonds and site occupancies …1100… (the singlet bonds in Fig. 1(c) are between the “occupied” sites). The reason for this is that unlike in $`\rho `$ = 1, the exchange integrals that describe the effective Heisenberg models in the SDW and the singlet are now different, in spite of the fact that both Heisenberg systems are derived from the same Hubbard Hamiltonian. In Fig. 1(c), we are considering isolated singlet bonds, with site occupancies of 1, and $`J`$ is clearly 2$`t^2/U`$, exactly as for $`\rho `$ = 1. In Fig. 1(b), on the other hand, the exchange integral has to correspond to a true $`\rho `$ = 1/2 system, since each site occupancy is now 1/2. The exchange integral J for arbitrary $`\rho `$ in 1D is 2($`t^2/U)\rho [1sin(2\pi \rho /)2\pi \rho ]`$ , so that for $`\rho `$ = 1/2 we have J = (1/2)J along each chain (the x-direction). This expression is strictly true only in the 1D undistorted chain, and for the distorted 1D chain or in 2D one needs to calculate J from comparing singlet-triplet gaps within the structure corresponding to Fig. 1(b) and within the 1/2-filled band. We have calculated these gaps for finite lattices separately for the longitudinal and transverse directions and have found that while J = (1/2)J is quite accurate for the longitudinal direction, the J in the transverse direction is even smaller (the difference between the longitudinal and transverse directions originates from dimerization along the longitudinal direction only), with the restriction that only interdimer hops lead to spin exchange. Even if we consider the largest possible value for J = J/2, the variational energy of the Néel state in Fig. 1(b) is then –(1/2)(N<sup>2</sup>/2)J = –(1/8)N<sup>2</sup>J, while that of the frozen valence bond state in Fig. 1(c) (with N<sup>2</sup>/4 singlet bonds) is –(3/4)N<sup>2</sup>/4J = –(3/16)N<sup>2</sup>J. Thus the frozen valence bond state dominates over the dimer SDW, implying that the dimerization of the dimer lattice is unconditional, and the difference from the simpler $`\rho `$ = 1 case arises from the smaller (by factor of 2) exchange integral in the uniform dimer lattice of Fig. 1(b). The above approach is obviously simplistic, but no more so than the physical argument for the dominance of the SDW in $`\rho `$ = 1.
## Appendix 2: Numerical Methodology
### A Contstrained Path Monte Carlo (CPMC)
The CPMC ground-state quantum Monte Carlo method uses a constraining trial wavefunction to eliminate exponential loss of signal due to the Fermion sign problem. Although the method has been thoroughly bench-marked against known results for the Hubbard model, the method is non-variational, and it is important to check its accuracy in every new system against exact results and to use a variety of different trial wavefunctions. The current application is different from previous ones in including the $`V`$ interaction, as well as in the choice of the bandfilling (previously tested cases were for bandfillings close to 1/2). Furthermore, previous work has shown that most accurate results are obtained when the trial noninteracting wavefunctions have “closed-shell” nondegenerate configurations. In the next subsection it is shown that the proper boundary conditions for simulating coupled 1/4-filled band chains involves having 4n electrons (where n is an integer) per chain. This implies degeneracy of the trial wavefunctions at $`t_{}=0`$ and again near $`t_{}`$ = 1. It is thus necessary to check the accuracy of the method for our purpose, and this was done by comparing CPMC results with the exact results for the 8 $`\times `$ 2 lattice.
Fig. 13 summarizes the results of the bench-mark energy calculations for an 8$`\times `$2 lattice, periodic in the x-direction, with $`U=6`$ and $`V=1`$. Both undistorted and and the 2k<sub>F</sub> ($`r_{2k_F}0,r_{4k_F}=0`$ in Eq. (7)) bond distorted systems were compared, where for the the uniform lattice all hopping integrals were taken to be 1.0, while for the distorted system they were 1.14, 1.0, 0.86 and 1.0 (as in Fig. 2(a)). For this amplitude of the 2k<sub>F</sub> distortion,
the absolute value of $`\mathrm{\Delta }E`$ is only 0.3% of the total energy (at $`t_{}=0.4`$). Such a small energy difference is not easy to measure within quantum Monte Carlo. We note that energy differences of this order of magnitude have also been calculated using CPMC to study hole binding in the the 3-band Hubbard model . The CPMC values are scaled for $`\mathrm{\Delta }\tau 0`$ from $`\mathrm{\Delta }\tau =0.05`$ and $`\mathrm{\Delta }\tau =0.1`$ to remove the Trotter discretization error. The trial wavefunctions used were either the free-electron wavefunction, or an Unrestricted Hartree-Fock (UHF) wavefunction with $`U=2`$ and $`V=0.5`$. Hartree-Fock wavefunctions with larger $`U`$ and $`V`$ gave less accurate results, probably due to the tendency of UHF to exaggerate AFM correlations. In Fig. 13 the UHF trial functions produced larger errors than the free-electron trial functions for the distorted system at small $`t_{}`$ because the SDW correlations there are exaggerated by the UHF approximation. The CPMC systematic errors are largest at small $`t_{}`$ ($`<0.2`$) and large $`t_{}`$ ($`>0.8`$) possibly due to the degeneracies in the one-electron occupancies at $`t_{}=0`$ and $`t_{}=t`$. However, at large $`t_{}`$, the UHF trial wavefunction produced slightly more accurate results for the 8$`\times `$2 distorted lattice possibly because the numerically-derived UHF wavefunction breaks some of the symmetry of the non-interacting wavefunction. In the intermediate $`t_{}0.4`$ regime, the CPMC energies are indistinguishable from the exact energies within the statistical error. The accuracy of the CPMC method in this region is very reassuring, since for the noninteracting case, at $`t_{}=0.4`$ the distortion has already vanished.
In addition to comparing energies, we have also compared charge densities and spin-spin correlation functions. Table I compares the charge densities and spin-spin correlations computed by CPMC for the 8$`\times `$2 distorted lattice at $`t_{}=0.4`$. The agreement with the exact result is not as good as for the energy (typically 1-5% for the charges and 5-10% for the spin-spin correlations), but is more than adequate to identify the presence and periodicity of the broken symmetry states. Thus in general, we find the CPMC results are close to the exact results
| $`\rho _j`$ | | | $`s_i^zs_j^z`$ | | |
| --- | --- | --- | --- | --- | --- |
| j | exact | CPMC | i,j | exact | CPMC |
| 1 | 0.4799 | 0.4756(6) | 1,9 | -0.06095 | -0.0585(7) |
| 2 | 0.5201 | 0.5250(6) | 1,10 | -0.03215 | -0.0312(7) |
| 3 | 0.5201 | 0.5240(6) | 1,11 | 0.01408 | 0.0161(7) |
| 4 | 0.4799 | 0.4772(6) | 1,12 | -0.02698 | -0.0231(6) |
| | | | 1,13 | -0.07299 | -0.0687(6) |
| | | | 1,14 | -0.03085 | -0.0268(5) |
| | | | 1,15 | 0.01408 | 0.0158(6) |
| | | | 1,16 | -0.02552 | -0.0239(7) |
Table I. Comparison of CPMC and exact charge density and spin-spin correlations for an 8$`\times `$2 system with $`U=6`$, $`V=1`$, t=0.4, with the same distortion of hopping integrals as in Figure (13). Sites on the first chain are numbered 1 – 8, those on the second chain 9 – 16.
for both energies and correlation functions, except for very small or large $`t_{}`$.
### B Boundary Conditions
As noted above, we determine the proper combinations of lattices and boundary conditions for the numerical simulations by the requirement that nonzero $`t_{}`$ destabilizes the BCDW for noninteracting electrons with those boundary conditions on that particular finite lattice: i.e., we require the finite lattices to reflect correctly the known behavior of the noninteracting case in the thermodynamic limit.
Consider an N $`\times `$ M lattice, with N sites along the chain and M chains. To avoid odd/even effects, consider an even number of electrons per chain. This number can then be either 4n or 4n + 2, where n is an integer. To obtain a 1/4-filled band, one can then have N = 4n $`\times `$ 2 = 8n or N = (4n + 2) $`\times `$ 2= 8n + 4. The proper N for our purpose is N = 8n (i.e., 4n electrons per chain). This follows from the one-electron energy levels of coupled chains with 4n and 4n + 2 electrons per chain. In Fig. 14 below we have shown the one-electron energy levels for the undistorted 8 $`\times `$ 2 (top panel, labeled a) and 12 $`\times `$ 2 (bottom panel, labeled b) lattices (both periodic in the x-direction), corresponding to $`t_{}`$ = 0 on the left and 0.1 on the right in both cases. In the 8 $`\times `$ 2 lattice, the degeneracy at $`t_{}`$ = 0 will lead to spontaneous distortion. For nonzero $`t_{}`$ and a $`\pi `$-phase shift between chains (which gives lower energy than phase shifts of 0 or $`\frac{\pi }{2}`$), the pairs of one-electron levels that are coupled by phonons with wave-vector (2k<sub>F</sub>, $`\pi `$) are (– $`\frac{2\pi }{8}`$, 0) and (+ $`\frac{2\pi }{8}`$, $`\pi `$); and (+ $`\frac{2\pi }{8}`$, 0) and (– $`\frac{2\pi }{8}`$, $`\pi `$). The finite gap that occurs for $`t_{}`$ 0 between each pair of one-electron
levels coupled by the (2k<sub>F</sub>, $`\pi `$) phonon indicates absence of nesting and the destabilization of the distortion. This energy gap increases with $`t_{}`$, leading to a decrease in $`\mathrm{\Delta }`$E with $`t_{}`$ for N = 8n (see Fig. 15 for details), as occurs in the thermodynamic limit. In contrast, consider the 12 $`\times `$ 2 lattice, in which the one-electron ground state is non-degenerate. There is now a nonzero energy gap between the levels coupled by the 2k<sub>F</sub> electron-phonon interaction already at $`t_{}`$ = 0 ($`k_x`$ = – $`\frac{2\pi }{12}`$ and $`k_x`$ = + $`\frac{4\pi }{12}`$; $`k_x`$ = + $`\frac{2\pi }{12}`$ and $`k_x`$ = – $`\frac{4\pi }{12}`$). With nonzero $`t_{}`$, and once again a $`\pi `$-phase shift between the chains, the energy gap between the levels (– $`\frac{2\pi }{12}`$, $`\pi `$) and (+ $`\frac{4\pi }{12}`$, 0), and similarly that between the levels (+ $`\frac{2\pi }{12}`$, $`\pi `$) and (– $`\frac{4\pi }{12}`$, 0), decreases, indicating that the tendency to distort here increases with inter-chain coupling, at least for small to moderate $`t_{}`$.
For large N, the difference between N = 8n and N = 8n+4 vanishes, as is shown in Fig. 15, where Figs. 15 (a) and (b) show the behavior of $`\mathrm{\Delta }E(t_{})`$ for N = 8n and 8n+4, respectively. The qualitative behavior (destabilization of the distortion) is the same for all N = 8n, and monotonically decreasing $`\mathrm{\Delta }`$E is also seen for N = 8n+4 for large N, but finite size effects (increasing $`\mathrm{\Delta }`$E at small
to intermediate $`t_{}`$) are strong even for N = 28, a chain length already too large for accurate 2D many-body calculations. The correct qualitative behavior of all N = 8n is the basis of our choice of these N.
In contrast to the choice of N, there is no immediate restriction on the choice of M, the number of chains, except that M should be even, to avoid even/odd effects. M = 4n and 4n + 2 both show the same qualitative behavior, as seen from the plots of $`\mathrm{\Delta }`$E versus $`t_{}`$ in Fig. 16, for several M = 4n lattices (M = 4n + 2 are included in Fig. 15). Thus both M = 4n and 4n + 2 are appropriate. Our choice of M = 4n + 2 is based on two reasons. First, exact diagonalization calculations on the 8$`\times `$2 lattice allows comparisons to results obtained within CPMC, and the exact diagonalizations cannot be done for the next larger appropriate lattice, viz. 8 $`\times `$ 4. Second, the M = 4n lattices are characterized by one-electron Fermi level degeneracies for $`t_{}`$ $``$ 0 (even though the degenerate levels are not coupled by (2k<sub>F</sub>, $`\pi `$) phonons), and the absence of a single well-defined one-electron wave-function would make the CPMC calculations considerably more difficult than for M = 4n + 2 lattices, which have non-degenerate one-electron levels for nonzero $`t_{}`$.
### C UHF calculations of bond distortion
As discussed in the subsection on methods in this Appendix, UHF trial wavefunctions for the CPMC calculations were constructed for regions where one-electron wavefunctions were degenerate. Since UHF calculations give reasonably correct results in the small $`U,V`$ range it is also of interest to determine the tendency of the 2D lattice to distort within the UHF approximation. One advantage of this procedure is that much larger lattices than those discussed in section IV can be tested. We report these results here. We have chosen relatively small $`U`$ and $`V`$ for two reasons: the UHF procedure does not converge well for larger interactions, and the smaller
values of $`U`$ and $`V`$ gave better results when used as a CPMC trial function (compared to a numerically exactly solved 8$`\times `$2 system). Fig. 17 shows the normalized energy gain from a 2k<sub>F</sub> distortion for two different lattices, within the UHF approximation. The UHF results show that $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ remains close to 1 for at least up to $`t_{}`$ 0.4, indicating a tendency to persistent distortion up to this $`t_{}`$. Although $`\mathrm{\Delta }E/\mathrm{\Delta }E_0`$ begins to decrease at still larger $`t_{}`$, these calculations are for a relatively small value of $`U`$, and as discussed in section III, the range of $`t_{}`$ over which the distortion should persist increases with $`U`$. Thus the qualitative effects of the e-e interaction are already visible within the UHF approach at small $`U`$, while a fully persistent broken symmetry state will occur only for larger values of the e-e interaction that are beyond the scope of the UHF. Given that the UHF approximation predicts a vanishing of the bond dimerization in the 1/2-filled band for a fairly small $`U_c`$ (the actual magnitude of $`U_c`$ depends on $`\alpha `$), in contrast to the correct result that there is an enhancement of the dimerization for $`0<U<4`$, the present results, showing a persistence of the distortion for moderate $`t_{}`$, is initially perplexing. The reason for the correct prediction in this case is that the UHF exaggerates the SDW, which destroys the BOW in the 1/2-filled band, but has a co-operative interaction with the 1/4-filled band BOW for small to moderate $`t_{}`$.
## Appendix 3: Spin character of the ground state
As discussed in Appendix 2, the proper boundary condition for the numerical evaluation of the electronic energy gained upon bond or site distortion in $`\rho `$ = 1/2 involves finite N $`\times `$ M lattices with N = 8n. This requires the number of electrons per chain to be 4n, and it is known that in 1D periodic undistorted rings with $`\rho `$ 1, the ground state has overall spin S = 1 instead of 0 for any nonzero $`U`$.
| $`t_{}`$ | undistorted | | 2k<sub>F</sub> distortion | |
| --- | --- | --- | --- | --- |
| | S=0 | S=1 | S=0 | S=1 |
| 0.01 | -9.335651 | -9.335637 | -9.352522 | -9.352228 |
| 0.025 | -9.337570 | -9.336944 | -9.354380 | -9.353739 |
| 0.05 | -9.344122 | -9.341546 | -9.361083 | -9.358425 |
Table II: The S=0 and S=1 energies of the 8$`\times `$2 undistorted and 2k<sub>F</sub> bond-distorted lattice for $`U=6`$ and $`V=1`$. The lowest energy is S=0 for both undistorted and distorted cases.
The spin of the ground state of the distorted periodic ring depends on its size and the magnitude of the Hubbard $`U`$. For the values of the correlation parameters and bond distortion parameter in Fig. 4, the ground state in the N = 8 distorted periodic ring has S = 1, while the N = 16 ground state has S = 0. Thus the $`\mathrm{\Delta }E_0`$ in Fig. 4 for nonzero e-e interaction corresponds to $`\mathrm{\Delta }E_{TT}`$ (i.e., the energy gained by the triplet state upon bond distortion) for N = 8, and to $`\mathrm{\Delta }E_{TS}`$ (undistorted state in S = 1, distorted state in S = 0) for N = 16. Whether or not the comparisons of the zero and nonzero $`t_{}`$ are then meaningful is an important question. We present here the detailed results of three different sets of calculations, each of which indicates that our interpretation of the results of Fig. 4 (viz., strong tendency of the interacting 1/4-filled lattice to distort at arbitrary $`t_{}`$) is correct.
First, we have calculated the exact ground states of the 8 $`\times `$ 2 lattice for $`t_{}`$ as small as 0.01. In Table II we have given the S = 0 and S = 1 energies of the 8 $`\times `$ 2 lattice for $`U`$ = 6 and $`V=1`$, for three small values of $`t_{}`$. The coupled chain system is in the S = 0 state for both zero and nonzero bond distortion for the smallest nonzero $`t_{}`$. The important point now is that instead of choosing the single isolated chain as the standard in Fig. 4, we could have also chosen the coupled chain system with $`t_{}`$ = 0.01 as the standard, provided the distortion of the $`t_{}`$ = 0.01 lattice is also unconditional. Even if the nesting ideas were valid, we believe that the coupled chain system with $`t_{}`$ = 0.01 is unconditionally distorted and then the results in Table II clearly show that $`\mathrm{\Delta }E`$ increases with further increase in $`t_{}`$, indicating enhanced distortion relative to $`t_{}`$ = 0.01. The error bars in the CPMC calculations prevent us from performing similar calculations for the 8 $`\times `$ 6 or the 16 $`\times `$ 6 lattices, but the overall similarities in the (a) occupancies of the one-electron levels for nonzero $`t_{}`$ and (b) $`\mathrm{\Delta }E`$ behavior, especially in the region $`t_{}0.4`$, preclude different behavior at small nonzero $`t_{}`$.
We performed a second set of calculations for the 8 $`\times `$ 2 lattice for very small values of $`U`$ (with $`V`$ = 0). Note that if the persistent distortion implied in Fig. 4 were merely due to our choosing the wrong reference point $`t_{}`$ = 0
(since exactly at this point $`\mathrm{\Delta }E_0`$ = $`\mathrm{\Delta }E_{TT}`$), an apparently enhanced distortion for nonzero $`t_{}`$ should occur for all nonzero $`U`$ (since the single chain is S = 1 for all nonzero $`U`$, while the coupled chain system has S = 0 for all nonzero $`t_{}`$ and $`U`$). On the other hand, if the results in Fig. 4 are due to the confinement effect discussed in section III.D, then enhanced/persistent distortion should occur only above a threshold e-e interaction: for weak e-e interaction the behavior should resemble that of the noninteracting lattice (with enhanced or persistent distortion occurring for a small range of $`t_{}`$ near $`t_{}`$ = 0). We show here the results of calculations at small $`U`$ for site distortion (as opposed to bond distortion), since we also report calculations for very large $`U`$ below, and the bond distortion pattern (the magnitude of $`r_{4k_F}`$) is $`U`$-dependent, but the site distortion pattern is not. The distorted lattice here has site energies + $`ϵ`$, \+ $`ϵ`$, – $`ϵ`$, – $`ϵ`$ (with $`ϵ=0.1`$) over four consecutive sites, and a $`\pi `$-phase shift between the two periodic rings. Since the 2k<sub>F</sub> CDW has a synergetic coexistence with both the r$`_{4k_F}`$ = 0 BOW (Fig. 2(a)) and the r$`{}_{4k_F}{}^{}`$ 0 BOW (Fig. 2(b)) a persistent CDW also implies persistent BOW; we have confirmed this by calculating the expectation values of the bond orders. In Fig. 18(a) we show the $`\mathrm{\Delta }E`$ behavior as a function of $`t_{}`$ for both $`U=0.5`$ and $`U=1`$. Decreasing $`\mathrm{\Delta }E`$ with $`t_{}`$ is a clear signature that the tendency to distortion here decreases with increasing two dimensionality, since confinement at these small $`U`$ is not sufficient to give persistent distortion. Even though these calculations are with fixed site energies, the expectation values of the charge densities depend on $`t_{}`$, and our calculated CDW amplitudes decrease with $`t_{}`$, as expected from Fig. 18(a). This behavior is exactly opposite to that in Fig. 5(b), indicating again a decrease in distortion with $`t_{}`$ at small $`U`$. Finally, we emphasize that similar calculations have also been done with fixed 2k<sub>F</sub> bond distortion, and once again we observe decreasing $`\mathrm{\Delta }E`$ and CDW amplitude with increasing $`t_{}`$.
We performed a third set of calculations with very large $`U=100`$, again with the same site distorted lattice but now with $`ϵ`$ = 0.2, since at this very large $`U`$, the energy gained upon distortion for $`ϵ`$ = 0.1 is very small. The resultant BOW here has strong 4k<sub>F</sub> component ($`r_{4k_F}0`$), and this is why the distorted lattice was chosen to be the 2k<sub>F</sub> CDW in this and the above calculations, such that meaningful comparisons between these extreme cases can be made. At this large $`U`$, the energy difference between S = 0 and S = 1 states is negligible. For example, for the 1D 8-site periodic ring $`\mathrm{\Delta }E_{SS}`$ (electronic energy gained in the S = 0 subspace, with both undistorted and distorted states in S = 0) = 0.06222, while $`\mathrm{\Delta }E_{TT}`$ (electronic energy gained in the S = 1 subspace, with both undistorted and distorted states in S = 1) = 0.06224. Fig. 18(b) shows the $`\mathrm{\Delta }E`$ behavior as a function of $`t_{}`$ (with $`ϵ`$ = 0.2 now). An enhanced CDW (and therefore BOW) is seen from as a function of $`t_{}`$, where the singlet and triplet data points at $`t_{}`$ = 0 are the same. As seen in Fig. 18(b), the $`\mathrm{\Delta }E`$ for nonzero $`t_{}`$ is weakly enhanced now even when compared to $`\mathrm{\Delta }E_{SS}`$ at $`t_{}`$ = 0. Once again, the behavior of the CDW amplitude is in complete agreement with the prediction from Fig. 18(b), viz., a weak enhancement of the CDW amplitude with $`t_{}`$.
Considering the above three different sets of results, we therefore conclude that the results in Fig. 4, Fig. 5(b) and Fig. 9 are not artifacts, and the persistent distortion is real and a true confinement effect, as would also be expected from the “variational” arguments in Appendix 1. |
warning/0003/cond-mat0003435.html | ar5iv | text | # Quantum Phase Transitions in the Shastry-Sutherland Model for 𝐒𝐫𝐂𝐮_𝟐(𝐁𝐎_𝟑)_𝟐
\[
## Abstract
We investigate the quantum phase transitions in the frustrated antiferromagnetic Heisenberg model for $`\mathrm{SrCu}_2(\mathrm{BO}_3)_2`$ by using the series expansion method. It is found that a novel spin-gap phase, which is adiabatically connected to the plaquette-singlet phase, exists between the dimer and the magnetically ordered phases known so far. When the ratio of the competing exchange couplings $`\alpha (=J^{}/J)`$ is varied, this spin-gap phase exhibits the first- (second-) order quantum phase transition to the dimer (the magnetically ordered) phase at the critical point $`\alpha _{c1}=0.677(2)`$ ($`\alpha _{c2}=0.86(1)`$). Our results shed light on some controversial arguments about the nature of the quantum phase transitions in this model.
\]
Two-dimensional (2D) antiferromagnetic quantum spin systems with the spin gap have been the subject of considerable interest. A typical compound found recently is $`\mathrm{SrCu}_2(\mathrm{BO}_3)_2`$, in which the characteristic lattice structure of the $`\mathrm{Cu}^{2+}`$ spins (see Fig. 1) stabilizes the singlet ground state. This system has been providing a variety of interesting phenomena such as the plateaus in the magnetization curve observed at $`1/3,1/4`$ and $`1/8`$ of the full moment. The spin system may be described by the 2D Heisenberg model on the square lattice with some diagonal bonds which is referred to as the Shastry-Sutherland model, as pointed out by Miyahara and Ueda. The key structure with the orthogonal dimers shown in Fig. 1 makes the system unique and particularly interesting among 2D spin-gap compounds. In this frustrated system, there may occur non-trivial quantum phase transitions when the nearest-neighbor coupling $`J`$ and the next-nearest-neighbor coupling $`J^{}`$ are varied. Albrecht and Mila discussed the possibility of a helical phase between the dimer and the magnetically ordered phases by means of the Schwinger boson mean-field theory. Recent theoretical studies, however, have suggested that there may not be such a helical phase, but the first-order phase transition occurs from the dimer to the ordered phases . Furthermore, more recent study claims that the phase transition should be of the second order with a non-trivial critical exponent $`\nu =0.45(2)`$. These controversial conclusions may come from the fact that the quantum phase transition in the Shastry-Sutherland model suffers from the strong frustration due to the competing exchange interactions $`J`$ and $`J^{}`$, and therefore a careful treatment should be necessary to figure out the correct nature of the phase transition. In particular, we have to keep in mind that such a strong frustration may possibly stabilize another spin-gap phase distinct from the dimer phase.
In this paper, by calculating the ground state energy, the staggered susceptibility and the spin gap by means of the series expansion method, we find that there should exist a novel spin-gap phase with the disordered ground state, which is stabilized by the strong frustration, between the dimer and the magnetically ordered phases. The spin-gap phase found in this paper undergoes the first- (second-) order quantum phase transition to the dimer (the ordered) phase, when the exchange couplings $`J`$ and $`J^{}`$ are varied. The existence of the new phase can resolve controversial conclusions deduced for the quantum phase transitions in this frustrated model. We also point out that the material $`\mathrm{SrCu}_2(\mathrm{BO}_3)_2`$ lies around the phase boundary between these two spin-gap phases, which may give a natural interpretation for the 1/8-plateau formation in the magnetization curve.
To investigate the frustrated spin system for the compound $`\mathrm{SrCu}_2(\mathrm{BO}_3)_2`$, we consider the 2D quantum Heisenberg model (Shastry-Sutherland model), which is described by the following Hamiltonian
$`H=J{\displaystyle \underset{n.n.}{}}𝐒_i𝐒_j+J^{}{\displaystyle \underset{n.n.n.}{}}𝐒_i𝐒_j,`$ (1)
where $`𝐒_i`$ is the $`s=1/2`$ spin operator at the $`i`$-th site and $`J`$ $`(J^{})`$ represents the nearest-neighbor (next-nearest-neighbor) antiferromagnetic exchange coupling.
For later convenience, we introduce the ratio $`\alpha =J^{}/J`$. In Fig. 1, we have drawn the 2D Heisenberg model schematically. We note that the system with only the next-nearest-neighbor coupling $`J^{}`$ is equivalent to the Heisenberg model on the square lattice which has a spontaneous staggered magnetization at $`T=0`$. From this point of view, the nearest-neighbor coupling $`J`$ is regarded as the coupling for a diagonal bond (see Fig. 2), which gives rise to the frustration together with $`J^{}`$ .
In order to study the quantum phase transitions in this spin system, we employ the series expansion method developed by Singh, Gelfand and Huse. We recall here that the quantum phase transitions in the Shastry-Sutherland model have been discussed by Weihong et al. and Müller-Hartmann et al., by means of the dimer and the Ising expansions, from which the critical point between the dimer phase and the magnetically ordered phase has been estimated as $`\alpha _c=0.691(6)`$ and $`0.697(2)`$, respectively. As mentioned above, however, there is a controversy to be resolved about the nature of the phase transitions. Also, in order to determine the complete phase diagram, it is crucial to figure out whether there may exist another spin-gap phase besides the above two phases. We will address this problem in the following by using the series expansion method.
To see our strategy clearly, we start with the 2D quantum spin model schematically shown in Fig. 2 , which is topologically equivalent to the original model in Fig. 1.
In this figure, we have introduced an auxiliary parameter $`\lambda `$, which parameterizes the antiferromagnetic couplings labeled by the bold solid, the thin solid and the dashed lines, respectively, as $`J^{}`$, $`\lambda J^{}`$ and $`\lambda J^{}/\alpha (=\lambda J)`$. Note that the system is reduced to the original Shastry-Sutherland model in the case of $`\lambda =1`$. An important point is that the introduction of $`\lambda `$ enables us to perform the cluster expansion starting from the isolated plaquette singlets ($`\lambda =0`$), which naturally describes the most likely spin-gap phase distinct from the dimer phase.
To proceed the analysis based on the series expansion, we divide the original Hamiltonian eq. (1) into two parts as $`H=J^{}\left[𝐒_i𝐒_j+\lambda \mathrm{\Gamma }_{ij}𝐒_i𝐒_j\right]`$, where $`\mathrm{\Gamma }_{ij}=1`$ or $`\alpha ^1`$ for each bond on the square lattice (see Fig. 2). The first term is the unperturbed Hamiltonian which stabilizes the isolated plaquette singlets with the spin excitation gap. The perturbed Hamiltonian labeled by $`\lambda `$ connects these isolated plaquette singlets, by which a 2D network develops. We expand the staggered susceptibility $`\chi _{\mathrm{AF}}`$, the spin-triplet excitation energy $`E(𝐤)`$ and the ground state energy $`E_\mathrm{g}`$ as a power series in $`\lambda `$. Here, to estimate the susceptibility, we introduce the Zeeman term $`H^{}=h\left[_{iA}S_i^z_{iB}S_i^z\right]`$, where $`h`$ is the staggered magnetic field and $`A(B)`$ denotes one of the two sublattices. Note that an asymptotic analysis of the series expansion is necessary to deduce the accurate phase boundary on which the susceptibility $`\chi _{\mathrm{AF}}`$ diverges and the spin gap $`\mathrm{\Delta }=E(𝐤=\mathrm{𝟎})`$ vanishes. For this purpose, we make use of the Padé approximants for both quantities obtained up to the finite order in $`\lambda `$. Besides ordinary Dlog Padé approximants, we also employ biased Padé approximants, for which we assume that the phase transition in our 2D quantum spin models should belong to the universality class of the 3D classical Heisenberg model. Then the critical value of $`\lambda _c`$ is determined by the formula $`\chi _{\mathrm{AF}}(\lambda _c\lambda )^\gamma `$ and $`\mathrm{\Delta }(\lambda _c\lambda )^\nu `$ with the known exponents $`\gamma =1.4`$ and $`\nu =0.71`$.
We first calculate the staggered susceptibility $`\chi _{\mathrm{AF}}`$ and the spin gap $`\mathrm{\Delta }`$ by means of the plaquette expansion up to the fourth and the fifth order in $`\lambda `$, respectively, for various values of $`\alpha `$. Using the Dlog and the biased Padé approximants, we end up with the phase diagram shown in Fig. 3.
In this figure, the solid (dashed) line represents the phase boundary obtained by the biased Padé approximants for the spin gap (the staggered susceptibility). When $`\alpha \mathrm{}`$ and $`\lambda =0`$, the system is reduced to an assembly of the isolated plaquettes with the spin gap. As $`\lambda `$ is increased, the correlation between these plaquettes grows up and the second-order quantum phase transition from the spin-gap phase to the magnetically ordered phase occurs at the critical point $`\lambda _c=0.56`$ for $`\alpha \mathrm{}`$, which has already been studied by several groups. On the other hand, decreasing $`\alpha `$ enhances the frustration, which in turn suppresses the antiferromagnetic correlation, thus shifting the phase boundary upward for smaller $`\alpha `$ in the phase diagram. It is seen that two lines obtained from the distinct quantities are in good agreement with each other, which implies that the obtained phase boundary is rather accurate in spite of the lower-order pertubative calculation. By exploiting the phase boundary determined by means of biased Padé approximants for the spin gap, the critical value is given by $`\alpha _{c2}=0.86(1)`$ for $`\lambda =1`$. Recall that the system is reduced to the original model only for $`\lambda =1`$. We thus find that the Shastry-Sutherland model has the disordered ground state in the region ($`0<\alpha <\alpha _{c2}`$) on the $`\lambda =1`$ line.
The above result does not necessarily imply that in the region $`0<\alpha <\alpha _{c2}`$ the system always belongs to the disordered phase which is continuously connected to isolated plaquettes. In fact, it is known that the orthogonal dimer phase appears in the vicinity of $`\alpha =0`$ . Therefore, it is necessary to clarify how these two spin-gap phases compete with each other by carefully comparing the ground state energy $`E_\mathrm{g}`$. To this end, performing the plaquette expansion up to the seventh order in $`\lambda `$ with $`\alpha `$ being fixed, we estimate the ground state energy $`E_\mathrm{g}`$ for the Shastry-Sutherland model $`(\lambda =1)`$ by means of the first order inhomogeneous differential method. The results are shown in Fig. 4.
As mentioned above, the system stabilizes the orthogonal dimer ground state for smaller $`\alpha `$. It is found, however, that the first-order transition to the novel spin-gap phase introduced here occurs at the critical point $`\alpha _{c1}=0.677(2)`$. It is also seen from this figure that further increase of $`\alpha `$ induces the antiferromagnetic order, whose transition point is determined by the crossing point of the ground-state energy obtained respectively by the Ising and plaquette expansions. The result confirms the second-order phase transition deduced above, and the transition point estimated from the figure is consistent with $`\alpha _{c2}=0.86(1)`$ obtained by the analysis of the susceptibility and the spin gap. Consequently, we end up with the phase diagram for the Shastry-Sutherland model as shown in Fig. 5. The present results shed light on the controversial arguments whether the quantum phase transition in this model is of the first or second order. In those previous studies, it was believed that the phase transition occurs only once between the dimer phase (I) and the ordered phase (III), giving rise to some confusions. Our phase diagram clearly resolves this problem by explicitly showing the existence of the new spin-gap phase (II) which undergoes the first- (I$``$II) as well as the second-order transitions (II$``$III).
To check the validity of the above phase diagram, we also show the results for the spin gap as a function of $`\alpha =J^{}/J`$ in Fig. 6.
In this figure, the results obtained by Weihong et al. are shown for the orthogonal dimer phase (I: $`0<\alpha <\alpha _{c1}`$). In the new phase (II: $`\alpha _{c1}<\alpha <\alpha _{c2}`$), we determine the values of the spin gap at $`𝐤=\mathrm{𝟎}`$ by means of the plaquette expansion up to the fifth order in $`\lambda `$ with the first order inhomogeneous differential method. The results are shown as the dots with the error bars. As seen in this figure, with the decrease of $`\alpha `$ from the second-order transition point $`\alpha _{c2}`$, the spin-gap continuously grows up to stabilize the disordered ground state. As $`\alpha `$ is further decreased, the first-order phase transition occurs at $`\alpha _{c1}`$.
In order to further confirm the present results, we have performed a different series expansion by choosing the isolated plaquettes with diagonal bonds as an initial configuration, which is different from the one shown in Fig. 2. The calculation of the susceptibility up to the fourth order yields second-order transition with $`\alpha _{c2}=0.87(3)`$, being consistent with the above results. Furthermore, to confirm the first-order phase transition between the two spin-gap states, we have checked how the first-order phase transition point known for the 1D orthogonal-dimer chain evolves with the increase of the inter-chain couplings. By performing the exact diagonalization studies for the $`4\times 4`$ system, we have found that the first-order transition point for 1D is continuously changed, and in the Shastry-Sutherland case, it coincides with the one found above within reasonable accuracy $`(\alpha _{c1}0.66)`$, providing further support to our conclusion on the phase diagram. Although our results still seem to be partly contradicted to the staggered magnetization obtained by Weihong et al., we believe that this could be resolved by further analysis of the results of the Ising expansion.
Before concluding the paper, a brief comment is in order for the plateau-formation in the magnetization curve. Experimentally, the plateaus in the magnetization curve have been observed for the compound $`\mathrm{SrCu}_2(\mathrm{BO}_3)_2`$ at 1/3, 1/4 and $`1/8`$ of the full moment. In the theoretical studies on the dimer phase, it has been clarified that the stripe order of the isolated dimer-triplets is important to understand the $`1/3`$ and $`1/4`$plateaus. On the other hand, it is not so trivial why the 1/8-plateau occurs in this compound, although a possible mechanism has been proposed . We think that the formation of the 1/8 plateau may reflect the fact that this compound is located around the first-order phase transition point between the two spin-gap phases and thereby possesses the dual properties inherent in two distinct phases implicitly. We note here that the new spin-gap phase belongs to the same phase as the Heisenberg model on the 1/5-depleted square lattice proposed for $`\mathrm{CaV}_4\mathrm{O}_9`$. Therefore it is likely that the 1/8-plateau could occur in the same origin discussed by Momoi and Totsuka for the plaquette system related to the 1/5-depleted Heisenberg model. It is interesting to further clarify the mechanism of the 1/8-plateau by taking into account the above dual properties explicitly, which is now under consideration.
In conclusion we have discussed the phase diagram for the Shastry-Sutherland model for the compound $`\mathrm{SrCu}_2(\mathrm{BO}_3)_2`$ by means of the series expansion method. Our analysis has shown that there exists a novel spin-gap phase with the disordered ground state, which is adiabatically connected to the plaquette-singlet phase, between the dimer and the magnetically ordered phases known so far. When the exchange coupling ratio $`\alpha =J^{}/J`$ is varied, the first-order phase transition occurs from the dimer state to the new spin-gap state, while the second-order phase transition occurs from this spin-gap state to the magnetically ordered state. This sheds light on the nature of the quantum phase transitions in this model, and resolves apparently controversial conclusions on this issue.
We would like to thank K. Ueda, S. Miyahara and K. Okunishi for useful discussions. The work is partly supported by a Grant-in-Aid from the Ministry of Education, Science, Sports, and Culture. A. K. is supported by the Japan Society for the Promotion of Science. A part of numerical computations in this work was carried out at the Yukawa Institute Computer Facility. |
warning/0003/cond-mat0003517.html | ar5iv | text | # Kinetic Theory of Collective Excitations and Damping in Bose-Einstein Condensed Gases
## I Introduction
The nonzero-temperature dependence of the frequencies and damping of collective modes in trapped atomic Bose gases has been investigated extensively both experimentally and theoretically . Above the Bose-Einstein transition temperature the lowest-lying collective modes have been calculated in the collisionless regime using a collisionless Boltzmann or Landau-Vlasov equation , and in the hydrodynamic regime either by using the conservation laws of hydrodynamics or by using the appropriate quantum kinetic equation . In the collisionless regime the frequency of the collective mode is large compared to the mean collision frequency, in contrast to the situation in the hydrodynamic regime. Several papers have also studied the regime intermediate between the collisionless and hydrodynamic regimes by taking into account interatomic collisions , which turn out to mostly lead to mode damping. These papers have indicated that the experiments were performed under conditions intermediate between collisionless and hydrodynamic. (This is particularly true for the experiments performed by Mewes et al. and Stamper-Kurn et al. ). For temperatures far below the transition temperature the collisionless modes can be described accurately by the time-dependent Gross-Pitaevskii equation . At higher temperatures the noncondensate fraction becomes substantial and the modes of the condensate are now coupled to those of the thermal cloud. In Refs. the temperature dependence of the mode frequencies has been calculated by employing the Popov approximation to include the static mean-field effects of the noncondensate atoms in the Gross-Pitaevskii equation. This was improved later by taking into account also the dynamics of the thermal cloud using a RPA approximation or the collisionless Boltzmann equation . For the hydrodynamic regime a two-fluid model has been developed in Refs. . Moreover, the theory of Zaremba, Griffin, and Nikuni was improved later by the same authors to include collisions between condensate and noncondensate atoms .
Using a combination of two previous papers, namely Refs. and , we aim in this paper at interpolating between the collisionless and hydrodynamic regimes for experiments below the transition temperature. The authors of Ref. have already performed such an interpolation for a gas above the critical temperature by using a Boltzmann equation with a relaxation time approximation. In contrast, the authors of Ref. use a collisionless Boltzmann equation for the thermal cloud coupled to a time-dependent nonlinear Schrödinger equation for the condensate to consider the collisionless dynamics below the critical temperature. Here, we thus combine these two approaches by using again the Boltzmann equation and time-dependent nonlinear Schrödinger equation, but now including also the effects of interatomic collisions in the manner as put forward in Refs. and . This implies that we have to add two collision terms to the Boltzmann equation. The first collision term represents collisions between two noncondensate atoms and the second describes collisions between a condensate and a noncondensate atom. We use for both these collision terms a relaxation time approximation, since this approximation leads above the transition temperature to a good agreement with microscopic calculations as well as with experimental data . Furthermore, for consistency reasons we also have to include a damping term in the time-dependent non-linear Schrödinger equation, which is due to collisions between the condensate and noncondensate atoms. As a result our calculation will essentially be characterized by two parameters, namely $`\tau _{22}`$ and $`\tau _{12}`$, denoting the mean collision time for collisions between the noncondensate atoms and between condensate and noncondensate atoms, respectively. Note that we use here the same notation as in Ref. . This will allow us to investigate the collisionless and hydrodynamic limits with respect to both $`\tau _{22}`$ and $`\tau _{12}`$, and enable us to fit our results for frequencies and damping with the experimental data.
To solve the complicated nonlinear dynamics of the gas, we employ a gaussian trial function for the wave function of the condensate atoms with three complex time-dependent variational parameters. For the thermal cloud we use a distribution function that incorporates deviations from the Bose-Einstein distribution function. The deviation function is a truncated power expansion in the momenta and coordinates of the atoms. Our system of coupled equations is then assembled from the Euler-Lagrange equations for the variational parameters in the condensate wavefunction and from all the linear and quadratic moments of the Boltzmann equation for the distribution function of the noncondensate atoms. The solution of the linearized version of this system of coupled equations results in a dispersion relation that gives the frequencies and damping of the coupled modes in terms of $`\tau _{22}`$ and $`\tau _{12}`$. Although we restrict our calculation to the axially symmetric traps used in the experiments, the generalization to fully anisotropic traps is straightforward.
The rest of this paper is organized as follows: In the next section we present the theoretical details of our calculation. Specifically, we write down the time-dependent nonlinear Schrödinger equation for the condensate wave function and the linearized Boltzmann equation for the thermal cloud, and show how to treat the two collision terms in a relaxation time approximation. In section II A we present our trial functions for the condensate wave function and noncondensate distribution function and in section II B we discuss how we obtain the equilibrium state of the gas around which we have to expand to find the collective modes. In section III we present the resulting dispersion relation and discuss its collisionless and hydrodynamic limits. We then compare our results with experiment. We end in section IV by summing up our main conclusions.
## II Coupled Dynamics of the condensate and the thermal cloud
It has been shown that the coupled time-dependent nonlinear Schrödinger equation and the quantum Boltzmann equation can both be derived starting from the equation of motion
$`i\mathrm{}{\displaystyle \frac{\widehat{\psi }(𝐫,t)}{t}}`$ $`=`$ $`\text{(}{\displaystyle \frac{\mathrm{}^2^2}{2m}}+V^{ext}(𝐫)`$ (1)
$`+`$ $`g\widehat{\psi }^{}(𝐫,t)\widehat{\psi }(𝐫,t)\text{)}\widehat{\psi }(𝐫,t),`$ (2)
for the Heisenberg field operator $`\widehat{\psi }(𝐫,t)`$. Here, $`g`$ is the effective two-body interaction, which is given in terms of the scattering length $`a`$ and the atomic mass $`m`$ as $`g=4\pi a\mathrm{}^2/m`$. The external potential we take here is a harmonic potential of the general form,
$$V^{ext}(𝐫)=\frac{1}{2}m(\omega _1^2x^2+\omega _2^2y^2+\omega _3^2z^2),$$
(3)
where $`\omega _i`$ is the characteristic frequency of the trap in the $`i`$th direction. To obtain the time-dependent nonlinear Schrödinger equation we write $`\widehat{\psi }(𝐫,t)`$ as
$$\widehat{\psi }(𝐫,t)=\mathrm{\Phi }(𝐫,t)+\widehat{\psi }^{}(𝐫,t),$$
(4)
where $`\mathrm{\Phi }(𝐫,t)`$ is the appropriate nonequilibrium expectation value of $`\widehat{\psi }(𝐫,t)`$ and the operator $`\widehat{\psi }^{}(𝐫,t)`$ describes the noncondensate atoms. The number density of condensate atoms $`n_c(𝐫,t)`$ is related to $`\mathrm{\Phi }(𝐫,t)`$ by $`n_c(𝐫,t)=|\mathrm{\Phi }(𝐫,t)|^2`$ while the number density of the noncondensate atoms $`n^{}`$ equals $`\widehat{\psi ^{}}^{}(𝐫,t)\widehat{\psi ^{}}(𝐫,t)`$, where the brackets represent the averaging over the nonequilibrium density matrix. Substituting this form for $`\widehat{\psi }(𝐫,t)`$ in the equation of motion and averaging, we obtain
$`i\mathrm{}{\displaystyle \frac{\mathrm{\Phi }(𝐫,t)}{t}}`$ $`=`$ $`\text{(}{\displaystyle \frac{\mathrm{}^2^2}{2m}}+V^{ext}(𝐫)+gn_c(𝐫,t)+2gn^{}(𝐫,t)`$ (5)
$``$ $`iR(𝐫,t){\displaystyle \frac{}{}}\text{)}\mathrm{\Phi }(𝐫,t),`$ (6)
where $`R(𝐫,t)`$ is given by
$`R(𝐫,t)`$ $`=`$ $`g(\widehat{\psi ^{}}(𝐫,t)\widehat{\psi ^{}}(𝐫,t)\mathrm{\Phi }^{}(𝐫,t)`$ (7)
$`+`$ $`\widehat{\psi ^{}}^{}(𝐫,t)\widehat{\psi ^{}}(𝐫,t)\widehat{\psi ^{}}(𝐫,t))/\mathrm{\Phi }(𝐫,t).`$ (8)
Next, the quantum Boltzmann equation can be derived by writing an equation of motion for the distribution function of the noncondensate atoms $`f(𝐩,𝐫,t)`$. This is usually done by writing $`f(𝐩,𝐫,t)`$ as a Wigner transform of $`\widehat{\psi ^{}}^{}(𝐫^{},t)\widehat{\psi ^{}}(𝐫,t)`$ and then determining the time evolution of $`f(𝐩,𝐫,t)`$ from the equation of motion in Eq. (2) . In the Hartree-Fock approximation the resulting Boltzmann equation takes the form
$$\left[\frac{f}{t}+_𝐩E_𝐫_𝐫E_𝐩\right]f=C_{22}[f]+C_{12}[f],$$
(9)
where $`C_{22}`$ is the contribution to the rate of change of $`f`$ due to collisions between noncondensate atoms, while $`C_{12}`$ is the contribution due to collisions between the condensate and the noncondensate atoms. The energy $`E(𝐩,𝐫,t)`$ of the noncondensate atoms is in this approximation given by
$$E(𝐩,𝐫,t)=p^2/2m+V^{ext}(𝐫)+2gn(𝐫,t),$$
(10)
where $`n=n_c+n^{}`$ is the total density.
In Ref. explicit forms of the two collision terms have been written down as follows
$`C_{22}[f]`$ $`=`$ $`{\displaystyle \frac{2g^2}{(2\pi )^5\mathrm{}^7}}{\displaystyle 𝑑𝐩_2𝑑𝐩_3𝑑𝐩_4\delta (𝐩+𝐩_2𝐩_3𝐩_4)}`$ (11)
$`\times `$ $`\delta (E+E_2E_3E_4)`$ (12)
$`\times `$ $`\left[(1+f)(1+f_2)f_3f_4ff_2(1+f_3)(1+f_4)\right],`$ (13)
and
$`C_{12}[f]`$ $`=`$ $`{\displaystyle \frac{2g^2n_c}{(2\pi )^2\mathrm{}^4}}{\displaystyle 𝑑𝐩_1𝑑𝐩_2𝑑𝐩_3}`$ (14)
$`\times `$ $`\delta (m𝐯_c+𝐩_1𝐩_2𝐩_3)`$ (15)
$`\times `$ $`\delta (E_c+E_1E_2E_3)`$ (16)
$`\times `$ $`\left[\delta (𝐩𝐩_1)\delta (𝐩𝐩_2)\delta (𝐩𝐩_3)\right]`$ (17)
$`\times `$ $`\left[(1+f_1)f_2f_3f_1(1+f_2)(1+f_3)\right].`$ (18)
Here $`E_c(𝐫,t)`$ and $`𝐯_c(𝐫,t)`$ are the local energy and velocity of the condensate atoms, $`E_i`$ is a shorthand for $`E(𝐩_i,𝐫,t)`$, and similarly $`f_i`$ is a shorthand for $`f(𝐩_i,𝐫,t)`$. As is shown in Ref. , the conservation of the total number of atoms constraints the collision term $`C_{12}`$ in the Boltzmann equation to be related to the damping term $`R(𝐫,t)`$ in the complex nonlinear Schrödinger equation as
$$2n_c(𝐫,t)R(𝐫,t)=\frac{d𝐩}{(2\pi )^3}C_{12}[f].$$
(19)
The coupled equations given in Eqs. (6), (9), and (19) in principle fully determine the dynamics of the Bose-Einstein condensed gas in the Hartree-Fock approximation. Since the dynamics of the thermal cloud is experimentally only important for temperatures which are larger or comparable to the mean-field interactions we believe that this is sufficiently accurate and we do not need to use the Popov approximation at first instance. Nevertheless, these equations are too complicated to be solved exactly and some approximation is called for.
### A Trial Functions
For the noncondensate atoms we start by linearizing the Boltzmann equation in small deviations of the distribution function around its equilibrium value, namely
$$f(𝐩,𝐫,t)=f^{(0)}(𝐩,𝐫)+f^{(0)}(𝐩,𝐫)(1+f^{(0)}(𝐩,𝐫))\psi (𝐩,𝐫,t),$$
(20)
where $`f^{(0)}(𝐩,𝐫)`$ is the equilibrium distribution function given by
$$f^{(0)}(𝐩,𝐫)=\left\{\mathrm{exp}\left[(E(𝐩,𝐫)\mu )/k_BT\right]1\right\}^1.$$
(21)
Here $`\mu `$ is the chemical potential and $`k_B`$ is Boltzmann’s constant. As a result the linearized Boltzmann equation takes the form
$$\left[\frac{}{t}+_𝐩E_𝐫_𝐫E_𝐩\right]\psi =C_{22}[\psi ]+C_{12}[\psi ].$$
(22)
It is shown in Ref. that for an uncondensed gas a trial function for $`\psi (𝐩,𝐫,t)`$ of the form
$`\psi `$ $`=`$ $`A_1x^2+B_1xp_x+C_1p_x^2+A_2y^2+B_2yp_y`$ (23)
$`+`$ $`C_2p_y^2+A_3z^2+B_3zp_z+C_3p_z^2,`$ (24)
where $`A_i(t)`$, $`B_i(t)`$, and $`C_i(t)`$ are nine time-dependent functions, is appropriate to describe the low-lying collisionless breathing modes with frequencies $`2\omega _i`$, some of which have been observed experimentally. We will therefore assume the above expansion to be also reasonably accurate below the critical temperature. For the collision integrals $`C_{22}`$ and $`C_{12}`$ we use a relaxation time approximation. In such an approximation one associates these collision integrals to mean relaxation times. However, the approximate expressions for $`C_{22}`$ and $`C_{12}`$ should still take into account the conservation laws associated with the collision processes exactly. In the case of $`C_{22}`$, which represents collisions between the noncondensate atoms, the number of atoms in the thermal cloud, their total momentum and their total kinetic energy should all be conserved. In our ansatz for $`\psi `$, terms like $`x^2`$ and $`xp_x`$ should therefore not be affected by such collisions since they correspond to two collision invariants, namely the number of atoms and the total momentum in the $`x`$ direction, respectively. On the other hand, terms like $`p_x^2`$ will be affected by collisions since it is only the sum $`p^2=p_x^2+p_y^2+p_z^2`$ which is conserved during the collision process. Therefore, we can write the following expression for the linear operator $`C_{22}`$
$`C_{22}[\psi _i]`$ $`=`$ $`{\displaystyle \frac{1}{\tau _{22}}}\{\begin{array}{ccc}(\psi _ip^2/3)& ,& \psi _i=p_x^2,p_y^2,p_z^2\\ 0& ,& \mathrm{otherwise}\end{array},`$ (27)
where $`\tau _{22}`$ is a mean collision time for the noncondensate-noncondensate atomic collisions. Note that in this expression $`C_{22}[p^2]=0`$, ensuring the conservation of the total kinetic energy.
For $`C_{12}`$, which represents collisions between the condensate and the noncondensate atoms, the number of atoms is not conserved, since the collision process involves transport of atoms back and forth from the condensate into the thermal cloud. This statement means mathematically that the zeroth moment of $`C_{12}`$, i.e., $`𝑑𝐩C_{12}`$, does not vanish in contrast to the case of $`C_{22}`$ where $`𝑑𝐩C_{22}=0`$. As a first attempt to associate with $`C_{12}`$ a mean collision time $`\tau _{12}`$ for noncondensate-condensate atomic collisions, we may write $`C_{12}1/\tau _{12}`$. However, we observe from Eq. (18) that $`C_{12}n_c(𝐫)`$. This dependence on $`n_c(𝐫)`$ requires us to assign to $`\tau _{12}`$ a position dependence that follows from $`n_c(𝐫)`$ as
$$\frac{1}{\tau _{12}(𝐫)}=\frac{1}{\tau _{12}(\mathrm{𝟎})}\frac{n_c(𝐫)}{n_c(\mathrm{𝟎})},$$
(29)
where $`\tau _{12}(\mathrm{𝟎})`$ and $`n_c(\mathrm{𝟎})`$ are the noncondensate-condensate mean atomic collision time and condensate density at the center of the trap, respectively. Generally, $`\tau _{12}(\mathrm{𝟎})`$ may, just like $`\tau _{22}`$, also have temperature dependence. By multiplying and dividing $`1/\tau _{12}(𝐫)`$ by $`\mathrm{}g`$ we can thus take $`C_{12}`$ as obeying
$$C_{12}\alpha \frac{2gn_c(𝐫)}{\mathrm{}},$$
(30)
where we introduced the dimensionless constant
$$\alpha =\frac{\mathrm{}}{2gn_c(\mathrm{𝟎})\tau (\mathrm{𝟎})}.$$
(31)
This form for $`C_{12}`$ is convenient for a reason that becomes clear lateron in this section. It should be emphasized here that the dimensionless parameter $`\alpha `$ is the second free parameter in our phenomenological calculation. The first dimensionless parameter being $`1/\overline{\omega }\tau _{22}`$ where $`\overline{\omega }=\sqrt[3]{\omega _1\omega _2\omega _3}`$. Therefore, the collisionless and hydrodynamic regimes, as well as the intermediate regime, will be described by only these two parameters. Finally, we make the relation between $`C_{12}`$ and $`\alpha `$ precise by taking
$`C_{12}[\psi _i]`$ $`=`$ $`\alpha {\displaystyle \frac{2gn_c(𝐫)}{\mathrm{}}}`$ (32)
$`\times `$ $`\{\begin{array}{ccc}(\psi _ip^2/3)& ,& \psi _i=p_{x}^{}{}_{}{}^{2},p_{y}^{}{}_{}{}^{2},p_{z}^{}{}_{}{}^{2}\\ \psi _i& ,& \psi _i=x^2,y^2,z^2\\ 0& ,& \mathrm{otherwise}\end{array}.`$ (36)
This completes our description of the treatment of the thermal cloud. Next we have to consider the condensate.
It is known that a time-dependent gaussian ansatz for the condensate wave function gives the correct frequencies of the lowest modes at zero temperature . Furthermore, the gaussian ansatz has also been used in Ref. for the whole temperature range below the transition temperature and leads to rather good agreement with experiment. Therefore, we employ here again a gaussian ansatz for the wave function of the condensate. It has the following form
$`\mathrm{\Phi }(𝐫,t)`$ $`=`$ $`\sqrt[4]{{\displaystyle \frac{8N_cb_{1r}b_{2r}b_{3r}}{\pi ^3}}}\mathrm{exp}\left[(b_1x^2+b_2y^2+b_3z^2)\right],`$ (37)
where $`b_1`$, $`b_2`$, and $`b_3`$ are complex time-dependent variational parameters and $`b_{1r}`$, $`b_{2r}`$, and $`b_{3r}`$ are their real parts, respectively. Similarly, we denote the imaginary parts, which will appear lateron, as $`b_{1i}`$, $`b_{2i}`$, and $`b_{3i}`$. The prefactor of $`\mathrm{\Phi }(𝐫,t)`$ guarantees its normalization $`𝑑𝐫|\mathrm{\Phi }(𝐫,t)|^2`$ to be equal to the number of condensate atoms $`N_c`$.
To obtain the first set of our coupled equations of motion, we start by writing down the energy functional that corresponds to the nonlinear Schrödinger equation in Eq. (6), namely
$`E_c[\mathrm{\Phi }]`$ $`=`$ $`{\displaystyle 𝑑𝐫\text{[}\frac{\mathrm{}^2\left|𝚽\right|^2}{2m}}+V^{ext}n_c+{\displaystyle \frac{1}{2}}g(n_{c}^{}{}_{}{}^{2}+4n_cn^{})`$ (39)
$``$ $`iRn_c{\displaystyle \frac{}{}}\text{]}`$ (40)
with $`n_c=|\mathrm{\Phi }|^2`$. We evaluate this energy functional using the expression for $`\mathrm{\Phi }`$ from Eq. (LABEL:phi), and by noting that the density of the thermal cloud
$$n^{}=\frac{d𝐩}{(2\pi )^3}f(𝐩,𝐫,t),$$
(41)
and the damping term $`R(𝐫,t)`$ can be calculated using Eqs. (19) and (36). We notice here that $`iR(𝐫,t)n_c`$ turns out to be proportional to the Hartree-Fock interaction term in Eq. (40), with a proportionality constant equals to $`i\alpha `$. This explains the reason for writing $`C_{12}`$ as in Eq. (30). The equations of motion for the condensate dynamics are now the Euler-Lagrange equations resulting from varying the lagrangian
$$L=\frac{1}{2}i\mathrm{}𝑑𝐫\left(\mathrm{\Phi }^{}\frac{}{t}\mathrm{\Phi }\mathrm{\Phi }\frac{}{t}\mathrm{\Phi }^{}\right)E_c[\mathrm{\Phi }]$$
(42)
with respect to the 6 variational parameters $`b_{kr}`$ and $`b_{ki}`$.
The second set involves the equations of motion for the constants $`A_i`$, $`B_i`$, and $`C_i`$. It is obtained from taking appropriate moments of the Boltzmann equation in Eq. (22). In detail, the moments are calculated by multiplying this equation by $`f^{(0)}(𝐩,𝐫)`$ and the various components of $`\psi `$ in Eq. (24), and then integrating over $`𝐩`$ and $`𝐫`$. This results in 9 equations of motion. In combination with the previous 6 ones, we thus have 15 coupled equations of motion. The coupling is provided on the one hand by the mean-field interaction $`2gn_cn^{}`$ and the imaginary damping term $`iRn_c`$ in the condensate energy functional $`E_c`$ in Eq. (40), and on the other hand by the contribution $`2gn_c`$ to the Hartree-Fock energy $`E`$ in Eq. (10) and the $`C_{12}`$ collision term in the Boltzmann equation.
### B Equilibrium
Finally, it is important to note that linearization of the equations of motion of the condensate is obligatory to be consistent with the equations of motion of the noncondensate part which are already linearized. Therefore, we need to calculate the equilibrium state of the gas. In principle, to obtain equilibrium properties we should minimize the free energy $`F=E_{tot}TS`$, where $`E_{tot}`$ is the total energy and $`S`$ is the entropy of the gas, with respect to some variational parameters that characterize the widths of the condensate and the thermal cloud. However, a simplified estimate of the total energy and the entropy contribution to the free energy above the transition temperature shows that for the experimental conditions of interest the former is dominant over the later. This simplified estimate can be made by using a one-parameter gaussian ansatz for the thermal cloud density, namely $`n^{}\mathrm{exp}(x^2/R^2)`$, where $`R`$ is the radius of the cloud. It turns out that in the Thomas-Fermi limit the entropy contribution to the free energy is of order $`k_BT\mathrm{ln}\gamma ^{3/5}`$, where $`\gamma =N^{}a/\overline{a}`$, $`N^{}`$ is the total number of atoms in the thermal cloud, and $`\overline{a}=\sqrt{\mathrm{}/m\overline{\omega }}`$ is the harmonic oscillator length, while the total energy contribution is of order $`k_BT\gamma ^{2/5}`$. Therefore, if $`\gamma 1`$ the entropy contribution is much less than the total energy contribution.
As we shall see in the next section, the experiments were performed with a temperature-dependent total number of atoms ranging from about $`6000`$ atoms at zero temperature to approximately $`40000`$ atoms at the transition temperature. If we extend this simple estimate to temperatures below the transition temperature it turns out that the condition $`\gamma 1`$ is satisfied only for high temperatures but not for low temperatures since the number of thermal atoms becomes small. However, this condition is then no longer important since the free energy of the thermal cloud at such temperatures is small compared to the condensate energy. Therefore, we will in the following only minimize the total energy $`E_{tot}`$ with respect to the variational parameters of the condensate and the thermal cloud. To be able to calculate $`E_{tot}`$ we assume that the distribution function of the noncondensate atoms has the same form of that of a noninteracting gas but with varying spatial widths that effectively take into account the mean-field effects of both the noncondensate and condensate atoms. This effective distribution function is written as
$`f_{eff}^{(0)}`$ $`=`$ $`\left[\mathrm{exp}\left({\displaystyle \frac{p^2/2m\mu }{k_BT}}+{\displaystyle \frac{x^2}{R_{1}^{}{}_{}{}^{2}}}+{\displaystyle \frac{y^2}{R_{2}^{}{}_{}{}^{2}}}+{\displaystyle \frac{z^2}{R_{3}^{}{}_{}{}^{2}}}\right)1\right]^1,`$ (43)
where $`R_1`$, $`R_2`$, and $`R_3`$ are the widths of the noncondensate cloud in the three directions. The total energy is now a function of 6 variational parameters:
$$E_{tot}=E_{tot}(b_{1r}^{(0)},b_{2r}^{(0)},b_{3r}^{(0)},R_1,R_2,R_3),$$
(45)
where $`b_{1r}^{(0)}`$, $`b_{2r}^{(0)}`$, and $`b_{3r}^{(0)}`$ are the equilibrium values of $`b_{1r}`$, $`b_{2r}`$, and $`b_{3r}`$, respectively. The equilibrium is obtained by minimizing this energy with respect to these variational parameters.
The results of such a minimization will be shown in the next section, where we present the dispersion relation that results from solving the above-described system of 15 linearized coupled equations of motion. We discuss also the collisionless and hydrodynamic limits of these results.
## III The Dispersion Relation: Frequencies and Damping Rates
Our calculation accounts, in fully anisotropic traps, for 9 modes of the gas. In axially symmetric traps this number reduces to 6 modes. These are the in-phase and out-of-phase combinations of the two monopole ($`m=0`$) modes and one quadrupole ($`m=2`$) mode of both the condensate and the thermal cloud. Here we denote with $`m`$ the projection of the angular momentum of the mode along the axis of symmetry of the trap . We focus in this paper on the two lowest-lying $`m=0`$ and $`m=2`$ modes observed experimentally. It turns out that for the experimentally relevant temperature range, the in-phase $`m=0`$, and $`m=2`$ modes correspond mostly to oscillations of the thermal cloud, whereas the out-of-phase modes are mostly condensate oscillations. Therefore, we shall often refer in this paper to the in-phase modes as the thermal cloud or noncondensate modes and the out-of-phase modes as the condensate modes. Although our calculation provides results for another, higher-lying, monopole mode, we shall not discuss it further here. Moreover, we emphasize that throughout the following we perform our calculations for parameters taken from the experiments of Jin et al. , i.e., with <sup>87</sup>Rb atoms in an axially-symmetric trap with anisotropy ratio $`\omega _3/\omega _1=\omega _3/\omega _2=\sqrt{8}`$.
In that particular experiment the measurements were performed with a temperature-dependent total number of atoms as a result of the loss of atoms during evaporative cooling. The total number of atoms $`N_{tot}`$, as well as the number of condensate atoms $`N_c`$, are measured in the temperature range $`T/T_{BEC}0.48`$ to $`T/T_{BEC}1.0`$, where $`T_{BEC}`$ is the Bose-Einstein transition temperature. These measurements can be easily fitted with polynomials in $`T/T_{BEC}`$ as shown by the dashed line in Fig. 1. An extrapolation of such a fit to temperatures below $`0.48T_{BEC}`$ leads, however, to nonphysical situations. In particular, the two curves for $`N_{tot}`$ and $`N_c`$ cross at least once before reaching zero temperature. We overcome this problem by fitting only $`N_{tot}`$ with the experimental data, and then using $`N_c`$=$`N_{tot}(1(T/T_{BEC})^2)`$ in analogy with the ideal gas relation $`N_c`$=$`N_{tot}(1(T/T_{BEC})^3)`$ . Fig. 1 shows with the solid line the results of this slightly less accurate fit. It should be noted that using the ideal gas relation will grossly overestimate all the experimental points of $`N_c`$ for $`T/T_{BEC}>0.7`$. It turns out that the difference between the calculated frequencies using the less accurate fit from those calculated using the best fit are much smaller than the uncertainties in the measured frequencies for the experimental range of temperatures. This is shown in Fig. 2. Since we want to show also results for the complete temperature interval from zero to $`T_{BEC}`$, we employ from now on always the former fit, i.e., the solid lines in Fig. 1.
Before starting with calculating the frequencies and damping rates of the collective modes, we show in Fig. 3 the result of the minimization of the total energy that is required to obtain the equilibrium conditions of the gas, as described in the previous section. We plot the equilibrium widths of both condensate and noncondensate clouds as a function of temperature. We notice from this figure that at zero temperature and at the transition temperature we obtain rather good estimates for the widths of the condensate and thermal clouds. At zero temperature, where the thermal cloud is absent, we observe that the radial width of the condensate equals approximately $`2a_1`$, a result that is consistent with the familiar property that due to the mean-field interactions the equilibrium condensate width is larger than the ideal gas result $`a_1`$. The axial width, which is roughly $`1/\lambda ^{1/2}`$ of the radial width, is suppressed due to the anisotropy of the trap. We can also see in this figure that at the transition temperature the radial width of the condensate is slightly larger than the ideal gas one. This slight expansion is caused by the presence of the thermal cloud, whose mean-field interaction has the effect of slightly reducing the spring constants of the effective trapping potential.
Returning to the problem of the collective mode frequencies and damping, we start by neglecting collisions between the condensate and noncondensate atoms, which means that $`\alpha =0`$. In this case the dispersion relation turns out to have the following general structure
$$iw\left(P_C^{(0)}(\omega \tau _{22})^2+iP_I^{(0)}\omega \tau _{22}P_H^{(0)}\right)=0,$$
(46)
where $`P_C^{(0)}`$, $`P_I^{(0)}`$, and $`P_H^{(0)}`$ are 6th order polynomials in $`\omega ^2`$ which can all be factorized as
$$P_C^{(0)}=\underset{k=1}{\overset{6}{}}(\omega ^2\omega _{Ck}^2),$$
(47)
$$P_I^{(0)}=\underset{k=1}{\overset{6}{}}(\omega ^2\omega _{Ik}^2),$$
(48)
and
$$P_H^{(0)}=\underset{k=1}{\overset{6}{}}(\omega ^2\omega _{Hk}^2).$$
(49)
Here, $`\omega _{Ck}`$, $`\omega _{Hk}`$, and $`\omega _{Ik}`$, $`k=1,2,3`$, are temperature-dependent collisionless, intermediate, and hydrodynamic frequencies, respectively. The superscripts in $`P_C^{(0)}`$, $`P_I^{(0)}`$, and $`P_H^{(0)}`$ indicate the value of $`\alpha `$.
The general structure of this dispersion relation agrees with the result of Refs. above the transition temperature, where this dispersion relation was studied in detail. Particularly, it was shown that damping rates calculated using this relation agree in order of magnitude with experiments and numerical calculations.
The collisionless regime is defined by the condition $`\omega \tau _{22}1`$. Therefore, we find from Eq. (46) that the collisionless frequencies are $`\omega _{Ck}`$. In the hydrodynamic limit $`\omega \tau _{22}1`$, and the hydrodynamic frequencies are $`\omega _{Hk}`$. Note that we use the word hydrodynamic here to denote that the thermal cloud is in the hydrodynamic regime though $`\alpha =0`$ and there are therefore no collisions between condensate and noncondensate atoms. In a sense this regime is thus precisely the limit discussed by Nikuni, Zaremba, and Griffin . Using the experimental parameters we calculate these frequencies for the whole temperature range below $`T_{BEC}`$. In Fig. 4 we present the results of this calculation for the $`m=0`$ and $`m=2`$ modes together with the experimental data. In this figure the collisionless curves agree with those of Bijlsma and Stoof for $`T/T_{BEC}>0.2`$, which is not surprising since we use similar ansatz functions. The discrepancy for $`T/T_{BEC}<0.2`$ is due to the different ways of treating the equilibrium state.
Above the transition temperature, the explanation of the mode damping was based on the fact that the measured frequencies were less than the theoretical collsionless frequencies. This shift in frequency is then interpreted to be due to the fact that the system is shifted from the collisionless regime towards the hydrodynamic regime. Collisional damping associated with this shift was calculated and compared to the experimental damping. It is clear from Fig. 4 that below the transition temperature this kind of explanation is not possible, since most of the experimental points are not located between the collisionless and hydrodynamic curves.
To investigate the possibility that this damping might be due to the condensate-noncondensate atomic collisions, we now study the effect of these collisions by taking $`\alpha `$ to be nonzero. For nonzero $`\alpha `$ the dispersion relation takes the general form
$`\left({\displaystyle \underset{k=0}{\overset{9}{}}}P_C^{(k)}\alpha ^k\right)(\omega \tau _{22})^2`$ $`+`$ $`i\left({\displaystyle \underset{k=0}{\overset{8}{}}}P_I^{(k)}\alpha ^k\right)\omega \tau _{22}`$ (50)
$``$ $`{\displaystyle \underset{k=0}{\overset{7}{}}}P_H^{(k)}\alpha ^k=0.`$ (51)
We note that the $`k=0`$ term in this expression corresponds to Eq. (46), where $`\alpha `$ was indeed taken to be zero. Frequencies and damping rates can now be obtained by writing $`\omega `$ in terms of a real and imaginary parts, $`\omega =\omega _r+i\omega _i`$, and then inserting this expression in the last equation. In general, one obtains expressions for $`\omega _r`$ and $`\omega _i`$ in terms of $`\overline{\omega }\tau _{22}`$ and $`\alpha `$.
We start our discussion of the frequencies and damping rates for $`T/T_{BEC}=0.79`$, since at this temperature the experiment provides data for both the condensate and noncondensate oscillations. We present the results of our calculation for the $`m=0`$ and $`m=2`$ modes of the thermal cloud in Fig. 5, where we have eliminated $`\overline{\omega }\tau _{22}`$ and plotted $`\omega _i`$ as a function of $`\omega _r`$ for $`\alpha =0`$ and $`\alpha =0.25`$. On the same plot we show the two experimental points corresponding to the $`m=0`$ and $`m=2`$ modes of the noncondensate oscillation. In this figure, the right end of the curves represent the collisionless regime ($`\overline{\omega }\tau _{22}1`$) and the left end is the hydrodynamic regime ($`\overline{\omega }\tau _{22}1`$). Therefore, the location where the dashed curves intersect with the horizontal axis can also be seen in Fig. 4 at $`T=0.79T_{BEC}`$. For the experimental point corresponding to the monopole mode, the error bars do not intersect with the dashed curve ($`\alpha =0`$), but they do with the solid curve ($`\alpha =0.25`$) indicating that a nonzero value of $`\alpha `$ is needed to account for the experimental damping. This indicates that condensate-noncondensate atomic collisions are mainly responsible for the observed damping of this mode. This is less clear for the quadrupole mode although the experimental data is certainly not inconsistent with this conclusion.
Next we show in Fig. 6 at the same temperature both the condensate as well as the noncondensate $`m=0`$ and $`m=2`$ modes for values of $`\alpha `$ ranging from 0 to 0.25. The effect of $`\alpha `$ on the noncondensate modes is a slight shift and rotation of the curves but their general structure is roughly preserved. This can be seen more clearly in Fig. 5. The effect of $`\alpha `$ on the condensate modes is much larger, it gives rise to a large upward shift for the whole curve. The effect of $`\overline{\omega }\tau _{22}`$ on these curves becomes smaller for larger values of $`\alpha `$, which can be seen by noting that the radius of the small semicircles becomes smaller for larger values of $`\alpha `$. Again, we see that a nonzero value of $`\alpha `$ provides sufficient damping to account for the experimental observations. However, there is a discrepancy between our calculation and the experiment in the value of the frequency of the $`m=0`$ mode, since our calculation predicts $`\omega /\omega _11.68\pm 0.4`$, whereas the experimental data give a frequency $`2.05\omega _1`$. This discrepancy was also present in the calculations of Refs. and is discussed in Ref. . The authors of the last reference suggested that a possible reason for this discrepancy may be that the observed mode is in fact an in-phase $`m=0`$ mode (the upper solid curve in Fig. 4) rather than an out-of-phase one. This suggestion was based on calculating the oscillator strenghts of the in-phase and out-of-phase $`m=0`$ modes. It turned out that it is indeed possible experimentally to excite both modes simultaneously in the temperature range $`T0.25T_{BEC}`$ to $`T0.5T_{BEC}`$. Although the calculation in Ref. was performed in the collisionless limit, where there is no damping, we expect that with damping this argument remains qualitatively correct.
For the rest of the experimental points we can, in principle, explain the data using a two-parameter fit, namely with $`\alpha `$ and $`\overline{\omega }\tau _{22}`$. In fact, the collisions between atoms from the condensate with those from the thermal cloud are the main cause of damping in the condensate modes, and we can even explain the data using a one-parameter fit, namely only $`\alpha `$, by assuming that the system is in the collisionless regime with respect to $`\overline{\omega }\tau _{22}`$, i.e., $`\overline{\omega }\tau _{22}1`$. We perform this calculation by using a function $`\alpha (T/T_{BEC})`$ that gives the best fit for the experimental damping rates. In Fig. 7 we plot the function $`\alpha (T/T_{BEC})`$ and the resulting temperature-dependent frequencies and damping rates. This figure contains the main results of this paper. First of all we notice that now, with a nonzero $`\alpha `$, we obtain better agreement with the experimental data for the quadrupole mode frequencies than before where $`\alpha `$ was taken to be zero. This can be seen by comparing this figure with Fig. 4. We note also that, depending on the value of $`\alpha `$, the mode frequency may shift upwards or downwards. By comparing Fig. 4 with Fig. 7 for the mode frequencies one can clearly see that for temperatures below approximatelly $`0.6T_{BEC}`$ the theoretical curves are shifted upwards, whereas for the higher temperatures the curves are shifted downwards, in the end giving rise to the good agreement of the quadrupole mode with experiment. Secondly, we notice that the same function $`\alpha (T/T_{BEC})`$ also gives a good fit for the damping rates of both the monopole and quadrupole modes. Finally, there is still a discrepancy between our predictions and the experimental findings for the frequencies of the monopole mode at higher temperatures, but this may be resolved by a reinterpretation of the observed mode, as explained previously.
Finally, we want to mention an interesting feature of the in-phase and out-of-phase $`m=0`$ modes which can already be seen in Fig. 4 at $`T/T_{BEC}0.05`$. It is the familiar anticrossing tendency of these modes obtained also by Bijlsma and Stoof and Zaremba et al. . It is more dramatic here than in these two works due to the presence of the new parameter $`\alpha `$. This behavior can be seen in Fig. 8 where we see the two modes come close to each other and eventually intersect at some ‘critical’ values of $`\alpha `$ twice.
## IV Conclusion
We have extended the method of calculating collisional damping, used previously above the Bose-Einstein condensation transition temperature , to below the transition temperature. Furthermore, we have included the effect of noncondensate-condensate collisions. By comparing with experiment, we conclude that it is presumably this collision process which is mainly responsible for the observed damping. Our theory provides a general dispersion relation that gives complex mode frequencies at a certain temperature as a function of two dimensionless parameters, namely $`1/\overline{\omega }\tau _{22}`$ and $`\alpha `$, that characterize the noncondensate-noncondensate and noncondensate-condensate atomic collisions, respectively. Our results for the fully collisionless frequencies agree with those of Ref. for most of the temperature range below the transition temperature. The $`m=0`$ and $`m=2`$ hydrodynamic frequencies are, to the best of our knowledge, calculated here below the transition temperature for the first time.
At present we have not carried out a microscopic calculation of the mean collision times $`\tau _{22}`$ and $`\tau _{12}\alpha ^1`$, which are treated as phenomenological parameters here with the possibility to investigate the intermediate regime with respect to either of the two collision processes. However, the homogeneous calculations of Zaremba, Nikuni, and Griffin indicate that the function $`\alpha (T/T_{BEC})`$ that we have obtained from a fit to the experimental data, has the correct order of magnitude and qualitatively also the correct temperature-dependence. Nevertheless such a microscopic calculation for the inhomogeneous experimental conditions of interest would be very desirable and is left for future work.
We obtain a rather good agreement with the experimental results for the mode frequencies and damping rates apart from the discrepancy for the $`m=0`$ out-of-phase mode for $`T>0.7T_{BEC}`$. While the experiment shows an upward shift in the frequency with respect to its zero temperature value, we predict a downward shift. Resolution of this discrepancy seems to require more accurate experimental data, which will also provide decisive comparison with the theory presented here.
## Acknowledgements
The authors would like to thank Michiel Bijlsma, Chris Pethick and Henrik Smith for useful suggestions and remarks. |
warning/0003/quant-ph0003139.html | ar5iv | text | # Stochastic dynamics of electronic wave packets in fluctuating laser fields
## I Introduction
The advancement of sophisticated trapping techniques and the development of powerful laser sources has stimulated numerous theoretical and experimental investigations on the dynamics of wave packets in elementary quantum systems. An understanding of their dynamics is important for our conception of quantum mechanics and its relation to classical mechanics. So far most of the research in this context has concentrated on coherent aspects of wave packet dynamics which may be traced back theoretically to semiclassical aspects originating from the smallness of the de Broglie wave lengths involved. However, for an understanding of the emergence of classical behaviour also a detailed understanding of the destruction of quantum coherence is required. Such a destruction of coherence may arise from the coupling of a quantum system to a reservoir or from stochastic external influences. Though by now many aspects of the coherent dynamics of wave packets are well understood still many questions concerning the influence of stochastic perturbations on elementary quantum systems with a high level density are open.
A paradigm of a quantum system in which many of these latter aspects can be investigated in great detail are Rydberg systems interacting with fluctuating laser fields. Due to the inherent stochastic nature of laser light a detailed understanding of optical excitation processes with fluctuating laser fields is of vital interest for laser spectroscopy. Rydberg systems are of particular interest in this context due to their high level density close to an ionization threshold. Any laser-induced excitation process which involves Rydberg and continuum states close to an ionization threshold typically leads to the preparation of a spatially localized electronic Rydberg wave packet . Under the influence of a fluctuating laser field the coherence of such an electronic wave packet is disturbed whenever it is close to the ionic core where the electron-laser interaction is localized . Eventually these random perturbations are expected to lead to a stochastically dominated Brownian motion of the electronic wave packet. Indeed it has been demonstrated recently that such a transition to a diffusive behaviour takes place and that this diffusive dynamics are dominated by characteristic power laws which govern the time evolution of the Rydberg system. However, these previous investigations were restricted to a particular type of phase fluctuations of laser fields which can be described by the so called phase diffusion model (PDM) . This PDM implies a Lorentzian spectrum for the fluctuating laser field. It is known from the dynamics of three level systems that the somewhat unrealistic asymptotic frequency dependence of a Lorentzian laser spectrum may lead to unphysical predictions . However, from our previous investigations it remained open to which extent these characteristic, diffusive long time dynamics of an excited Rydberg electron depend on details of the fluctuations of the exciting laser field. Such a dependence might be expected on intuitive grounds as the diffusion of the excited Rydberg electron in energy space eventually also reaches the far-off resonant regions of the laser spectrum.
In this paper we tackle these open questions by generalizing our previous results to arbitrary types of laser fluctuations. For this purpose we derive rate equations for the relevant density matrix elements of the excited Rydberg electron which are averaged over the fluctuations of the laser field. These rate equations are based on a decorrelation of the relevant electron-field averages. This decorrelation approximation (DCA) is valid as long as the characteristic correlation time of the fluctuating laser field, i.e. its inverse bandwidth, is much smaller than all other relevant intrinsic dynamical time scales. Within this framework it will become apparent that it is the laser spectrum only which determines the time evolution of the excited Rydberg electron. On the basis of this approach it will be demonstrated which aspects of the diffusive long time dynamics of an excited electronic Rydberg wave packet depend on which details of the laser spectrum.
The range of validity of these Pauli-type rate equations is investigated in detail for a special class of phase fluctuations of the exciting laser field. This special class of phase fluctuations implies non-Lorentzian spectra which for large laser frequencies decrease more rapidly than a Lorentzian. These phase fluctuations might be considered as a realistic model for a single mode laser field which is operated well above the laser threshold. In order to access the range of validity of the Pauli-type rate equations for this special class of laser fluctuations a more general master equation is derived for the averaged density operator of the Rydberg electron. This more general approach is also capable of dealing with all coherent aspects of the laser excitation process.
The paper is organized in the following way: In Sec. II the theoretical models for describing the laser fluctuations and the dynamics of the Rydberg electron are presented. For the sake of simplicity we restrict our subsequent discussion to Rydberg systems which can be described within the framework of a one-channel approximation . Typically this approximation is well satisfied for Alkali atoms. In Sec.III we derive rate equations which describe the dynamics of the excited Rydberg electron averaged over the laser fluctuations. Self consistent validity conditions for the applicability of the decorrelation approximation (DCA) are discussed on which these rate equations are based on. In Sec. IV characteristic aspects of the time evolution predicted by the rate equations of Sec. III are exemplified. The different dynamical long time regimes and their characteristic power laws are discussed in detail. From the resulting analytical expressions for these power laws it is apparent to which extent they depend on details of the laser spectrum. In Sec.V we derive a more sophisticated master equation for the averaged dynamics of the excited Rydberg electron. This master equation is capable of describing also coherent aspects of the dynamics of the excited Rydberg electron but its validity is restricted to a particular class of phase fluctuations only. In Sec. VI solutions of this master equation are compared with the corresponding results of the rate equations. Thus we are able to determine the range of validity of the DCA. In our subsequent discussions we use atomic units ($`m_e=e=\mathrm{}=1`$).
## II Theoretical framework
In this section the theoretical models are introduced with which the fluctuating laser field and the excited Rydberg system are described.
### A The fluctuating laser field
We consider an atomic or molecular Rydberg system which is driven by a laser field with electric field strength
$$𝑬(t)=𝒆\epsilon (t)e^{i\omega t}+\text{c.c..}$$
(1)
We assume that this laser field can be described by a classical stochastic process . The mean frequency of this laser field is denoted $`\omega `$ and $`𝒆`$ is its polarization vector. The fluctuations of this laser field are described by the envelope function $`\epsilon (t)`$ which is assumed to be slowly varying on time scales of the order of $`1/\omega `$. The associated spectrum of this laser field is defined by
$$S(\mathrm{\Omega })=\frac{1}{\pi }\text{Re}_0^{\mathrm{}}𝑑\tau K(\tau )e^{i\mathrm{\Omega }\tau }$$
(2)
with the two-time correlation function of the slowly varying amplitude
$$K(\tau )=\epsilon (t+\tau )\epsilon ^{}(t).$$
(3)
Thereby $`\mathrm{}`$ denotes statistical averaging over the fluctuations of the laser field.
For a single mode laser which is operated well above laser threshold to a good degree of approximation the amplitude of the laser field is stable. In these cases fluctuations of a realistic laser field can be described by a classical electromagnetic field whose phase is fluctuating, i.e.
$$\epsilon (t)=\epsilon _0e^{i\mathrm{\Phi }(t)}.$$
(4)
The fluctuating phase $`\mathrm{\Phi }(t)`$ obeys the (Ito-)stochastic differential equation
$$d\varphi (t)=\varphi (t)\beta dt+\sqrt{2b}\beta dW(t)$$
(5)
with $`\varphi (t)=\dot{\mathrm{\Phi }}(t)`$. In Eq.(5) $`1/\beta `$ determines the correlation time of the stochastic frequency $`\varphi (t)`$ and $`dW(t)`$ is the differential of a real-valued Wiener process with zero mean and unit variance, i.e. $`dW(t)=0`$, $`dW(t)^2=dt`$ . Eqs.(4) and (5) imply the relation
$$K(\tau )=|\epsilon _0|^2\mathrm{exp}\left[b\tau +b/\beta \left(1e^{\tau \beta }\right)\right].$$
(6)
In the limit of large values of $`\beta `$ $`\mathrm{\Phi }(t)`$ itself approaches a real-valued Wiener process, i.e.
$$d\mathrm{\Phi }(t)=\sqrt{2b}dW(t).$$
(7)
This limiting case constitutes the so called phase diffusion model (PDM). It implies a Lorentzian laser spectrum of the form
$$S(\mathrm{\Omega })=|\epsilon _0|^2\frac{1}{\pi }\frac{b}{b^2+\mathrm{\Omega }^2}.$$
(8)
For $`\beta b`$ the parameter $`\beta `$ may be interpreted as a cut-off parameter of the laser spectrum. This becomes apparent by noting that for frequencies $`\mathrm{\Omega }\beta `$ the spectrum is always approximately Lorentzian whereas for large frequencies, i.e. $`\mathrm{\Omega }\beta `$, it tends to zero more rapidly as can be seen in Fig.1. More precisely, for $`\beta b`$ we obtain from Eqs.(6) and (2) the approximate relation
$`S(\mathrm{\Omega })`$ $`=`$ $`{\displaystyle \frac{|\epsilon _0|^2}{\pi }}{\displaystyle \frac{b}{\mathrm{\Omega }^2+b^2}}{\displaystyle \frac{1}{1+\left(\frac{\mathrm{\Omega }}{\beta }\right)^2}}.`$ (9)
### B The interaction Hamiltonian
Let us assume that the considered atom or molecule is prepared initially in an energetically low lying bound state $`|g`$ with energy $`ϵ_g`$. Furthermore, it is excited resonantly by the fluctuating laser field to Rydberg and/or continuum states $`|n`$ with energies $`ϵ_n`$ close to an ionization threshold. In the dipole and rotating wave approximation this excitation process can be described by the Hamiltonian
$$H=\underset{j=g,n}{}|jj|ϵ_j\underset{n}{}\left[\epsilon (t)e^{i\omega t}d_{ng}|ng|+\text{h.c.}\right].$$
(10)
The dipole matrix elements between the initial state $`|g`$ and the excited Rydberg states $`|n`$ are denoted $`d_{ng}=n\left|𝒅\mathbf{}𝒆\right|g`$. It is understood that the sum over the excited states $`|n`$ appearing in Eq.(10) includes also an integration over the adjacent continuum states. The energies of the excited Rydberg states and the energy dependence of the dipole matrix elements entering the Hamiltonian of Eq.(10) can be determined with the help of quantum defect theory (QDT) . In the case of excited Rydberg states which can be described within the framework of the one-channel approximation we find, for example,
$`ϵ_n`$ $`=`$ $`{\displaystyle \frac{1}{2(n\alpha )^2}},`$ (11)
$`d_{ng}`$ $`=`$ $`d_{ϵg}(n\alpha )^{3/2}d_{ϵg}|ϵ_nϵ_{n+1}|^{1/2}.`$ (12)
Thereby $`\alpha `$ denotes the quantum defect of the excited Rydberg series and $`d_{ϵg}`$ is the dipole matrix element between the initial state $`|g`$ and an energy normalized continuum state $`|ϵ`$ with energy $`ϵ0`$. Within the framework of QDT $`\alpha `$ as well as $`d_{ϵg}`$ are approximately energy independent close to the ionization threshold, i.e. for energies $`|ϵ|1`$. A one channel approximation is appropriate for all cases in which excited states of the ionic core are located far away from the excited energy region. Typically this condition is fulfilled for Alkali atoms.
The main problem is to solve the stochastic Schrödinger equation associated with the Hamiltonian of Eq.(10). In general this is a complicated task due to the simultaneous presence of the laser fluctuations and of the threshold effects arising from the infinitely many bound Rydberg states converging to the ionization threshold. By the resulting intricate interplay between laser fluctuations and threshold phenomena it is difficult to apply stochastic simulation methods which typically become unreliable due to numerical inaccuracies in particular for long interaction times.
## III Decorrelation approximation (DCA)
In this section an approximation method is developed for determining the dynamics of the Rydberg system in the fluctuating laser field. This approximation method is based on a decorrelation of atom-field averages and leads to a Pauli-type master equation for the density operator of the Rydberg system which is averaged over the laser fluctuations. In this master equation all coherences between different energy levels have been eliminated adiabatically. This decorrelation approximation (DCA) is valid for arbitrary types of laser fluctuations provided these fluctuations are sufficiently fast (compare with conditions (23), (24) and (26) derived below).
Let us start by determining first of all an approximate equation of motion for the probabilities $`\rho _{nn}(t)=n|\psi (t)\psi (t)|n`$ of observing the excited Rydberg system in one of the Rydberg states $`|n`$. Neglecting coherences $`\rho _{nn^{}}(t)`$ with $`nn^{}`$ we find from the stochastic Schrödinger equation with Hamiltonian (10) the relation
$`\underset{nn}{\overset{\mathbf{}}{\rho }}(t)`$ $`=`$ $`2d_{ng}^2\mathrm{Re}\{{\displaystyle _{t_0}^t}𝑑t^{}\epsilon (t)\epsilon ^{}(t^{})e^{i(\overline{ϵ}ϵ_n)(tt^{})}\}`$ (15)
$`\times \left[\rho _{gg}(t^{})\rho _{nn}(t^{})\right]`$
$`\left[id_{ng}\epsilon (t)e^{i(\overline{ϵ}ϵ_n)(tt_0)}\rho _{gn}(t_0)+\text{h.c.}\right]`$
with $`0t_0t`$ and with $`\rho _{kj}(t)=k|\psi (t)\psi (t)|j\mathrm{exp}[it(ϵ_kϵ_j)]`$. The mean excited energy is denoted $`\overline{ϵ}=ϵ_g+\omega +\delta \omega `$ with the quadratic Stark shift contribution of all other (non-resonant) states. In the subsequent discussion we assume for the sake of simplicity that the intensity dependence of $`\delta \omega `$ does not affect the dynamics of the excited Rydberg system. This is valid either for pure phase fluctuations of the laser field or in the case of arbitrary laser fluctuations for laser bandwidths which are much larger than $`\delta \omega `$. The self-consistency condition for the omission of the coherences $`\rho _{nn^{}}(t)`$ with $`nn^{}`$ will be discussed later (compare with Eq.(24)). Taking the integration interval $`[t_0,t]`$ to be smaller than the characteristic time scale over which $`[\rho _{nn}(t^{})\rho _{gg}(t^{})]`$ changes significantly this latter term can be approximated by its value at time $`t`$. If on the other hand the interval $`[t_0,t]`$ is assumed to be larger than the correlation time of the fluctuating laser field we can replace the lower integration limit $`t_0`$ by $`\mathrm{}`$ in Eq.(15). Thus we obtain
$`\underset{nn}{\overset{\mathbf{}}{\rho }}(t)`$ $`=`$ $`2|d_{ng}|^2\left[\rho _{gg}(t)\rho _{nn}(t)\right]`$ (18)
$`\times \text{Re}{\displaystyle _0^{\mathrm{}}}𝑑\tau \epsilon (t)\epsilon ^{}(t\tau )e^{i\tau (\overline{ϵ}ϵ_n)}`$
$`\left\{id_{ng}\epsilon (t)e^{i(tt_0)(\overline{ϵ}ϵ_n)}\rho _{gn}(t_0)+\text{h.c.}\right\}.`$
Now we are able to carry out the statistic average $`\mathrm{}`$ over the laser fluctuations. Due to the above mentioned conditions on the integration interval the involved density matrix elements and the laser field $`\epsilon (t)`$ decorrelate. As $`\epsilon (t)=0`$ the contribution of the last term on the right hand side of Eq.(18) vanishes. The remaining terms yield the rate equation
$$\underset{nn}{\overset{\mathbf{}}{\rho }}(t)=_{ng}[\rho _{gg}(t)\rho _{nn}(t)]$$
(19)
with the time independent rates
$$_{ng}=2\pi |d_{ng}|^2S(\overline{ϵ}ϵ_n)$$
(20)
and with the laser spectrum $`S(\mathrm{\Omega })`$ as defined by Eq.(2).
In order to work out quantitative criteria for the validity of Eq.(19) let us define an effective bandwidth $``$ of the fluctuating laser field $`S(\mathrm{\Omega })`$ by the relation
$$S(0)\pi _{\mathrm{}}^{\mathrm{}}𝑑\mathrm{\Omega }S(\mathrm{\Omega })=|\epsilon |^2.$$
(21)
The quantity $`1/`$ measures the correlation time of the fluctuating laser field. In the case of the PDM, for example, this effective bandwidth equals the width of the Lorentzian spectrum $`b`$. In all other cases which are described by Eq.(5) it characterizes the effective frequency width of the laser spectrum of Eq.(2). According to Eq.(19) the inverse rates $`1/_{ng}`$ define the characteristic time scale over which $`[\rho _{gg}(t)\rho _{nn}(t)]`$ varies significantly. Thus, the above decorrelation approximation applies to cases only for which $`_{ng}`$. Typically one finds $`_{n_{res}g}_{ng}`$ with $`ϵ_{n_{res}}=\overline{ϵ}`$ so that one of the validity conditions for the decorrelation condition becomes
$$2\pi |d_{n_{res}g}|^2S(0).$$
(22)
Eliminating $`S(0)`$ by Eq.(21) we finally arrive at the equivalent condition
$$^2\frac{1}{2}\mathrm{\Omega }_R^2$$
(23)
with the average Rabi-frequency $`\mathrm{\Omega }_R=2|d_{n_{res}g}|\sqrt{|\epsilon |^2}`$. What remains to be found is a validity condition for neglecting the coherences $`\rho _{nn^{}}(t)`$ with $`nn^{}`$ in Eq.(15). As long as
$$|ϵ_{n_{res}}ϵ_{n_{res}+1}|_{n_{res}g}=\frac{\mathrm{\Omega }_R^2}{2}$$
(24)
the coherences $`\rho _{nn^{}}(t)`$ with $`nn^{}`$ are rapidly oscillating functions in comparison with the slowly varying probabilities $`\rho _{nn}(t)`$ and $`\rho _{gg}(t)`$ entering Eq.(15) so that their influence averages to zero approximately. Therefore the inequalities (23) and (24) are the required conditions for the validity of the DCA. According to these conditions we may distinguish two limiting cases. In the limit of small laser bandwidths for which $`|ϵ_{n_{res}}ϵ_{n_{res}+1}|`$ they reduce to the requirement $`^2\mathrm{\Omega }_R^2/2`$. In two-level systems which are excited resonantly by a fluctuating laser field this is the well known limit of large laser bandwidths in which the dynamics are dominated by rate equations. In the opposite limit where the bandwidth is large enough to affect many excited Rydberg states, i.e. $`|ϵ_{n_{res}}ϵ_{n_{res}+1}|`$, the conditions for the applicability of the DCA reduce to the relation $`\gamma /\pi `$. Thereby we have introduced the laser-induced rate
$$\gamma =2\pi |d_{ϵg}|^2|\epsilon |^2\frac{\pi }{2}\mathrm{\Omega }_R^2|ϵ_{n_{res}}ϵ_{n_{res}+1}|^1$$
(25)
which characterizes ionization of the initial state $`|g`$ into continuum states close to the ionization threshold according to Fermi’s Golden rule.
Up to now, our arguments for the derivation of the rate equation (19) and of conditions (23) and (24) apply for discrete excited states only. However, our previous arguments can be generalized easily also to continuum states by viewing these continuum states as infinitesimally spaced discrete energy levels. According to quantum defect theory close to an ionization threshold the energy dependence of the discrete dipole matrix elements $`d_{ng}`$ is described by Eq.(12). Thus condition (24) reduces to
$$2|d_{ϵg}|^2|\epsilon |^2\frac{\gamma }{\pi }.$$
(26)
In the limit of an infinitesimally small level spacing between the excited states Eqs. (19) and (12) imply that the probability of finding the excited Rydberg system in a continuum state becomes vanishingly small. Thus we find
$$\dot{\rho }_{ϵϵ}(t)=_{ϵg}\rho _{gg}(t)$$
(27)
with $`_{ϵg}=2\pi |d_{ϵg}|^2S(\overline{ϵ}ϵ)`$. Integration over the whole electron continuum finally yields
$$\underset{ion}{\overset{\mathbf{}}{P}}(t)=\mathrm{\Gamma }\rho _{gg}(t)$$
(28)
with the mean ionization probability $`P_{ion}(t)=_0^{\mathrm{}}𝑑ϵ\rho _{ϵϵ}(t)`$ and with the effective ionization rate
$$\mathrm{\Gamma }=2\pi _0^{\mathrm{}}𝑑ϵ|d_{ϵg}|^2S(\overline{ϵ}ϵ).$$
(29)
If the mean excited energy $`\overline{ϵ}`$ is located well above threshold and if $`d_{ϵg}`$ is still energy independent over the energy region over which $`S(\overline{ϵ}ϵ)`$ is significant, this effective ionization $`\mathrm{\Gamma }`$ reduces to the previously introduced ionization rate $`\gamma `$ of Eq.(25).
The rate Eqs. (19) and (28) together with the conservation of probability, i.e.
$$\rho _{gg}(t)=1P_{ion}(t)\underset{n}{}\rho _{nn}(t)$$
(30)
and together with the initial condition $`\rho (t=0)=|gg|`$ determine the time evolution of a laser excited Rydberg electron within the framework of the DCA.
## IV Stochastic dynamics of Rydberg systems within the DCA
In this section the dynamics of a Rydberg system is investigated with the help of the DCA on the basis of Eqs.(19), (28) and (30).
Within the framework of a one-channel approximation the excited energies and the energy dependence of the relevant dipole matrix elements of a Rydberg system can be described by Eqs.(12). They are characterized by a quantum defect $`\alpha `$ and by an energy-normalized dipole matrix element $`d_{ϵg}`$ which are both approximately energy independent for $`|ϵ|1`$. The laser-induced coupling between the initial state $`|g`$ and the excited Rydberg- and continuum states is characterized by the ionization rate $`\gamma `$ of Eq.(25). Typically this description is adequate for Rydberg states of Alkali atoms.
The rate equations for the averaged density operator of the Rydberg system (compare with Eqs.(19),(28) and (30)) can be analyzed in a convenient way with the help of Laplace transformations. Defining the Laplace transformed density operator by
$$\stackrel{~}{\rho }(z)=_0^{\mathrm{}}𝑑te^{izt}\rho (t)$$
(31)
the associated inverse transformation is given by
$$\rho (t)\frac{1}{2\pi }_{\mathrm{}+i0}^{\mathrm{}+i0}𝑑ze^{izt}\stackrel{~}{\rho }(z).$$
(32)
Thus the Laplace transformed rate equations (19), (28) and (30) imply the relations
$`\stackrel{~}{\rho }_{gg}(z)`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{\Gamma }iz\sigma (z)}}`$ (33)
$`\text{and }\stackrel{~}{P}_{ion}(z)`$ $`=`$ $`{\displaystyle \frac{i\mathrm{\Gamma }}{z\left[\mathrm{\Gamma }iz\sigma (z)\right]}}`$ (34)
with
$$\sigma (z)=\underset{n}{}\frac{_{ng}}{_{ng}iz}.$$
(35)
The rates $`_{ng}`$ entering Eq.(35) characterize the transitions between states $`|g`$ and $`|n`$ within the DCA and are defined by Eq.(20). In the derivation of Eqs.(33) and (34) $`\rho _{gg}(t)`$ has been neglected in comparison with $`_n\rho _{nn}(t)`$ and $`P_{ion}(t)`$ in Eq.(30).
We may distinguish various dynamics regimes which are treated subsequently.
### A Asymptotic long time behaviour
The time evolution of the averaged density operator of the Rydberg system can be obtained from Eqs.(33) and (34) and from the inversion formula (32). In general the time evolution will exhibit both exponential decays originating from poles of the Laplace transforms (33) and (34) in the complex $`z`$-plane and power law decays which originate from cut contributions starting from the branch point of $`\sigma (z)`$ at $`z=0`$. As the asymptotic long time behaviour will be dominated by power law decays we have to investigate the structure of the characteristic kernel $`\sigma (z)`$ around the branch point $`z=0`$ in more detail. From Eqs.(35) and (20) it follows that for $`z0`$ its main contributions arise from the infinitely many Rydberg states very close to the ionization threshold. Hence in the long time limit we may approximate $`S(\overline{ϵ}ϵ_n)`$ by $`S(\overline{ϵ})`$ in expression (20). Furthermore we may replace the sum over all Rydberg states in Eq.(35) by an integration. So finally in the limit $`z0`$ we obtain the relation
$$\sigma (z)\frac{2\pi }{3\sqrt{3}}\left(\frac{i2\pi |d_{ϵg}|^2S(\overline{ϵ})}{z}\right)^{1/3}(z0).$$
(36)
Inserting Eq.(36) into Eqs.(32), (33) and (34) one obtains the asymptotic long time behaviour
$`\rho _{gg}(t)`$ $`=`$ $`\left({\displaystyle \frac{S(\overline{ϵ})}{|\epsilon |^2}}\right)^{1/3}{\displaystyle \frac{\mathrm{\Gamma }(\frac{5}{3})}{3(\mathrm{\Gamma }/\gamma )^2}}(t\gamma )^{5/3}`$ (37)
$`P_{ion}(t)`$ $`=`$ $`1\left({\displaystyle \frac{S(\overline{ϵ})}{|\epsilon |^2}}\right)^{1/3}{\displaystyle \frac{\mathrm{\Gamma }(\frac{2}{3})}{3(\mathrm{\Gamma }/\gamma )}}(t\gamma )^{2/3}`$ (38)
with $`\mathrm{\Gamma }(x)=_0^{\mathrm{}}𝑑ue^uu^{x1}`$ denoting the gamma function . Eqs.(37) and (38) describe the time evolution of the mean initial state probability and of the mean ionization probability for sufficiently long interaction times. They are generalizations of our previous results of Ref. which were only valid for phase fluctuations of the PDM. Within the framework of the DCA these asymptotic laws are valid for arbitrary fluctuations of the laser field provided $`\gamma /\pi `$ (compare with Eq.(26)). Obviously this asymptotic long time behaviour is independent of the quantum defect which characterizes the influence of the ionic core of the Rydberg system. Furthermore, the characteristic exponents of the long time behaviour are a peculiar property of the Coulomb problem and do not depend on details of the laser spectrum. However, the time independent pre-factors of these power laws depend on $`S(\mathrm{\Omega })`$ and on the effective ionization rate $`\mathrm{\Gamma }`$ of Eqs.(2) and (29).
After which interaction time do we expect the asymptotic power laws of Eqs.(37) and (38) to become valid? As apparent from Eq.(38) this diffusive long time dynamics finally leads to complete ionization of the Rydberg system. Thus it is reasonable to characterize the onset of this asymptotic long time dynamics by a stochastic ionization time $`t_c`$ which is defined by the condition $`P_{ion}(t_c)=1/2`$ which yields
$$t_c=\frac{1}{\gamma }\left[\left\{\frac{\gamma \mathrm{\Gamma }(\frac{2}{3})}{\mathrm{\Gamma }}\right\}^3\frac{8}{27}\frac{S(\overline{ϵ})}{|\epsilon |^2}\right]^{1/2}.$$
(39)
### B Intermediate interaction times
In this section we deal with characteristic aspects of the dynamics described by Eqs.(19) and (28) in cases in which the interaction times are large enough so that the initial state $`|g`$ is depleted significantly but which are still much smaller than the stochastic ionization time $`t_c`$ of Eq.(39). For these interaction times we may distinguish two characteristic regimes depending on whether the excited states are close to the ionization threshold or above or whether they are located well below this threshold.
#### 1 Excitation at or above threshold
In this dynamical regime the significantly excited states are located at the ionization threshold or above. This implies that all rates $`_{ng}`$ which describe the coupling between states $`|g`$ and $`|n`$ are small in comparison with the total ionization rate $`\gamma `$. Thus considering interaction times $`t`$ which are not very much larger than $`1/\gamma `$ implies that we may take $`t_{ng}1`$ for all quantum numbers $`n`$. Hence, considering the formal solution of Eq.(19)
$$\rho _{nn}(t)=_{ng}_0^t𝑑se^{(ts)_{ng}}\rho _{gg}(s)$$
(40)
we may replace the exponential by unity. Performing the summation over all Rydberg states we find with the help of $`_n_{ng}\gamma \mathrm{\Gamma }`$ (the sum over all Rydberg states has been replaced by an integral)
$$\underset{n}{}\underset{nn}{\overset{\mathbf{}}{\rho }}(t)=(\gamma \mathrm{\Gamma })\rho _{gg}(t)\gamma ^2\mathrm{\Lambda }_{Sp}_0^t𝑑s\rho _{gg}(s)$$
(41)
where we used the quantity
$`\mathrm{\Lambda }_{Sp}`$ $`=`$ $`{\displaystyle \frac{1}{|\epsilon |^2^2}}{\displaystyle _{\mathrm{}}^0}𝑑ϵ_n({\displaystyle \frac{dϵ_n}{dn}})S^2(\overline{ϵ}ϵ_n)`$ (43)
$`{\displaystyle \frac{1}{|\epsilon |^2^2}}{\displaystyle _0^{\mathrm{}}}𝑑ϵS^2(\overline{ϵ}+ϵ)(2ϵ)^{3/2}`$
which characterizes the spectral properties of the fluctuating laser field.
We take the time-derivative of Eq.(30) and eliminate the ionization probability with Eq.(28). Inserting Eq.(41) in this equation we find an integro-differential equation for $`\rho _{gg}(t)`$, namely
$$\gamma \rho _{gg}(t)\gamma ^2\mathrm{\Lambda }_{Sp}_0^t𝑑s\rho _{gg}(s)+\underset{gg}{\overset{\mathbf{}}{\rho }}(t)=0.$$
(44)
As $`\mathrm{\Lambda }_{Sp}1`$ one finally obtains the relations
$$\rho _{gg}(t)=\frac{1}{1+\mathrm{\Lambda }_{Sp}}\left[\mathrm{exp}(\gamma t)+\mathrm{\Lambda }_{Sp}\mathrm{exp}(\gamma \mathrm{\Lambda }_{Sp}t)\right]$$
(45)
and
$$P_{ion}(t)=\frac{\mathrm{\Gamma }}{\gamma (1+\mathrm{\Lambda }_{Sp})}\left[\mathrm{exp}(\gamma \mathrm{\Lambda }_{Sp}t)\mathrm{exp}(\gamma t)\right].$$
(46)
These equations even apply to interaction times $`t<1/\gamma `$. Note that consistent with our approximations the interaction times always fulfill the inequality $`\gamma \mathrm{\Lambda }_{Sp}t1`$. In the special case of the PDM $`\mathrm{\Lambda }_{Sp}`$ reduces to
$$\mathrm{\Lambda }_{Sp}=\sqrt{\frac{b}{2\pi ^2}}\text{Re}\left\{\left(12i\frac{\overline{ϵ}}{b}\right)\sqrt{\frac{\overline{ϵ}}{b}i}\right\}.$$
(47)
Very close to threshold, i.e. for $`\overline{ϵ}0`$, one obtains $`\mathrm{\Lambda }_{Sp}=\sqrt{b}/(2\pi )`$ so that in this special case we obtain again our previous results of Ref. .
#### 2 Excitation well below threshold
If the fluctuating laser field excites Rydberg states well below the ionization threshold, i.e. $`|\overline{ϵ}|`$ and $`\overline{ϵ}<0`$, the considerations of Sec.IV B 1 have to be modified. Since the interaction time is assumed to be smaller than the stochastic ionization time $`t_c`$ we can approximate the characteristics of the dominantly excited Rydberg states by
$`ϵ_n`$ $``$ $`\overline{ϵ}+2\pi (nn_{res})/T_{\overline{ϵ}},`$ (48)
$`d_{ng}`$ $``$ $`d_{ϵg}\sqrt{2\pi /T_{\overline{ϵ}}}.`$ (49)
Thereby $`T_{\overline{ϵ}}2\pi (2\overline{ϵ})^{3/2}`$ denotes the classical Kepler period of the mean excited Rydberg state of energy $`\overline{ϵ}<0`$. Whereas in the previous subsection the stochastic influence could be characterized by a single average spectral property for arbitrary types of fluctuations, namely by $`\mathrm{\Lambda }_{Sp}`$, the excitation dynamics well below threshold turns out to be much more sensitive to the details of the laser spectrum. This is easily demonstrated by considering phase fluctuations which can be described by the laser spectrum of Eq.(9) as a particular example. This spectrum describes fluctuations of a single mode laser field well above the laser threshold in the limit $`\beta b`$. In addition, if only Rydberg states well below threshold are excited significantly we may neglect the effective ionization rate $`\mathrm{\Gamma }`$ in the denominator of Eqs.(33) and (34). In the limit $`\beta b`$ we thus arrive at the relations
$`\rho _{gg}(t)`$ $`=`$ $`{\displaystyle \frac{2}{T_{\overline{ϵ}}\beta }}f^{}\left({\displaystyle \frac{2\gamma b}{T_{\overline{ϵ}}\beta ^2}}t\right)`$ (50)
$`P_{ion}(t)`$ $`=`$ $`{\displaystyle \frac{1}{\pi }}\left[{\displaystyle \frac{\beta }{|\overline{ϵ}|}}\mathrm{arctan}\left({\displaystyle \frac{\beta }{|\overline{ϵ}|}}\right)\right]f\left({\displaystyle \frac{2\gamma b}{T_{\overline{ϵ}}\beta ^2}}t\right).`$ (51)
Thus within this limit for arbitrary values of $`\beta `$ and $`b`$ the influence of the phase fluctuations of the laser field is described by the single scaling function $`f(\tau )`$ which is defined by the equation
$$\frac{df(\tau )}{d\tau }=\text{Im}_0^{\mathrm{}}\frac{d\zeta }{\zeta }e^{i\zeta \tau }\left\{_{\mathrm{}}^{\mathrm{}}\frac{dx}{1i\zeta (x^2+x^4)}\right\}^1.$$
(52)
To end up with Eq.(52) we had to apply the further approximation $`\mathrm{\Omega }b`$ in the laser spectrum of Eq.(9). Physically speaking this approximation means that we consider cases in which the essential dynamics are dominated by energy states which are located in the wings of the laser spectrum.
In the limits $`\tau 1`$ and $`\tau 1`$ asymptotic expressions are easily obtained from Eq.(52). The limit of small values of $`\tau `$ is realized in the PDM where $`\beta \mathrm{}`$ and where the spectrum of Eq.(9) reduces to a Lorentzian form. In this case one obtains the expression
$$f(\tau )2\sqrt{\frac{\tau }{\pi }}(\tau 1).$$
(53)
Consequently Eqs.(50) and (51) yield
$`\rho _{gg}(t)`$ $`=`$ $`\sqrt{{\displaystyle \frac{2}{\pi bT_{\overline{ϵ}}\gamma t}}},`$ (54)
$`P_{ion}(t)`$ $`=`$ $`2\left[{\displaystyle \frac{1}{|\overline{ϵ}|}}{\displaystyle \frac{1}{\beta }}\mathrm{arctan}\left({\displaystyle \frac{\beta }{|\overline{ϵ}|}}\right)\right]\sqrt{{\displaystyle \frac{2t\gamma b}{\pi ^3T_{\overline{ϵ}}}}}.`$ (55)
From the numerical data shown in Fig.2 it is apparent that Eqs.(54) and (55) are good estimates for interaction times $`t<t_{\text{PDM}}=T_{\overline{ϵ}}\beta ^2/(200b\gamma )`$.
In the extreme opposite limit of large values of $`\tau `$ we obtain the relations
$$f(\tau )\frac{4}{3\pi }\mathrm{\Gamma }(\frac{1}{4})\tau ^{3/4}(\tau 1)$$
(56)
and
$`\rho _{gg}(t)={\displaystyle \frac{\mathrm{\Gamma }(\frac{1}{4})}{\pi }}\left({\displaystyle \frac{8}{T_{\overline{ϵ}}^3\beta ^2t\gamma b}}\right)^{1/4},`$ (57)
$`P_{ion}(t)=`$ (58)
$`{\displaystyle \frac{4}{3\pi ^2}}\mathrm{\Gamma }(1/4)\left[{\displaystyle \frac{\beta }{|\overline{ϵ}|}}\mathrm{arctan}\left({\displaystyle \frac{\beta }{|\overline{ϵ}|}}\right)\right]\left({\displaystyle \frac{2\gamma bt}{T_{\overline{ϵ}}\beta ^2}}\right)^{3/4}.`$ (59)
In this case the power law decays which characterize the diffusive dynamics of the excited Rydberg electron differ from the corresponding results of the PDM significantly. Even the characteristic exponents are changed. According to Fig.2 this dynamical regime is realized for interaction times $`t`$ which fulfill the relation $`t_{1/4}<tt_c`$ with $`t_{1/4}=T_{\overline{ϵ}}\beta ^2/(2b\gamma )`$.
## V Full master-equation
Due to their simplicity and their applicability to all types of laser spectra the DCA rate equations are ideal for understanding the dynamics of Rydberg systems in the case of large laser bandwidths (compare with Eqs.(23),(24) and (26)). However the DCA approximation is not capable of describing coherent aspects of the laser-induced excitation process. In order to investigate the limits of applicability of the DCA rate equations in this section a more general approach is developed which is also capable of describing all coherent aspects of the excitation process. For the sake of simplicity we shall restrict our subsequent discussion to the case of phase fluctuations of the exciting laser field which deviate only slightly from a Lorentzian spectrum and which can be modelled by Eqs.(4), (5) and (9). For these type of laser fluctuations we shall derive an approximate master equation involving the density matrix elements of the excited Rydberg electron which are averaged over the fluctuations of the laser field. This procedure is a generalization of previous approaches which so far have been applied to atomic few level system only . In the special case of the PDM these subsequently derived density matrix equations reduce to our previous results of Ref..
We start from the Schrödinger equation with Hamiltonian (10) with a fluctuating laser field as given by Eqs. (4) and (5). For the effective density operator
$$\rho (t)=\underset{k,jg,n}{}|kj|k|\psi (t)\psi (t)|je^{i(\mathrm{\Phi }(t)+\omega t)(\delta _{jg}\delta _{kg})}$$
(60)
we obtain the equation of motion
$$\stackrel{\mathbf{}}{\rho }(t)=i[H_{dr.},\rho (t)]i\varphi (t)[|gg|,\rho (t)].$$
(61)
The self-adjoint Hamiltonian
$$H_{dr.}=\underset{n,g}{}|nn|ϵ_n+|gg|\overline{ϵ}\epsilon _0\underset{n}{}\left(d_{ng}|ng|+\text{h.c.}\right)$$
(62)
describes the dynamics of the Rydberg system in the absence of phase fluctuations and the stochastic process $`\varphi (t)`$ is defined by Eq.(5). In order to average Eq.(61) over all possible realizations of the stochastic process $`\varphi (t)`$ it is convenient to introduce the averaged operators
$$\rho ^{(n)}(t)=\left(\frac{\beta }{b}\right)^{n/2}\frac{(i)^n}{\sqrt{n!}}_{\mathrm{}}^{\mathrm{}}𝑑\varphi Q_n(\varphi )\rho (t)p(\varphi ,t)$$
(63)
with $`n=0,1,2,3,\mathrm{}`$ . The (conditional) probability distribution $`p(\varphi ,t)`$ obeys the Fokker-Planck equation
$$[\frac{}{t}+]p(\varphi ,t)=0$$
(64)
with the Fokker-Planck operator
$$=\beta \frac{}{\varphi }\varphi +b\beta \frac{^2}{\varphi ^2}.$$
(65)
This equation has to be solved with the initial condition
$$p(\varphi ,0)=\frac{1}{\sqrt{2\beta b\pi }}\mathrm{exp}\left(\frac{\varphi ^2}{2\beta b}\right)$$
(66)
which represents the stationary solution of the Fokker-Planck equation. The quantities
$$Q_n(\varphi )=H_n\left(\varphi \left[2^{n+1}n!\beta b\right]^{1/2}\right)$$
(67)
with the Hermite polynomials $`H_n`$ are eigenfunctions of the adjoined Fokker-Planck operator $`^{}`$ with eigenvalues $`\mathrm{\Lambda }_n=n\beta `$ . Starting from Eq.(61) we obtain a set of coupled differential equations for the operators $`\rho ^{(n)}(t)`$, namely
$`\stackrel{\mathbf{}}{\rho ^{(n)}}(t)`$ $`=`$ $`i[H_{dr.},\rho ^{(n)}(t)]n\beta \rho ^{(n)}(t)`$ (68)
$`+`$ $`b(n+1)[|gg|,\rho ^{(n+1)}(t)]\beta [|gg|,\rho ^{(n1)}(t)].`$ (69)
These equations have to solved subject to the initial condition
$$\rho ^{(n)}(0)=\delta _{n0}\rho (0).$$
(70)
According to Eqs.(63) and (67) $`\rho ^{(0)}(t)`$ is the required density operator which is averaged over the phase fluctuations of the laser field.
We may derive an approximate equation of motion for the averaged density operator $`\rho ^{(0)}(t)`$. As a first step we Laplace transform Eqs.(68), i.e.
$`iz\stackrel{~}{\rho }^{(n)}(z)`$ $`=`$ $`\rho ^{(n)}(0)n\beta \stackrel{~}{\rho }^{(n)}(z)`$ (71)
$``$ $`i[H_{dr.},\stackrel{~}{\rho }^{(n)}(z)]+`$ (72)
$`+`$ $`[|gg|,b(n+1)\stackrel{~}{\rho }^{(n+1)}(z)\beta \stackrel{~}{\rho }^{(n1)}(z)].`$ (73)
In view of the large values of $`\beta `$ we are interested in we may neglect terms containing $`\epsilon _0`$ in comparison with terms containing $`\beta `$. Thus with the definition
$$\stackrel{~}{\alpha }_l^n(z)=\frac{\stackrel{~}{\rho }_{lg}^{n+1}(z)}{\stackrel{~}{\rho }_{lg}^n(z)}$$
(74)
we arrive at the recursion relations
$$n+1+\frac{i}{\beta }(ϵ_l\overline{ϵ}z)=\frac{b}{\beta }(n+2)\stackrel{~}{\alpha }_l^{n+1}(z)+\frac{1}{\stackrel{~}{\alpha }_l^n(z)}.$$
(75)
From these recursion relations we find
$$\stackrel{~}{\alpha }_k^0(z)=\frac{1}{1+i{\displaystyle \frac{ϵ_k\overline{ϵ}z}{\beta }}+{\displaystyle \frac{2b/\beta }{2+i{\displaystyle \frac{ϵ_k\overline{ϵ}z}{\beta }}+{\displaystyle \frac{3b/\beta }{3+\mathrm{}}}}}}.$$
(76)
Using Eqs.(74) and (76) we may now eliminate $`\stackrel{~}{\rho }^{(1)}(z)`$ in Eq.(72) for $`n=0`$. Performing the Laplace back transformation (32) and using the definition $`\rho (t)=\rho ^{(0)}(t)`$ we finally obtain the master equation
$`\stackrel{\mathbf{}}{\rho }(t)`$ $`=`$ $`i[H_{dr.},\rho (t)]`$ (77)
$``$ $`b{\displaystyle _0^t}𝑑\tau {\displaystyle \underset{kg}{}}\left\{\alpha _k^0(\tau )|kk|\rho (t\tau )|gg|+\text{h.c.}\right\}`$ (78)
with the memory function
$$\alpha _k^0(\tau )=\frac{1}{2\pi }_{\mathrm{}+i0}^{\mathrm{}+i0}𝑑ze^{iz\tau }\stackrel{~}{\alpha }_k^0(z).$$
(79)
In the limit of the PDM, i.e. for $`\beta \mathrm{}`$, this master equation reduces to the well known form
$`\stackrel{\mathbf{}}{\rho }(t)`$ $`=`$ $`i[H_{dr.},\rho (t)]`$ (80)
$`+`$ $`{\displaystyle \frac{1}{2}}\left\{[L,\rho (t)L^{}]+[L\rho (t),L^{}]\right\}`$ (81)
with the Lindblad operator
$$L=\sqrt{2b}|gg|.$$
(82)
## VI Numerical results
In this section numerical solutions of the master equation (77) are compared with the corresponding solutions of the DCA rate equations (19),(28) and (30). Details of the numerical technique for solving Eq.(77) are summarized in the appendix. On the basis of these comparisons the validity conditions for the applicability of the DCA and its accuracy can be tested. For this purpose we consider the laser excitation of a Rydberg system which can be described by quantum defect theory in a one-channel approximation (compare with Eqs.(12)). Typically this is a good approximation for Alkali atoms.
The time evolution of the mean initial state probability $`\rho _{gg}(t)`$ and of the mean ionization probability $`P_{ion}(t)`$ are depicted in Fig.3 for excitation at and well below the ionization threshold for different values of $`\beta `$. In both cases it is assumed that the exciting laser field has a well defined amplitude and a fluctuating phase.
Let us first turn to the case depicted in Fig.3a: The spectrum of the laser field is close to Lorentzian ($`\beta b`$), so that the asymptotic form Eq.(9) applies well. Thus the effective bandwidth $``$ as defined by Eq.(21) is approximately equal to the parameter $`b`$ which characterizes the spectrum of Eq.(9) and the parameter $`\beta `$ might be interpreted as an effective cut-off frequency of the laser spectrum. Rydberg states are excited by the fluctuating laser field well below the ionization threshold. The mean excited energy corresponds to a quantum number $`n_{res}=(2\overline{ϵ})^{1/2}=200`$. The laser bandwidth $`b`$ and the laser-induced rate $`\gamma `$ are so small that the excited Rydberg states are located well below threshold, i.e. $`\overline{ϵ}b,\gamma `$. However, the values of $`b`$ and $`\gamma `$ are large enough so that more than one Rydberg state around energy $`\overline{ϵ}`$ is affected significantly by the laser field, i.e. $`T_{\overline{ϵ}}\gamma ,T_{\overline{ϵ}}b>1`$. The three curves of Fig.3a (solid, dashed and long dashed) correspond to different values of the effective cut-off frequency $`\beta `$ of the laser spectrum. As many excited states are involved in the depletion of state $`|g`$ the initial stage of the time evolution is governed by an approximate exponential decay of state $`|g`$ with rate $`\gamma `$ . This initial stage of the time evolution is independent of the fluctuations of the laser field. At larger interaction times with $`t>1/\gamma `$ a coherent oscillation starts to appear in $`\rho _{gg}(t)`$ with the classical Kepler period $`T_{\overline{ϵ}}`$. This oscillation reflects the time evolution of the electronic Rydberg wave packet which has been prepared by the fast depletion of the initial state $`|g`$. With each return to the core region this Rydberg wave packet might undergo a transition to state $`|g`$ thus increasing $`\rho _{gg}(t)`$. These coherent oscillations cannot be described by the DCA rate equations. However, due to laser fluctuations after a few Kepler periods these coherent oscillations are damped out and merge into diffusive dynamics which is characterized by power law decay of the initial state $`|g`$. From this time on the dynamics of the Rydberg system under the influence of the fluctuating laser field is well described by the DCA rate equations. This is apparent from Fig.3a by comparing the numerical solutions of the master equation (solid, dashed and long dashed curves) with the asymptotic solutions of the DCA rate equations (circles and thin dashed curves). According to the discussion presented in Sec. IV B 2 this diffusive dynamical behaviour appears when all coherent effects are damped out and disappears again at interaction times $`t>t_c`$ at which stochastic ionization starts to dominate. Physically speaking for these intermediate interaction times the excited electronic Rydberg wave packet starts to diffuse in energy space towards the ionization threshold. It reaches the ionization threshold roughly at time $`t_c`$ at which the ionization probability rises significantly from vanishingly small values to values close to unity. The early stages of this diffusion towards the ionization threshold are governed by the power law decay of Eq.(54) for $`\rho _{gg}(t)`$ which is characterized by the exponent $`(1/2)`$. In the case of laser fluctuations which can be described by the PDM to a good degree of approximation this power law decay governs the time evolution up to the stochastic ionization time $`t_c`$. However, for non-Lorentzian spectra this is no longer the case. In cases in which the non-Lorentzian effects can be described by the spectrum of Eq.(9) the characteristic exponent of this power law decay is changed to a value of $`(1/4)`$ as soon as the interaction times become larger than the characteristic time $`t_{1/4}=T_{\overline{ϵ}}\beta ^2/(2b\gamma )`$ provided $`t_{1/4}<t_c`$ (compare with Eq.(57)). This non-Lorentzian effect is clearly apparent in Fig.3a where the characteristic times $`t_{1/4}`$ are indicated for $`b/\beta =0.03`$ and $`b/\beta =0.2`$. With increasing values of $`\beta `$ this characteristic time increases and this cross over phenomenon disappears for sufficiently large values of $`\beta `$ as soon as $`t_{1/4}>t_c`$. At interaction times exceeding the stochastic ionization time $`t_c`$ the excited Rydberg wave packet has already reached the ionization threshold and the mean ionization probability rises to a value of unity. This asymptotic long time behaviour of the excitation dynamics is described to a good degree of approximation by Eqs.(37) and (38). That is apparent from a comparison of the thin dashed lines of Fig.3a with the corresponding numerical solutions of the master equation (77).
In Fig.3b the laser bandwidth is so large that the significantly excited energy region $`[\overline{ϵ},\overline{ϵ}+]`$ (before the onset of the electronic diffusion process) contains already the ionization threshold. Thus the excited Rydberg system is ionized significantly already in the early stages of the time evolution. As $`\gamma `$ this early stage of the ionization process is well described by the DCA rate equations which yield (compare to Eqs.(45) and (46))
$`\rho _{gg}(t)`$ $`=`$ $`e^{\gamma t},`$ (83)
$`P_{ion}(t)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Gamma }}{\gamma }}(1e^{\gamma t}).`$ (84)
These approximate solutions are obtained from Eqs.(19) and (28) by neglecting $`\rho _{nn}(t)`$ in comparison with $`\rho _{gg}(t)`$. According to Eqs.(45) and (46) this initial ionization process saturates as soon as the mean initial state probability and the mean ionization probability have reached the values $`\mathrm{\Lambda }_{Sp}`$ and $`\mathrm{\Gamma }/\gamma `$. At these interaction times we still have $`\gamma \mathrm{\Lambda }_{Sp}t1`$. These characteristic aspects of the laser-induced excitation process are clearly apparent in Fig.3b. Physically speaking in this initial stage of the excitation process the Rydberg electron is ionized with probability $`\mathrm{\Gamma }/\gamma `$. With a probability of $`(1\mathrm{\Gamma }/\gamma )`$ an excited electronic Rydberg wave packet is prepared after a time of the order of $`1/\gamma `$. This wave packet is formed by a coherent superposition of all Rydberg states within the dominantly excited energy interval $`[\overline{ϵ},0]`$. Depending on the actual value of the laser bandwidth the coherent dynamics of this electronic wave packet is damped sooner or later. After the destruction of all coherences to a good degree of approximation the subsequent dynamics is governed by the DCA rate equations. In Fig. 3b small coherence oscillations are visible in the time evolution of $`\rho _{gg}(t)`$. As soon as the interaction time exceeds the stochastic ionization time the excited Rydberg electron starts to ionize significantly. The time evolution of this stochastic ionization process is well described by Eqs.(37) and (38) within the framework of the DCA rate equations.
## VII Summary and conclusion
The dynamics of an electronic Rydberg wave packet under the influence of a fluctuating cw-laser field has been discussed. It has been shown that for large laser bandwidths its dynamics can be described by Pauli-type rate equations for the relevant density matrix elements of the excited Rydberg electron averaged over the laser fluctuations. These rate equations are valid for arbitrary types of laser fluctuations and their dynamics is determined by the spectrum of the laser field only and not by any of the higher order correlation functions. The validity of these rate equations has been investigated in detail for a special class of phase fluctuations of the laser field.
With the help of these rate equations we have investigated the dynamics of a laser excited Rydberg electron for long interaction times. At these interaction times the dynamics of the Rydberg electron are dominated by stochastic diffusion in energy space towards the ionization threshold which leads finally to stochastic ionization. This diffusion process is accompanied by a characteristic scenario of power law decays. Analytical expressions have been derived for these power laws and their associated characteristic exponents. These analytical expressions exhibit in a clear way that the asymptotic power laws are independent of the quantum defect of the excited Rydberg states and to which extent they depend on details of the laser spectrum. In particular, it has been demonstrated that the characteristic exponents which describe the process of stochastic ionization are completely independent of the laser spectrum. However, the initial stages of the diffusion of the excited Rydberg electron depend on details of the laser spectrum.
Support by the Deutsche Forschungsgemeinschaft within the SPP ‘Zeitabhängige Phänomene und Methoden’ is acknowledged.
## A Numerical solution technique for the Master equation
In this appendix an efficient numerical method for solving the master equation (77) is outlined.
First, we make a rearrangement of Eq.(77) splitting it into a PDM-part and an additional part that disappears in the limit $`\beta \mathrm{}`$, namely
$$\stackrel{\mathbf{}}{\rho }(t)=i\left[H_{eff}\rho (t)\rho (t)H_{eff}^{}\right]+𝐋\rho (t).$$
(A1)
Thereby we have introduced the effective non-Hermitian Hamiltonian
$$H_{eff}=H_{dr.}ib|gg|,$$
(A2)
the damping operator
$`𝐋\rho (t)`$ $`=`$ $`2b|gg|\rho (t)|gg|`$ (A3)
$`+`$ $`b{\displaystyle _0^t}𝑑\tau {\displaystyle \underset{ng}{}}\left\{|nn|\rho (t)|gg|w_n(\tau )+\text{h.c.}\right\}`$ (A4)
and a memory function
$$w_n=\underset{\mathrm{\Omega }\mathrm{}}{lim}\mathrm{\Omega }e^{\mathrm{\Omega }\tau }\alpha _n^0(\tau ).$$
(A5)
In the PDM-limit, i.e. $`\beta \mathrm{}`$, the second term on the right side of Eq.(A3) disappears and the master equation reduces to Eq.(80). Integration of Eq.(A1) yields
$`\rho (t)=U(t)\rho (0)U^{}(t)+`$ (A6)
$`{\displaystyle _0^{\mathrm{}}}𝑑t^{}\mathrm{\Theta }(tt^{})U(tt^{})𝐋\rho (t^{})U^{}(tt^{})`$ (A7)
where $`U(t)=\mathrm{exp}[iH_{eff}t]`$ is a non unitary time evolution operator. If the initial condition is taken to be $`\rho (0)=|gg|`$, the Laplace transformed matrix elements of the density operator $`\stackrel{~}{\rho }_{ij}(z)_z\rho _{ij}(t)_0^{\mathrm{}}𝑑te^{izt}\rho _{ij}(t)`$ become
$`\stackrel{~}{\rho }_{lk}(z)=H_{lggk}(z)\left[1+2b\stackrel{~}{\rho }_{gg}(z)\right]+`$ (A8)
$`b{\displaystyle \underset{ng}{}}\left\{H_{lngk}(z)\stackrel{~}{w}_n(z)\stackrel{~}{\rho }_{ng}(z)+H_{lgnk}(z)\stackrel{~}{w}_n^{}(z)\stackrel{~}{\rho }_{gn}(z)\right\},`$ (A9)
$`H_{abcd}(z)=`$ (A10)
$`{\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑z_1_{z_1+i0}a\left|U(t)\right|b\left[_{z_1z+i0}d\left|U(t)\right|c\right]^{}`$ (A11)
with $`\stackrel{~}{w}_n(z)=1\stackrel{~}{\alpha }_n^0(z)`$. The Laplace transformed transition amplitudes $`_zi\left|U(t)\right|j`$ appearing in Eq.(A10) are easily calculated .
$`_zg\left|U(t)\right|g={\displaystyle \frac{i}{z+ib\overline{ϵ}\mathrm{\Sigma }(z)}},`$ (A12)
$`_zg\left|U(t)\right|n=_zn\left|U(t)\right|g`$ (A13)
$`={\displaystyle \frac{i\epsilon _0d_{ng}}{(zϵ_n)[z+ib\overline{ϵ}\mathrm{\Sigma }(z)]}},`$ (A14)
$`_zn\left|U(t)\right|l=`$ (A15)
$`{\displaystyle \frac{1}{zϵ_n}}\left\{i\delta _{nl}+{\displaystyle \frac{i\epsilon _0^2d_{ng}d_{lg}}{(zϵ_l)[z+ib\overline{ϵ}\mathrm{\Sigma }(z)]}}\right\}`$ (A16)
where $`\mathrm{\Sigma }(z)=_{ng}\frac{|\epsilon _0d_{ng}|^2}{zϵ_n}`$ is the self energy of state $`|g`$ and the Kronecker-symbol $`\delta _{nl}`$ turns into a Dirac-delta function $`\delta (ϵ_nϵ_l)`$ for energy normalized continuum states $`|ϵ_n`$ and $`|ϵ_l`$. For the subsequent treatment it is convenient to introduce the expressions
$`\alpha _n(z)`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{\rho }_{ng}(z)}{[1+2b\stackrel{~}{\rho }_{gg}(z)]\epsilon _0d_{ng}}},`$ (A17)
$`\beta _n(z)`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{\rho }_{gn}(z)}{[1+2b\stackrel{~}{\rho }_{gg}(z)]\epsilon _0d_{ng}}},`$ (A18)
$`E_n(z)`$ $`=`$ $`{\displaystyle \frac{H_{gngg}(z)}{\epsilon _0d_{ng}}},F_n(z)={\displaystyle \frac{H_{ggng}(z)}{\epsilon _0d_{ng}}}.`$ (A19)
Thus Eq.(A8) and Laplace transformation of Eq.(A1) yield
$`\stackrel{~}{\rho }_{gg}(z)={\displaystyle \frac{𝒦(z)}{12b𝒦(z)}},`$ (A20)
$`\stackrel{~}{P}_{ion}(z)=`$ (A21)
$`{\displaystyle \frac{1}{2\pi (z+i0)(12b𝒦(z))}}{\displaystyle _0^{\mathrm{}}}𝑑ϵ\left[\alpha _ϵ(z)\beta _ϵ(z)\right]`$ (A22)
with the characteristic kernel
$$𝒦(z)=H_{gggg}(z)+b\left(\underset{ng}{}+_0^{\mathrm{}}𝑑n(ϵ)\right)|\epsilon _0d_{ng}|^2\left\{E_n(z)\stackrel{~}{w}_n(z)\alpha _n(z)+F_n(z)\stackrel{~}{w}_n^{}(z)\beta _n(z)\right\}$$
(A23)
and with
$`\alpha _l(z)`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{w}_l^{}(z)C_l(z)|\epsilon _0d_{lg}|^2\left(F_l(z)+S_l(z)\right)+\left(1J_l(z)\right)\left(E_l(z)+T_l(z)\right)}{\left(1G_l(z)\right)\left(1J_l(z)\right)(\stackrel{~}{w}_l(z)\stackrel{~}{w}_l^{}(z))^2C_l(z)^2|d_{lg}|^4}},`$ (A24)
$`\beta _l(z)`$ $`=`$ $`{\displaystyle \frac{w_l(z)C_l(z)|\epsilon _0d_{lg}|^2\left(E_l(z)+T_l(z)\right)+\left(1G_l(z)\right)\left(F_l(z)+S_l(z)\right)}{\left(1G_l(z)\right)\left(1J_l(z)\right)(\stackrel{~}{w}_l(z)\stackrel{~}{w}_l^{}(z))^2C_l(z)^2|d_{lg}|^4}}.`$ (A25)
The non-diagonal couplings $`(\alpha _l,\beta _l\alpha _n,\beta _n,nl)`$ due to the sum in Eq.(A8) give rise to the expressions $`S_l(z)`$ and $`T_l(z)`$ appearing in Eqs.(A24) and (A25), namely
$`S_l(z)`$ $`=`$ $`b\left\{{\displaystyle \underset{n\{g,l\}}{}}+{\displaystyle _0^{\mathrm{}}}𝑑n(ϵ)\right\}|\epsilon _0d_{ng}|^2\left[\beta _n(z){\displaystyle \frac{F_n(z)F_l(z)}{ϵ_lϵ_n}}\alpha _n(z){\displaystyle \frac{F_l(z)E_n(z)}{ϵ_nϵ_lz2i0}}\right],`$ (A26)
$`T_l(z)`$ $`=`$ $`b\left\{{\displaystyle \underset{n\{g,l\}}{}}+{\displaystyle _0^{\mathrm{}}}𝑑n(ϵ)\right\}|\epsilon _0d_{ng}|^2\left[\alpha _n(z){\displaystyle \frac{E_n(z)E_l(z)}{ϵ_lϵ_n}}\beta _n(z){\displaystyle \frac{E_l(z)F_n(z)}{ϵ_nϵ_l+z+2i0}}\right].`$ (A27)
The diagonal couplings of the $`\alpha _n`$ and $`\beta _n`$ yield $`J_l(z),G_l(z)`$ and $`C_l(z)`$, i.e.
$`C_l(z)`$ $`=`$ $`b\mathrm{\Theta }(ϵ_l){\displaystyle \frac{E_l(z)F_l(z)}{z+2i0}},`$ (A28)
$`G_l(z)`$ $`=`$ $`{\displaystyle \frac{b\stackrel{~}{w}_l(z)}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{dz_1\left\{1+\mathrm{\Theta }(ϵ_l)|\epsilon _0d_{lg}|^2\left[z_1+ib\overline{ϵ}\mathrm{\Sigma }(z_1+i0)\right]^1(z_1ϵ_l+i0)^1\right\}}{\left[z_1zib\overline{ϵ}\mathrm{\Sigma }(z_1zi0)\right](z_1ϵ_l+i0)}},`$ (A29)
$`J_l(z)`$ $`=`$ $`{\displaystyle \frac{b\stackrel{~}{w}_l^{}(z)}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{dz_1\left\{1+\mathrm{\Theta }(ϵ_l)|\epsilon _0d_{lg}|^2\left[z_1zib\overline{ϵ}\mathrm{\Sigma }(z_1zi0)\right]^1(z_1zϵ_li0)^1\right\}}{\left[z_1+ib\overline{ϵ}\mathrm{\Sigma }(z_1+i0)\right](z_1zϵ_li0)}}.`$ (A30)
All information about the quantum system is contained in the self energy $`\mathrm{\Sigma }(z)`$. In order to solve Eqs.(A24-A27) for a given value of $`z`$, the coefficients $`H_{gggg},E_l,F_l,J_l`$ and $`G_l`$ have to be calculated. Using the energy levels and dipole matrix elements of Eqs.(12) the self energy becomes
$`\mathrm{\Sigma }(z)`$ $`=`$ $`\delta \omega i{\displaystyle \frac{\gamma }{2}}+i\gamma {\displaystyle \frac{1}{1\mathrm{exp}(2\pi i\nu (z))]}}`$ (A31)
$`\text{with}\nu (z)`$ $`=`$ $`(2z)^{1/2}+\alpha `$ (A32)
and with the (non-resonant) quadratic Stark-shift contribution $`\delta \omega `$. In we calculated the quantity $`H_{gggg}`$ ($`f(z)`$ in that work) by contour integration. In an analogous way the quantities $`E_l,F_l,J_l`$ and $`G_l`$ can be calculated but for sake of brevity we do not give them here explicitly. Starting with $`S_n=T_n=0`$, Eqs.(A20-A27) are solved by iteration. Actually we found the non-diagonal coupling terms $`S_l,T_l`$ to be very small in comparison with the diagonal couplings so that this iteration converges very rapidly. |
warning/0003/nlin0003035.html | ar5iv | text | # Instabilities of dispersion-managed solitons in the normal dispersion regime
## 1 Introduction
New ways in optimization of existing telecommunication systems based on dispersion management technology attracted recently wide research interest from soliton-based groups (see reviews ). The main idea was to combine a high local group-velocity dispersion with a low path-average dispersion. The former feature results in the reduction of the four-wave mixing while the latter one reduces the Gordon–Haus timing jitter effects. When the path-average dispersion is small and normal, i.e. the defocussing segment in the fiber is dominant over the focussing one, a new phenomenon of branching of soliton solutions was discovered . The soliton propagation in this regime is not supported by an uniform-dispersion optical fiber and seems to be one of the remarkable achievement of the dispersion management with sufficiently high local dispersion.
The stability of branching soliton solutions in the normal dispersion regime was a subject of intense studies which lead to contradictory conclusions. Grigoryan and Menyuk announced the linear and nonlinear stability of both the branches , while Berntson et al. conjectured instability of one of the branches .
In this paper, we intend to shed light on the complicated issue of existence and stability of soliton signals in the normal regime of the dispersion map. We find, in the small-amplitude approximation, that there exist two branches of soliton solutions for different levels of energy and different pulse durations at a fixed propagation constant. The short pulses with larger energy are proved to be linearly unstable, while the other (long) pulses with smaller energy are neutrally stable. We show that the transition from large-energy unstable solitons to the stable soliton signals occurs via long-term transient oscillations. The two branches of soliton solutions correspond to a single (small-energy) branch B in Fig. 1 of . Depending on a normalization condition (see Section 4), this branch may be either stable or unstable.
Our strategy to develop the small-amplitude approximation is based on the combination of two analytical approaches: the Gaussian variational approximation and the integral evolution model.
The Gaussian variational approximation, being inaccurate in details, is still useful for a quick and rough analysis (see for review and references). Also it was shown that the method can be extended to a rigorous Gauss-Hermite expansion of the basic model . We improve the previous results summarized in by deriving a new dynamical system from the variational equations of a Gaussian pulse. The system clearly displays the linear and nonlinear instability of the short Gaussian pulse with larger energy.
More rigorous analysis of the problem is based on the integral evolution model obtained by Gabitov and Turitsyn and by Ablowitz and Biondini . Although this model is more complicated from computational point of view (see recent papers ), we managed to study numerically the construction of the linear spectrum of dispersion-managed solitons. Our results confirm the instability and transition scenarios predicted within the variational model. We also deduce from this model that the soliton signals in the normal regime of the dispersion map are in resonance with the wave continuum of linear excitations of the map. The resonance implies isually the generation of wave packets from stable pulses oscillating in time. The latter effects are beyond the accuracy of the analytical model and are left for further studies.
## 2 Gaussian approximation: New dynamical model
We study the NLS model in the dimensionless form ,
$$iu_z+\frac{1}{2}D(z)u_{tt}+ϵ\left(\frac{1}{2}D_0u_{tt}+|u|^2u\right)=0,$$
(1)
where $`u(z,t)`$ is the envelope of an optical pulse in the retarded reference frame of the fiber. The small parameter $`ϵ`$ measures the length of the dispersion’s map and the inverse variance of the local dispersion. After normalization, $`D_0`$ and $`D(z)`$ are assumed to be of order of $`\mathrm{O}(1)`$, and
$$D=_0^1D(z)𝑑z=0,D(z+1)=D(z).$$
(2)
Further physical motivations for derivation and verification of the model (1) can be found in . Soliton-like optical pulses are solutions of the model in the form,
$$u(z,t)=\psi (z,t)e^{iϵ\mu z},$$
(3)
where $`\mu `$ is the propagation constant and $`\psi (z,t)`$ is a soliton profile satisfying the boundary conditions,
$$\psi (z+1,t)=\psi (z,t)$$
(4)
and
$$\underset{t\mathrm{}}{lim}\psi (z,t)=0.$$
(5)
One of the conventional approximation for soliton solutions of NLS-type equations is based on averaging the Gaussian anzatz in the Lagrangian density and further varying the Lagrangian density with respect to parameters of the Gaussian pulse (see for review). The Gaussian approximation is the first term of the Gauss-Hermite expansions when solving the NLS equation (1) in the limit $`ϵ0`$ . We apply the Gaussian anzatz in the form,
$$u(z,t)=c(z)\mathrm{exp}\left(\frac{(\alpha (z)2i\beta (z))}{\alpha (z)^2+4\beta (z)^2}t^2+i\varphi (z)\right).$$
(6)
Here the four parameters of the Gaussian pulse are: $`c(z)`$ \- the amplitude, $`\varphi (z)`$ \- the gauge parameter, $`\alpha (z)`$ \- the pulse duration, and $`\beta (z)`$ \- the chirp. It was found that the four equations for variations of the Lagrangian density can be decoupled into a system for $`\alpha (z)`$ and $`\beta (z)`$ of the form,
$`{\displaystyle \frac{d\alpha }{dz}}`$ $`=`$ $`{\displaystyle \frac{4ϵE\alpha ^{5/2}\beta }{(\alpha ^2+4\beta ^2)^{3/2}}},`$ (7)
$`{\displaystyle \frac{d\beta }{dz}}`$ $`=`$ $`D(z)+ϵ\left(D_0{\displaystyle \frac{E\alpha ^{3/2}(\alpha ^24\beta ^2)}{2(\alpha ^2+4\beta ^2)^{3/2}}}\right).`$ (8)
The phase factor $`\varphi (z)`$ is expressed in terms of $`\alpha (z)`$ and $`\beta (z)`$,
$$\frac{d}{dz}\left(\varphi +\frac{1}{2}\mathrm{arctan}\frac{2\beta }{\alpha }\right)=\frac{ϵE\alpha ^{1/2}(3\alpha ^2+20\beta ^2)}{4(\alpha ^2+4\beta ^2)^{3/2}},$$
(9)
while the amplitude $`c(z)`$ is given in terms of the input energy constant $`E`$ as
$$E=\frac{\sqrt{\alpha ^2+4\beta ^2}}{\sqrt{2\alpha }}c^2=\frac{1}{\sqrt{\pi }}_0^1𝑑z_{\mathrm{}}^{\mathrm{}}𝑑t|u|^2(z,t)>0.$$
(10)
The stationary pulse (3) - (5) corresponds to the periodic solutions of the system (7) and (8) in the form,
$$\alpha (z+1)=\alpha (z),\beta (z+1)=\beta (z),\varphi (z+1)=\varphi (z)+ϵ\mu .$$
(11)
For simplicity, we study the periodic solutions in the limit $`ϵ0`$ by using a two-step dispersion map with zero average,
$$D(z)=\{\begin{array}{c}D_1,\mathrm{\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}0}<z<L\\ D_2,L<z<1\end{array},$$
(12)
where
$$m=D_1L=D_2(1L)>0.$$
The asymptotic solution in the limit $`ϵ0`$ can be sought in the regular form,
$$\alpha (z)=\alpha _s+\mathrm{O}(ϵ),\beta (z)=_0^zD(z^{})𝑑z^{}+\beta _s+\mathrm{O}(ϵ),$$
where $`\alpha _s`$, $`\beta _s`$ are constant. The periodic solutions appear when $`\beta _s=m/2`$ and $`\alpha _s`$ is a root of the equation,
$$D_0=E\alpha _s^{3/2}\left[\frac{1}{(m^2+\alpha _s^2)^{1/2}}\frac{1}{2m}\mathrm{log}\left(\frac{m+(m^2+\alpha _s^2)^{1/2}}{\alpha _s}\right)\right].$$
(13)
In addition, the propagation constant $`\mu `$ can be obtained as a function of $`E`$ and $`\alpha _s`$ according to the equation,
$$\mu =\frac{1}{4}E\alpha _s^{1/2}\left[\frac{2}{(m^2+\alpha _s^2)^{1/2}}+\frac{5}{m}\mathrm{log}\left(\frac{m+(m^2+\alpha _s^2)^{1/2}}{\alpha _s}\right)\right].$$
(14)
These equations have been already derived in the literature, see for (13) and for (14). However, the relations (13) and (14) were viewed typically under the normalization condition,
$$m=1,E=\frac{1}{\sqrt{2S}},\alpha _s=\frac{1}{S},$$
(15)
where $`S`$ is called the map strength. In this normalization, the expression (13) gives a small-amplitude limit of the results of , i.e. the slope $`E/D_0`$ is a function of $`S`$. The existence of solitons was identified both for $`D_0>0`$ (when $`S<S_{thr}`$) and for $`D_0<0`$ (when $`S<S_{thr}`$), where $`S_{thr}3.32`$.
In this paper, we develop a different frame to view the soliton solutions (13) and (14). Guided by the stability analysis of solitons in generalized NLS equations , we fix the parameters of the model ($`D_0,m`$) and construct periodic solutions as a one-parameter family in terms of the propagation constant $`\mu `$. As a result, the parameters $`\alpha `$ and $`E`$ can be found from (13)-(14) as $`\alpha _s=\alpha _s(\mu )`$ and $`E=E(\mu )`$. These functions are shown in Fig. 1(a,b) for $`D_0=0.02`$ and $`m=2`$ and in Fig. 2(a,b) for $`D_0=0.02`$ and $`m=2`$. Obviously, the branching occurs at $`D_0<0`$ (i.e. at the normal regime of the dispersion map), when the dispersion map is defocussing on average. The two solutions coexist for $`\mu >\mu _{thr}(D_0,m)`$ and $`E>E_{thr}(D_0,m)`$. Both the branches I and II correspond to a single branch B in the small-amplitude approximation under the normalization condition (15) .
In order to describe non-stationary dynamics of the Gaussian pulse near the periodic solutions, we derive a dynamical model from Eqs. (7) and (8) by setting,
$$\alpha (z)=\alpha _0(\zeta )+ϵ\alpha _1(z,\zeta )+\mathrm{O}(ϵ^2)$$
and
$$\beta (z)=_0^zD(z^{})𝑑z^{}+\beta _0(\zeta )+ϵ\beta _1(z,\zeta )+\mathrm{O}(ϵ^2).$$
Here $`\zeta =ϵz`$ is the distance to measure the evolution of a Gaussian pulse over many map’s periods. The coupled system (7) and (8) can be averaged over the map’s period subject to the periodic conditions: $`\alpha _1(z+1,\zeta )=\alpha _1(z,\zeta )`$ and $`\beta _1(z+1,\zeta )=\beta _1(z,\zeta )`$. Then, the non-stationary system reduces to the dynamical model for $`\alpha _0(\zeta )`$ and $`\beta (\zeta )`$,
$`{\displaystyle \frac{d\alpha _0}{d\zeta }}=F_\alpha (\alpha _0,\beta _0)`$ $``$ $`{\displaystyle \frac{E\alpha _0^{5/2}}{m}}\left[{\displaystyle \frac{1}{(\alpha _0^2+4\beta _0^2)^{1/2}}}{\displaystyle \frac{1}{(\alpha _0^2+4(\beta _0+m)^2)^{1/2}}}\right],`$ (16)
$`{\displaystyle \frac{d\beta _0}{d\zeta }}=F_\beta (\alpha _0,\beta _0)`$ $``$ $`D_0{\displaystyle \frac{E\alpha _0^{3/2}}{4m}}[{\displaystyle \frac{4(m+\beta _0)}{(\alpha _0^2+4(m+\beta _0)^2)^{1/2}}}{\displaystyle \frac{4\beta _0}{(\alpha _0^2+4\beta _0^2)^{1/2}}}`$ (17)
$`+`$ $`\mathrm{log}\left({\displaystyle \frac{2\beta _0+(\alpha _0+4\beta _0^2)^{1/2}}{2(m+\beta _0)+(\alpha _0^2+4(m+\beta _0)^2)^{1/2}}}\right)].`$
This system has of course the same stationary solutions $`\alpha _0=\alpha _s`$ and $`\beta _0=\beta _s=m/2`$ as those given in (13). The stationary solutions appear as equilibrium states in the dynamical system, whose stability can be found by linearizing,
$$\alpha _0(\zeta )=\alpha _s+\mathrm{\Delta }\alpha e^{i\lambda \zeta },$$
$$\beta _0(\zeta )=\beta _s+\mathrm{\Delta }\beta e^{i\lambda \zeta },$$
where the eigenvalue $`\lambda `$ is
$$\lambda ^2(\mu )=\frac{F_\alpha }{\beta }(\alpha _s)\frac{F_\beta }{\alpha }(\alpha _s)=\frac{2E\alpha _s^{3/2}}{(m^2+\alpha _s^2)^{3/2}}\left[3D_0+\frac{E\alpha _s^{3/2}(m^2\alpha _s^2)}{(m^2+\alpha _s^2)^{3/2}}\right].$$
(18)
We plot $`\lambda ^2(\mu )`$ in Fig.3 to confirm that $`\lambda ^2>0`$ for branch I of the periodic solutions and $`\lambda ^2<0`$ for branch II (cf. Fig. 2). Thus, the linear analysis predicts the instability of the short Gaussian pulses with larger energy at a fixed propagation constant $`\mu `$ (branch II). In the limit $`\mu \mu _{thr}(D_0,m)`$, the instability disappears, i.e.
$$\underset{\mu \mu _{thr}}{lim}\lambda ^2(\mu )=0.$$
This property follows from Eq. (17) in the limit $`\mu \mu _{thr}(D_0,m)`$, when
$$\frac{F_\beta }{E}(\alpha _s,\beta _s)\frac{dE}{d\mu }+\frac{F_\beta }{\alpha }(\alpha _s,\beta _s)\frac{d\alpha }{d\mu }=0.$$
(19)
Connecting Eqs. (18) and (19), we find the following asymptotic approximation,
$$\lambda ^2\left(\frac{F_\beta /EF_\alpha /\beta }{d\alpha /d\mu }\right)\frac{dE}{d\mu }.$$
(20)
Taking into account that $`dE/d\mu `$, $`F_\beta /E`$, and $`F_\alpha /\beta `$ are all positive for $`D_0<0`$, and $`\alpha (\mu \mu _{thr})^{1/2}`$ (see Fig. 2(a,b)), the asymptotic approximation (20) produces the result, $`\lambda (\mu \mu _{thr})^{1/4}`$.
The nonlinear dynamics of the system (16) and (17) is shown in Fig. 4 for $`D_0=0.02`$ and $`m=2`$. At a fixed value of the energy $`E`$, there are two stationary Gaussian pulses of different durations: a short pulse is a saddle point, while a long one is a center. Inside the separatrix loop, there are oscillations of the pulse trapped by the center point. Outside the separatrix, the Gaussian pulse transfers to chirped linear waves.
We notice that the transition scenario resembles the nonlinear dynamics of unstable solitons in generalized NLS equations . The only difference is that the unstable branch in generalized NLS equations is located for those values of soliton propagation constant $`\mu `$, where $`dE/d\mu <0`$. Although this conventional stability criterion failed for the dispersion-managed solitons (see Fig. 2(a)), the instability development shows up to be alike (cf. Fig. 4 here and Fig. 2(b) in ).
## 3 Integral evolution model: Numerical analysis
The Gaussian approximation of the optical pulse in the NLS model (1) can be improved by summating all higher-order Gauss-Hermite solutions of the linear equation, $`iu_z+0.5D(z)u_{tt}=0`$ as shown in . However, this analysis results in a complicated infinite-dimensional system of algebraic equations for parameters and coefficients of the Gauss-Hermite expansion. Instead, we adopt a direct asymptotic method to deduce an integral evolution model valid in the limit $`ϵ0`$. This method is based on Fourier expansion of solutions of the linear equation above as well as on the asymptotic expansion,
$$u(z,t)=u_0(z,t)+ϵu_1(z,t)+\mathrm{O}(ϵ^2),$$
where $`u_0(z,t)`$ is given in the Fourier form as
$$u_0=\frac{1}{2\pi }_{\mathrm{}}^{\mathrm{}}𝑑\omega W(\omega ,\zeta )\mathrm{exp}\left(\frac{i}{2}\omega ^2\left(_0^zD(z^{})𝑑z^{}\right)+i\omega t\right).$$
(21)
Here $`W(\omega ,\zeta )`$ is a complex Fourier coefficient and $`\zeta =ϵz`$ is the distance to measure the pulse evolution over many map’s periods. By supplying the periodic condition $`u_1(z+1,t)=u_1(z,t)`$ in the Fourier form, the NLS equation (1) can be reduced to the integral evolution model,
$$iW_\zeta \frac{1}{2}D_0\omega ^2W+_{\mathrm{}}^{\mathrm{}}𝑑\omega _1𝑑\omega _2r(\omega _1\omega _2)W(\omega +\omega _1)W(\omega +\omega _2)\overline{W}(\omega +\omega _1+\omega _2)=0,$$
(22)
where
$$r(\omega _1\omega _2)=\frac{1}{4\pi ^2}_0^1𝑑z\mathrm{exp}\left(i\omega _1\omega _2_0^zD(z^{})𝑑z^{}\right).$$
For the two-step dispersion map (12), the integral kernel $`r(x)`$ can be computed explicitly as
$$r(x)=\frac{1}{4\pi ^2}\frac{\mathrm{sin}\left(\frac{mx}{2}\right)}{\frac{mx}{2}}.$$
(23)
It is obvious that the dynamical system (16) and (17) studied in the previous section can be deduced from (22) within the same Gaussian approximation. This correspondence implies that the qualitative results on instability of short Gaussian pulses for $`D_0<0`$ can be reconfirmed within a more systematic theory.
In this section we present numerical results consisting of three subsections. In the first subsection, we construct a numerical solution of the stationary problem identifying optical solitons in the normal regime, when $`D_0<0`$. In the second part, we analyze the linearized problem and locate numerically the linear spectrum in the problem, indicating possible instability of optical solitons. Then, we simulate numerically the non-stationary problem described by (22) and discuss the transformation routes of the unstable dispersion-managed solitons.
### 3.1 Stationary solutions
The periodic-type localized solutions of the NLS equation in the form (3) are equivalent to stationary solutions of (22) in the form,
$$W(\omega ,\zeta )=\mathrm{\Phi }(\omega )e^{i\mu \zeta },$$
(24)
where $`\mathrm{\Phi }(\omega )`$ is the real Fourier coefficient which defines $`\psi (z,t)`$ according to (21). This function satisfies a nonlinear integral boundary-value problem,
$$\left(\mu +\frac{1}{2}D_0\omega ^2\right)\mathrm{\Phi }(\omega )=R(\omega )_{\mathrm{}}^{\mathrm{}}𝑑\omega _1𝑑\omega _2r(\omega _1\omega _2)\mathrm{\Phi }(\omega +\omega _1)\mathrm{\Phi }(\omega +\omega _2)\mathrm{\Phi }(\omega +\omega _1+\omega _2),$$
(25)
where $`\mathrm{\Phi }(\omega )=\mathrm{\Phi }(\omega )`$ (the symmetry condition) and $`lim_\omega \mathrm{}\mathrm{\Phi }(\omega )=0`$ (the boundary condition).
For numerical analysis, we intend to use the Petviashvili’s iteration scheme :
$$\mathrm{\Phi }^{(n)}(\omega )\mathrm{\Phi }^{(n+1)}(\omega )$$
for $`n=0,1,2,\mathrm{}`$. Within this scheme, the right-hand-side $`R(\omega )`$ can be approximated at the $`n`$-th approximation by $`\mathrm{\Phi }^{(n)}(\omega )`$ provided a certain stabilizing factor is introduced for convergence (see (27) and (28) below). However, the numerical scheme breaks down for $`D_0<0`$ due to resonances at $`\omega =\pm \omega _{res}`$, where
$$\omega _{res}=\sqrt{\frac{2\mu }{|D_0|}}.$$
(26)
Indeed for $`\omega =\pm \omega _{res}`$, the left-hand-side of (25) vanishes. \[Here we notice that $`\mu >0`$ for the Gaussian pulse solutions (6) of the NLS model (1).\] In order to avoid resonances in the numerical scheme, we add and subtract a dummy positive dispersion term $`0.5|D_0|\omega ^2\mathrm{\Phi }(\omega )`$ to the left-hand-side of (25). As a result, the scheme converts to the following map,
$$\mathrm{\Phi }^{(n)}(\omega )\mathrm{\Phi }^{(n+1)}(\omega )=S_n^{3/2}\left(\frac{R^{(n)}(\omega )+\frac{1}{2}(|D_0|D_0)\omega ^2\mathrm{\Phi }^{(n)}(\omega )}{\mu +\frac{1}{2}|D_0|\omega ^2}\right),D_0<0,$$
(27)
where $`S_n`$ is the Petviashvili’s stabilizing factor given by
$$S_n=\frac{_{\mathrm{}}^{\mathrm{}}𝑑\omega (\mu +\frac{1}{2}|D_0|\omega ^2)\mathrm{\Phi }^{(n)}(\omega )}{_{\mathrm{}}^{\mathrm{}}𝑑\omega \mathrm{\Phi }^{(n)}(\omega )\left(R^{(n)}(\omega )+\frac{1}{2}(|D_0|D_0)\omega ^2\mathrm{\Phi }^{(n)}(\omega )\right)}.$$
(28)
The factor $`S_n`$ is unity at the stationary solution and serves therefore as an indicator for termination of the iterating procedure. We stop iterations when $`|S_n1|<10^5`$.
To use the map (27), we apply the Simpson’s integration method, reducing complexity due to the symmetry: $`\mathrm{\Phi }(\omega )=\mathrm{\Phi }(\omega )`$. As a starting solution, the profile $`\mathrm{\Phi }(\omega )`$ can be approximated by the Gaussian pulse with parameters corresponding to the periodic solution (13) and (14),
$$\mathrm{\Phi }^{(0)}(\omega )=\sqrt{\pi E\sqrt{2\alpha _s}}\mathrm{exp}\left(\frac{1}{4}\alpha _s\omega ^2\right).$$
(29)
| Number of iterations | $`S_n`$: $`D_0=0.02`$ | $`S_n`$: $`D_0=0.02`$ (I) | $`S_n`$: $`D_0=0.02`$ (II) |
| --- | --- | --- | --- |
| 1 | 0.9897 | 0.9957 | 0.9920 |
| 2 | 0.9971 | 0.9981 | 0.9917 |
| 3 | 0.9994 | 0.9988 | 0.9921 |
| 4 | 0.9998 | 0.9991 | 0.9942 |
| 5 | 0.9999 | 0.9993 | 0.9989 |
| 6 | | 0.9994 | 1.0069 |
| 7 | | 0.9995 | 1.0167 |
| 8 | | 0.9996 | 1.0173 |
| 9 | | 0.9997 | 0.9642 |
| 10 | | 0.9997 | 0.8133 |
| 11 | | 0.9998 | 0.7025 |
| 12 | | 0.9998 | 0.6833 |
| 13 | | 0.9999 | 0.6773 |
| 14 | | 0.9999 | 0.6695 |
| 15 | | 0.9999 | 0.6601 |
| 16 | | | 0.6499 |
| 17 | | | 0.6394 |
| 18 | | | 0.6292 |
| 19 | | | 0.6192 |
Table I. Iterations of the stabilizing factor $`S_n`$ for $`D_0=0.02`$ and $`D_0=0.02`$ (branches I and II).
Table 1 shows iterations for the stabilizing factor $`S_n`$ in the three different cases: (i) $`D_0=0.02`$, (ii) $`D_0=0.02`$ (branch I), and (iii) $`D_0=0.02`$ (branch II). For all the cases, the other parameters are $`\mu =1`$ and $`m=2`$. In the first case, the convergence is monotonic and the profile for stationary soliton $`\mathrm{\Phi }(\omega )`$ is shown in Fig. 5(a). The numerical value for energy of the stationary soliton is shown in Fig. 1(a) by a bullet. In the second case, the iterations converge slowly to the stationary soliton shown in Fig. 5(b). Sometimes, the convergence is accompanied by a single oscillation of $`S_n`$ near unity. The numerical value for the energy is shown in Fig. 2(a) by a bullet at branch I. In the last case, however, the iterations oscillate and finally diverge. Inspection of the profile $`\mathrm{\Phi }^{(n)}(\omega )`$ at a final iteration shows that the iterations change the initial pulse drastically leading to its disappearance. Two conjectures follow from this fact. Either the shorter soliton with larger energy at branch II does not exist as a stationary solution of (25) or it is unstable within the iterational scheme (27). Since the short Gaussian pulse does exist (see Fig. 2), we incline to work along the second conjecture. The iterational scheme (27) is not relevant for the time-evolution problem and rigorous analysis of linearized problem is needed to confirm predictions of the instability of the short stationary pulse.
### 3.2 Linear spectrum
There are several forms of the linear problem associated to the NLS-type equations. We will use the matrix form which was studied in our previous paper subject to certain simplifications. The matrix form appears upon perturbations of the stationary solutions of (22) as
$$W(\omega ,\zeta )=e^{i\mu \zeta }\left[\mathrm{\Phi }(\omega )+w_1(\omega )e^{i\lambda \zeta }+\overline{w}_2(\omega )e^{i\lambda \zeta }\right].$$
(30)
The vector $`𝐰(\omega )=(w_1,w_2)^T`$ can be shown to satisfy the matrix linear problem,
$$\lambda 𝐰(\omega )=\left(\mu +\frac{1}{2}D_0\omega ^2\right)\sigma _3𝐰(\omega )+_{\mathrm{}}^{\mathrm{}}𝑑\omega _1\left(2K_1(\omega ,\omega _1)\sigma _3+K_2(\omega ,\omega _1)\sigma _2\right)𝐰(\omega _1),$$
(31)
where
$$\sigma _3=\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right],\sigma _2=\left[\begin{array}{cc}0& 1\\ 1& 0\end{array}\right],$$
and the integral kernels are
$`K_1(\omega ,\omega _1)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\omega _2r[(\omega \omega _1)(\omega \omega _2)]\mathrm{\Phi }(\omega _2)\mathrm{\Phi }(\omega _1+\omega _2\omega ),`$
$`K_2(\omega ,\omega _1)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\omega _2r[(\omega _2\omega _1)(\omega _2\omega )]\mathrm{\Phi }(\omega _2)\mathrm{\Phi }(\omega _1\omega _2+\omega ).`$
Using the stationary solution $`\mathrm{\Phi }(\omega )`$ from the previous subsection and implementing the Simpson’s integration method again, we solve the linear problem by using the linear algebra packages built in Matlab 5.2. We identify two types of modes of the linear spectrum: symmetric eigenfunctions, when $`𝐰(\omega )=𝐰(\omega )`$, and anti-symmetric eigenfunctions, when $`𝐰(\omega )=𝐰(\omega )`$.
The linear spectrum for the soliton of Fig. 5(a) is shown in Fig. 6(a,b). It consists of three main parts: (i) continuous spectrum, (ii) neutral (zero) modes, and (iii) internal (oscillatory) modes.
When $`D_0>0`$ (the anomalous regime of the dispersion map), the continuous spectrum is located at the real axis for $`|\lambda |>\mu `$ (see Fig. 6(a,b) where $`\mu =1`$). Indeed, the continuous modes are singular in the Fourier representation, i.e. $`𝐰(\omega )\delta (\omega \mathrm{\Omega })`$. Then, the linear problem (31) has the continuous spectrum at $`\lambda =\pm \lambda _\mathrm{\Omega }`$, where
$$\lambda _\mathrm{\Omega }=\mu +\frac{1}{2}D_0\mathrm{\Omega }^2,$$
(32)
provided the following integral kernels are not singular,
$`\underset{\omega \mathrm{\Omega }}{lim}K_1(\omega ,\mathrm{\Omega }\omega )`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\omega \mathrm{\Phi }^2(\omega ),`$
$`\underset{\omega \mathrm{\Omega }}{lim}K_2(\omega ,\mathrm{\Omega }\omega )`$ $`=`$ $`{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑\omega r(\omega ^2)\mathrm{\Phi }(\omega +\mathrm{\Omega })\mathrm{\Phi }(\omega \mathrm{\Omega }).`$
The neutral (zero) modes always appear at $`\lambda =0`$ as double degenerate states for both symmetric and asymmetric eigenfunctions. However, the inaccuracy of the numerical method destroys the degeneracy of the zero modes. As a result, the two zero modes may appear either for small real or for small imaginary values of $`\lambda `$. Fig. 7(b) displays two imaginary eigenvalues of order of $`\mathrm{O}(10^2)`$ which appear to be shifted from the origin of $`\lambda `$ due to this numerical effect. Since the zero modes are not of interest from stability point of view, we neglect this effect and leave the scheme without any additional modification.
The internal (oscillatory) modes are located in the gap of the continuous spectrum as $`0<|\lambda |<\mu `$. The set of internal modes may contain different number of eigenvalues. We have shown in previous paper that the set is empty in the NLS limit (which corresponds to the limit $`\mu 0`$ at fixed $`D_0>0`$ and $`m`$). \[Note that in we used the Gaussian pulse (29) for approximating $`\mathrm{\Phi }(\omega )`$ while here we substitute the numerical result from Eq. (25).\] Then, we showed that the number of internal modes increases with the map’s strength (if the parameter $`\alpha _s`$ is set to unity, the map’s strength is proportional to $`m`$ and vice versa). In Fig. 6(a,b) for $`\mu =1`$, we identify 14 internal modes for symmetric eigenfunctions and $`12`$ internal modes for anti-symmetric eigenfunctions. Still complex eigenvalues are absent for $`D_0>0`$ which confirms stability of dispersion-managed solitons in the anomalous regime.
The linear spectrum for the soliton of Fig. 5(b) is shown in Fig. 7(a,b). The continuous spectrum is seen to have changed drastically. When $`D_0<0`$ (normal regime of the dispersion map), the continuous spectrum covers the segment $`|\lambda |<\mu `$ twice according to (32). As a result, neutral and internal modes, if any, become embedded in the wave continuum as seen in Fig. 7(a,b). This indicates a resonance of stationary soliton with the linear spectrum in the normal regime of the dispersion map. However, this resonance does not result to any instability of solitons of branch I within the linear theory. We discuss the resonance issue in Section 4. The two imaginary eigenvalues on Fig. 7(b) appear from the origin as artifacts of the numerical scheme as it was explained above.
At last, we would like to construct the linear spectrum for the solitons of branch II. However, the stationary solutions were not identified within the Petviashvili’s numerical method. Therefore, assuming that the solutions still exist, the profile $`\mathrm{\Phi }(k)`$ can only be approximated by the Gaussian pulse (29) as we did in . The linear spectrum in this approximation is shown in Fig. 8(a,b) for $`D_0=0.02`$, $`m=2`$, and $`\mu =1`$. The same type of the continuous spectrum is clearly seen not to possess any gap in the origin. In addition to this, we identify new complex eigenvalues both for symmetric and anti-symmetric eigenfunctions. These complex eigenvalues has relatively large, order of $`\mathrm{O}(10^1)`$, imaginary part and they are associated with the instability of the solitons of branch II. The numerical result for the instability eigenvalues is in a reasonable comparison with the asymptotic predictions which follow from the dynamical model (16) and (17) (see Fig. 3, branch II). The eigenvectors $`𝐰(\omega )`$ for the unstable (imaginary) eigenvalues are shown in Fig. 8(c,d) for the symmetric and anti-symmetric eigenfunctions, respectively. The numerical approximations of the eigenvectors, being inaccurate in details, display clearly that the unstable modes are localized at the intermediate wave frequencies $`\omega `$ of the pulse spectrum (at $`\omega 4.5`$). Thus, the development of the unstable eigenvectors would affect the duration of the soliton pulse in the nonlinear stage as described in the next subsection.
### 3.3 Non-stationary evolution
To confirm the transition scenario, we simulate the non-stationary dynamics of unstable solitons in the integral model (22) by using the central-difference scheme,
$$\frac{V^{(n+1)}(\omega )V^{(n1)}(\omega )}{2\mathrm{\Delta }\zeta }=i_{\mathrm{}}^{\mathrm{}}𝑑\omega _1𝑑\omega _2r(\omega _1\omega _2)e^{iD_0\omega _1\omega _2\zeta _n}V^{(n)}(\omega +\omega _1)V^{(n)}(\omega +\omega _2)\overline{V}^{(n)}(\omega +\omega _1+\omega _2)$$
(33)
where
$$V^{(n)}(\omega )=W(\omega ,\zeta _n)\mathrm{exp}\left(\frac{i}{2}D_0\omega ^2\zeta _n\right)$$
and $`\zeta _n=n\mathrm{\Delta }\zeta `$. An initial iteration can be done within a forward scheme starting with the initial Gaussian pulse (29) with the parameter $`\alpha _s`$ and $`E`$ corresponding to branches I and II at $`\mu =1`$ on Fig. 2(a,b). Evolution of a stable long pulse (branch I) is shown in Fig. 9(a,b), while that of an unstable short pulse (branch II) is shown in Fig. 9(c,d). The stable long Gaussian pulse quickly transits to the stationary pulse given by Eq. (24) (Fig. 5(b)) which propagates later without visible distortions. The second and third peeks in the signal spectrum and profile (Fig. 9(a,b)) appear in complete agreement with the profile of the stationary DM soliton (cf. Fig. 1 from ). Some oscillations along the distance $`\zeta `$ are excited due to the difference between the Gaussian pulse (29) and the exact stationary solution of Eq. (25). These oscillations are small compared to the soliton profile and they do not change the duration of the soliton pulse (see Fig. 9(a)). The nonlinear resonance at $`\omega _{res}`$ ($`\omega _{res}=10`$ for Fig. 9(a)) is not seen to be excited during the signal propagation.
Evolution of the short (unstable) pulse differs drastically from the previous picture. The pulse is being broaden during the evolution, it generates the strong radiation tail and tends to the long (stable) signal which has the first node at $`\omega 4.5`$ (cf. Fig. 9(a) and Fig. 9(c)). This transformation is accompanied by the intermediate oscillations around the soliton’s shape. Thus, we confirm the analytical predictions that the short pulses are linearly unstable and switch into long stable solitons via long-term oscillatory dynamics. The unstable eigenvectors at Fig. 8(c,d) clearly match at Fig. 9(c) with the growing deformations of the localization of the pulse spectrum.
## 4 Discussion: Resonance of dispersion-managed solitons
We have shown that both the Gaussian variational approximation and the integral model prescribe the instability of short nonlinear signals in the normal regime of the dispersion map. This instability broadens the signal’s profile through some intermediate oscillations. The long signals propagate stably then.
We point out that the small-amplitude approximation considered here corresponds to the asymptotic limit $`E/D_00`$ (see branch B at Fig. 1 in ). The stability of the small-amplitude branch for $`D_0<0`$ was previously under discussion in the literature . This discussion can be resolved if one takes into account the normalization condition (15), which was used in the previous works. We have checked that the normalization condition (15) selects the pulse solution along the stable branch I for $`S_{thr}<S<S_{stab}`$, where $`S_{stab}6.76`$. For $`S>S_{stab}`$, it selects solutions along the unstable branch II (see Figs. 2(a,b)). Thus, for the intermediate map strengths (when $`S<S_{stab}`$), the soliton signal propagates stably in the limit of small energies, as reported in . However, if the map strength $`S`$ exceeds the value $`S_{stab}`$, the soliton signal breaks down and switches to a longer signal, as conjectured in .
For both the branches, we observe the resonance appearing between the linear wave spectrum and the stationary signals. This resonance is related to the fact that the origin $`\lambda =0`$ is absorbed in the continuous spectrum of the linear problem. Another way to find the resonance is to construct the linear spectrum for the integral model (22), $`W(\omega ,\zeta )W_0e^{i\mathrm{\Omega }(\omega )\zeta }`$, where
$$\mathrm{\Omega }=\frac{1}{2}D_0\omega ^20$$
and $`D_0<0`$. Therefore, for any $`\mu >0`$ there exists $`\omega _{res}`$ such that $`\mathrm{\Omega }(\omega _{res})=\mu `$. This resonance does not show up in the linear theory since the discrete and continuous spectra are separated. However, in the nonlinear stage, the resonance generally leads to emission of wave packets and soliton’s decay.
Since the transformation or decay of long stable solitons for $`D_0<0`$ have never been observed numerically (nor in our simulations reported in Figs. 9), it is likely that the effective gap in the spectrum still appears in the nonlinear theory. Also the truncated approximation given by the integral model (22) may not be valid for a correct description of the resonance. The latter issues remain open for further analytical consideration.
## Aknowledgements.
The author thanks S. Turitsyn and G. Biondini for bringing the present problem to his attention as well as T. Lakoba and C. Sulem for collaboration and valuable remarks. Fig. 9 was prepared by using computational capacities of Institute of Optics at Rochester, NY. |
warning/0003/physics0003083.html | ar5iv | text | # References
Plasma comprises over 99.9 per cent of known matter in the Universe. However, among the different states of matter its physical properties are the least understood. This is largely due to a highly complex and nonlinear behaviour, which makes theoretical investigations quite difficult. Particularly notorious are the instabilities that hamper plasma confinement in thermonuclear fusion energy experiments .
In the present Letter we consider the electromagnetic interactions within a charge neutral plasma, with an equal number of negative and positive charge carriers. We propose a first principles field theory model to describe the fluid dynamical properties of this plasma, and find results that challenge certain widely held views on plasma behaviour. In particular, we argue that stable self-confining plasma filaments can exist, and are described by topologically nontrivial knotted solitons.
In magnetohydrodynamics the geometrical properties of an electrically neutral plasma are conventionally described using a single-fluid approximation. The individual charged particles contribution is described collectively by the hydrostatic pressure $`p`$, which according to standard kinetic theory relates to the kinetic energies of the individual particles $`pmv^2`$. The equation of motion then follows from the properties of the pertinent energy-momentum tensor $`T_{\mu \nu }`$, the spatial part of its divergence coincides with the external dissipative force which leads to the Navier-Stokes equation
$$\rho \frac{d\stackrel{}{𝐔}}{dt}=p+(\times \stackrel{}{𝐁})\times \stackrel{}{𝐁}+\eta ^2\stackrel{}{𝐔}^2$$
(1)
Here $`\stackrel{}{𝐔}`$ is the bulk (center of mass) velocity of the plasma, and $`\eta `$ is the coefficient of viscosity. The plasma evolves according to (1), dissipating its kinetic energy by the viscous force. This force is present whenever the plasma is in motion but ceases when the plasma reaches a magnetostatic equilibrium configuration. In that limit the Navier-Stokes equation reduces to a balance relation between the gradient of the hydrostatic pressure and the magnetic force,
$$p=(\times \stackrel{}{𝐁})\times \stackrel{}{𝐁}$$
Ideally, one might expect that under proper conditions a plasma in isolation becomes self-confined due to the currents that flow entirely within the plasma itself. But this appears to be excluded by a simple virial theorem which suggests that any static plasma configuration in isolation is dissipative. As a consequence of such apparently inborn instabilities, strong external currents are then commonly introduced to confine a plasma in laboratory experiments.
We now argue that there are important non-linear effects which are not accounted for by a structureless mean field variable such as the pressure $`p`$. These nonlinearities have their origin in the electromagnetic interactions between the charged particles within the plasma. They remain hidden when the energy-momentum tensor relates to the kinetic energies of the individual particles, but become visible once we recall the familiar but nontrivial relation between the kinetic momentum $`m\stackrel{}{𝐯}`$ and the canonical momentum $`\stackrel{}{𝐩}`$ of a charged point particle,
$$m\stackrel{}{𝐯}=\stackrel{}{𝐩}e\stackrel{}{𝐀}$$
where $`\stackrel{}{𝐀}`$ is the electromagnetic vector potential. We propose that when these electromagnetic forces within the plasma are properly accounted for, the ensuing field theory model has the potential of supporting stable soliton-like configurations which describe helical, self-confined structures within the plasma medium.
Our starting point is a natural kinetic field theory model of a two-component plasma of electromagnetically interacting charged point particles such as electrons and deuterons. In natural units the classical action is
$$S=dtd^3x[i\psi _{e}^{}{}_{}{}^{}(_t+ieA_t)\psi _e+i\psi _{i}^{}{}_{}{}^{}(_tieA_t)\psi _i\frac{1}{2m}|(_k+ieA_k)\psi _e|^2$$
$$\frac{1}{2M}|(_kieA_k)\psi _i|^2\frac{1}{4}F_{\mu \nu }^2]$$
(2)
As usual $`F_{\mu \nu }=_\mu A_\nu _\nu A_\mu `$. The $`\psi _e`$ and $`\psi _i`$ are two (complex) non-relativistic fields for electrons and ions with masses $`m`$ and $`M`$ and electric charges $`\pm e`$, respectively. Notice that we describe both charged fields by macroscopic (Hartree-Fock) wave functions. This is adequate in the classical Bolzmannian limit which is relevant in conventional plasma scenarios . The action (2) determines our first principles description of a non-relativistic plasma. Its magnetohydrodynamical properties are governed by the pertinent energy-momentum tensor $`T_{\mu \nu }`$, which can be constructed from (2) in a standard manner. When we include the contributions that account for the bulk motion of the plasma medium, this leads to an appropriate version of the Navier-Stokes equation (1). Here we are interested in the ensuing static equilibrium configurations. These configurations are local minima of the internal energy $`E`$, which is determined by the temporal $`T_{00}`$ component of the energy-momentum tensor. For a stationary plasma fluid (2) we get from (2)
$$E=d^3x[\frac{1}{2\mu }\{\mathrm{sin}^2\alpha |(_k+ieA_k)\psi _e|^2+\mathrm{cos}^2\alpha |(_kieA_k)\psi _i|^2\}$$
$$+\frac{1}{2}B_i^2+g(\psi _{e}^{}{}_{}{}^{}\psi _e\psi _{i}^{}{}_{}{}^{}\psi _i)^2]$$
(3)
Here $`\mu =m\mathrm{sin}^2\alpha =M\mathrm{cos}^2\alpha `$ is the reduced mass and $`B_i=\frac{1}{2}ϵ_{ijk}F_{jk}`$ is the magnetic field. The quartic potential is the remnant of the Coulomb interaction with $`g`$ an effective coupling constant. It emerges when we first use Gauss’ law to eliminate the electric field, and then recall that in any realistic plasma the Debye screening radius is small in comparison to any characteristic length scale of interest.
The free energy (3) is subjected to the conditions that the plasma is electrically neutral with an equal (large) number $`n_e`$ of electrons and $`n_i`$ of ions, $`n_e=n_i`$ and the total number of charge carriers in the volume $`V`$ remains intact $`n_e+n_i=N`$. These conditions can be implemented by adding appropriate chemical potential terms to (3) in the usual fashion. But for simplicity we here account for them as constraints, imposed by appropriate boundary conditions. Besides the terms that we have displayed in (3) there can also be additional interaction terms for the charged fields. Such terms are usually induced by thermal fluctuations and finite density effects, or by gravitational interactions. However, according to standard universality arguments we expect the main features of (3) to persist at temperatures and distance scales which are relevant in conventional plasma scenarios.
We propose that (3) yields an adequate approximation for a non-relativistic plasma in a kinetic regime where the thermal energy is sufficiently high to prevent the formation of charge neutral bound states, which correspond to hydrogen atoms in the case of electrons and deuterons. Such bound states are present at lower temperatures, and their presence can be accounted for by terms of the form
$$E_{bs}=d^3x\left[\frac{1}{2}\frac{1}{m+M}(_k\mathrm{\Phi })^2+\lambda \mathrm{\Phi }\psi _e\psi _i+\overline{\lambda }\mathrm{\Phi }\psi _e^{}\psi _i^{}\right]$$
Here $`\mathrm{\Phi }`$ a real scalar field that describes a charge neutral bound state of $`\psi _e`$ and $`\psi _i`$. At a sufficiently high temperature this bound state degree of freedom decouples, and (3) becomes adequate for describing the bulk properties of the plasma.
Since $`n_e=n_i`$ we have overall charge neutrality. However, there can be local charge density fluctuations that should not be ignored. Indeed, we now proceed to argue that static charge density fluctuations are naturally present in (3). These fluctuations accompany stable, static solitons which describe filamental self-confined structures within the plasma. For this we first note that the different contributions in (3) respond differently to a scaling $`\stackrel{}{𝐱}\lambda \stackrel{}{𝐱}`$. The kinetic terms scale in proportion to $`\lambda `$ and the Coulomb potential in proportion to $`\lambda ^3`$, but the magnetic energy scales like $`\lambda ^1`$. Consequently the existence of nontrivial, non-dissipative plasma configurations in (3) can not be excluded by simple virial arguments, quantitative investigations become necessary.
We start by observing that the vector potential $`A_k`$ enters at most quadratically. Consequently it can be eliminated: We vary (3) w.r.t. $`A_k`$ and get
$$A_k=\frac{1}{2e}\frac{1}{\mathrm{sin}^2\alpha |\psi _e|^2+\mathrm{cos}^2\alpha |\psi _i|^2}[i\mathrm{sin}^2\alpha (\psi _{e}^{}{}_{}{}^{}_k\psi _e_k\psi _{e}^{}{}_{}{}^{}\psi _e)$$
$$i\mathrm{cos}^2\alpha (\psi _{i}^{}{}_{}{}^{}_k\psi _i_k\psi _{i}^{}{}_{}{}^{}\psi _i)\frac{2\mu }{e}ϵ_{kij}_iB_j]$$
(4)
which determines $`A_k`$ in terms of an iterative gradient expansion, in powers of derivatives in the charged fields. We introduce new variables by
$$(\psi _e,\psi _i)=\rho (\mathrm{cos}\alpha \mathrm{sin}\frac{\theta }{2}e^{i\phi },\mathrm{sin}\alpha \mathrm{cos}\frac{\theta }{2}e^{i\chi })$$
(5)
For reasons that will soon become obvious we have chosen these variables so that they are natural for describing tubular field configurations, with $`\phi `$ and $`\chi `$ related to the toroidal and poloidal angles and $`\theta `$ a shape function that measures the distance away from the centerline of the tube. We compute the free energy (3) to the leading order in a self-consistent gradient expansion, where we keep only terms which are at most fourth order in the derivatives of the variables (5). This approximation is adequate in conventional plasma scenarios where the fields are relatively slowly varying. We start by determining $`A_k`$ from (4) iteratively in the variables (5). We substitute the result in (3), and by defining a three-component unit vector $`\stackrel{}{𝐧}=(\mathrm{cos}(\chi +\phi )\mathrm{sin}\theta ,\mathrm{sin}(\chi +\phi )\mathrm{sin}\theta ,\mathrm{cos}\theta )`$ we finally get for the free energy
$$E=d^3x\left[\frac{1}{2}\frac{1}{m+M}\left\{(_k\rho )^2+\rho ^2|_k\stackrel{}{𝐧}|^2\right\}+\frac{1}{4e^2}(\stackrel{}{𝐧}_i\stackrel{}{𝐧}\times _j\stackrel{}{𝐧})^2+\frac{g\rho ^4}{4}(n_3\mathrm{cos}2\alpha )^2\right]$$
(6)
We note that since $`m`$ and $`M`$ are both nonvanishing, overall charge neutrality implies that asymptotically $`\theta 2\alpha n\pi `$. Since $`\rho const.0`$ asymptotically (see below), the Coulomb interaction then yields a mass term for the variable $`\theta `$. We also note that (6) naturally embodies a helical structure, described by the Hopf invariant . To the relevant order in our gradient expansion
$$Q_H=\frac{1}{e^24\pi ^2}d^3x\stackrel{}{𝐁}\stackrel{}{𝐀}=d^3x\mathrm{cos}\theta \frac{\phi }{2\pi }\times \frac{\chi }{2\pi }=\mathrm{\Delta }\phi \mathrm{\Delta }\chi $$
(7)
Here $`\mathrm{\Delta }\phi `$ resp. $`\mathrm{\Delta }\chi `$ denotes the ($`2\pi `$) change in the pertinent variable over the (would-be) tube, when we cover it once in the toroidal and poloidal directions over a magnetic flux surface with constant $`\theta `$.
The field $`\rho `$ is a measure of the particle density in the bulk of the plasma. If its average (asymptotic) value $`<\rho ^2>=\rho _0^2`$ becomes too small, the collective behaviour of the plasma will be lost and instead we have an individual-particle behaviour of the charged constituents, interacting via Coulomb collisions. Consequently we select the average $`\rho _0^2`$ so that it acquires a sufficiently large value in the medium. Local charge fluctuations then occur in regions where the unit vector $`\stackrel{}{𝐧}`$ becomes a variable so that $`\theta 2\alpha `$. According to our adiabatic approximation $`|_k\stackrel{}{𝐧}|`$ is a slowly varying bounded function over the entire charge fluctuation region, and in particular it vanishes outside of the fluctuation region. When we inspect the $`\rho `$-equation of motion that follows from (6) we find that it can be related to a Schroedinger equation for the lowest energy scattering state in an external potential $`|_k\stackrel{}{𝐧}|^2`$. From this we then conclude that $`|\rho (\stackrel{}{𝐱})|`$ never vanishes; it is bounded from below by a non-vanishing positive value which is related to the ensuing scattering length. This implies that if we average the free energy (6) over $`\rho (\stackrel{}{𝐱})`$, to the relevant order in our gradient expansion the result can be related to the universality class determined by the Hamiltonian
$$H=d^3x\left[\gamma |_k\stackrel{}{𝐧}|^2+\frac{1}{4e^2}(\stackrel{}{𝐧}_i\stackrel{}{𝐧}\times _j\stackrel{}{𝐧})^2+\lambda (n_3\mathrm{cos}2\alpha )^2\right]$$
(8)
where $`\gamma ,\lambda `$ are nonvanishing positive constants, proportional to the scattering length of our Schroedinger equation. This Hamiltonian is known to support stable knotlike solitons . In particular, since the third (Coulomb) term is positive it does not interfere with the lower bound estimate derived in . This estimate states that the first two terms in (8) are bounded from below by the fractional power $`|Q_H|^{3/4}`$ of the Hopf invariant. Even though we do not expect that in the case of (6) this lower bound estimate remains valid as such, we nevertheless conclude that when $`Q_H0`$ the energy (6) admits a nontrivial lower bound; the conclusions from the virial theorem in should not be adapted too hastily.
The properties of (8) with $`\lambda =0`$ have been studied in -. In particular, the numerical simulations in , clearly confirm the existence of stable, knotted and linked solitons with a nontrivial Hopf invariant . The present considerations firmly suggest that the conclusions in - prevail also in the case of (6). Indeed, we have tentatively verified that similar solitons are present in (6), by numerically constructing a line vortex soliton in this model; we describe our solution in figure 1. These solitons then become natural candidates for describing filamental and toroidal structures in the plasma, including coronal loops above the solar photosphere and the design of magnetic geometries in thermonuclear fusion energy experiments. The numerical simulations reported in - are very extensive, and clearly reveal the complexity of the problem. Accordingly the interest has thus far mainly concentrated on the identification of soliton geometries, very little is still known about the solitons detailed physical properties. Consequently at this time we are not in a position to present definite physical predictions in the context of actual applications, high precision numerical methods still remain under active development , and we have to limit ourselves to a few general remarks: In the numerical simulations that have been completed thus far, it has been found that for generic integer values $`(\mathrm{\Delta }\phi ,\mathrm{\Delta }\chi )=(n,m)`$ in (7) the $`\lambda =0`$ solitons of (8) form involved knotted and linked structures. Such complex geometries might be natural in a number of applications, for example when modelling coronal loops. But they might not be of any immediate practical interest for the design of plasma geometries in fusion energy experiments, where planar toroidal configurations are preferable. Indeed, there are also a few torus-shaped solitons which are essentially planar. These occur for values $`(n,m)=(1,1),(2,1),(1,2),(2,2)`$ . The simplest one is $`(1,1)`$ but it appears to have an energy density that peaks at the toroidal symmetry axis. As such this may be an advantage in designing actual fusion reactors. But it could also become problematic, as it may interfere with the construction of an external torus-shaped coiling system which should be needed to create the soliton. On the other hand, the $`(2,1)`$ soliton seems to have a torus-shaped energy density distribution which vanishes at the symmetry axis and peaks at the centerline of the torus (see ). Since this soliton is also quite sturdy , it is a natural candidate e.g. for designing magnetic geometries for thermonuclear fusion energy purposes. In particular, this configuration strongly suggests that for a stable, toroidal planar geometry the safety factor in the bulk of the plasma should not exceed $`q2`$. A configuration with a higher value for $`q`$ tends to adjust itself towards a geometrical shape which is not planar; see the computer animations in the www-address of reference .
In conclusion, we have argued that an electrically neutral conducting plasma can form stable, self-confining structures. This is due to soliton-like solutions, which we have shown will appear when we properly account for the nontrivial electromagnetic interactions within the plasma. We have proposed that our solitons can become relevant in a number of practical scenarios, including coronal loops and the design of magnetic geometries in thermonuclear fusion energy experiments. However, in order to assess the impact of our findings, detailed numerical investigations are necessary. Unfortunately the simulations remain highly complex, even with the present day supercomputers. Consequently we have not been able to reliably confirm that parameters such as the asymptotic density $`\rho _0`$ and the coupling $`g`$ can indeed be selected appropriately for the solitons to have direct technological relevance for example in the design of magnetic geometries for energy producing thermonuclear fusion reactors. But since over 99.9 per cent of all known matter in the Universe exists in the plasma state, there are no doubt numerous scenarios where our results can become important. Besides astrophysical applications or quark-gluon plasma experiments, these might include even an explanation to the highly elusive ball lightning.
We thank A. Alekseev, E. Babaev, A. Bondeson, H. Hansson, E. Langmann, V. Maslov, H.K. Moffatt, S. Nasir, A. Polychronakos, R. Ricca and G. Semenoff for discussions. We are particularly indebted to M. Lübcke for his help, and to J. Hietarinta for communicating the results in prior to publication. We thank the Center for Scientific Computing in Espoo, Finland for the use of their computers. The work of L.F. has been supported by grants RFFR 99-01-00101 and INTAS 9606, and the work of A.J.N. has been supported by NFR Grant F-AA/FU 06821-308.
Figure Caption
figure 1: An example of a numerically constructed tubular line vortex solution of (6), with energy density plotted as a function of the distance from the tubular center-line. We use standard cylindrical coordinates $`(r,\varphi ,z)`$ so that the tubular center-line coincides with the $`z`$-axis. For simplicity we have taken a limit of large ion mass which sends $`2\alpha \pi `$. All numerical parameters in (6) are $`𝒪(1)`$ and the helical structure is characterized by $`\phi +\chi =\varphi +0.6z`$. |
warning/0003/cs0003045.html | ar5iv | text | # Termination Proofs for Logic Programs with TablingThis article is a collection and integration of a number of results that appeared—sometimes in weaker forms—in the conference papers [27] and [28].
## 1 Introduction
Tabled logic programming is receiving increasing attention in the Logic Programming community. It avoids many of the shortcomings of SLD(NF) execution and provides a more flexible and often extremely efficient execution mechanism for logic programs. Furthermore, tabled execution of logic programs terminates more often than execution based on SLD-resolution. In particular, all programs that terminate under SLD also terminate under tabled execution. So, if a program can be proven to terminate under SLD-resolution (by one of the existing automated techniques surveyed in ), then the program will trivially also terminate under SLG-resolution, the resolution principle of tabling; see . But, since there are SLG-terminating programs which are not SLD-terminating, more effective proof techniques need to and can be found.
The idea underlying *tabling* is quite simple. Essentially, under a tabled execution mechanism, answers for selected tabled atoms as well as these atoms are stored in a table. When an identical (up to renaming of variables) such atom is recursively called, the selected atom is not resolved against program clauses; instead, all corresponding answers computed so far are looked up in the table and the corresponding answer substitutions are applied to the atom. This process is repeated for all subsequent computed answer substitutions that correspond to the atom.
We study *universal* termination of *definite* tabled logic programs executed under SLG-resolution using a fixed *left-to-right selection rule* (we drop the “S” in SLD and SLG whenever we refer to the left-to-right selection rule). We introduce a first basic notion of termination under tabled execution, called quasi-termination. Quasi-termination captures the property that, under an LD-computation, a given atomic query leads to only finitely many different non-variant calls to tabled predicates and there is no infinite derivation consisting of queries with only selected non-tabled atoms. In a broader context, the notion of quasi-termination and techniques for proving it are of independent interest; they can be used to e.g. ensure termination of off-line specialisation of logic programs, whether tabled or not; see . However, the notion of quasi-termination only partially corresponds to our intuitive notion of a “terminating computation”. This is because an atom can have infinitely many computed answers (which does not have to lead to infinitely many new calls). Therefore, we also introduce the stronger notion of LG-termination. A program $`P`$ LG-terminates w.r.t. a given atomic query iff $`P`$ quasi-terminates w.r.t. the query and the set of all computed answers for calls in the LD-computation of the query is finite.
We present *sufficient conditions* for these two notions of termination under tabled execution: namely, quasi-acceptability for quasi-termination and LG-acceptability for LG-termination. We show that these conditions are also *necessary* in case the set of tabled predicates is *well-chosen*; see Section 5. Our termination conditions are adapted from the acceptability notion for LD-termination defined in , and not from the more “standard” definition of acceptability by Apt and Pedreschi in . The reason for this choice is that the quasi-termination as well as the LG-termination property of a tabled program and query is *not* closed under substitution. The acceptability notion in is expressed in terms of ground instances of clauses and its associated notion of LD-termination is expressed in terms of the set of all queries that are bounded under the given level mapping. Such sets are closed under substitution. Because quasi-termination and LG-termination lack invariance under substitution, we use a stronger notion of acceptability, capable of treating *any* set of queries.
Besides a characterisation of the two notions of universal termination under tabled execution, we also give *modular termination conditions*, i.e., conditions on two programs $`P`$ and $`R`$, where $`P`$ extends $`R`$, ensuring termination of the union $`PR`$. Such modular proofs were already motivated in the literature in the context of termination under SLD-resolution (see for instance ). Indeed, for programming in the large, it is important to have modular termination proofs, i.e., proofs that are capable of combining termination proofs of separate programs to obtain termination proofs of combined programs.
Finally, we present easy to *automate*, sufficient conditions for quasi-termination and LG-termination. To this end, we use mode information: we consider simply moded, well-moded programs and queries. We point out how these termination conditions could be automated, by extending the recently developed, constraint-based, automatic termination analysis for SLD-resolution of .
All the above mentioned results are developed and presented for a *mixed tabled/non-tabled* execution mechanism. This means that, in the execution, only a subset of the predicates (specified by the programmer) will be tabled, while standard LD-resolution steps are applied to all others. In Section 3, we discuss the benefits of having such a mixed execution mechanism. This focus on mixed execution considerably strengthens our results. In particular, our results both introduce new termination conditions for (fully) tabled logic programs, and at the same time generalize existing termination conditions for LD-resolution. Of course, this choice also makes the results more technically involved.
The rest of the article is structured as follows. In Section 2, we define some preliminary concepts, in particular the notion of finitely partitioning level mapping, which plays a central role in our termination conditions. Next, in Section 3, we recall the execution mechanism of LG-resolution, the tabled-based resolution strategy used in this article. We first present examples from context-free grammar recognition and parsing which motivate the need to freely mix untabled and tabled execution and then we formally define the resolution principle of tabling, called SLG-resolution. Next, in Section 4, two notions of termination of LG-resolution are introduced: quasi-termination and the stronger notion of LG-termination. We also define a transformation on programs which reduces the problem of proving LG-termination to the problem of proving quasi-termination. In Section 5, sufficient (and also necessary in case the tabling is well-chosen) conditions for the two notions of termination are given: the condition of quasi-acceptability for quasi-termination (Subsection 5.1) and the condition of LG-acceptability for LG-termination (Subsection 5.2). Modular termination conditions, i.e., conditions that are capable of combining termination proofs of separate programs to obtain termination proofs of combined programs, are given in Section 6: in Subsection 6.1 for quasi-termination, and in Subsection 6.2 for LG-termination. In Subsection 6.3, more detailed modular termination conditions for quasi-termination are given, which also provide an incremental construction of an appropriate level mapping. Finally, in Section 7, we investigate conditions for termination of LG-resolution which are easy to automate. In particular, our eventual goal is to extend the constraint-based automatic approach towards LD-termination of , in order to prove termination of tabled logic programs in an automatic way. Our extension is restricted to the class of simply moded, well-moded programs and queries, which we recall from . Only quasi-termination is considered in Section 7; the results for LG-termination carry over in the same way. We end with some concluding remarks, a discussion on related work and with some topics for future research.
## 2 Preliminaries
We assume familiarity with the basic concepts of logic programming; see . Throughout the article, $`P`$ will denote a definite logic program. By $`Pred_P`$, $`Fun_P`$ and $`Const_P`$ we denote the set of predicate, function and constant symbols occurring in $`P`$. We assume that these sets are finite. By $`Def_P`$ we denote the set of predicates defined in $`P`$ (i.e., predicates occurring in the head of a clause of $`P`$). By $`Rec_P`$, resp. $`NRec_P`$, we denote the set of (directly or indirectly) recursive, resp. non-recursive, predicates of the program $`P`$ (so $`NRec_P=Pred_PRec_P`$). If $`A=p(t_1,\mathrm{},t_n)`$, then we denote by $`Rel(A)`$ the predicate symbol $`p`$ of $`A`$; i.e., $`Rel(A)=p`$. We call $`A=p(t_1,\mathrm{},t_n)`$ a $`p`$-atom.
The extended Herbrand Universe, $`U_P^E`$, and the extended Herbrand Base, $`B_P^E`$, associated with a program $`P`$, were introduced in . They are defined as follows. Let $`Term_P`$ and $`Atom_P`$ denote the set of respectively all terms and atoms that can be constructed from the alphabet underlying $`P`$. The variant relation, denoted $``$, defines an equivalence. $`U_P^E`$ and $`B_P^E`$ are respectively the quotient sets $`Term_P/`$ and $`Atom_P/`$. For any term $`t`$ (or atom $`A`$), we denote its class in $`U_P^E`$ ($`B_P^E`$) as $`\stackrel{~}{t}`$ ($`\stackrel{~}{A}`$). However, when no confusion is possible, we omit the tildes. For $`\mathrm{\Pi }Pred_P`$, we denote with $`B_\mathrm{\Pi }^E`$ the subset of $`B_P^E`$ consisting of (equivalence classes of) atoms based on the predicate symbols of $`\mathrm{\Pi }`$. So $`B_P^E`$ can be seen as an abbreviation of $`B_{Pred_P}^E`$.
Let $`P`$ be a program and $`p,qPred_P`$. We say that $`p`$ refers to $`q`$ in $`P`$ iff there is a clause in $`P`$ with $`p`$ in the head and $`q`$ occurring in the body. We say that $`p`$ depends on $`q`$ in $`P`$, and write $`pq`$, iff $`(p,q)`$ is in the reflexive, transitive closure of the relation refers to. Note that, by definition, each predicate depends on itself. We write $`pq`$ iff $`pq`$, $`qp`$ ($`p`$ and $`q`$ are mutually recursive or $`p=q`$). The dependency graph $`G_P`$ of a program $`P`$ is a graph where the nodes are labeled with the predicates of $`Pred_P`$. There is a directed arc from $`p`$ to $`q`$ in $`G_P`$ iff $`p`$ refers to $`q`$. A program $`P`$ extends a program $`R`$ iff no predicate defined in $`P`$ occurs in $`R`$.
As mentioned and used in the introduction, in analogy with , we will refer to SLD-derivations (see ) following the left-to-right selection rule as LD-derivations. Other concepts adopt this naming accordingly.
###### Definition 1 (call set associated to $`S`$)
Let $`P`$ be a program and $`SB_P^E`$. By $`Call(P,S)`$ we denote the subset of $`B_P^E`$ such that $`BCall(P,S)`$ whenever a representant of $`B`$ is a selected atom in an LD-derivation for some $`P\{A\}`$, with $`\stackrel{~}{A}S`$.
Throughout the article we assume that in any derivation of a query w.r.t. a program, representants of equivalence classes are systematically provided with fresh variables, to avoid the necessity of renaming apart. In the sequel, we abbreviate most general unifier with $`mgu`$ and LD-computed answer substitution with $`cas`$.
The concepts defined in the following Definitions 2, 3 and 4, will be used in the proofs of some theorems and propositions of this article.
###### Definition 2 (direct descendant)
Let $`P`$ be a program and $`\stackrel{~}{A},\stackrel{~}{B}B_P^E`$. We call $`\stackrel{~}{B}`$ a direct descendant of $`\stackrel{~}{A}`$ iff there exists a clause $`HB_1,\mathrm{},B_n`$ in $`P`$ such that $`mgu(A,H)=\theta `$ exists and, there is an $`i[1,n]`$ such that there is an LD-refutation for $`(B_1,\mathrm{},B_{i1})\theta `$ with $`cas`$ $`\theta _{i1}`$ and $`BB_i\theta \theta _{i1}`$.
###### Definition 3 (directed subsequence of an LD-derivation)
Let $`P`$ be a program and $`\stackrel{~}{A}B_P^E`$. Let $`A=G_0,`$ $`G_1,\mathrm{}`$ be an LD-derivation of $`A`$ in $`P`$. A subsequence $`G_{i_0},G_{i_1},\mathrm{}`$, with $`G_{i_j}=A_{i_j},𝒜_{i_j}`$, is called a directed subsequence iff for all $`j0`$, $`\stackrel{~}{A}_{i_{j+1}}`$ is a direct descendant of $`\stackrel{~}{A}_{i_j}`$ in the LD-derivation.
###### Definition 4 (call graph associated to $`S`$)
Let $`P`$ be a program and $`SB_P^E`$. The call graph $`Call`$-$`Gr(P,S)`$ associated to $`P`$ and $`S`$ is a graph such that:
* its set of nodes is $`Call(P,S)`$,
* there exists a directed arc from $`\stackrel{~}{A}`$ to $`\stackrel{~}{B}`$ iff $`\stackrel{~}{B}`$ is a direct descendant of $`\stackrel{~}{A}`$.
We recall the definitions of norm and level mapping, which are useful in the context of termination analysis (see for a survey on termination analyses for (S)LD-resolution).
###### Definition 5 (norm)
A norm is a function $`.:U_P^E\mathrm{}`$.
###### Definition 6 (level mapping)
A level mapping is a function $`|.|:B_P^E\mathrm{}`$.
A level mapping or norm is said to be trivial if it is the constant $`0`$-mapping.
Our termination conditions are based on the following concept of a finitely partitioning level mapping.
###### Definition 7 (finitely partitioning level mapping)
Let $`P`$ be a program and $`CB_P^E`$. A level mapping $`|.|`$ is finitely partitioning on $`C`$ iff for all $`n\mathrm{}:\mathrm{}(|.|^1(n)C)<\mathrm{}`$, where $`\mathrm{}`$ is the cardinality function.
So, a level mapping $`|.|`$ is finitely partitioning on $`CB_P^E`$ if it does not map an infinite set of atoms of $`C`$ to the same natural number. That is, $`|.|`$ partitions $`C`$ into finite subsets. In particular, we have that every level mapping is finitely partitioning on a finite set $`C`$.
## 3 Tabling in Logic Programs
Our experience is that tabled execution is used *selectively* in practice. Thus, before formally defining the resolution principle of tabling, called SLG-resolution, we first present some examples which motivate the need to freely mix LD-resolution and tabled execution.
### 3.1 Mixing Tabled and LD Execution: Motivating Examples
It has long been noted in the literature , that tabled evaluation can be used for context-free grammar recognition and parsing: tabling eliminates redundancy and handles grammars that would otherwise infinitely loop under Prolog-style execution (e.g. left-recursive ones). The following program, where all predicates are tabled, provides such an example.
$$\{\begin{array}{ccc}expr(Si,So)\hfill & \hfill & expr(Si,S1),S1=[^{}+^{}|S2],term(S2,So)\hfill \\ expr(Si,So)\hfill & \hfill & term(Si,So)\hfill \\ term(Si,So)\hfill & \hfill & term(Si,S1),S1=[^{}^{}|S2],primary(S2,So)\hfill \\ term(Si,So)\hfill & \hfill & primary(Si,So)\hfill \\ primary(Si,So)\hfill & \hfill & Si=[^{}(^{}|S1],expr(S1,S2),S2=[^{})^{}|So]\hfill \\ primary(Si,So)\hfill & \hfill & Si=[I|So],integer(I)\hfill \end{array}$$
This grammar, recognizing arithmetic expressions containing additions and multiplications over the integers, is left recursive—left recursion is used to give the arithmetic operators their proper associativity—and would be non-terminating for Prolog-style execution. Under tabled execution, left recursion is handled correctly. In fact, one only needs to table predicates $`expr/2`$ and $`term/2`$ to get the desired termination behaviour; we can and will safely drop the tabling of $`primary/2`$ in the sequel. However, this integration of non-tabled (LD) and tabled execution is perhaps a trivial one.
To see why a non-trivial mix of tabled with LD execution is desirable in practice, suppose that we want to extend the above recognition grammar to handle exponentiation. The most natural way to do so is to introduce a new nonterminal, named $`factor`$, for handling exponentiation and make it right recursive, since the exponentiation operator is right associative. The resulting grammar is as below where only the predicates $`expr/2`$ and $`term/2`$ are tabled.
$$\{\begin{array}{ccc}expr(Si,So)\hfill & \hfill & expr(Si,S1),S1=[^{}+^{}|S2],term(S2,So)\hfill \\ expr(Si,So)\hfill & \hfill & term(Si,So)\hfill \\ term(Si,So)\hfill & \hfill & term(Si,S1),S1=[^{}^{}|S2],factor(S2,So)\hfill \\ term(Si,So)\hfill & \hfill & factor(Si,So)\hfill \\ factor(Si,So)\hfill & \hfill & primary(Si,S1),S1=[^{}^{}|S2],factor(S2,So)\hfill \\ factor(Si,So)\hfill & \hfill & primary(Si,So)\hfill \\ primary(Si,So)\hfill & \hfill & Si=[^{}(^{}|S1],expr(S1,S2),S2=[^{})^{}|So]\hfill \\ primary(Si,So)\hfill & \hfill & Si=[I|So],integer(I)\hfill \end{array}$$
Note that, at least as far as termination is concerned, there is no need to table the new nonterminal $`factor`$. Indeed, Prolog’s evaluation strategy handles right recursion in grammars finitely. In fact, Prolog-style evaluation of right recursion is more efficient than its tabled-based evaluation: Prolog has linear complexity for a simple right recursive grammar, but with tabling implemented as in XSB the evaluation could be quadratic as calls need to be recorded in the tables using explicit copying. Thus, it is important to allow tabled and non-tabled predicates to be freely intermixed, and be able to choose the strategy that is most efficient for the situation at hand.
By using tabling in context-free grammars, one gets a recognition algorithm that is a variant of Early’s algorithm (also known as active chart recognition algorithm) whose complexity is polynomial in the size of the input expression/string . However, often one wants to construct the parse tree(s) for a given input string. The usual approach is to introduce an extra argument to the nonterminals of the input grammar—representing the portion of the parse tree that each rule generates—and naturally to also add the necessary code that constructs the parse tree. This approach is straightforward, but as noticed by Warren in , using the same program for recognition as well as parsing may be extremely unsatisfactory from a complexity standpoint: in context-free grammars, recognition is polynomial while parsing is exponential, since there can be exponentially many parse trees for a given input string. The obvious solution is to use two interleaved versions of the grammar as in the following program, which recognizes and parses the language $`a^nb`$.
$$\begin{array}{cc}R:\hfill & \{\begin{array}{ccc}s(Si,So)\hfill & \hfill & a(Si,S),S=[b|So]\hfill \\ a(Si,So)\hfill & \hfill & a(Si,S),a(S,So)\hfill \\ a(Si,So)\hfill & \hfill & Si=[a|So]\hfill \end{array}\hfill \\ & \\ P:\hfill & \{\begin{array}{ccc}s(Si,So,PT)\hfill & \hfill & a(Si,S),S=[b|So],PT=spt(PTa,b),a(Si,S,PTa)\hfill \\ a(Si,So,PT)\hfill & \hfill & a(Si,S),a(S,So),PT=apt(PT1,PT2),a(Si,S,PT1),\hfill \\ & & a(S,So,PT2)\hfill \\ a(Si,So,PT)\hfill & \hfill & Si=[a|So],PT=a\hfill \end{array}\hfill \end{array}$$
Note that only $`a/2`$, i.e., the recursive predicate of the ‘recognition’ part, $`R`$, of the program (consisting of predicates $`s/2`$ and $`a/2`$), needs to be tabled. This action allows recognition to terminate and to have polynomial complexity. Furthermore, the recognizer can now be used as a filter for the parsing process in the following way: only after knowing that a particular part of the input belongs to the grammar and having computed the exact substring that each nonterminal spans, do we invoke the parsing routine on the nonterminal to construct its (possibly exponentially many) parse trees. Doing so, avoids e.g. cases where it may take exponential time to fail on an input string that does not belong in the given language: an example for the grammar under consideration is the input string $`a^n`$. On the other hand, tabling the ‘parsing’ part of the program (consisting of predicates $`s/3`$ and $`a/3`$) does not affect the efficiency of the process complexity-wise and incurs a small performance overhead due to the recording of calls and their answers in the tables. Finally, note that the construction is modular in the sense that the ‘parsing’ part of the program, $`P`$, depends on the ‘recognition’ part, $`R`$, but not vice versa; we say that $`P`$ *extends* $`R`$.
### 3.2 SLG-Resolution
In this article, we consider termination of SLG-resolution (see ), using a fixed left-to-right selection rule, for a given set of atomic (top level) queries with atoms in $`SB_P^E`$. We will abbreviate SLG-resolution under the left-to-right selection rule by LG-resolution. For definite programs LG-resolution is similar to OLDT-resolution , modulo the fact that OLDT specifies a more fixed control strategy and uses subsumption checking and term-depth abstraction instead of variant checking. We present a non-constructive definition of SLG-resolution that is sufficient for our purposes, and refer to for more constructive formulations of (variants) of tabled resolution.
By fixing a tabling for a program $`P`$, we mean choosing a set of predicates of $`P`$ which are tabled. The set of tabled predicates for a given tabling of a program $`P`$ is denoted with $`Tab_P`$. The complement of this set is denoted with $`NTab_P=Pred_PTab_P`$.
###### Definition 8 (pseudo SLG-tree, pseudo LG-tree)
Let $`P`$ be a definite program, $`Tab_PPred_P`$, $``$ a selection rule and $`A`$ an atom. A pseudo SLG-tree w.r.t. $`Tab_P`$ for $`P\{A\}`$ under $``$ is a tree $`\tau _A`$ such that:
1. the nodes of $`\tau _A`$ are labeled with queries along with an indication of the selected atom according to $``$,
2. the root of $`\tau _A`$ is $`A`$,
3. the children of the root $`A`$ are obtained by resolution against all matching program clauses in $`P`$, the arcs are labeled with the corresponding $`mgu`$ used in the resolution step,
4. the children of a non-root node labeled with the query $`𝐐`$ where $`(𝐐)=B`$ are obtained as follows:
1. if $`Rel(B)Tab_P`$, then
the (possibly infinitely many) children of the node can only be obtained by resolving the selected atom $`B`$ of the node with clauses of the form $`B\theta `$ (not necessarily in $`P`$), the arcs are labeled with the corresponding $`mgu`$ used in the resolution step (i.e., $`\theta `$),
2. if $`Rel(B)NTab_P`$, then
the children of the node are obtained by resolution of $`B`$ against all matching program clauses in $`P`$, and the arcs are labeled with the corresponding $`mgu`$ used in the resolution step.
If $``$ is the leftmost selection rule, $`\tau _A`$ is called a pseudo LG-tree w.r.t. $`Tab_P`$ for $`P\{A\}`$.
We say that a pseudo SLG-tree $`\tau _A`$ w.r.t. $`Tab_P`$ for $`P\{A\}`$ is smaller than another pseudo SLG-tree $`\tau _A^{^{}}`$ w.r.t. $`Tab_P`$ for $`P\{A\}`$ iff $`\tau _A^{^{}}`$ can be obtained from $`\tau _A`$ by attaching new sub-branches to nodes in $`\tau _A`$.
A (computed) answer clause of a pseudo SLG-tree $`\tau _A`$ w.r.t. $`Tab_P`$ for $`P\{A\}`$ is a clause of the form $`A\theta `$ where $`\theta `$ is the composition of the substitutions found on a branch of $`\tau _A`$ whose leaf is labeled with the empty query.
Intuitively, a pseudo SLG-tree (in an SLG-forest, see Definition 9 below) represents the tabled computation (w.r.t. $`Tab_P`$) of all answers for a given subquery labeling the root node of the tree. The trees in the above definition are called pseudo SLG-trees because there is no condition yet on which clauses $`B\theta `$ exactly are to be used for resolution in point 4a. These clauses represent the answers found (possibly in another tree of the forest) for the selected tabled atom. This interaction between the trees in an SLG-forest is captured in the following definition.
###### Definition 9 (SLG-forest, LG-forest)
Let $`P`$ be a definite program, $`Tab_PPred_P`$, $``$ be a selection rule and $`T`$ be a (possibly infinite) set of atoms such that no two different atoms in $`T`$ are variants of each other. $``$ is an SLG-forest w.r.t. $`Tab_P`$ for $`P`$ and $`T`$ under $``$ iff $``$ is a set of minimal pseudo SLG-trees $`\{\tau _A|AT\}`$ w.r.t. $`Tab_P`$ where
1. $`\tau _A`$ is a pseudo SLG-tree w.r.t. $`Tab_P`$ for $`P\{A\}`$ under $``$,
2. every selected tabled atom $`B`$ of each node in every $`\tau _A`$ is a variant of an element $`B^{^{}}`$ of $`T`$, such that every clause resolved with $`B`$ is a variant of an answer clause of $`\tau _B^{^{}}`$ and vice versa, for every answer clause of $`\tau _B^{^{}}`$ there is a variant of this answer clause which is resolved with $`B`$.
Let $`S`$ be a set of atoms. An SLG-forest for $`P`$ and $`S`$ w.r.t. $`Tab_P`$ under $``$ is an SLG-forest w.r.t. $`Tab_P`$ for a minimal set $`T`$ with $`\stackrel{~}{S}\stackrel{~}{T}`$. If $`S=\{A\}`$, then we also talk about the SLG-forest for $`P\{A\}`$.
An LG-forest is an SLG-forest containing only pseudo LG-trees.
Point 2 of Definition 9, together with the imposed minimality of trees in a forest, now uniquely determines these trees. So we can henceforth drop the designation “pseudo” and refer to (S)LG-trees in an (S)LG-forest.
Note that, selected atoms which are not tabled (i.e., of predicates belonging to $`NTab_P`$) are resolved against program clauses, as in (S)LD-resolution. So, if $`Tab_P=\mathrm{}`$, the (S)LG-forest of $`P\{A\}`$ consists of one tree: the (S)LD-tree of $`P\{A\}`$.
We use the following small, tabled program to illustrate the notions that we introduced so far. Variations of it will also be used throughout this article to exemplify concepts related to the termination aspects of tabled logic programs.
###### Example 1
The following program $`P`$ computes the paths from a given node to the reachable nodes in a given graph. The graph is represented as a list of terms $`e(n_1,n_2)`$, indicating that there is an edge from node $`n_1`$ to node $`n_2`$; this list is passed as an input argument to predicate $`\mathrm{𝑝𝑎𝑡ℎ}/4`$ and the predicate $`\mathrm{𝑒𝑑𝑔𝑒}/3`$ is used to retrieve edges of the graph with a specific source node.
$$\{\begin{array}{ccc}path(X,Ed,Y,[Y])\hfill & \hfill & edge(X,Ed,Y)\hfill \\ path(X,Ed,Z,[Y|L])\hfill & \hfill & edge(X,Ed,Y),path(Y,Ed,Z,L)\hfill \\ edge(X,[e(X,Y)|L],Y)\hfill & \hfill & \\ edge(X,[e(X_1,X_2)|L],Y)\hfill & \hfill & edge(X,L,Y)\hfill \end{array}$$
Let $`S=\{\mathrm{𝑝𝑎𝑡ℎ}(a,[e(a,b),e(b,a)],Y,L)\}`$ and $`Tab_P=\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$. Then,
$`Call(P,S)`$ $`=`$ $`\{\mathrm{𝑝𝑎𝑡ℎ}(a,[e(a,b),e(b,a)],Y,L),\mathrm{𝑝𝑎𝑡ℎ}(b,[e(a,b),e(b,a)],Y,L),`$
$`\mathrm{𝑒𝑑𝑔𝑒}(a,[e(a,b),e(b,a)],Y),\mathrm{𝑒𝑑𝑔𝑒}(a,[e(b,a)],Y),\mathrm{𝑒𝑑𝑔𝑒}(a,[],Y),`$
$`\mathrm{𝑒𝑑𝑔𝑒}(b,[e(a,b),e(b,a)],Y),\mathrm{𝑒𝑑𝑔𝑒}(b,[e(b,a)],Y),\mathrm{𝑒𝑑𝑔𝑒}(b,[],Y)\}`$
The LG-forest w.r.t. $`Tab_P`$ for $`P`$ and $`S`$ is shown in Figure 1. Note that there are two LG-trees (only 2 tabled atoms are called), both with finite branches, but both trees have an infinitely branching node. Due to the last argument of the $`\mathrm{𝑝𝑎𝑡ℎ}/4`$ predicate, each of these selected tabled atoms has infinitely many computed answers.
As proven in e.g. \[20, Theorem 2.1\], the set of call patterns and the set of computed answer substitutions are not influenced by tabling. Thus, we can use the notions of call set, $`Call(P,S)`$, and LD-computed answer substitution, $`cas`$, even in the context of SLG-resolution.
The notion of a call graph (Definition 4) has the following particularly interesting property, described in the proposition below, which is useful in the study of termination. We will use this property in the proof of Theorem 5.1, which gives a necessary and sufficient condition for quasi-termination of tabled logic programs.
###### Proposition 1 (call graph: paths and selected atoms)
Let $`P`$ be a program, $`Tab_PPred_P`$ and $`SB_P^E`$. Let $`p`$ be any directed path in $`Call`$-$`Gr(P,S)`$. Then there exists an LG-derivation for some element of $`Call(P,S)`$, such that all the nodes in $`p`$ occur as selected atoms in the derivation.
###### Proof
By definition of $`Call`$-$`Gr(P,S)`$, for every arc from $`\stackrel{~}{A}`$ to $`\stackrel{~}{B}`$ in $`Call`$-$`Gr(P,S)`$, there exists a sequence of consecutive LG-derivation steps, starting from $`A`$ and having a variant of $`B`$ as its selected atom at the end. Because (a variant of) $`B`$ is selected at the end-point, any two such derivation-step sequences, corresponding to consecutive arcs in $`Call`$-$`Gr(P,S)`$, can be composed to form a new sequence of LG-derivation steps. In this sequence, all 3 nodes of the consecutive arcs remain selected atoms in the new sequence of derivation steps. Transitively exploiting the above argument yields the result. ∎
## 4 Two Notions of Termination of Tabled Logic Programs
We start by introducing a first notion of universal termination of tabled logic programs, called *quasi-termination*. A program $`P`$ with a tabling $`Tab_P`$ is said to be quasi-terminating w.r.t. a query $`A`$ iff the LG-forest of $`P\{A\}`$ consists of a finite number of LG-trees which all have finite branches. Quasi-termination captures the property that, under LD-computation, a given atomic query leads to only finitely many different (nonvariant) calls to tabled predicates and there is no infinite derivation consisting of queries with only selected non-tabled atoms. As mentioned in the introduction, techniques for proving quasi-termination can be used to ensure termination of off-line specialisation of logic programs (whether tabled or not). Currently, in all off-line partial evaluation methods for logic programs (e.g. ) termination has to be ensured manually. In the context of off-line partial evaluation, quasi-termination (when tabling the whole set of predicates) is actually *identical* to termination of the partial evaluator; see e.g. the discussion in . Thus, given a technique to establish quasi-termination, one can also establish whether a given binding time annotation will ensure termination or whether further abstraction is called for. This idea has already been successfully applied in the context of functional programming , using the termination criterion of .
Despite its usefulness, the notion of quasi-termination only partially corresponds to our intuitive notion of a terminating execution of a query against a tabled program. This is because this notion only requires that the LG-forest consists of only a finite number of LG-trees, without infinite branches, yet these trees can have infinitely branching nodes. In order to capture this source of non-termination for a tabled computation, we also introduce the stronger notion of LG-termination. A program $`P`$ with a tabling $`Tab_P`$ is said to be LG-terminating w.r.t. a query $`A`$ iff the LG-forest of $`P\{A\}`$ consists of a finite number of finite LG-trees. So, a program $`P`$ is LG-terminating w.r.t. a query $`A`$ iff it is quasi-terminating w.r.t. $`A`$ and all atoms in the call set $`Call(P,\{A\})`$ have only a finite number of computed answers.
In the next two subsections, we formally introduce these two notions of termination of LG-resolution with Tabling, we give examples and discuss some of their properties.
### 4.1 Quasi-Termination
A first basic notion of universal termination under a tabled execution mechanism is quasi-termination (a term borrowed from , defining a similar notion in the context of termination of off-line partial evaluation of functional programs). It is formally defined as follows.
###### Definition 10 (quasi-termination)
Let $`P`$ be a program, $`Tab_PPred_P`$, and $`SB_P^E`$.
$`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$ iff for all $`A`$ such that $`\stackrel{~}{A}S`$, the LG-forest w.r.t. $`Tab_P`$ for $`P\{A\}`$ consists of a finite number of LG-trees without infinite branches.
Also, $`P`$ quasi-terminates w.r.t. $`S`$ iff $`P`$ quasi-terminates w.r.t. $`Pred_P`$ and $`S`$.
Note that quasi-termination does not require that the LG-trees are finitely branching in their nodes.
###### Example 2
Recall the program $`P`$ and set $`S=\{\mathrm{𝑝𝑎𝑡ℎ}(a,[e(a,b),e(b,a)],Y,L)\}`$ of Example 1. The LG-forest w.r.t. $`Tab_P=\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ was shown in Figure 1. $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$.
Many works address the problem of termination of logic programs executed under LD-resolution (see for a survey): A program $`P`$ is said to be LD-terminating w.r.t. a set $`SB_P^E`$ iff for all $`A`$ such that $`\stackrel{~}{A}S`$, the LD-tree of $`P\{A\}`$ is finite. In the next lemma, we show that the notion of LD-termination is stronger than the notion of quasi-termination. Taking Example 2 into account, it then follows that the notion of LD-termination is strictly stronger than the notion of quasi-termination.
###### Lemma 1
Let $`P`$ be a program, $`Tab_PPred_P`$, and $`SB_P^E`$.
If $`P`$ LD-terminates w.r.t. $`S`$, then $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$.
###### Proof
Let $`A`$ be an atom such that $`\stackrel{~}{A}S`$. Let $``$ be the LG-forest w.r.t. $`Tab_P`$ for $`P\{A\}`$. If $`P`$ LD-terminates w.r.t. $`S`$, it is easy to see that $`Call(P,\{A\})`$ is finite. Hence, $`Call(P,\{A\})B_{Tab_P}^E`$ is finite, and $``$ consists of a finite number of LG-trees.
Now we prove that no tree in $``$ has an infinite branch. Suppose this is not the case and there is a tree in $``$ with an infinite branch. Let $`H`$ be the leftmost atom of a query labeling a node in this infinite branch. Then, $`H`$ has an infinite LD-derivation (just plug in, for each tabled atom $`G`$ in the infinite branch which is resolved with an answer, the branch of the tree with root $`G`$ which leads to this answer). This gives a contradiction. ∎
Note that by definition, $`P`$ quasi-terminates w.r.t. $`Tab_P=\mathrm{}`$ and $`S`$ iff $`P`$ LD-terminates w.r.t. $`S`$.
Consider next the special case where all predicates occurring in $`P`$ are tabled. If $`Tab_P=Pred_P`$, an LG-tree cannot have infinite branches. So, $`P`$ quasi-terminates w.r.t. a set $`S`$ iff for all $`A`$ such that $`\stackrel{~}{A}S`$, the LG-forest for $`P\{A\}`$ consists of a finite number of LG-trees. The following equivalence holds.
###### Lemma 2
Let $`P`$ be a program, $`Tab_P=Pred_P`$, and $`SB_P^E`$. SW
$`P`$ quasi-terminates w.r.t. $`S`$ iff for all $`A`$ such that $`\stackrel{~}{A}S`$, $`Call(P,\{A\})`$ is finite.
###### Proof
Since $`Tab_P=Pred_P`$, an LG-tree cannot have infinite branches. The equivalence then follows from the fact that for every $`A`$ such that $`\stackrel{~}{A}S`$, $`B`$ is the root of an LG-tree in the LG-forest of $`P\{A\}`$ iff $`\stackrel{~}{B}Call(P,\{A\})`$. ∎
When all predicates are tabled, from the above lemma, it follows that in case the Herbrand Universe $`U_P^E`$ associated to a program $`P`$ is finite, $`P`$ quasi-terminates w.r.t. any set of queries $`S`$.
Lemma 2 does not hold in case that the tabled predicates of a program are a strict subset of the set of predicates occurring in the program. A counterexample for the if-direction is given by the program $`P=\{pq,qp\}`$, the set $`S=\{p\}`$ and the empty set of tabled predicates, $`Tab_P=\mathrm{}`$. The LG-forest consists of one tree, namely the LD-tree of $`P\{p\}`$ (so quasi-termination is the same as LD-termination). $`P`$ does not quasi-terminate w.r.t. $`Tab_P`$ and $`S`$, whereas $`Call(P,\{p\})=\{p,q\}`$ is a finite set. Also the only-if direction of Lemma 2 does not hold in case $`Tab_PPred_P`$. We provide a counterexample.
###### Example 3
Consider the following program $`P`$:
$$\{\begin{array}{ccc}p(a)\hfill & \hfill & \\ p(f(X))\hfill & \hfill & p(X),q(X)\hfill \\ q(X)\hfill & \hfill & \end{array}$$
with set of tabled predicates $`Tab_P=\{p/1\}`$ and $`S=\{p(X)\}`$. The LG-forest is shown in Figure 2.
$`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$. There is only one LG-tree in the LG-forest for $`P\{p(X)\}`$ without infinite branches. Note that the LG-tree has an infinitely branching node. But the call set $`Call(P,\{p(X)\})=\{p(X),q(a),\mathrm{},q(f^n(a)),\mathrm{}\}`$ is infinite.
Note however that, since quasi-termination requires that there are only finitely many LG-trees in the LG-forest of a query, there can only be a finite number of tabled atoms in the call set of that query. Hence, in general, the following holds.
###### Lemma 3
Let $`P`$ be a program, $`Tab_PPred_P`$ and $`SB_P^E`$.
If $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$, then, for all $`A`$ such that $`\stackrel{~}{A}S`$, $`Call(P,\{A\})B_{Tab_P}^E`$ is finite.
###### Proof
The implication follows from the fact that for every $`A`$ such that $`\stackrel{~}{A}S`$, $`B`$ is the root of an LG-tree in the LG-forest w.r.t. $`Tab_P`$ of $`P\{A\}`$ iff $`\stackrel{~}{B}(Call(P,\{A\})B_{Tab_P}^E)\{\stackrel{~}{A}\}`$. ∎
###### Example 4
Recall program $`P`$ and set $`S`$ of Example 3. We already know that $`P`$ quasi-terminates w.r.t. $`Tab_P=\{p/1\}`$ and $`S`$. Indeed, $`Call(P,\{p(X)\})B_{Tab_P}^E=\{p(X)\}`$ is finite.
### 4.2 LG-Termination
As already noted, the notion of quasi-termination only partially corresponds to our intuitive notion of a terminating execution of a query against a tabled program. Therefore, the following stronger notion of LG-termination is introduced.
###### Definition 11 (LG-termination)
Let $`P`$ be a program, $`Tab_PPred_P`$ and $`SB_P^E`$.
$`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$ iff for every atom $`A`$ such that $`\stackrel{~}{A}S`$, the LG-forest w.r.t. $`Tab_P`$ for $`P\{A\}`$ consists of a finite number of finite LG-trees.
Also, $`P`$ LG-terminates w.r.t. $`S`$ iff $`P`$ LG-terminates w.r.t. $`Pred_P`$ and $`S`$.
Note that by definition, $`P`$ LG-terminates w.r.t. $`Tab_P=\mathrm{}`$ and $`S`$ iff $`P`$ LD-terminates w.r.t. $`S`$.
Recall the program $`P`$ and set $`S`$ of Example 1. The LG-forest of $`P`$ and $`S`$ w.r.t. $`Tab_P=\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ was shown in Figure 1. Note that there are infinitely branching nodes in the LG-trees. Hence, $`P`$ does not LG-terminate w.r.t. $`Tab_P`$ and $`S`$.
Observe that if the program $`P`$ is called with an acyclic graph as input, we have LG-termination and even LD-termination. The program $`P^{^{}}`$ of the following example is obtained from $`P`$ by removing the last argument of the $`\mathrm{𝑝𝑎𝑡ℎ}/4`$ predicate in which the path is computed; the resulting predicate is named $`\mathrm{𝑟𝑒𝑎𝑐ℎ𝑎𝑏𝑙𝑒}/3`$. When $`P^{^{}}`$ is called with a cyclic graph as input, we have LG-termination (but no LD-termination).
###### Example 5
The following program $`P^{^{}}`$ computes the reachable nodes from a given node in a given graph. As in Example 1, the graph is represented as a list of terms $`e(n_1,n_2)`$, indicating that there is an edge from node $`n_1`$ to node $`n_2`$. Note that, contrary to program $`P`$ of Example 1, $`P^{^{}}`$ does not compute the paths leading from the given node to the reachable nodes.
$$\{\begin{array}{ccc}reachable(X,Ed,Y)\hfill & \hfill & edge(X,Ed,Y)\hfill \\ reachable(X,Ed,Z)\hfill & \hfill & edge(X,Ed,Y),reachable(Y,Ed,Z)\hfill \\ edge(X,[e(X,Y)|L],Y)\hfill & \hfill & \\ edge(X,[e(X_1,X_2)|L],Y)\hfill & \hfill & edge(X,L,Y)\hfill \end{array}$$
Let $`S^{^{}}=\{\mathrm{𝑟𝑒𝑎𝑐ℎ𝑎𝑏𝑙𝑒}(a,[e(a,b),e(b,a)],Y)\}`$ and $`Tab_P^{^{}}=\{\mathrm{𝑟𝑒𝑎𝑐ℎ𝑎𝑏𝑙𝑒}/3\}`$.. Then,
$`Call(P^{^{}},S^{^{}})`$ $`=`$ $`\{\mathrm{𝑟𝑒𝑎𝑐ℎ𝑎𝑏𝑙𝑒}(a,[e(a,b),e(b,a)],Y),\mathrm{𝑟𝑒𝑎𝑐ℎ𝑎𝑏𝑙𝑒}(b,[e(a,b),e(b,a)],Y),`$
$`\mathrm{𝑒𝑑𝑔𝑒}(a,[e(a,b),e(b,a)],Y),\mathrm{𝑒𝑑𝑔𝑒}(a,[e(b,a)],Y),\mathrm{𝑒𝑑𝑔𝑒}(a,[],Y),`$
$`\mathrm{𝑒𝑑𝑔𝑒}(b,[e(a,b),e(b,a)],Y),\mathrm{𝑒𝑑𝑔𝑒}(b,[e(b,a)],Y),\mathrm{𝑒𝑑𝑔𝑒}(b,[],Y)\}`$
The LG-forest w.r.t. $`Tab_P^{^{}}`$ for $`P^{^{}}`$ and $`S^{^{}}`$ is shown in Figure 3. Note that there are 2 LG-trees (only 2 tabled atoms are called), both with finite branches and finitely branching nodes (the selected tabled atoms have a finite number of computed answers). $`P^{^{}}`$ LG-terminates w.r.t. $`Tab_P^{^{}}`$ and $`S^{^{}}`$. Observe that $`P^{^{}}`$ does not LD-terminate w.r.t. $`S^{^{}}`$.
As illustrated by the above examples, the notion of LG-termination is strictly stronger than the notion of quasi-termination. Also, LD-termination implies (and is strictly stronger than) LG-termination.
###### Lemma 4
Let $`P`$ be a program, $`Tab_PPred_P`$, and $`SB_P^E`$.
If $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$, then $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$.
If $`P`$ LD-terminates w.r.t. $`S`$, then $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$.
###### Proof
The first statement is trivial by definition. For the second statement, this is a corollary of the following Proposition 2 with $`Tab_1=\mathrm{}`$ and $`Tab_2=Tab_P`$. ∎
Note that, if a program quasi-terminates w.r.t. a tabling and a set $`S`$ and the program does not LG-terminate w.r.t. that tabling and $`S`$, then there does not exist a tabling such that the program LG-terminates w.r.t. that tabling and $`S`$. This is proven in the following lemma.
###### Lemma 5
Let $`P`$ be a program and $`SB_P^E`$ a set of queries. Suppose there exists a tabling $`Tab_P^{}Pred_P`$ such that $`P`$ quasi-terminates w.r.t. $`Tab_P^{}`$ and $`S`$ and $`P`$ does not LG-terminate w.r.t. $`Tab_P^{}`$ and $`S`$.
Then for all tablings $`Tab_PPred_P`$, $`P`$ does not LG-terminate w.r.t. $`Tab_P`$ and $`S`$.
###### Proof
Let $`Tab_P^{}Pred_P`$ be such that $`P`$ quasi-terminates w.r.t. $`Tab_P^{}`$ and $`S`$ and $`P`$ does not LG-terminate w.r.t. $`Tab_P^{}`$ and $`S`$. Then, there exists a predicate $`pTab_P^{}Rec_P`$ such that there is a $`p`$-atom in $`Call(P,S)`$ which has infinitely may different (nonvariant) computed answers. Since tabling does not influence the set of call patterns nor the set of computed answer substitutions (see e.g. \[20, Theorem 2.1\]), there cannot exist a tabling such that $`P`$ LG-terminates w.r.t. that tabling and the set $`S`$. ∎
Consider two tablings $`Tab_1,Tab_2Pred_P`$ for a program $`P`$. Suppose $`Tab_1Tab_2`$ (hence $`NTab_1NTab_2`$). The next proposition studies the relationship between the LG-termination of $`P`$ w.r.t. these two tablings.
###### Proposition 2
Let $`P`$ be a program. Let $`Pred_P=Tab_1NTab_1`$ and $`Pred_P=Tab_2NTab_2`$. Suppose $`Tab_1Tab_2`$. Let $`SB_P^E`$.
If $`P`$ $`LG`$-terminates w.r.t. $`Tab_1`$ and $`S`$, then $`P`$ $`LG`$-terminates w.r.t. $`Tab_2`$ and $`S`$.
###### Proof
Let $`A`$ be an atom such that $`\stackrel{~}{A}S`$. Let $`_1`$ be the LG-forest w.r.t. $`Tab_1`$ of $`P\{A\}`$ and let $`_2`$ be the LG-forest w.r.t. $`Tab_2`$ of $`P\{A\}`$. We know that $`_1`$ consists of a finite number of finite LG-trees. So, $`\mathrm{}Call(P,\{A\})<\mathrm{}`$, hence, $`\mathrm{}(Call(P,\{A\})B_{Tab_2}^E)<\mathrm{}`$ and $`_2`$ consists of a finite number of LG-trees. We prove that the LG-trees of $`_2`$ are finite. Since each LG-tree in $`_2`$ can be extended to obtain an LG-tree in $`_1`$, this follows from the finiteness of the LG-trees in $`_1`$. ∎
Note that this proposition does not hold for quasi-termination as is shown in the following example.
###### Example 6
Recall the program $`P`$ and set $`S=\{p(X)\}`$ of Example 3. Let $`Tab_1=\{p/1\}`$ (as in Example 3) and $`Tab_2=\{p/1,q/1\}`$. Then, $`P`$ quasi-terminates w.r.t. $`Tab_1`$ and $`S`$ (the LG-forest in this case was shown in Figure 2). But, as is shown in Figure 4, $`P`$ does not quasi-terminate w.r.t. $`Tab_2`$ and $`S`$.
### 4.3 Characterization of LG-termination through quasi-termination
We now relate the notions of quasi-termination and LG-termination in a more detailed way. By definition, quasi-termination only corresponds to part of the LG-termination notion; it fails to capture non-termination caused by an infinitely branching node in an LG-tree. Note that if an LG-forest contains a tree with an infinitely branching node, then there is an LG-tree in the forest which is infinitely branching in a node which contains a query with a selected atom which is tabled and recursive. This observation leads to the following lemma. We denote the set of tabled, recursive predicates in a program $`P`$ with $`TR_P`$:
$$TR_P=Tab_PRec_P.$$
###### Lemma 6
Let $`P`$ be a program, $`Tab_PPred_P`$, and $`SB_P^E`$.
$`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$ iff $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$ and for all $`A`$ such that $`\stackrel{~}{A}S`$, the set of LD-computed answers for atoms in $`Call(P,\{A\})B_{TR_P}^E`$ is finite.
###### Proof
$`:`$ Suppose $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$. Then $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$. It is easy to see that, since for every $`A`$ such that $`\stackrel{~}{A}S`$ the LG-forest for $`P\{A\}`$ consists of a finite number of finite trees, the set of computed answers for atoms in $`Call(P,\{A\})`$ is finite.
$`:`$ Suppose that $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$ and for all $`A`$ such that $`\stackrel{~}{A}S`$ the set of LD-computed answers for atoms in $`Call(P,\{A\})B_{TR_P}^E`$ is finite. We prove that $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$. Let $`A`$ be an atom such that $`\stackrel{~}{A}S`$. We already know that the LG-forest $``$ of $`P\{A\}`$ consists of a finite number of LG-trees without infinite branches. We prove by contradiction that these LG-trees are finitely branching. Suppose there is an LG-tree in $``$ which is infinitely branching. Then, there is an LG-tree in $``$ with an infinitely branching node, which contains a query which has a tabled, recursive atom at the leftmost position. That is, there is an atom in $`Call(P,\{A\})B_{TR_P}^E`$ which has infinitely many computed answers. This gives a contradiction. ∎
It follows from the proof of Lemma 6 that, if $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$, the set of computed answers for atoms in $`Call(P,\{A\})`$ is finite for all $`A`$ such that $`\stackrel{~}{A}S`$.
Based on the observation in Lemma 6, we next define a transformation on programs, called the answer-transformation, such that LG-termination of a program $`P`$ is equivalent to the quasi-termination of the program $`P^a`$ obtained by applying the answer-transformation on $`P`$.
###### Definition 12 (a(nswer)-transformation)
Let $`P`$ be a program and $`Tab_PPred_P`$. The a-transformation on $`P`$ and $`Tab_P`$ is defined as follows:
* For a clause $`C=HB_1,\mathrm{},B_n`$ in $`P`$, we define
$$C^a=HB_1,B_1^{},\mathrm{},B_n,B_n^{}$$
with $`B_i^{}`$ defined as follows (suppose $`B_i=p(t_1,\mathrm{},t_n)`$):
if $`pTab_P`$ and $`pRel(H)`$ then $`B_i^{}=p^a(t_1,\mathrm{},t_n)`$, where $`p^a/n`$ is a new predicate, else $`B_i^{}=\mathrm{}`$.
Let $`TR_P^a=\{p^a/n|p/nTR_P\}`$ (recall that $`TR_P=Tab_PRec_P`$).
* For the program $`P`$, we define
$$P^a=\{C^a|CP\}\{p^a(X_1,\mathrm{},X_n)|p^a/nTR_P^a\}.$$
* The set of tabled predicates of the program $`P^a`$ is defined as
$$Tab_{P^a}=Tab_PTR_P^a.$$
###### Example 7
Let $`P`$ be the program of Example 1, with $`Tab_P=\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$. The a-transformation, $`P^a`$, of $`P`$ is the following program:
$$\{\begin{array}{ccc}path(X,Ed,Y,[Y])\hfill & \hfill & edge(X,Ed,Y)\hfill \\ path(X,Ed,Z,[Y|L])\hfill & \hfill & edge(X,Ed,Y),path(Y,Ed,Z,L),\hfill \\ & & path^a(Y,Ed,Z,L)\hfill \\ edge(X,[e(X,Y)|L],Y)\hfill & \hfill & \\ edge(X,[e(X_1,X_2)|L],Y)\hfill & \hfill & edge(X,L,Y)\hfill \\ path^a(X,Ed,Y,L)\hfill & \hfill & \end{array}$$
with $`Tab_{P^a}=\{\mathrm{𝑝𝑎𝑡ℎ}/4,\mathrm{𝑝𝑎𝑡ℎ}^a/4\}`$.
It is easy to see that $`Call(P,S)=Call(P^a,S)B_P^E`$. Also, if we denote with $`cas(P,\{p(\overline{t})\})`$ the set of computed answer substitutions for $`p(\overline{t})`$ in $`P`$, then $`cas(P,\{p(\overline{t})\})`$ $`=cas(P^a,\{p(\overline{t})\})`$ for all $`p(\overline{t})B_P^E`$. It is important to note that, if we have a query $`p(\overline{t})B_{TR_P}^E`$ to the program $`P`$, then $`p(\overline{t})\sigma `$ is a computed answer if $`p^a(\overline{t})\sigma `$ $`Call(P^a,\{p(\overline{t})\})`$. This is in fact the main purpose of the transformation.
###### Theorem 4.1 (characterisation of LG-termination in terms of quasi-termination)
Let $`P`$ be a program, $`Tab_PPred_P`$ and $`SB_P^E`$.
$`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$ iff $`P^a`$ quasi-terminates w.r.t. $`Tab_{P^a}`$ and $`S`$.
###### Proof
$`:`$ Suppose $`P^a`$ is quasi-terminating w.r.t. $`Tab_{P^a}`$ and $`S`$. Let $`A`$ be an atom such that $`\stackrel{~}{A}S`$. Let $``$ be the LG-forest w.r.t. $`Tab_P`$ of $`P\{A\}`$. We prove that $``$ consists of a finite number of finite LG-trees.
We know that the LG-forest $`^a`$ w.r.t. $`Tab_{P^a}`$ of $`P^a\{A\}`$ is a finite set of LG-trees, without infinite branches. It is easy to see that hence, $``$ consists also of a finite number of trees without infinite branches. We prove that the LG-trees in $``$ are finitely branching. Suppose this is not the case, i.e. there is an LG-tree in $``$ which is infinitely branching. Then, there is an LG-tree $`T`$ in $``$ which is infinitely branching in a non-root node, which is a query with leftmost atom $`p(t_1,\mathrm{},t_n)`$, with $`pTR_P`$, which is directly descending from an atom $`q(s_1,\mathrm{},s_m)`$, with $`pq`$, via a recursive clause $`C=q(u_1,\mathrm{},u_m)\mathrm{},p(v_1,\mathrm{},v_n),\mathrm{}`$. Let $`T^a`$ be the LG-tree in $`^a`$ corresponding to $`T`$. Note that the clause $`C^a`$ instead of $`C`$ is used in $`T^a`$. Because of this, the atom to the right of $`p(t_1,\mathrm{},t_n)`$ in the infinitely branching node is $`p^a(t_1,\mathrm{},t_n)`$. Thus, $`^a`$ consists of a infinite number of LG-trees (there are an infinite number of LG-trees with predicate $`p^a`$ in the root). This gives a contradiction.
$`:`$ Suppose $`P`$ is LG-terminating w.r.t. $`Tab_P`$ and $`S`$. Let $`A`$ be an atom such that $`\stackrel{~}{A}S`$. Let $``$ be the LG-forest w.r.t. $`Tab_P`$ of $`P\{A\}`$. Then, $``$ consists of a finite number of finite LG-trees. Let $`^a`$ be the LG-forest w.r.t. $`Tab_{P^a}`$ of $`P^a\{A\}`$. By definition of the $`a`$-transformation, we see that $`^a`$ also consists of a finite number of finite LG-trees. Hence, $`P^a`$ quasi-terminates (and even LG-terminates) w.r.t. $`Tab_{P^a}`$ and $`S`$. ∎
###### Example 8 (Example 7 continued)
The LG-forest w.r.t. $`Tab_P`$ of $`P`$ and
$`\{path(a,[e(a,b),e(b,a)],Y,L)\}`$ was shown in Figure 1. Note that the trees are infinitely branching and hence $`P`$ does not LG-terminate w.r.t. $`Tab_P`$ and $`\{\mathrm{𝑝𝑎𝑡ℎ}(a,[e(a,b),e(b,a)],Y,L)\}`$.
In Figure 5, the LG-forest of the program $`P^a`$ and $`\{\mathrm{𝑝𝑎𝑡ℎ}(a,[e(a,b),e(b,a)],Y,L)\}`$ w.r.t. $`Tab_{P^a}`$ is shown. Note that there are infinitely many LG-trees in the forest; $`P^a`$ does not quasi-terminate w.r.t. $`Tab_{P^a}`$ and $`\{\mathrm{𝑝𝑎𝑡ℎ}(a,[e(a,b),e(b,a)],Y,L)\}`$.
## 5 Conditions for Termination of Tabled Logic Programs
In this section, we give sufficient conditions for the notions of quasi-termination and LG-termination. We prove that these conditions are also necessary in case the tabling satisfies the property of being well-chosen. First, we want to note that the termination conditions are adapted from the acceptability notion for LD-termination defined in , and not from the more “standard” definition of acceptability by Apt and Pedreschi in . The reason for this choice is that the quasi-termination as well as the LG-termination property of a tabled program and query is *not* closed under substitution. To see this, consider the following example from .
###### Example 9
Let $`p/2`$ be a tabled predicate defined by the following clause.
$$p(f(X),Y)p(X,Y)$$
Then, the query $`p(X,Y)`$ terminates while $`p(X,X)`$ does not.
The acceptability notion in is expressed in terms of ground instances of clauses and its associated notion of LD-termination is expressed in terms of the set of all queries that are bounded under the given level mapping. Such sets are closed under substitution. Because quasi-termination lacks invariance under substitution, we need a stronger notion of acceptability, capable of treating *any* set of queries.
We next introduce the notion of well-chosen tabling w.r.t. a program. If the tabling is well-chosen, we are able to give a necessary and sufficient condition for quasi-termination and for LG-termination. If the tabling is not well-chosen, the condition is still sufficient.
We first introduce some notation. Let $`P`$ be a program and let $`G_P`$ be the dependency graph of the predicates of $`P`$. For a tabling $`Tab_P`$ for $`P`$ and predicates $`p,qNTab_P`$ with $`pq`$, let $`C_1(p,q),`$ $`C_2(p,q)`$ and $`C_3(p,q)`$ denote the following disjoint cases:
No cycle of directed arcs in $`G_P`$ containing $`p`$ and $`q`$ contains a predicate from $`Tab_P`$.
All cycles of directed arcs in $`G_P`$ containing $`p`$ and $`q`$ contain at least one predicate from $`Tab_P`$.
There is a cycle of directed arcs in $`G_P`$ containing $`p`$ and $`q`$ which contains no predicate from $`Tab_P`$ and there is a cycle of directed arcs in $`G_P`$ containing $`p`$ and $`q`$ which contains a predicate from $`Tab_P`$.
Note that $`C_1(p,q),`$ $`C_2(p,q)`$ and $`C_3(p,q)`$ depend on the program $`P`$ (more precisely on the dependency graph $`G_P`$ of $`P`$) and on the tabling $`Tab_P`$ for $`P`$. When referring to one of these three cases, it will always be clear from the context which program and tabling are under consideration. Given a program $`P`$ and tabling $`Tab_P`$, for all predicates $`p,qNTab_P`$ with $`pq`$, exactly one of the cases $`C_1(p,q),`$ $`C_2(p,q)`$ or $`C_3(p,q)`$ holds.
###### Example 10
Consider the following three propositional programs $`P`$, $`P^{^{}}`$ and $`P^{^{\prime \prime }}`$:
$$\begin{array}{ccccc}P:\{\begin{array}{ccc}a\hfill & \hfill & b\hfill \\ b\hfill & \hfill & c\hfill \\ c\hfill & \hfill & b\hfill \end{array}\hfill & & P^{^{}}:\{\begin{array}{ccc}a\hfill & \hfill & b\hfill \\ b\hfill & \hfill & c\hfill \\ c\hfill & \hfill & a\hfill \end{array}\hfill & & P^{^{\prime \prime }}:\{\begin{array}{ccc}a\hfill & \hfill & b\hfill \\ b\hfill & \hfill & c\hfill \\ c\hfill & \hfill & a\hfill \\ c\hfill & \hfill & b\hfill \end{array}\hfill \end{array}$$
with $`Tab_P=Tab_P^{^{}}=Tab_{P^{^{\prime \prime }}}=\{a/0\}`$.
For the program $`P`$, we have that $`C_1(b,c)`$ holds. For the program $`P^{^{}}`$, we have that $`C_2(b,c)`$ holds. For the program $`P^{^{\prime \prime }}`$, we have that $`C_3(b,c)`$ holds.
We next define the notion of well-chosen tabling w.r.t. a program $`P`$. A tabling for $`P`$ is well-chosen w.r.t. $`P`$ if it is such that the third case $`C_3`$ never occurs.
###### Definition 13 (well-chosen tabling (w.r.t. a program))
Let $`P`$ be a program. The tabling $`Tab_P`$ is called well-chosen w.r.t. the program $`P`$ iff for every $`p,qNTab_P`$ such that $`pq`$, either $`C_1(p,q)`$ or $`C_2(p,q)`$ holds.
Note that in case $`Tab_P`$ is well-chosen w.r.t. $`P`$, we have that if $`p,q,rNTab_P`$ and $`pqr`$ and $`C_1(p,q)`$ (resp. $`C_2(p,q)`$) holds, then $`C_1(q,r)`$ (resp. $`C_2(q,r)`$) holds. In the special case that $`NTab_P\{pPred_P|p`$ is a non-recursive or only directly recursive predicate$`\}`$ or that $`NTab_P=\mathrm{}`$ (i.e. $`Tab_P=Pred_P`$), the tabling $`Tab_P`$ is well-chosen w.r.t. $`P`$.
###### Example 11
Recall the programs $`P`$, $`P^{^{}}`$ and $`P^{^{\prime \prime }}`$ of Example 10. The tabling $`\{a/0\}`$ is well-chosen w.r.t. $`P`$ and $`P^{^{}}`$, but not w.r.t. $`P^{^{\prime \prime }}`$.
### 5.1 Quasi-Termination
We now introduce the notion of quasi-acceptability, in general a sufficient condition for quasi-termination. In case the tabling is well-chosen, quasi-acceptability is also a necessary condition for quasi-termination.
###### Definition 14 (quasi-acceptability)
Let $`P`$ be a program, $`Tab_PPred_P`$, and $`SB_P^E`$. $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$ and $`S`$ iff there is a level mapping $`|.|`$ on $`B_P^E`$ such that for all $`A`$ such that $`\stackrel{~}{A}S`$, $`|.|`$ is finitely partitioning on $`Call(P,\{A\})B_{Tab_P}^E`$ and such that
* for every atom $`A`$ such that $`\stackrel{~}{A}Call(P,S)`$,
* for every clause $`HB_1,\mathrm{},B_n`$ in $`P`$, such that $`mgu(A,H)=\theta `$ exists,
* for every $`1in`$,
* for every $`cas`$ $`\theta _{i1}`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$\begin{array}{cc}|A||B_i\theta \theta _{i1}|\hfill & \\ \text{and}\hfill & \\ |A|>|B_i\theta \theta _{i1}|\hfill & \text{if}Rel(A)Rel(B_i)NTab_P\text{and}\hfill \\ & C_2(Rel(A),Rel(B_i))\text{does not hold}.\hfill \end{array}$$
###### Theorem 5.1 ((necessary and) sufficient condition for quasi-termination)
Let $`P`$ be a program, $`Tab_PPred_P`$ and $`SB_P^E`$.
If $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$ and $`S`$, then $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$.
If the tabling $`Tab_P`$ is well-chosen w.r.t. $`P`$, then also the converse holds, i.e. $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$ and $`S`$ iff $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$.
###### Proof
$`:`$ Suppose that $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$, $`S`$ and a level mapping $`|.|`$. We prove that $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$. Let $`A`$ be an atom such that $`\stackrel{~}{A}S`$, let $``$ be the LG-forest w.r.t. $`Tab_P`$ of $`P\{A\}`$.
* $``$ consists of a finite number of LG-trees, i.e. $`\mathrm{}(Call(P,\{A\})B_{Tab_P}^E)<\mathrm{}`$.
Due to the quasi-acceptability condition, any call in $`Call(P,\{A\})`$ directly descending from $`A`$, say $`B`$, is such that $`|A||B|`$. The same holds recursively for the atoms descending from $`B`$. Thus, the level mapping of any call, recursively descending from $`A`$, is smaller than or equal to $`|A|\mathrm{}`$. Since $`|.|`$ is finitely partitioning on $`Call(P,\{A\})B_{Tab_P}^E`$, we have that: $`\mathrm{}(_{n|A|}|.|^1(n)Call(P,\{A\})B_{Tab_P}^E)<\mathrm{}`$. Hence, $`\mathrm{}(Call(P,\{A\})B_{Tab_P}^E)<\mathrm{}`$, i.e. $``$ consists of a finite number of trees.
* The LG-trees in $``$ have finite branches.
Suppose there is a tree in $``$ with an infinite branch. This infinite branch contains an infinite directed subsequence $`G_0,G_1,\mathrm{}`$. It is easy to see that the leftmost atoms in the nodes of this infinite directed subsequence all are $`NTab_P`$-atoms (because $`Tab_P`$-atoms are resolved using answers). There is a $`n\mathrm{}`$, such that each $`G_i`$, $`in`$, has as leftmost atom $`A_i`$ and for all $`in`$, $`Rel(A_i)Rel(A_{i+1})`$ and $`C_2(Rel(A_i),Rel(A_{i+1}))`$ does not hold. Because of the quasi-acceptability condition, $`|A_i|>|A_{i+1}|`$, for all $`in`$. This gives a contradiction.
$`:`$ Suppose that the tabling $`Tab_P`$ is well-chosen w.r.t. $`P`$ and suppose that $`P`$ quasi-terminates w.r.t. $`S`$. We have to construct a level mapping $`|.|`$ such that $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$, $`S`$ and this level mapping $`|.|`$. We will only define $`|.|`$ on elements of $`Call(P,S)`$. On elements of the complement of $`Call(P,S)`$ in $`B_P^E`$, $`|.|`$ can be assigned any value, as these elements do not turn up in the quasi-acceptability condition.
In order to define $`|.|`$ on $`Call(P,S)`$, consider the $`Call`$-$`Gr(P,S)`$-graph (Definition 4). Consider a strongly connected component $`C`$ in $`Call`$-$`Gr(P,S)`$.
Then, there is at least one $`Tab_P`$-atom in $`C`$. To see this, suppose this is not the case. Consider a cyclic path $`p`$ in $`C`$. This consists only of $`NTab_P`$-atoms. But then, because of Proposition 1, there is an infinite branch in a tree of the LG-forest of an element of $`S`$. This gives a contradiction.
Also, there is only a finite number of $`Tab_P`$-atoms in $`C`$. To see this, suppose this is not the case. Then there is an infinitely long path $`p`$ through infinitely many $`Tab_P`$-atoms of $`C`$. Because of Proposition 1, there is an infinite number of $`Tab_P`$-atoms selected in a derivation of an element of $`S`$, i.e. there are infinitely many trees in the LG-forest of that element of $`S`$. This gives a contradiction.
For every two non-tabled atoms, say $`p(\overline{t})`$ and $`q(\overline{s})`$, in $`C`$ (note that thus $`pq`$), $`C_1(p,q)`$ does not hold (since there is at least one $`Tab_P`$-atom in $`C`$). Thus, since the tabling is well-chosen, $`C_2(p,q)`$ holds.
Define $`\overline{CG}`$ as the graph obtained from $`Call`$-$`Gr(P,S)`$ by replacing any strongly connected component by a single contracting node and replacing any arc from $`Call`$-$`Gr(P,S)`$ pointing to (resp. from) any node in that strongly connected component by an arc to (resp. from) that contracting node. $`\overline{CG}`$ does not have any (non-trivial) strongly connected components. Moreover, any strongly connected component from $`Call`$-$`Gr(P,S)`$ that was collapsed into a contracting node of $`\overline{CG}`$ necessarily contains at least one and at most a finite number of $`Tab_P`$-atoms.
Note now that each path in $`\overline{CG}`$ which is not cyclic (there are only trivial cycles in $`\overline{CG}`$) is finite. This also follows directly from Proposition 1.
Note also that it is possible that $`\overline{CG}`$ has an infinitely branching (possibly contracting) node. Let $`A`$ be an atom in that infinitely branching node. It follows from Lemma 3 that, because $`P`$ quasi-terminates w.r.t. $`S`$, $`\mathrm{}(\{B|B`$ is a descendant of $`A`$ in $`\overline{CG}\}`$ $``$ $`B_{Tab_P}^E)<`$ $`\mathrm{}`$.
We now construct $`\overline{\overline{CG}}`$ from $`\overline{CG}`$ starting from the top nodes $`N_1`$ downwards as follows:
* replace all direct descendants of $`N_1`$ in $`\overline{CG}`$ different from $`N_1`$, by a single contracting node $`N_2`$;
* replace any arc from $`\overline{CG}`$ pointing to (resp. from) any node in that (possibly infinite) set of direct descendants by an arc to (resp. from) that contracting node $`N_2`$;
* repeat this for the nodes $`N_2`$.
This process stops because, as we already noted, each path in $`\overline{CG}`$ which is not cyclic is finite. It is easy to see that $`\overline{\overline{CG}}`$ is a graph in which each node has at most one direct descendant different from itself. Also, each node in $`\overline{\overline{CG}}`$ consists of a (possibly infinite) set of nodes of $`Call`$-$`Graph(P,S)`$ which contains only finitely many $`Tab_P`$-atoms.
We define the level mapping $`|.|`$ as follows. Consider the layers of $`\overline{\overline{CG}}`$ (there are only a finite number of layers). Let layer-0 be the set of leaves in $`\overline{\overline{CG}}`$. We assign to these nodes a number in $`\mathrm{}`$, such that all nodes get a different number. Then, we move up to the next layer in $`\overline{\overline{CG}}`$. This layer, layer-1, consists of all nodes $`N`$ such that the path starting from $`N`$ has length 1. We assign to each such node $`N`$ a natural number, such that the number assigned to $`N`$ is strictly larger than the number assigned to its descendant (in the previous step). We continue this process layer by layer. The value of the level mapping $`|.|`$ on elements of $`Call(P,S)`$ is defined as follows: all calls contained in the node $`N`$ receive the number assigned to the node $`N`$.
We prove that $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$, $`S`$ and this level mapping $`|.|`$.
* for every $`AS`$, $`|.|`$ is finitely partitioning on $`B_{Tab_P}^ECall(P,\{A\})`$.
Note that $`|.|`$ is even finitely partitioning on $`B_{Tab_P}^ECall(P,S)`$. This is because each (contracting) node of $`\overline{\overline{CG}}`$ contains only a finite number of $`Tab_P`$-atoms and because of the construction of $`|.|`$.
* Let $`A`$ be an atom such that $`\stackrel{~}{A}Call(P,S)`$, let $`HB_1,\mathrm{},B_n`$ be a clause in $`P`$, such that $`mgu(A,H)=\theta `$ exists, let $`\theta _{i1}`$ be a $`cas`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
+ then $`|A||B_i\theta \theta _{i1}|`$.
This is because there is a directed arc from $`A`$ to $`B_i\theta \theta _{i1}`$ in $`Call`$-$`Graph(P,S)`$ and because of the construction of $`|.|`$.
+ then $`|A|>|B_i\theta \theta _{i1}|`$ if $`Rel(A)Rel(B_i)`$ $`NTab_P`$ and $`C_2(Rel(A),Rel(B_i))`$ does not hold (i.e. $`C_1(Rel(A),Rel(B_i))`$ holds).
There is a directed arc in $`Call`$-$`Graph(P,S)`$ from $`A`$ to $`B_i\theta \theta _{i1}`$. Note that $`A`$ and $`B_i\theta \theta _{i1}`$ do not belong to the same strongly connected component of $`Call`$-$`Graph(P,S)`$. This is because $`C_1(Rel(A),Rel(B_i))`$ holds. So, $`A`$ and $`B_i\theta \theta _{i1}`$ belong to a different layer and $`B_i\theta \theta _{i1}`$ is a direct descendant of $`A`$. Hence, because of the construction of $`|.|`$, $`|A|>|B_i\theta \theta _{i1}|`$. ∎
###### Example 12
Recall the programs $`P`$ and $`P^{^{}}`$ with $`Tab_P=Tab_P^{^{}}=\{a/0\}`$ of Example 10. Let $`S=\{a\}`$. The LG-forests for $`P\{a\}`$ and $`P^{^{}}\{a\}`$ are shown in Figure 6.
$`P`$ does not quasi-terminate w.r.t. $`\{a/0\}`$ and $`S`$, whereas $`P^{^{}}`$ quasi-terminates w.r.t. $`\{a/0\}`$ and $`S`$.
This can be proven by Theorem 5.1. Recall from Example 11 that for both programs, the tablings are well-chosen. Also note that, because the programs are propositional, every level mapping is finitely partitioning on the whole Herbrand base.
Let’s first consider program $`P`$. Recall that for this program and tabling $`\{a/0\}`$ the condition $`C_1(b,c)`$ holds. Note that there is no level mapping $`|.|`$ such that $`|b|>|c|`$ and $`|c|>|b|`$ holds. Hence, the condition in Theorem 5.1 can not be satisfied and $`P`$ does not quasi-terminate w.r.t. $`\{a/0\}`$ and $`S`$.
Consider next program $`P^{^{}}`$. Recall that for this program and tabling $`\{a/0\}`$ the condition $`C_2(b,c)`$ holds. Let $`|.|`$ be the following level mapping $`|a|=|b|=|c|=0`$. With this level mapping, $`P^{^{}}`$ satisfies the condition of Theorem 5.1 and hence, $`P^{^{}}`$ quasi-terminates w.r.t. $`\{a/0\}`$ and $`S`$.
The quasi-acceptability condition is necessary only in case the tabling is well-chosen. We next give an example of a program $`P`$, a tabling $`Tab_P`$ which is not well-chosen w.r.t. $`P`$, and a set of queries $`S`$, such that $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$, but $`P`$ is not quasi-acceptable w.r.t. $`Tab_P`$ and $`S`$.
###### Example 13
Let $`P`$ be the following program:
$$\{\begin{array}{ccc}p(X)\hfill & \hfill & q(X)\hfill \\ q(X)\hfill & \hfill & r(X)\hfill \\ r(s(X))\hfill & \hfill & q(X)\hfill \\ r(X)\hfill & \hfill & p(X)\hfill \end{array}$$
with tabling $`Tab_P=\{p/1\}`$. Notice that $`Tab_P`$ is not well-chosen w.r.t. $`P`$. Let $`S=\{p(0)\}`$. $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$. We show that $`P`$ is not quasi-acceptable w.r.t. $`Tab_P`$ and $`S`$. Suppose that there exists a level mapping $`|.|`$ such that $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$, $`S`$ and this level mapping $`|.|`$ (we prove a contradiction). Then, for this level mapping, the following inequalities must hold: $`|p(0)||q(0)|`$, $`|q(0)|>|r(0)|`$ (since $`C_3(q,r)`$ holds, and so $`C_2(q,r)`$ does not hold), and $`|r(0)||p(0)|`$. Hence, $`|p(0)|>|p(0)|`$ must hold, but this gives a contradiction.
### 5.2 LG-Termination
In analogy to quasi-termination, we now present a necessary and sufficient condition for LG-termination in case the tabling is well-chosen. In case the tabling is not well-chosen, the condition is still sufficient.
Note that Theorem 4.1 already provides us with a characterisation of LG-termination of a program in terms of quasi-termination. That is, to prove the LG-termination of $`P`$ w.r.t. $`Tab_P`$ and $`S`$, it suffices to prove the quasi-termination of $`P^a`$, the a-transformation of the program $`P`$, w.r.t. $`Tab_{P^a}`$ and $`S`$. To prove quasi-termination, we can use the results of Subsection 5.1. Namely, it is sufficient (and also necessary in case the tabling is well-chosen<sup>1</sup><sup>1</sup>1 Note that if $`Tab_P`$ is well-chosen w.r.t. $`P`$, then also $`Tab_{P^a}`$ is well-chosen w.r.t. $`P^a`$.) to prove the quasi-acceptability of $`P^a`$ w.r.t. $`Tab_{P^a}`$ and $`S`$. However, the condition of quasi-acceptability on $`P^a`$ can be weakened; i.e. some of the decreases “$`|A||B_i\theta \theta _{i1}|`$” need not be checked because they can always be fulfilled. In particular, we only have to require the non-strict decrease for recursive, tabled body atoms $`B_i`$ (to obtain an LG-forest with only finitely many LG-trees) or for body atoms $`B_i`$ of the form $`p^a(t_1,\mathrm{},t_n)`$ (to obtain LG-trees which are finitely branching); the conditions on non-tabled predicates remain the same. The following notion of LG-acceptability gives this optimised condition for LG-termination of a program.
###### Definition 15 (LG-acceptability)
Let $`P`$ be a program, $`Tab_PPred_P`$ and $`SB_P^E`$. $`P`$ is LG-acceptable w.r.t. $`Tab_P`$ and $`S`$ iff
there is a level mapping $`|.|`$ on $`B_{P^a}^E`$ such that for all $`A`$ such that $`\stackrel{~}{A}S`$, $`|.|`$ is finitely partitioning on $`Call(P^a,\{A\})B_{TR_PTR_P^a}^E`$, and such that
* for every atom $`A`$ such that $`\stackrel{~}{A}Call(P^a,S)`$,
* for every clause $`HB_1,\mathrm{},B_n`$ in $`P^a`$, such that $`mgu(A,H)=\theta `$ exists,
* for every $`B_i`$ such that $`Rel(B_i)Rel(H)`$ or $`Rel(B_i)TR_P^a`$,
* for every $`cas`$ $`\theta _{i1}`$ in $`P^a`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$\begin{array}{cc}|A||B_i\theta \theta _{i1}|\hfill & \\ \text{and}\hfill & \\ |A|>|B_i\theta \theta _{i1}|\hfill & \text{if}Rel(A)Rel(B_i)NTab_P\text{and}\hfill \\ & C_2(Rel(A),Rel(B_i))\text{does not hold}.\hfill \end{array}$$
###### Theorem 5.2 ((necessary and) sufficient condition for LG-termination)
Let $`P`$ be a program, $`Tab_PPred_P`$ and $`SB_P^E`$.
If $`P`$ is LG-acceptable w.r.t. $`Tab_P`$ and $`S`$, then $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$.
If the tabling $`Tab_P`$ is well-chosen w.r.t. $`P`$, then also the converse holds, i.e. $`P`$ is LG-acceptable w.r.t. $`Tab_P`$ and $`S`$ iff $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$.
###### Proof
$`:`$ Suppose that $`P`$ is LG-acceptable w.r.t. $`Tab_P`$ and $`S`$. We prove that $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$.
Let $`A`$ be an atom such that $`\stackrel{~}{A}S`$. Let $``$ be the LG-forest w.r.t. $`Tab_P`$ of $`P\{A\}`$. We prove that $``$ consists of a finite number of finite LG-trees.
* The LG-trees in $``$ are finitely branching.
Suppose this is not the case, i.e. there is an LG-tree in $``$ which is infinitely branching. Then, there is an LG-tree $`T`$ in $``$ which is infinitely branching in a non-root node, which is a query with leftmost atom $`p(t_1,\mathrm{},t_n)`$, with $`pTR_P`$, which is directly descending from an atom $`q(s_1,\mathrm{},s_m)`$, with $`pq`$, via a recursive clause $`C=q(u_1,\mathrm{},u_m)\mathrm{},p(v_1,\mathrm{},v_n),\mathrm{}`$. Now, consider the LG-forest $`^a`$ of $`P^a\{A\}`$. Let $`T^a`$ be the LG-tree in $`^a`$ corresponding to $`T`$. Note that the clause $`C^a`$ instead of $`C`$ is used in $`T^a`$. Because of this, the atom on the right of $`p(t_1,\mathrm{},t_n)`$ in the infinitely branching node is $`p^a(t_1,\mathrm{},t_n)`$. Thus, $`^a`$ consists of a infinite number of LG-trees (there are an infinite number of LG-trees with predicate $`p^a`$ in the root). But, all these $`p^a`$-atoms directly descend from the node $`q(s_1,\mathrm{},s_m)`$ via the clause $`C^a`$ in $`P^a`$ and hence, because of the LG-acceptability condition, their value under the level mapping $`|.|`$ is smaller or equal to $`|q(s_1,\mathrm{},s_m)|`$. Because $`|.|`$ is finitely partitioning on $`Call(P^a,\{A\})B_{TR_P^a}^E`$, this gives a contradiction.
* $``$ consists of a finite number of LG-trees, i.e. $`\mathrm{}(Call(P,\{A\})B_{Tab_P}^E)<\mathrm{}`$.
Suppose this is not the case. A first possible reason for an infinite number of LG-trees in $``$ is an infinitely branching LG-tree in $``$. But we already proved that this does not occur. The other possibility is that there exists an infinite LD-derivation of $`A`$ in $`P`$ which contains an infinite directed subsequence, such that this infinite directed subsequence has a tail $`G_n,G_{n+1},\mathrm{}`$ with $`G_i=A_i,𝒜_i`$, $`in`$, such that $`\{\stackrel{~}{A}_i|in\}B_{Tab_P}^E`$ is an infinite set and $`Rel(A_i)Rel(A_{i+1})`$ for all $`in`$. So, $`\{\stackrel{~}{A}_i|in\}B_{TR_P}^E`$. Since $`\stackrel{~}{A}_iCall(P,\{A\})B_{TR_P}^E`$ $``$ $`Call(P^a,\{A\})B_{TR_PTR_P^a}^E`$, and since $`|.|`$ is finitely partitioning on this set and $`|A_i||A_{i+1}|`$ for all $`in`$ (by the LG-acceptability condition), this gives a contradiction.
* The LG-trees in $``$ have finite branches.
The same argumentation as in the proof of Theorem 5.1 can be applied here.
$`:`$ Suppose that the tabling $`Tab_P`$ is well-chosen w.r.t. $`P`$ and suppose that $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$. We prove that there exists a level mapping $`|.|`$ such that $`P`$ is LG-acceptable w.r.t. $`Tab_P`$, $`S`$ and this level mapping $`|.|`$.
Since $`P`$ LG-terminates w.r.t. $`Tab_P`$ and $`S`$, we know by Theorem 4.1 that $`P^a`$ quasi-terminates w.r.t. $`Tab_{P^a}`$ and $`S`$. Note that, since $`Tab_P`$ is well-chosen w.r.t. $`P`$, $`Tab_{P^a}`$ is well-chosen w.r.t. $`P^a`$. By Theorem 5.1, there exists a level mapping $`|.|`$ such that $`P^a`$ is quasi-acceptable w.r.t. $`Tab_{P^a}`$, $`S`$ and this level mapping $`|.|`$. It is straightforward to verify that $`P`$ is LG-acceptable w.r.t. $`Tab_P`$, $`S`$ and this level mapping $`|.|`$. (Note that, as we already discussed in the beginning of this subsection, the level mapping obtained in this way satisfies more conditions than required by the notion of LG-acceptability.) ∎
###### Example 14
Recall the part $`R`$ of the grammar program (Section 3.1) which recognizes the language $`a^nb`$:
$$\begin{array}{cc}R:\hfill & \{\begin{array}{ccc}s(Si,So)\hfill & \hfill & a(Si,S),S=[b|So]\hfill \\ a(Si,So)\hfill & \hfill & a(Si,S),a(S,So)\hfill \\ a(Si,So)\hfill & \hfill & Si=[a|So]\hfill \end{array}\hfill \end{array}$$
with $`Tab_R=\{a/2\}`$. We show that $`R`$ LG-terminates w.r.t. $`\{a/2\}`$ and $`S=\{s(si,So)|si`$ is a ground list consisting of constants $`a,b`$ and $`So`$ is a variable$`\}`$. Consider the a-transformation of $`R`$:
$$\begin{array}{cc}R^a:\hfill & \{\begin{array}{ccc}s(Si,So)\hfill & \hfill & a(Si,S),S=[b|So]\hfill \\ a(Si,So)\hfill & \hfill & a(Si,S),a^a(Si,S),a(S,So),a^a(S,So)\hfill \\ a(Si,So)\hfill & \hfill & Si=[a|So]\hfill \\ a^a(Si,So)\hfill & \hfill & \end{array}\hfill \end{array}$$
with $`Tab_{R^a}=\{a/2,a^a/2\}`$. When applying Theorem 5.2, we only have to consider the second clause of $`R^a`$. Note that, for all $`a(t1,t2)Call(R^a,\{s(si,So)\})`$, $`t1`$ is a sublist of $`si`$ and $`t2`$ is a variable. Also, for all $`a^a(v1,v2)Call(R^a,\{s(si,So)\})`$, $`v1`$ is a sublist of $`si`$ and $`v2`$ is a (strict) sublist of $`v1`$. Let $`|.|`$ be the following level mapping:
$$\begin{array}{cc}|a(t1,t2)|\hfill & =2t1_l\hfill \\ |a^a(v1,v2)|\hfill & =v1_l+v2_l\hfill \end{array}$$
where $`._l`$ is the list-length norm<sup>2</sup><sup>2</sup>2The list-length norm is defined as follows:
$$\{\begin{array}{ccc}[h|t]_l\hfill & =1+t_l\hfill & \\ u_l\hfill & =0\hfill & \text{if}u[h|t].\hfill \end{array}$$
. The level mapping $`|.|`$ is finitely partitioning on the whole set $`Call(R^a,S)B_{\{a/2,a^a/2\}}^E`$. It can be easily verified that $`R`$ and $`S`$, together with $`|.|`$, satisfy the conditions of Theorem 5.2. Hence, $`R`$ LG-terminates w.r.t. $`\{a/2\}`$ and $`S`$.
## 6 Modular Termination Proofs for Tabled Logic Programs
In the context of programming in the large, it is important to be able to obtain modular termination proofs, i.e. proofs built by combining termination proofs of separate components of the program. Starting from the quasi- and LG-acceptability conditions, we present modular proofs for quasi-termination in Subsections 6.1 and 6.3, and for LG-termination in Subsection 6.2. We consider the union $`PR`$ of two programs $`P`$ and $`R`$, where $`P`$ extends<sup>3</sup><sup>3</sup>3Recall that a program $`P`$ extends a program $`R`$ iff no predicate defined in $`P`$ occurs in $`R`$. $`R`$, and we prove the quasi/LG-termination of $`PR`$ by imposing conditions on the two components $`P`$ and $`R`$.
In order to fix a notation, for $`Pred_{PR}=Tab_{PR}NTab_{PR}`$, let
$$\begin{array}{ccc}Tab_P=Tab_{PR}Pred_P\hfill & ,\hfill & NTab_P=NTab_{PR}Pred_P\hfill \\ Tab_R=Tab_{PR}Pred_R\hfill & ,\hfill & NTab_R=NTab_{PR}Pred_R.\hfill \end{array}$$
So the tabling of the union $`PR`$ determines the tabling of the components $`P`$ and $`R`$. Note that $`Tab_P`$ also contains predicates which are tabled in $`PR`$ but defined in $`R`$.
In the following we give modular termination proofs for the union $`PR`$ of two programs $`P`$ and $`R`$ where:
1. $`P`$ extends $`R`$.
2. $`P`$ extends $`R`$ and no defined predicate in $`P`$ is tabled ($`Def_PNTab_P`$).
3. $`P`$ extends $`R`$ and all defined predicates in $`P`$ are tabled ($`Def_PTab_P`$).
4. $`P`$ extends $`R`$ and $`R`$ extends $`P`$.
Note that points 2, 3 and 4 are special cases of the first one. The reason for treating them separately is because they occur quite often in practice and, more importantly, because in these special cases, simpler modular termination conditions can be given.
### 6.1 Modular Conditions for Quasi-Termination
Throughout this subsection, we will consider the following example.
###### Example 15
Consider the following union of programs $`U=TPRP^{^{}}`$ with $`Tab_U=\{path/4\}`$. Let $`S=\{reachable(rome,X)\}`$, then $`U`$ will compute the cities belonging to the same region $`r`$ as $`rome`$ and which are reachable from $`rome`$ making use of the list of connections of the region $`r`$. The program $`P^{^{}}`$ contains facts giving the region to which each city belongs and the list of connections in each region (a connection between city $`c_1`$ and city $`c_2`$ is given by the term $`e(c_1,c_2)`$).
$$\begin{array}{cc}T:\hfill & \{\begin{array}{ccc}reachable(X,Y)\hfill & \hfill & inregion(X,R),connections(R,Ed),\hfill \\ & & path(X,Ed,Y,L)\hfill \end{array}\hfill \\ & \\ P:\hfill & \{\begin{array}{ccc}path(X,Ed,Y,[Y])\hfill & \hfill & edge(X,Ed,Y)\hfill \\ path(X,Ed,Z,[Y|L])\hfill & \hfill & edge(X,Ed,Y),path(Y,Ed,Z,L)\hfill \end{array}\hfill \\ & \\ R:\hfill & \{\begin{array}{ccc}edge(X,[e(X,Y)|L],Y)\hfill & \hfill & \\ edge(X,[e(X_1,X_2)|L],Y)\hfill & \hfill & edge(X,L,Y)\hfill \end{array}\hfill \\ & \\ P^{^{}}:\hfill & \{\begin{array}{ccc}inregion(city,region).\hfill & & \\ \mathrm{}\hfill & & \\ connections(region,list\mathrm{\_}of\mathrm{\_}connections).\hfill & & \\ \mathrm{}\hfill & & \end{array}\hfill \end{array}$$
We will prove that $`U`$ quasi-terminates w.r.t. $`Tab_U`$ and $`S`$. We will do this in a modular way:
* In Example 16, after Proposition 3, we prove that $`U=T(PRP^{^{}})`$ quasi-terminates, given that $`(PRP^{^{}})`$ quasi-terminates.
* In Example 17, after Proposition 4, we prove that $`PR`$ quasi-terminates (recall that $`PR`$ is the program of Example 1 in Subsection 3.2).
* In Example 18, after Proposition 5, we prove that $`(PR)P^{^{}}`$ quasi-terminates, given that $`(PR)`$ and $`P^{^{}}`$ quasi-terminate.
###### Proposition 3
Suppose $`P`$ and $`R`$ are two programs, such that $`P`$ extends $`R`$. Let $`SB_{PR}^E`$. If
* $`R`$ quasi-terminates w.r.t. $`Tab_R`$ and $`Call(PR,S)`$,
* there is a level mapping $`|.|`$ on $`B_P^E`$ such that for all $`A`$ such that $`\stackrel{~}{A}S`$, $`|.|`$ is finitely partitioning on $`Call(PR,\{A\})B_{Tab_P}^E`$, and such that
+ for every atom $`A`$ such that $`\stackrel{~}{A}Call(PR,S)`$,
+ for every clause $`HB_1,\mathrm{},B_n`$ in $`P`$ such that $`mgu(A,H)=\theta `$ exists,
+ for every $`1in`$,
+ for every $`cas`$ $`\theta _{i1}`$ in $`PR`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$\begin{array}{cc}|A||B_i\theta \theta _{i1}|\hfill & \\ \text{and}\hfill & \\ |A|>|B_i\theta \theta _{i1}|\hfill & \text{if}Rel(A)Rel(B_i)NTab_P\text{and}\hfill \\ & C_2(Rel(A),Rel(B_i))\text{does not hold}.\hfill \end{array}$$
then, $`PR`$ quasi-terminates w.r.t. $`Tab_{PR}`$ and $`S`$.
###### Proof
Let $`A`$ be an atom such that $`\stackrel{~}{A}S`$. Let $``$ be the LG-forest w.r.t. $`Tab_{PR}`$ of $`PR\{A\}`$. We prove that $``$ consists of a finite number of LG-trees without infinite branches.
If $`A`$ is defined in $`R`$, this follows directly from the fact that $`P`$ extends $`R`$ and that $`R`$ quasi-terminates w.r.t. $`Tab_R`$ and $`Call(PR,S)`$.
So, suppose $`A`$ is defined in $`P`$. Because of the second condition in the proposition statement, every call directly descending from $`A`$, say $`B`$, is such that $`|B||A|`$. This holds recursively for atoms descending from $`A`$ using clauses of $`P`$. Because $`|.|`$ is finitely partitioning on $`B_{Tab_P}^ECall(PR,\{A\})`$, the set of tabled atoms, descending from $`A`$, using clauses of $`P`$, is finite. For atoms $`C`$, defined in $`R`$ and descending from $`A`$ using clauses of $`P`$, we know that $`\mathrm{}(Call(PR,\{C\})B_{Tab_R}^E)<\mathrm{}`$. So, $`\mathrm{}(Call(PR,\{A\})B_{Tab_{PR}}^E)<\mathrm{}`$.
We now prove that there is no tree in $``$ with an infinite branch. Suppose this is not the case, and there is a tree in $``$ with an infinite branch. Because, $`R`$ quasi-terminates w.r.t. $`Tab_R`$ and $`Call(PR,S)`$, and because $`P`$ extends $`R`$, this infinite branch contains an infinite directed subsequence $`G_0,G_1,\mathrm{}`$, with leftmost atoms $`A_0,A_1,\mathrm{}`$, belonging to $`B_{NTab_PDef_P}^E`$. This infinite directed subsequence has a tail, such that for all $`i`$ such that $`G_i`$ belongs to this tail, $`Rel(A_i)Rel(A_{i+1})`$ and $`C_2(Rel(A_i),Rel(A_{i+1}))`$ does not hold. But because of the condition in the proposition statement, $`|A_i|>|A_{i+1}|`$ and this gives a contradiction. ∎
###### Example 16 (Example 15 continued)
We illustrate the above proposition by proving that $`U=T(PRP^{^{}})`$ quasi-terminates w.r.t. $`Tab_U=\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ and $`S=\{\mathrm{𝑟𝑒𝑎𝑐ℎ𝑎𝑏𝑙𝑒}(\mathrm{𝑟𝑜𝑚𝑒},X)\}`$, given that $`PRP^{^{}}`$ quasi-terminates w.r.t. $`\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ and $`Call(U,S)`$. The quasi-termination of $`PRP^{^{}}`$ will be shown in the following examples of this subsection.
The trivial level mapping (mapping every atom to $`0`$) satisfies the condition of the proposition; there is no recursive call to a non-tabled predicate in $`T`$ and the set of called $`\mathrm{𝑝𝑎𝑡ℎ}`$-atoms is finite (since the database $`P^{^{}}`$ is finite).
The case of two programs $`P`$ and $`R`$, such that $`P`$ extends $`R`$ and such that no defined predicate in $`P`$ is tabled (mentioned as point 2 in the introduction of Section 6), does not give rise to a simpler modular termination condition than the condition in Proposition 3. We want to note already here that regarding LG-termination, this special case (point 2) will give rise to a simpler modular termination condition than in the general case.
The next proposition considers the special case of two programs $`P`$ and $`R`$, such that $`P`$ extends $`R`$ and such that all the defined predicates in $`P`$ are tabled.
###### Proposition 4
Suppose $`P`$ and $`R`$ are two programs, such that $`P`$ extends $`R`$, and such that $`Def_PTab_P`$. Let $`SB_{PR}^E`$. If
* $`R`$ quasi-terminates w.r.t. $`Tab_R`$ and $`Call(PR,S)`$,
* there is a level mapping $`|.|`$ on $`B_P^E`$ such that for all $`A`$ such that $`\stackrel{~}{A}S`$, $`|.|`$ is finitely partitioning on $`Call(PR,\{A\})B_{Tab_P}^E`$, and such that
+ for every atom $`A`$ such that $`\stackrel{~}{A}Call(PR,S)`$,
+ for every clause $`HB_1,\mathrm{},B_n`$ in $`P`$ such that $`mgu(A,H)=\theta `$ exists,
+ for every $`1in`$,
+ for every $`cas`$ $`\theta _{i1}`$ in $`PR`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$|A||B\theta \theta _{i1}|$$
then, $`PR`$ quasi-terminates w.r.t. $`Tab_{PR}`$ and $`S`$.
###### Proof
This is a direct corollary of Proposition 3 (every recursive predicate in $`P`$ is defined in $`P`$ and hence tabled). ∎
###### Example 17 (Example 15 continued)
We illustrate the above proposition by proving that $`PR`$ quasi-terminates w.r.t. $`\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ and $`Call(U,S)B_{PR}^E`$.
* First we prove that $`R`$ quasi-terminates w.r.t. $`\mathrm{}`$ and $`Call(U,S)B_R^E`$ (or, since there are no tabled atoms in $`R`$, that $`R`$ LD-terminates w.r.t. $`Call(U,S)B_R^E`$). We use Theorem 5.1 and show that $`R`$ is quasi-acceptable w.r.t. $`\mathrm{}`$ and $`Call(U,S)B_R^E`$. Consider the following level mapping:
$$|edge(t_1,t_2,t_3)|=t_2_l$$
It can be easily seen that we have a strict decrease between the head and the body atom of the recursive clause for $`edge`$ in $`R`$. Hence, the quasi-acceptability condition is satisfied.
* The trivial level mapping on $`B_P^E`$ satisfies the second condition in the proposition statement. Indeed, $`\mathrm{𝑝𝑎𝑡ℎ}`$ is tabled so a strict decrease is never required, and the set of called $`\mathrm{𝑝𝑎𝑡ℎ}`$-atoms is finite since the database of facts comprising $`P^{^{}}`$ is finite.
Finally, we consider the case of two programs $`P_1`$ and $`P_2`$ extending each other.
###### Proposition 5
Let $`P_1,P_2`$ be two programs such that $`P_1`$ extends $`P_2`$ and $`P_2`$ extends $`P_1`$. Let $`SB_{P_1P_2}^E`$. If
* $`P_1`$ quasi-terminates w.r.t. $`Tab_{P_1}`$ and $`SB_{P_1}^E`$,
* $`P_2`$ quasi-terminates w.r.t. $`Tab_{P_2}`$ and $`SB_{P_2}^E`$,
then $`P_1P_2`$ quasi-terminates w.r.t. $`Tab_{P_1P_2}`$ and $`S`$.
###### Proof
Because $`P_1`$ extends $`P_2`$ and $`P_2`$ extends $`P_1`$, $`Call(P_1P_2,S)B_{P_i}^E=Call(P_i,SB_{P_i}^E)`$ for $`i=1,2`$. The proposition follows then by definition of quasi-termination. ∎
###### Example 18 (Example 15 continued)
We prove that $`(PR)P^{^{}}`$ quasi-terminates w.r.t. $`\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ and $`Call(U,S)B_{PRP^{^{}}}^E`$, given that $`PR`$ quasi-terminates w.r.t. $`\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ and $`Call(U,S)B_{PR}^E`$ (which was shown in Example 17) and that $`P^{^{}}`$ quasi-terminates w.r.t. $`\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ and $`Call(U,S)B_P^{^{}}^E`$ (which is obvious since it consists of a finite set of facts). We can apply Proposition 5, since $`PR`$ extends $`P^{^{}}`$ and vice versa, $`P^{^{}}`$ extends $`PR`$.
The above modular conditions for the quasi-termination of $`PR`$ are proven to be sufficient, but in many cases they are also necessary. In particular, the modular conditions of Proposition 3 are also necessary for the quasi-termination of $`PR`$ in case the tabling $`Tab_P`$ is well-chosen w.r.t. $`P`$. Because in all cases where $`Def_PTab_P`$, $`Tab_P`$ is well-chosen w.r.t. $`P`$, it follows that the modular conditions of Proposition 4 are necessary in general. Finally, it can be easily seen that the modular conditions of Proposition 5 are also necessary in general.
Note that all the above modular termination conditions prove the quasi-termination of $`PR`$ without constructing a level mapping $`|.|`$ such that $`PR`$ is quasi-acceptable w.r.t. this level mapping. In Subsection 6.3, modular termination conditions for quasi-termination are given which construct (from simpler level mappings) a level mapping such that $`PR`$ is quasi-acceptable w.r.t. this level mapping. This construction will be illustrated in Example 20 on the program $`PR`$ of Example 15 (see also Example 17).
### 6.2 Modular Conditions for LG-Termination
Similarly to the case of quasi-termination, we want to have modular termination proofs for the LG-termination of the union $`PR`$ of two programs $`P`$ and $`R`$, where $`P`$ extends $`R`$. Note that, because of Theorem 4.1 and because $`(PR)^a=`$ $`P^aR^a`$ (if $`P`$ extends $`R`$), we can use the modular proofs for quasi-termination of Subsection 6.1. However, as we already noted in Subsection 5.2, we can give simpler conditions which require less checks of decreases between the levels of successive calls. These conditions are given below.
###### Proposition 6
Let $`P`$ and $`R`$ be two programs, such that $`P`$ extends $`R`$. Let $`SB_{PR}^E`$. If
* $`R`$ LG-terminates w.r.t. $`Tab_R`$ and $`Call(PR,S)`$, and
* there is a level mapping $`|.|`$ on $`B_{P^a}^E`$ such that for all $`A`$ such that $`\stackrel{~}{A}S`$, $`|.|`$ is finitely partitioning on $`Call(P^aR,\{A\})B_{TR_PTR_P^a}^E`$, and such that
+ for every atom $`A`$ such that $`\stackrel{~}{A}Call(P^aR,S)`$,
+ for every clause $`HB_1,\mathrm{},B_n`$ in $`P^a`$ such that $`mgu(A,H)=\theta `$ exists,
+ for every $`B_i`$ such that $`Rel(B_i)Rel(H)`$ or $`Rel(B_i)TR_P^a`$,
+ for every $`cas`$ $`\theta _{i1}`$ in $`P^aR`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$\begin{array}{cc}|A||B_i\theta \theta _{i1}|\hfill & \\ \text{and}\hfill & \\ |A|>|B_i\theta \theta _{i1}|\hfill & \text{if}Rel(A)Rel(B_i)NTab_P\text{and}\hfill \\ & C_2(Rel(A),Rel(B_i))\text{does not hold}.\hfill \end{array}$$
then $`PR`$ LG-terminates w.r.t. $`Tab_{PR}`$ and $`S`$.
###### Proof
The proof is a simple adaptation of the proof of the if-direction of Theorem 5.2; the adaptation is similar to the adaptation needed to transform the proof of the if-direction of Theorem 5.1 into a proof of Proposition 3. ∎
We next consider three special cases of Proposition 6. In the following proposition we consider the case in which no defined predicate in $`P`$ is tabled.
###### Proposition 7
Let $`P`$ and $`R`$ be two programs, such that $`P`$ extends $`R`$ and such that $`\mathrm{𝐷𝑒𝑓}_P\mathrm{𝑁𝑇𝑎𝑏}_P`$. Let $`SB_{PR}^E`$. If
* $`R`$ LG-terminates w.r.t. $`\mathrm{𝑇𝑎𝑏}_R`$ and $`Call(PR,S)`$,
* there is a level mapping $`|.|`$ on $`B_P^E`$ such that
+ for every atom $`A`$ such that $`\stackrel{~}{A}Call(PR,S)`$,
+ for every clause $`HB_1,\mathrm{},B_n`$ in $`P`$ such that $`mgu(A,H)=\theta `$ exists,
+ for every $`B_i`$ such that $`Rel(B_i)Rel(A)`$,
+ for every $`cas`$ $`\theta _{i1}`$ in $`PR`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$|A|>|B_i\theta \theta _{i1}|$$
then $`PR`$ LG-terminates w.r.t. $`\mathrm{𝑇𝑎𝑏}_{PR}`$ and $`S`$.
###### Proof
Because no defined predicate in $`P`$ is tabled, $`P^a=P`$. Also, for all $`p,q\mathrm{𝑁𝑇𝑎𝑏}_P\mathrm{𝐷𝑒𝑓}_P`$ with $`pq`$, $`C_1(p,q)`$ holds. The proposition follows then from Proposition 6. ∎
###### Example 19
Recall program $`R`$ of Example 14. Let $`P`$ be the following program which parses the language $`a^nb`$ (see also Subsection 3.1):
$$\begin{array}{cc}P:\hfill & \{\begin{array}{ccc}s(\mathrm{𝑆𝑖},\mathrm{𝑆𝑜},\mathrm{𝑃𝑇})\hfill & \hfill & a(\mathrm{𝑆𝑖},S),S=[b|\mathrm{𝑆𝑜}],\mathrm{𝑃𝑇}=\mathrm{𝑠𝑝𝑡}(\mathrm{𝑃𝑇𝑎},b),a(\mathrm{𝑆𝑖},S,\mathrm{𝑃𝑇𝑎})\hfill \\ a(\mathrm{𝑆𝑖},\mathrm{𝑆𝑜},\mathrm{𝑃𝑇})\hfill & \hfill & a(\mathrm{𝑆𝑖},S),a(S,\mathrm{𝑆𝑜}),\mathrm{𝑃𝑇}=\mathrm{𝑎𝑝𝑡}(\mathit{PT1},\mathit{PT2}),a(\mathrm{𝑆𝑖},S,\mathit{PT1}),\hfill \\ & & a(S,\mathrm{𝑆𝑜},\mathit{PT2})\hfill \\ a(\mathrm{𝑆𝑖},\mathrm{𝑆𝑜},\mathrm{𝑃𝑇})\hfill & \hfill & \mathrm{𝑆𝑖}=[a|\mathrm{𝑆𝑜}],\mathrm{𝑃𝑇}=a\hfill \end{array}\hfill \end{array}$$
As already noted, $`P`$ extends $`R`$. Let $`a/2`$ be the only tabled predicate in $`PR`$; see Subsection 3.1 for why this tabling is sufficient. Let $`S=\{s(\mathrm{𝑠𝑖},\mathrm{𝑆𝑜},\mathrm{𝑃𝑇})|si`$ is a ground list consisting of constants $`a,b`$, and $`So,PT`$ are distinct variables$`\}`$. We show, using Proposition 7, that $`PR`$ LG-terminates w.r.t. $`\{a/2\}`$ and $`S`$.
* $`R`$ LG-terminates w.r.t. $`\{a/2\}`$ and $`Call(PR,S)`$.
Note that, if $`a(t1,t2)Call(PR,s(si,So,PT))`$, then either $`t1`$ is a sublist of $`si`$ and $`t2`$ is a variable, or $`t1`$ and $`t2`$ are both sublists of $`si`$. In Example 14, we proved that $`R`$ LG-terminates w.r.t. this first kind of queries. To prove that $`R`$ LG-terminates w.r.t. the second kind of queries, we can again apply Theorem 5.2. Since the proof is similar to the one given in Example 14, we omit it here.
* Note first that, if $`a(t1,t2,P)Call(PR,\{s(si,So,PT)\})`$, then $`t2`$ is a (strict) sublist of $`t1`$, $`t1`$ is a sublist of $`si`$ and $`P`$ is a variable. Let $`|.|`$ be the following level mapping on $`Call(PR,S)B_{\{a/3\}}^E`$: $`|a(t1,t2,P)|=t1_lt2_l`$. Because of the remark above, $`|.|`$ is well-defined. Note that we only have to consider the recursive clause for $`a/3`$ in the analysis.
+ First consider the fourth body atom in the recursive clause for $`a/3`$. If this clause is called with $`a(ti,to,PT)`$, with $`to`$ a (strict) sublist of $`ti`$, then the fourth body atom is called as $`a(ti,t,PT1)`$ where $`to`$ is a (strict) sublist of $`t`$ and $`t`$ is a (strict) sublist of $`ti`$. Hence,
$$|a(ti,to,PT)|=ti_lto_l>ti_lt_l=|a(ti,t,PT1)|.$$
+ Now consider the last body atom. If the recursive clause is called with $`a(ti,to,PT)`$, with $`to`$ a (strict) sublist of $`ti`$, then the last body atom is called as $`a(t,to,PT2)`$ where $`to`$ is a (strict) sublist of $`t`$ and $`t`$ is a (strict) sublist of $`ti`$. Hence,
$$|a(ti,to,PT)|=ti_lto_l>t_lto_l=|a(t,to,PT2)|.$$
We conclude that $`PR`$ and $`S`$ satisfy the condition of Proposition 7, so $`PR`$ LG-terminates w.r.t. $`\{a/2\}`$ and $`S`$.
In the next proposition, a modular termination proof for the LG-termination of the union $`PR`$ is given, where $`P`$ extends $`R`$ and all defined predicates in $`P`$ are tabled.
###### Proposition 8
Let $`P`$ and $`R`$ be two programs, such that $`P`$ extends $`R`$ and such that $`\mathrm{𝐷𝑒𝑓}_P\mathrm{𝑇𝑎𝑏}_P`$. Let $`SB_{PR}^E`$. If
* $`R`$ LG-terminates w.r.t. $`Tab_R`$ and $`Call(PR,S)`$, and
* there is a level mapping $`|.|`$ on $`B_{P^a}^E`$ such that for all $`A`$ such that $`\stackrel{~}{A}S`$, $`|.|`$ is finitely partitioning on $`Call(P^aR,\{A\})B_{TR_PTR_P^a}^E`$, and such that
+ for every atom $`A`$ such that $`\stackrel{~}{A}Call(P^aR,S)`$,
+ for every clause $`HB_1,\mathrm{},B_n`$ in $`P^a`$ such that $`mgu(A,H)=\theta `$ exists,
+ for every $`B_i`$ such that $`Rel(B_i)Rel(H)`$ or $`Rel(B_i)TR_P^a`$,
+ for every $`cas`$ $`\theta _{i1}`$ in $`P^aR`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$|A||B_i\theta \theta _{i1}|$$
then $`PR`$ LG-terminates w.r.t. $`\mathrm{𝑇𝑎𝑏}_{PR}`$ and $`S`$.
###### Proof
This is a direct corollary of Proposition 6 (every recursive predicate in $`P`$ is defined in $`P`$ and hence tabled). ∎
Finally, we consider the case of two programs $`P_1`$ and $`P_2`$ extending each other.
###### Proposition 9
Let $`P_1,P_2`$ be two programs such that $`P_1`$ extends $`P_2`$ and $`P_2`$ extends $`P_1`$. Let $`SB_{P_1P_2}^E`$. If
* $`P_1`$ LG-terminates w.r.t. $`Tab_{P_1}`$ and $`SB_{P_1}^E`$,
* $`P_2`$ LG-terminates w.r.t. $`Tab_{P_2}`$ and $`SB_{P_2}^E`$,
then $`P_1P_2`$ LG-terminates w.r.t. $`Tab_{P_1P_2}`$ and $`S`$.
###### Proof
Because $`P_1`$ extends $`P_2`$ and $`P_2`$ extends $`P_1`$, $`Call(P_1P_2,S)B_{P_i}^E=Call(P_i,SB_{P_i}^E)`$, for $`i=1,2`$. The proposition follows then by definition of LG-termination. ∎
Similar as in the case of quasi-termination, the above modular, sufficient conditions for the LG-termination of $`PR`$ are in many cases also necessary. In particular, the modular conditions of Proposition 6 are also necessary for the LG-termination of $`PR`$ in case the tabling $`Tab_P`$ is well-chosen w.r.t. $`P`$. Because in all cases where $`Def_PNTab_P`$, respectively $`Def_PTab_P`$, $`Tab_P`$ is well-chosen w.r.t. $`P`$, it follows that the modular conditions of Proposition 7, respectively Proposition 8, are necessary in general. Also the modular conditions of Propositions 9 are necessary in general.
### 6.3 Construction of Level Mappings in Modular Termination Proofs
We now take a closer look at the modular termination proofs of the previous subsections. We follow the approach of , where modular proofs for SLD-termination (i.e. termination of SLD-resolution w.r.t. all selection rules) and LD-termination are given. In , (S)LD-termination of a program $`PR`$, where $`P`$ extends $`R`$, is proven by constructing a level mapping $`|.|`$ for $`PR`$ which satisfies some acceptability condition. The level mapping $`|.|`$ is constructed from simpler level mappings for the separate components $`P`$ and $`R`$. Namely, $`|.|`$ is constructed from $`|.|_P`$, $`|.|_R`$ and $`._P`$, where $`|.|_P`$, respectively $`|.|_R`$, is a level mapping for $`P`$, respectively $`R`$, satisfying the acceptability condition, and where $`._P`$ is a level mapping for $`P`$ serving as the connecting part between the two components. The level mapping $`|.|`$ for $`PR`$ is then defined as $`|.|_P+._P`$ on the atoms defined in $`P`$ and as $`|.|_R`$ on the atoms defined in $`R`$. It is proven that such a construction always returns a level mapping satisfying the acceptability condition for the whole program $`PR`$.
We follow the same approach for the case of quasi-termination (we do not consider LG-termination since it can be dealt with in a similar way). In particular, we give modular proofs of the quasi-termination of a program $`PR`$, where $`P`$ extends $`R`$, by constructing a level mapping such that $`PR`$ is quasi-acceptable w.r.t. this level mapping (see Definition 14 and Theorem 5.1). The construction of such a level mapping is done in a way similar to , which we explained above. We first need the following lemma, which gives sufficient, modular conditions on a level mapping in order to be finitely partitioning on some subset of the extended Herbrand base. We use a slightly more general definition of a level mapping, namely, a level mapping is a mapping from a *subset* of the extended Herbrand base to the natural numbers.
###### Lemma 7
1. Let $`P`$ be a program and $`LB_P^E`$. Let $`|.|,.:L\mathrm{}`$ be level mappings. If $`|.|`$ is finitely partitioning on $`CL`$, then $`|.|+.:L\mathrm{}:A(|.|+.)(A)=|A|+A`$ is finitely partitioning on $`C`$.
2. Let $`P_1,P_2`$ be two programs and $`L_1B_{P_1}^E`$, $`L_2B_{P_2}^E`$. Let $`|.|_1:L_1\mathrm{}`$ and $`|.|_2:L_2\mathrm{}`$ be level mappings. If $`|.|_1`$, respectively $`|.|_2`$, is finitely partitioning on $`C_1L_1`$, respectively $`C_2L_2`$, then $`m(|.|_1,|.|_2):L_1L_2\mathrm{}:`$
$$Am(|.|_1,|.|_2)(A):=\{\begin{array}{ccc}min(|A|_1,|A|_2)\hfill & ,\hfill & AL_1L_2\hfill \\ |A|_1\hfill & ,\hfill & AL_1L_2\hfill \\ |A|_2\hfill & ,\hfill & AL_2L_1\hfill \end{array}$$
is finitely partitioning on $`C_1C_2`$.
###### Proof
1. Let $`n\mathrm{}`$. We prove that $`\mathrm{}((|.|+.)^1(n)C)<\mathrm{}`$.
$$\begin{array}{ccc}(|.|+.)^1(n)C\hfill & =\hfill & \{AC|(|.|+.)(A)=n\}\hfill \\ & \hfill & \{AC||A|n\}\hfill \\ & =\hfill & _{0mn}\{AC||A|=m\}\hfill \end{array}$$
and this last set is finite.
2. Let $`n\mathrm{}`$. We prove that $`\mathrm{}(m(|.|_1,|.|_2)^1(n)`$ $``$ $`(C_1C_2))<`$ $`\mathrm{}`$.
$$\begin{array}{ccc}m(|.|_1,|.|_2)^1(n)(C_1C_2)\hfill & =\hfill & \{AC_1C_2|m(|.|_1,|.|_2)(A)=n\}\hfill \\ & =\hfill & \{AC_1C_2||A|_1=n\}\hfill \\ & & \{AC_2C_2||A|_2=n\}\hfill \\ & & \{AC_1C_2|min(|A|_1,|A|_2)=n\}\hfill \end{array}$$
It is obvious that the first two sets in the union are finite ($`|.|_1`$, resp. $`|.|_2`$, is finitely partitioning on $`C_1`$, resp. $`C_2`$). The set $`\{AC_1C_2|min(|A|_1,|A|_2)`$ $`=n\}`$ is finite, because it is a subset of the finite set $`\{AC_1C_2||A|_1=n\}`$ $``$ $`\{AC_1C_2||A|_2=n\}`$.
We next give a modular termination condition for the quasi-termination of $`PR`$ where $`P`$ extends $`R`$, by constructing a level mapping, from simpler ones, such that $`PR`$ is quasi-acceptable w.r.t. this level mapping. Notice that we consider the same case as in Proposition 3.
###### Proposition 10
Let $`P`$ and $`R`$ be two programs such that $`P`$ extends $`R`$ and let $`SB_{PR}^E`$. If
1. $`R`$ is quasi-acceptable w.r.t. $`Tab_R`$, $`Call(PR,S)`$ and the level mapping $`|.|_R`$, defined on $`B_R^E`$ and finitely partitioning on $`Call(PR,S)B_{Tab_R}^E`$,
2. there is a level mapping $`|.|_P`$ defined on $`B_P^EB_R^E`$ and finitely partitioning on $`Call(PR,S)B_{Tab_PTab_R}^E`$ such that
* for every atom $`A`$ such that $`\stackrel{~}{A}Call(PR,S)`$,
* for every clause $`HB_1,\mathrm{},B_n`$ in $`P`$, such that $`mgu(A,H)=\theta `$ exists,
* for every $`1in`$ such that $`B_iB_P^EB_R^E`$,
* for every $`cas`$ $`\theta _{i1}`$ in $`PR`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$\begin{array}{cc}|A|_P|B_i\theta \theta _{i1}|_P\hfill & \\ \text{and}\hfill & \\ |A|_P>|B_i\theta \theta _{i1}|_P\hfill & \text{if}Rel(A)Rel(B_i)NTab_P\text{and}\hfill \\ & C_2(Rel(A),Rel(B_i))\text{does not hold}.\hfill \end{array}$$
3. there exists a level mapping $`._P`$ on $`B_P^EB_R^E`$ such that
* for every atom $`A`$ such that $`\stackrel{~}{A}Call(PR,S)`$,
* for every clause $`HB_1,\mathrm{},B_n`$ in $`P`$, such that $`mgu(A,H)=\theta `$ exists,
* for every $`1in`$,
* for every $`cas`$ $`\theta _{i1}`$ in $`PR`$ for $`(B_1,\mathrm{},B_{i1})\theta `$:
$$A_P\{\begin{array}{ccc}B_i\theta \theta _{i1}_P\hfill & ,\hfill & B_i\theta \theta _{i1}B_P^EB_R^E\hfill \\ |B_i\theta \theta _{i1}|_R\hfill & ,\hfill & B_i\theta \theta _{i1}B_R^E\hfill \end{array}$$
then, the following level mapping $`|.|`$, defined on $`B_{PR}^E`$, is finitely partitioning on $`Call(PR,S)B_{Tab_{PR}}^E`$:
$$|A|=\{\begin{array}{cc}|A|_P+A_P\hfill & \text{if}AB_P^EB_R^E,\hfill \\ |A|_R\hfill & \text{if}AB_R^E,\hfill \end{array}$$
and $`PR`$ is quasi-acceptable w.r.t. $`Tab_{PR}`$, $`S`$, and the level mapping $`|.|`$. Hence, $`PR`$ quasi-terminates w.r.t. $`Tab_{PR}`$ and $`S`$.
###### Proof
Because of Lemma 7(1,2), the level mapping $`|.|`$ is finitely partitioning on $`Call(PR,S)B_{Tab_{PR}}^E`$. We prove that $`PR`$ is quasi-acceptable w.r.t. $`Tab_{PR}`$, $`S`$ and the level mapping $`|.|`$ (see Definition 14).
Let $`A`$ be an atom such that $`\stackrel{~}{A}Call(PR,S)`$. Let $`HB_1,\mathrm{},B_n`$ be a clause of $`PR`$ such that $`mgu(A,H)=\theta `$ exists. Let $`\theta _{i1}`$ be a $`cas`$ in $`PR`$ for $`(B_1,\mathrm{},B_{i1})\theta `$. There are two cases to consider:
* $`A`$ is defined in $`R`$, $`|A|=|A|_R`$.
Then, because of condition 1 in the proposition statement, $`|A|_R|B_i\theta \theta _{i1}|_R`$ (note that because $`P`$ extends $`R`$, for a clause $`HB_1,\mathrm{},B_n`$ in $`R`$ and $`mgu(A,H)=\theta `$, a $`cas`$ for $`(B_1,\mathrm{},B_{i1})\theta `$ in $`PR`$ is the same as a $`cas`$ for $`(B_1,\mathrm{},B_{i1})\theta `$ in $`R`$ only). Since $`A,B_i\theta \theta _{i1}`$ $`B_R^E`$, $`|A|=|A|_R`$ $`|B_i\theta \theta _{i1}|_R=|B_i\theta \theta _{i1}|`$. In case $`Rel(A)Rel(B_i)NTab_R`$ and $`C_2(Rel(A),Rel(B_i))`$ does not hold, $`|A|=|A|_R>`$ $`|B_i\theta \theta _{i1}|_R=|B_i\theta \theta _{i1}|`$.
* $`A`$ is defined in $`P`$, $`|A|=|A|_P+A_P`$.
+ $`B_iB_R^E`$, $`|B_i\theta \theta _{i1}|=|B_i\theta \theta _{i1}|_R`$.
Because of condition 3 in the proposition statement, $`A_P`$ $``$ $`|B_i\theta \theta _{i1}|_R`$. Hence, $`|A|=`$ $`|A|_P+A_P`$ $``$ $`|B_i\theta \theta _{i1}|_R`$ $`=|B_i\theta \theta _{i1}|`$.
Note that in this case we always have that $`Rel(A)\simeq ̸Rel(B_i)`$ (because $`P`$ extends $`R`$).
+ $`B_iB_P^EB_R^E`$, $`|B_i\theta \theta _{i1}|=`$ $`|B_i\theta \theta _{i1}|_P+`$ $`B_i\theta \theta _{i1}_P`$.
Because of condition 2 in the proposition statement, $`|A|_P`$ $``$ $`|B_i\theta \theta _{i1}|_P`$. Also, because of condition 3, $`A_P`$ $``$ $`B_i\theta \theta _{i1}_P`$. Hence, $`|A|_P+`$ $`A_P`$ $``$ $`|B_i\theta \theta _{i1}|_P+`$ $`B_i\theta \theta _{i1}_P`$. In case $`Rel(A)Rel(B_i)NTab_P`$ and $`C_2(Rel(A),Rel(B_i))`$ does not hold, we have that $`|A|_P`$ $`>`$ $`|B_i\theta \theta _{i1}|_P`$, hence $`|A|_P+A_P`$ $`>`$ $`|B_i\theta \theta _{i1}|_P+`$ $`B_i\theta \theta _{i1}_P`$.
In each case, we conclude that $`|A|`$ $``$ $`|B_i\theta \theta _{i1}|`$ and that, in case $`Rel(A)Rel(B_i)NTab_{PR}`$ and $`C_2(Rel(A),Rel(B_i))`$ does not hold, $`|A|`$ $`>`$ $`|B_i\theta \theta _{i1}|`$. ∎
###### Example 20
Recall the program $`PR`$ of Example 15 (see also Examples 17 and 1).
$$\begin{array}{cc}P:\hfill & \{\begin{array}{ccc}path(X,Ed,Y,[Y])\hfill & \hfill & edge(X,Ed,Y)\hfill \\ path(X,Ed,Z,[Y|L])\hfill & \hfill & edge(X,Ed,Y),path(Y,Ed,Z,L)\hfill \end{array}\hfill \\ & \\ R:\hfill & \{\begin{array}{ccc}edge(X,[e(X,Y)|L],Y)\hfill & \hfill & \\ edge(X,[e(X_1,X_2)|L],Y)\hfill & \hfill & edge(X,L,Y)\hfill \end{array}\hfill \end{array}$$
Let $`Tab_{PR}=\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ and $`S=\{\mathrm{𝑝𝑎𝑡ℎ}(a,[e(a,b),e(b,a)],Y,L)\}`$. We prove that $`PR`$ quasi-terminates w.r.t. $`\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$ and $`S`$ using the above proposition. The first two conditions of this proposition were already tackled in Example 17 (using a different set $`S`$; the arguments remain the same however).
1. $`R`$ is quasi-acceptable w.r.t. $`\mathrm{}`$, $`Call(PR,S)`$ and the level mapping $`|.|_R`$:
$$|edge(t_1,t_2,t_3)|_R=t_2_l.$$
2. The trivial level mapping, $`|.|_P`$, on $`B_P^EB_R^E=B_{\{\mathrm{𝑝𝑎𝑡ℎ}\}}^E`$ satisfies the second condition of the proposition; $`\mathrm{𝑝𝑎𝑡ℎ}/4`$ is tabled so a strict decrease is never required and there are a finite number of $`\mathrm{𝑝𝑎𝑡ℎ}`$-atoms in the call set.
3. The following level mapping, $`._P`$, on $`B_P^EB_R^E=B_{\{\mathrm{𝑝𝑎𝑡ℎ}\}}^E`$ satisfies the third condition of the proposition:
$$path(t_1,t_2,t_3,t_4)_P=t_2_l.$$
Hence, the level mapping $`|.|`$, on $`B_{PR}^E`$, is defined as follows:
$$\begin{array}{ccc}|path(t_1,t_2,t_3,t_4)|\hfill & =\hfill & t_2_l\hfill \\ |edge(t_1,t_2,t_3)|\hfill & =\hfill & t_2_l\hfill \end{array}$$
and by the proposition, the program $`PR`$ is quasi-acceptable w.r.t. $`Tab_{PR}=\{\mathrm{𝑝𝑎𝑡ℎ}/4\}`$, $`S=\{\mathrm{𝑝𝑎𝑡ℎ}(a,[e(a,b),e(b,a)],Y,L)\}`$ and the level mapping $`|.|`$.
In the special case where two programs use disjoint sets of predicates is also considered. We next consider a more general case in which the two programs may use the same predicates but may not define the same predicate; that is, both programs extend each other. This case was already considered in Proposition 5.
###### Proposition 11
Let $`P_1,P_2`$ be programs such that $`P_1`$ extends $`P_2`$ and $`P_2`$ extends $`P_1`$. Let $`SB_{P_1P_2}^E`$. Suppose that
1. $`P_1`$ is quasi-acceptable w.r.t. $`Tab_{P_1}`$, $`SB_{P_1}^E`$ and a level mapping $`|.|_1`$ on $`B_{P_1}^E`$ which is finitely partitioning on $`Call(P_1,SB_{P_1}^E)B_{Tab_{P_1}}^E`$,
2. $`P_2`$ is quasi-acceptable w.r.t. $`Tab_{P_2}`$, $`SB_{P_2}^E`$ and a level mapping $`|.|_2`$ on $`B_{P_2}^E`$ which is finitely partitioning on $`Call(P_2,SB_{P_2}^E)B_{Tab_{P_2}}^E`$,
then $`m(|.|_1,|.|_2)`$ (see Lemma 7, point 2) is a finitely partitioning level mapping on $`Call(P_1P_2,S)B_{Tab_{P_1P_2}}^E`$, and $`P_1P_2`$ is quasi-acceptable w.r.t. $`Tab_{P_1P_2}`$, $`S`$ and $`m(|.|_1,|.|_2)`$. Hence, $`P_1P_2`$ quasi-terminates w.r.t. $`Tab_{P_1P_2}`$ and $`S`$.
###### Proof
Note that because $`P_1`$ extends $`P_2`$ and vice versa, $`Call(P_1P_2,S)B_{P_i}^E`$ $`=`$ $`Call(P_i,SB_{P_i}^E)`$, $`i\{1,2\}`$. By Lemma 7(2), $`m(|.|_1,|.|_2)`$ is finitely partitioning on $`Call(P_1P_2,S)B_{Tab_{P_1P_2}}^E`$. Also, if $`HB_1,\mathrm{},B_n`$ is a clause in $`P_i`$ and $`mgu(A,H)=\theta `$, then a $`cas`$ in $`P_i`$ for $`(B_1,\mathrm{},B_{i1})\theta `$ is a $`cas`$ in $`P_iP_j`$ ($`\{i,j\}=\{1,2\}`$) for $`(B_1,\mathrm{},B_{i1})\theta `$ and vice versa. Then it directly follows that $`P_1P_2`$ is quasi-acceptable w.r.t. $`Tab_{P_1P_2}`$, $`S`$ and $`m(|.|_1,|.|_2)`$. ∎
As was shown in for the case of (S)LD-termination, the above modular conditions provide us with an incremental, bottom up method for proving termination of tabled logic programs.
## 7 Towards Automated Termination Proofs for Tabled Logic Programs
Having described the basic framework for proving termination of tabled logic programs, in this section we examine issues related to the automation of the termination conditions. We will only consider quasi-termination in this section; the results for LG-termination carry over in the same way. We show how to extend the constraint-based, automatic approach for proving LD-termination of Decorte, De Schreye and Vandecasteele , in order to prove quasi-termination of tabled logic programs in an automatic way. Our results hold for the class of simply moded, well-moded programs and queries.
We first recall the main ideas of . In , a new strategy for automatically proving LD-termination of logic programs w.r.t. sets of queries is developed. A symbolic termination condition is introduced, called *rigid acceptability*, by parametrising the concepts of norm, level mapping and model. The rigid acceptability condition is translated into a system of constraints on the values of the introduced symbols only. A system of constraints identifies sets of suitable norms, level mappings and models which can be used in the termination condition. In other words, if a solution for the constraint system exists, termination can be proved. The solving of constraint sets enables the different components of a termination proof to communicate with one another and to direct the proof towards success (if there is). The method of is both efficient and precise.
This section<sup>4</sup><sup>4</sup>4For the referees, we want to mention that the material presented in this section is very related to results presented in the article “Termination of simply moded well-typed programs under a tabled execution mechanism” (S. Verbaeten and D. De Schreye), which is submitted to the Journal of Applicable Algebra in Engineering, Communication and Computing (AAECC). There, we present similar results for the special case in which all the predicates of the program are tabled. Here we extend those results to allow a mix of tabled and non-tabled predicates. In spite of the partial overlap with this article, we choose to include this section anyway, because we feel that it is important to explicitly discuss the prospects of automating the results presented in the previous sections. In the article submitted to AAECC, we consider the larger class of simply moded well-typed programs, instead of simply moded well-moded programs that we consider here. With respect to the automation of the method, this larger class adds nothing new, so we do not include these results here. If the referees think this section should not be included, we are willing to remove it. is structured as follows. We first reformulate the quasi-acceptability condition into a condition at the clause level, which is needed in the constraint-based termination analysis framework of . This gives us the rigid quasi-acceptability condition. Then, we recall the symbolic forms for norm, level mapping and model, as introduced in . We introduce the class of simply moded, well-moded programs and queries, for which we translate the rigid quasi-acceptability condition into a system of constraints on the introduced symbols.
### 7.1 Rigid Quasi-Acceptability Condition
In order to prove termination in an automatic, constraint-based way as in , it is important to have a termination condition which is stated at the clause level (and not on sets of calls as the quasi-acceptability condition of Theorem 5.1 is). In most automatic approaches, and in particular in that of , this is obtained by requiring that the level mapping is *rigid* on the call set. A level mapping is rigid on the call set iff the value of an atom in the call set is invariant under substitutions. If a level mapping is rigid on the call set, the atoms in the call set can be considered as ground w.r.t. the level mapping. In this way, the problem of back-propagation of bindings in the calls is dealt with, and this allows the termination condition to be stated at the clause level (see also ).
###### Definition 16 (rigid level mapping)
Let $`P`$ be a program and $`CB_P^E`$. A level mapping $`|.|`$ is rigid on $`C`$ iff for all atoms $`AC`$, for all substitutions $`\psi `$, $`|A|=|A\psi |`$.
The following condition of rigid quasi-acceptability is derived from the quasi-acceptability condition, and will serve as the basis for a constraint-based, symbolic condition for quasi-termination.
###### Proposition 12 (rigid quasi-acceptability condition)
Let $`P`$ be definite program, $`Tab_PPred_P`$ and $`SB_P^E`$ be a set of queries.
If there exists a level mapping $`|.|`$, such that $`|.|`$ is rigid on $`Call(P,S)`$, $`|.|`$ is finitely partitioning on $`Call(P,S)B_{Tab_P}^E`$, and such that
* for every clause $`HB_1,\mathrm{},B_n`$ in $`P`$,
* for every atom $`B_i`$, $`i\{1,..,n\}`$,
* for every substitution $`\psi `$ such that $`PB_1\psi ,\mathrm{},B_{i1}\psi `$:
$$\begin{array}{cc}|H\psi ||B_i\psi |\hfill & \\ \text{and}\hfill & \\ |H\psi |>|B_i\psi |\hfill & \text{if}Rel(H)Rel(B_i)NTab_P\text{and}\hfill \\ & C_2(Rel(A),Rel(B_i))\text{does not hold},\hfill \end{array}$$
then $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$ and $`S`$. Hence, $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$.
###### Proof
Suppose the above condition is satisfied for $`P`$. We prove that $`P`$ is quasi-acceptable w.r.t. $`Tab_P`$, $`S`$ and the level mapping $`|.|`$. Let $`A`$ be an atom such that $`\stackrel{~}{A}Call(P,S)`$. Let $`HB_1,\mathrm{},B_n`$ be a clause in $`P`$ such that $`mgu(A,H)=\theta `$ exists. Let $`\theta _{i1}`$ be an LD- computed answer substitution for $`(B_1,\mathrm{},B_{i1})\theta `$. Then, $`PB_1\theta \theta _{i1},\mathrm{},B_{i1}\theta \theta _{i1}`$. We prove that $`|A||B_i\theta \theta _{i1}|`$. By the condition in the proposition, we know that $`|H\theta \theta _{i1}||B\theta \theta _{i1}|`$. Because $`A\theta =H\theta `$, $`|A\theta \theta _{i1}|=|H\theta \theta _{i1}|`$. Now, since $`ACall(P,S)`$ and $`|.|`$ is rigid on $`Call(P,S)`$, $`|A\theta \theta _{i1}|=|A|`$. Thus, $`|A|=|H\theta \theta _{i1}|`$, and therefore $`|A||B_i\theta \theta _{i1}|`$. The proof that $`|A|>|B_i\theta \theta _{i1}|`$ in case $`Rel(A)Rel(B_i)NTab_P`$ and $`C_2(Rel(A),Rel(B_i))`$ does not hold, is analogous. ∎
### 7.2 Symbolising the Concepts of Norm, Level Mapping and Model
We recall the symbolic forms for norms, level mappings and interargument relations (which are abstractions of models), as introduced in . These will form the basis for the symbolic termination condition. The condition will be formulated as a search for suitable values for all introduced symbols. We refer to for motivation and more details.
We will need the following notions.
###### Definition 17 (functor, predicate and extended predicate coefficients)
The set of functor coefficients, respectively predicate coefficients, respectively extended predicate coefficients associated to a program $`P`$ are the sets of symbols
$$\begin{array}{c}FC(P)=\{f_i|f/nFun_Pi\{0,\mathrm{},n\}\},\hfill \\ PC(P)=\{p_i|p/nPred_Pi\{1,\mathrm{},n\}\},\hfill \\ EC(P)=\{p_i^e|p/nPred_Pi\{0,\mathrm{},n\}\}.\hfill \end{array}$$
Let $`𝒞`$ denote the set of symbols $`FC(P)PC(P)EC(P)`$. The symbol $`L_{<𝒞;+,.;>}`$ denotes the language containing the symbols in the set $`𝒞`$ as constants, the infix functor $`+/2`$, the infix functor $`./2`$, the relation symbol $`/2`$ and the set of variables in the first order language of the program $`P`$. Terms in that language are defined in the usual way. The relation symbols $`=/2`$ and $`</2`$ are defined in terms of $`/2`$ as usual and considered as additional primitive predicates. We denote the set of all possible atoms in that language by $`S_{<𝒞;+,.;>}`$. We call such atoms symbolic expressions. The set of all logical formulae over $`S_{<𝒞;+,.;>}`$ is denoted as $`F_{<𝒞;+,.;>}`$. Such formulae are called symbolic formulae. By natural formulae, we denote formulae in $`F_{<\{0,1\};+,.;>}`$. For a formula $`F`$, $`F`$ denotes the universal closure over the free variables occurring in $`F`$.
We introduce symbolic (semi-linear) norms.
###### Definition 18 (symbolic norm $`.^s`$)
Let $`FC(P)`$ be a set of functor coefficients.
$$\begin{array}{ccccc}.^s:\hfill & Term_P& \hfill & S_{<𝒞;+,.;>}& \\ & t& \hfill & f_0+_{i=1}^nf_it_i^s& \text{ if }t=f(t_1,\mathrm{},t_n),n>0,\hfill \\ & t& \hfill & 0& \text{ if }t=cConst_P,\hfill \\ & t& \hfill & X& \text{ if }t=X\text{ is a variable}.\hfill \end{array}$$
Note that the symbolic norm of a variable is the variable itself. So, the symbolic norm includes information about the instantiation level of the term. That is, the symbolic norm takes into account those parts of the concrete term whose size may still change under instantiation.
In the same way, we can define a symbolic level mapping by symbolising its coefficients.
###### Definition 19 (symbolic level mapping $`|.|^s`$)
Let $`PC(P)`$ be a set of predicate coefficients and $`.^s`$ a symbolic norm.
$$\begin{array}{cccc}|.|^s:\hfill & Atom_P& \hfill & S_{<𝒞;+,.;>}\hfill \\ & p(t_1,\mathrm{},t_n)& \hfill & _{i=1}^np_it_i^s.\hfill \end{array}$$
Finally, we want to abstract the notion of model. Norms allow to abstract models by specifying relations which hold between the size of certain arguments of their member atoms. This leads to the notion of interargument relation.
###### Definition 20 ((valid) interargument relation)
Let $`p/nPred_P`$.
An interargument relation for $`p/n`$ is a relation $`R^{p/n}\mathrm{}^n`$.
An interargument relation $`R^{p/n}`$ for $`p/n`$ is a valid interargument relation w.r.t. a norm $`.`$ iff $`p(t_1,\mathrm{},t_n)Atom_P`$: if $`Pp(t_1,\mathrm{},t_n)`$, then $`(t_1,\mathrm{},t_n)`$ $`R^{p/n}`$.
As in , we will allow interargument relations which express an inequality relation: $`R^{p/n}=\{(x_1,\mathrm{},x_n)|_{iI_p}k_ix_i_{jO_p}k_jx_j+k_0\}`$, with $`k_i\mathrm{}`$, $`i\{0,1,\mathrm{},n\}`$, depending on $`p/n`$, $`I_pO_p\{1,\mathrm{},n\}`$ and $`I_pO_p=\mathrm{}`$. The sets $`I_p`$ and $`O_p`$ are assumed fixed for each predicate. In , it is argued that these sets can best be seen as some kind of a generalisation of the sets of input and output arguments. In the following subsection, we will assume a fixed mode for each predicate, and then the set $`I_p`$, respectively $`O_p`$, will be taken as the set of input, respectively output, positions of the predicate $`p`$.
Finally, we can abstract success sets by abstracting interargument relations.
###### Definition 21 (symbolic size expression $`𝒜^s`$)
Let $`EC(P)`$ be a set of extended predicate coefficients and $`.^s`$ a symbolic norm.
$$\begin{array}{cccc}𝒜^s:\hfill & Atom_P& \hfill & S_{<𝒞;+;>}\hfill \\ & t_1=t_2& \hfill & t_1^s=t_2^s,\hfill \\ & p(t_1,\mathrm{},t_n)& \hfill & _{iI_p}p_i^et_i^s_{jO_p}p_j^et_j^s+p_0^e,\text{ where }p=.\hfill \end{array}$$
###### Definition 22 (symbol mapping)
A symbol mapping is a mapping $`s:𝒞\mathrm{}`$.
Expressions involving only symbols from $`𝒞`$ are mapped into the natural numbers by substituting the symbols by their mapped value. With abuse of notation, if $`F`$ is a symbolic formula and $`s`$ a symbol mapping, we denote the associated natural formula as $`s(F)`$.
Each symbol mapping induces in a natural way a norm, level mapping and interargument relations. More precisely, the symbol mapping $`s`$ induces the following norm $`._s`$:
$$\{\begin{array}{cc}X_s=c_s=0\hfill & \text{if}X\text{is a variable},cConst_P,\hfill \\ f(t_1,\mathrm{},t_n)_s=s(f_0)+_{i=1}^ns(f_i)t_i_s,\hfill & \end{array}$$
level mapping $`|.|_s`$:
$$|p(t_1,\mathrm{},t_n)|_s=\underset{i=1}{\overset{n}{}}s(p_i)t_i_s,$$
and interargument relations $`R_s^{p/n}`$:
$$\{(t_1_s),\mathrm{},t_n_s)|t_1,\mathrm{},t_nTerm_P\text{and}s(𝒜^s(p(t_1,\mathrm{},t_n)))\text{ holds }\}.$$
Our aim is to formulate the rigid quasi-acceptability condition in a constraint-based way. In particular this means that we have to find syntactical conditions on a symbol mapping $`s`$ such that
* $`|.|_s`$ is rigid on $`Call(P,S)`$, and
* $`|.|_s`$ is finitely partitioning on $`Call(P,S)B_{Tab_P}^E`$.
As shown in e.g. , a level mapping $`|.|_s`$ is rigid on the call set if it does not take into account too many argument positions of predicates and functors in its linear combination<sup>5</sup><sup>5</sup>5More precisely, $`s`$ is equal to $`0`$ on all functor and predicate coefficients for which there is an atom in the call set which contains a variable on such a position. For a more formal discussion on this topic, we refer to .. On the other hand, a level mapping $`|.|_s`$ is finitely partitioning on $`Call(P,S)B_{Tab_P}^E`$, if enough argument positions are taken into account in its linear combination<sup>6</sup><sup>6</sup>6That is, the symbol mapping $`s`$ is different from $`0`$ on enough functor and predicate coefficients. For a more formal discussion on this topic, we again refer to .. We will show that, for the class of simply moded, well-moded programs and queries, we are able to combine these two—at first sight contradictory— conditions. We first introduce the class of simply moded, well-moded programs and queries and then formulate, for this class, the symbolic termination condition.
### 7.3 The Class of Simply Moded, Well-Moded Programs and Queries
###### Definition 23 (mode for a predicate)
Let $`p`$ be an $`n`$-ary predicate symbol. A mode for $`p`$ is a function $`m_p:`$ $`\{1,\mathrm{},n\}`$ $`\{In,Out\}`$. If $`m_p(i)=In`$ (respectively $`Out`$), then we say that $`i`$ is an input (respectively output) position of $`p`$ (w.r.t. $`m_p`$).
We assume that each predicate symbol has a unique mode. For predicates that have multiple modes, we can assume that a straightforward renaming process taking each mode into account has already been performed. In examples, we will write the mode $`m_p`$ for the predicate $`p`$ as follows: $`p(m_p(1),\mathrm{},m_p(n))`$. Given a predicate $`pPred_P`$ with mode $`m_p`$, we denote by $`I_p=\{i|m_p(i)=In\}`$ the set of input positions of $`p`$ according to $`m_p`$, and by $`O_p=\{i|m_p(i)=Out\}`$ the set of output positions of $`p`$ according to $`m_p`$.
We recall the notion of well-modedness (see e.g. ). For simplifying the notation, when writing an atom as $`p(𝐮,𝐯)`$, we assume that $`𝐮`$ is the sequence of terms filling in the input positions of $`p`$ and $`𝐯`$ is the sequence of terms filling in the output positions of $`p`$. For a term $`t`$, we denote by $`Var(t)`$ the set of variables occurring in $`t`$. Similar notation is used for sequences of terms.
###### Definition 24 (well-modedness)
A clause $`p_0(𝐭_\mathrm{𝟎},𝐬_{𝐧+\mathrm{𝟏}})`$ $`p_1(𝐬_\mathrm{𝟏},𝐭_\mathrm{𝟏}),\mathrm{},p_n(𝐬_𝐧,𝐭_𝐧)`$ is called well-moded iff for $`i[1,n+1]`$:
$$Var(𝐬_𝐢)\underset{j=0}{\overset{i1}{}}Var(𝐭_𝐣).$$
A program is called well-moded iff every clause of it is well-moded.
A query $`p_1(𝐬_\mathrm{𝟏},𝐭_\mathrm{𝟏}),\mathrm{},p_n(𝐬_𝐧,𝐭_𝐧)`$ is well-moded iff the clause $`p`$ $`p_1(𝐬_\mathrm{𝟏},`$ $`𝐭_\mathrm{𝟏}),`$ $`\mathrm{},`$ $`p_n(𝐬_𝐧,𝐭_𝐧)`$ is well-moded, where $`p`$ is any zero-ary predicate symbol.
An example of a well-moded program and query is given in the next subsection (Example 21). In the persistence of the notion of well-modedness was proven; i.e. an LD-resolvent of a well-moded query and a well-moded clause that is variable-disjoint with it, is well-moded. Note that in a well-moded query, the terms that occur in the input positions of the leftmost atom are all ground. Hence, as a consequence of the persistence of well-modedness, we have that well-moded programs are data driven; i.e. all atoms selected in an LD-derivation of a well-moded query in a well-moded program contain ground terms in their input positions.
Next, we introduce the notion of simply-modedness . A family (or multiset) of terms is called linear iff every variable occurs at most once in it.
###### Definition 25 (simply modedness)
A clause $`p_0(𝐬_\mathrm{𝟎},𝐭_{𝐧+\mathrm{𝟏}})`$ $`p_1(𝐬_\mathrm{𝟏},𝐭_\mathrm{𝟏}),\mathrm{},p_n(𝐬_𝐧,𝐭_𝐧)`$ is called simply moded iff $`𝐭_\mathrm{𝟏},\mathrm{},𝐭_𝐧`$ is a linear family of variables and for $`i[1,n]`$:
$$Var(𝐭_𝐢)(\underset{j=0}{\overset{i}{}}Var(𝐬_𝐣))=\mathrm{}.$$
A program is called simply moded iff every clause of it is simply moded.
A query $`p_1(𝐬_\mathrm{𝟏},𝐭_\mathrm{𝟏}),\mathrm{},p_n(𝐬_𝐧,𝐭_𝐧)`$ is simply moded iff the clause $`p`$ $`p_1(𝐬_\mathrm{𝟏},𝐭_\mathrm{𝟏}),`$ $`\mathrm{},`$ $`p_n(𝐬_𝐧,𝐭_𝐧)`$ is simply moded, where $`p`$ is any zero-ary predicate symbol.
The program and query of Example 21 of the next subsection, are simply moded. The notion of simply modedness is persistent ; i.e. an LD-resolvent of a simply moded query and a simply moded clause that is variable-disjoint with it, is simply moded. An atom is called input/output disjoint if the family of terms occurring in its input positions has no variable in common with the family of terms occurring in its output positions. As a consequence of the persistence of simply modedness, we have that all atoms selected in an LD-derivation of a simply moded query in a simply moded program are input/output disjoint and such that each of the output positions is filled in by a distinct variable.
In , it is argued that most programs are simply moded, and that often non-simply moded programs can be naturally transformed into simply moded ones. In , the class of simply moded, well-moded programs and queries is shown to be unification-free, that is, in the execution, unification can be replaced by iterated matching.
We want to note that our results for the class of simply moded well-moded programs and queries carry over to the bigger class of simply moded well-typed programs and queries such that the heads of the program clauses are input safe. We refer to , where it is shown why the results also hold for these programs and queries. We also refer to , where this class of programs and queries is introduced and shown to be unification-free.
### 7.4 The Symbolic Condition for Quasi-Termination
We will now show how the rigid quasi-acceptability condition (Proposition 12) is translated into a symbolic termination condition. The symbolic condition is a system of constraints on the introduced symbols for norm, level mapping and interargument relations (i.e. the functor, predicate and extended predicate coefficients). We first need the following concepts.
###### Definition 26 (measuring only/all input)
Let $`CB_P^E`$. Let $`|.|_s`$ be a level mapping induced by the symbol mapping $`s`$.
We say that $`|.|_s`$ measures only input positions in $`C`$ iff
* for every predicate $`p/n`$ occurring in $`C`$: if $`iO_p`$, then $`s(p_i)=0`$,
We say that $`|.|_s`$ measures all input positions in $`C`$ iff
* for every predicate $`p/n`$ occurring in $`C`$: if $`iI_p`$, then $`s(p_i)0`$, and
* for every functor $`f/m`$, $`m>0`$, occurring in an input position of an atom in $`C`$: $`s(f_i)0`$ for all $`i\{0,\mathrm{},m\}`$.
The next lemma follows from the fact that, for a well-moded program $`P`$ and a set $`S`$ of well-moded queries, the input positions of atoms in the call set $`Call(P,S)`$ are ground.
###### Lemma 8
Let $`P`$ be a well-moded program and $`SB_P^E`$ be a set of well-moded queries. Let $`|.|_s`$ be a level mapping which measures only input positions in $`Call(P,S)`$. Then, $`|.|_s`$ is rigid on $`Call(P,S)`$.
The following lemma is a corollary of the fact that, for a simply moded program $`P`$ and a set $`S`$ of simply moded queries, the output positions of atoms in the call set $`Call(P,S)`$ consist of distinct variables.
###### Lemma 9
Let $`P`$ be a simply moded program and $`SB_P^E`$ be a set of simply moded queries. Let $`|.|_s`$ be a level mapping which measures all input positions in $`CCall(P,S)`$. Then, $`|.|_s`$ is finitely partitioning on $`C`$.
In the following proposition from , a condition on a symbol mapping $`s`$ is given which ensures that the interargument relations induced by $`s`$ are valid w.r.t. the norm induced by $`s`$. We include its proof since it is essential for understanding the proposition.
###### Proposition 13
Let $`P`$ be a program and $`s`$ a symbol mapping on $`𝒞`$. If for each clause $`HB_1,\mathrm{},B_nP`$ it holds that
$$s(:[𝒜^s(B_1)\mathrm{}𝒜^s(B_n)𝒜^s(H)])$$
then for all $`p/n`$, $`R_s^{p/n}`$ is valid w.r.t. $`._s`$.
###### Proof
The union of the relations
$$R_s^{p/n}=\{(||t_1||_s),\mathrm{},||t_n||_s)|t_1,\mathrm{},t_nTerm_P\text{ and }s(𝒜^s(p(t_1,\mathrm{},t_n)))\text{ holds }\}$$
$`p/nP`$, define an interpretation of $`P`$ on the domain $`\mathrm{}`$. The condition expresses that for this interpretation, $`T_P(I)I`$ holds. Thus, the interpretation is a model and therefore each $`R_s^{p/n}`$ is a valid interargument relation. ∎
Finally, we can formulate the rigid quasi-acceptability condition in a constraint-based way. We use Lemma 8 and Lemma 9, and thus restrict our results to simply moded well-moded programs and queries.
###### Proposition 14
Let $`P`$ be a simply moded well-moded program. Let $`SB_P^E`$ be a set of simply moded well-moded queries. Let $`Tab_PPred_P`$ be a tabling for $`P`$. $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$ if
there exists a symbol mapping $`s`$ such that
1. $`|.|_s`$ measures only input positions in $`Call(P,S)`$:
* for every predicate $`p/n`$ occurring in $`Call(P,S)`$:
if $`iO_p`$, then $`s(p_i)=0`$,
2. $`|.|_s`$ measures all input positions in $`Call(P,S)B_{Tab_P}^E`$:
* for every predicate $`p/n`$ occurring in $`Call(P,S)B_{Tab_P}^E`$:
if $`iI_p`$, then $`s(p_i)0`$, and
* for every functor $`f/m`$, $`m>0`$, occurring in an input position of an atom in $`Call(P,S)B_{Tab_P}^E`$:
$`s(f_i)0`$ for all $`i\{0,\mathrm{},m\}`$,
3. all interargument relations induced by $`s`$ are valid w.r.t. the norm $`._s`$ induced by $`s`$:
$`HB_1,\mathrm{},B_nP`$:
$$s(:[𝒜^s(B_1)\mathrm{}𝒜^s(B_n)𝒜^s(H)]),$$
4. the quasi-acceptability condition w.r.t. the norm, level mapping and interargument relations induced by $`s`$ must hold:
$`HB_1,\mathrm{},B_nP`$, $`B_i,i\{1,\mathrm{},n\}`$:
$$s(:[𝒜^s(B_1)\mathrm{}𝒜^s(B_{i1})|H|^s|B_i|^s])$$
and if $`Rel(H)Rel(B_i)NTab_P`$ and $`C_2(Rel(H),Rel(B_i))`$ does not hold, then
$$s(:[𝒜^s(B_1)\mathrm{}𝒜^s(B_{i1})|H|^s>|B_i|^s]).$$
###### Proof
This symbolic condition for quasi-termination is derived from the rigid quasi-acceptability condition in a way analogous to the derivation of the symbolic condition for LD-termination from the rigid acceptability condition (see ). In order for this article to be self-contained, we include the proof. Suppose that there exists a symbol mapping $`s`$ satisfying the above condition. We prove that $`P`$ is rigid quasi-acceptable w.r.t. $`Tab_P`$ and $`S`$, and hence that $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$.
We propose as a level mapping the level mapping $`|.|_s`$ induced by $`s`$ (based on the norm $`._s`$ induced by $`s`$). Because $`|.|_s`$ measures only input positions in $`Call(P,S)`$, we have by Lemma 8 (and by the fact that $`P`$ and $`S`$ are well-moded) that $`|.|_s`$ is rigid on $`Call(P,S)`$. Also, because $`|.|_s`$ measures all input positions in $`Call(P,S)B_{Tab_P}^E`$, we have by Lemma 9 (and by the fact that $`P`$ and $`S`$ are simply moded) that $`|.|_s`$ is finitely partitioning on $`Call(P,S)B_{Tab_P}^E`$.
Take any clause $`HB_1,\mathrm{},B_n`$ in $`P`$, and any body atom $`B_i`$, $`i\{1,\mathrm{},n\}`$. Let $`\psi `$ be a substitution such that $`PB_1\psi ,\mathrm{},B_{i1}\psi `$. We prove that $`|H\psi |_s|B_i\psi |_s`$ (the proof that $`|H\psi |_s>|B_i\psi |_s`$ in case $`Rel(H)Rel(B_i)NTab_P`$ and $`C_2(Rel(H),Rel(B_i))`$ does not hold, is analogous). Condition 4 of this proposition holds for any instantiation of it, so
$$s(:[𝒜^s(B_1\psi )\mathrm{}𝒜^s(B_{i1}\psi )|H\psi |^s|B_i\psi |^s])()$$
holds. Now, we prove that for any $`1ji1`$, $`s(:𝒜^s(B_j\psi ))`$ holds. By Proposition 13 and condition 3 of this proposition, we have that, for all $`p/n`$, $`R_s^{p/n}`$ is valid w.r.t. $`._s`$. Then, since for all $`1ji1`$, $`PB_j\psi `$ holds, we have that $`s(:𝒜^s(B_j\psi ))`$ holds. So, by $`()`$ we conclude that $`s(:|H\psi |^s|B_i\psi |^s)`$ holds, which implies that $`|H\psi |_s|B_i\psi |_s`$ holds. ∎
Given a program $`P`$, a set of atoms $`S`$, and a tabling $`Tab_P`$, we can set up a symbolic condition for quasi-termination using the above proposition. By solving the generated constraints, we get a demand-driven solution for all the concepts involved in the termination analysis (norm, level mapping and model). More precisely, if a norm, level mapping and interargument relations of the given generic forms exist such that the program can be proven to quasi-terminate, then our generated set of constraints has these required instances of the generic forms as a solution.
###### Example 21
Let $`P`$ be the following program, computing the paths from a given node to the reachable nodes in a given cyclic graph:
$$\{\begin{array}{ccc}edge(a,b)\hfill & \hfill & \\ edge(b,a)\hfill & \hfill & \\ path(X,Y,[Y])\hfill & \hfill & edge(X,Y)\hfill \\ path(X,Y,[Z|L])\hfill & \hfill & edge(X,Z),path(Z,Y,L)\hfill \end{array}$$
Let $`Tab_P=\{path/3\}`$ and $`S=\{path(a,Y,L)\}`$. Then, $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$. We consider the following modes: $`edge(In,Out),path(In,Out,Out)`$. Then the program $`P`$ and query $`S`$ are simply moded and well-moded.
We prove, using the constraint-based approach of Proposition 14, that $`P`$ quasi-terminates w.r.t. $`Tab_P`$ and $`S`$. We set up the constraints:
1. First for the output positions of the predicates:
$$s(edge_2)=0,s(path_2)=0,s(path_3)=0.$$
2. For the input position of $`path/3`$:
$$s(path_1)0.$$
3. We only need a linear size expression for the $`edge/2`$ predicate. Its two clauses both give rise to the following constraint:
$$s(edge_1^e)0s(edge_2^e)0+s(edge_0^e).$$
After simplification, we get: $`s(edge_0^e)=0`$.
4. The non-recursive clause for $`path/2`$ gives rise to the following constraint:
$$s(X:[path_1Xedge_1X]).$$
This constraint has to hold for all possible values (in $`\mathrm{}`$) for $`X`$. Hence, we derive the following constraint on the symbols $`path_1`$ and $`edge_1`$:
$$s(path_1)s(edge_1).$$
The recursive clause gives rise to the following two constraints:
$$\begin{array}{c}s(X:[path_1Xedge_1X]),\hfill \\ s(X,Z:[edge_1^eXedge_2^eZpath_1Xpath_1Z]).\hfill \end{array}$$
The first constraint is the same as the one for the non-recursive clause and reduces to $`s(path_1)s(edge_1)`$. We refer to for a general methodology for solving constraints as generated by Proposition 14. Such constraints involve two types of variables: the symbolic coefficients for which we aim to fix a symbol mapping and the universally quantified variables, which express that the derived conditions should hold for any value of these; the point is to eliminate the latter variables. We briefly explain how the second constraint reduces to a system of constraints on the symbolic coefficients only. We first rewrite the second constraint into the following equivalent form:
$$s(X,Z:[edge_1^eXedge_2^eZ0path_1Xpath_1Z0]).$$
Then the idea is to derive the right hand side as a positive linear combination of the assumption in the left hand side. We do this by subtracting the left hand side of the implication from the right hand side, and by requiring that the resulting coefficients of the variables are greater than or equal to $`0`$. Doing so, we obtain the following constraints:
$$s(path_1)s(edge_1^e)0,s(edge_2^e)s(path_1)0.$$
One solution to this system of constraints is ($``$ stands for the list constructor):
$$\begin{array}{c}s(_0)=s(_1)=s(_2)=0,\hfill \\ s(edge_1)=1,s(edge_2)=0,\hfill \\ s(path_1)=1,s(path_2)=s(path_3)=0,\hfill \\ s(edge_1^e)=s(edge_2^e)=1,s(edge_0^e)=0.\hfill \end{array}$$
This gives us the following concrete norm and level mapping:
$$\begin{array}{c}t_s=0tU_P^E(._s\text{is the trivial norm}),\hfill \\ |edge(t_1,t_2)|_s=t_1_s,\hfill \\ |path(t_1,t_2,t_3)|_s=t_1_s.\hfill \end{array}$$
The interargument relation for $`\mathrm{𝑒𝑑𝑔𝑒}(t_\mathit{1},t_\mathit{2})`$ is $`t_1_st_2_s`$.
The rigid quasi-acceptability condition is satisfied using these concrete norm, level mapping and valid interargument relation. Hence, we have proven that $`P`$ quasi-terminates w.r.t. $`\{\mathrm{𝑝𝑎𝑡ℎ}/2\}`$ and $`S`$.
## 8 Conclusions, Related Work and Topics for Future Research
In this article we studied termination of tabled logic programs. We introduced two notions of universal termination under a tabled execution mechanism: quasi-termination and (the stronger notion of) LG-termination. We presented sufficient conditions (which are also necessary in case the tabling is well-chosen) for quasi-termination and LG-termination: namely quasi-acceptability and LG-acceptability. We extended the applicability by presenting modular termination conditions, i.e. conditions ensuring termination of the union $`PR`$ of two programs $`P`$ and $`R`$, where $`P`$ extends $`R`$. Finally, we investigated the problem of automatically proving quasi-termination and LG-termination. We showed that for simply moded, well-moded programs, a sufficient condition for quasi-termination and LG-termination can be given, which is formulated fully at the clause level. We pointed out how these sufficient conditions can be automated by extending the constraint-based, automatic approach towards LD-termination of .
Since all programs that terminate under LD-resolution, are quasi-terminating and LG-terminating as well, verification of termination under LD-resolution using an existing automated termination analysis (such as those surveyed in e.g. ) is a sufficient proof of the program’s quasi-termination and LG-termination. However, since there are quasi-terminating and LG-terminating programs, which are not LD-terminating, better proof techniques can and should be found. There are only relatively few works studying termination under a tabled execution mechanism. In , the special case where all predicates of the program are tabled is considered and the two notions of universal termination of a tabled logic program w.r.t. a set of queries is introduced and characterised. In , in the context of well-moded programs, Plümer presents a sufficient condition for the bounded term-size property of programs, which implies LG-termination. Holst, in , provides another sufficient condition for quasi-termination in the context of functional programming.
Our modular conditions for termination of tabled logic programs and more precisely the modular conditions which incrementally construct a level mapping for the whole program, are inspired by the modular conditions for (S)LD-resolution as given by Apt and Pedreschi in . More specifically, in , the notions of semi-recurrent program (for SLD-resolution) and of semi-acceptable program (for LD-resolution) are introduced, and modular termination proofs are presented which are based on these notions.
In the constraint-based approach towards quasi- and LG-termination, we used mode information in the presentation of the sufficient conditions. In a recent article, , Etalle et al study how mode information can be used for characterizing properties of LD-termination. They define and study the class of well-terminating programs, i.e., programs for which all well-moded queries have finite LD-derivations. They introduce the notion of well-acceptability which is based on the concept of moded level mapping; that is, a level mapping which only measures input positions. It is then shown that for well-moded programs, well-acceptability implies well-termination. Furthermore, it is proven that for simply moded well-moded programs, the notions of well-acceptability and well-termination are equivalent.
A topic for future research is to extend our results to *normal* logic programs executed under such a mixed tabled/non-tabled execution. Another topic, with an arguably more practical flavour, is to investigate how the termination conditions presented here can form the basis of a compiler that automatically decides on—or at least guides a programmer in choosing—a tabling (i.e. a set of tabled predicates) for an input program such that quasi-termination of the program is ensured. We plan to implement the constraint-based technique for automatically proving quasi-termination and LG-termination (note that a prototype implementation for automatically proving LD-termination exists). Also, it remains to be studied how our results can be extended to automatically prove quasi-termination and LG-termination for larger classes of programs and queries (i.e. for programs and queries which are not simply moded, well-moded). |
warning/0003/hep-ph0003161.html | ar5iv | text | # Perspectives on Finding the Neutrino Nature
\[
## Abstract
The possibility of determining the neutrino nature is considered in view of the most recent experimental observations. The analysis combines schemes with three and four neutrinos.The data on oscillations is put together with that from the search of neutrinoless double beta decay and results on tritium beta decay. All solar neutrino oscillation solutions are taken into account. The sensitivity of the problem on future experimental bounds from GENIUS is studied. Dirac neutrinos are shown to be unavoidable already at present in some schemes and the constraints will quickly become more stringent with future data. The consequences of including bounds from Cosmology on the neutrino content of Hot Dark Matter are commented.
\]
At present, particle physics seems to have a long awaited opportunity to discover the path towards an extension of the Standard Model. This opportunity is given by the recent confirmation of neutrino oscillations . With the currently accumulated experimental data, we have many hints on the size of neutrino masses and mixing angles. This information imposes serious constraints on the structure of the model. Unfortunately there remains a major unknown, which is the neutrino nature. As a neutral particle it can be described either by means of the Dirac equation or by means of the Majorana equation . This problem is of the utmost importance for model builders.
It has been proved many times, that with the current precision, the only experiment able to show which of the two descriptions is valid, is the search for the neutrinoless double beta decay $`(\beta \beta )_{0\nu }`$. Should this process be observed, we would be left with no other choice but to acknowledge neutrinos to be Majorana particles. As this has not been the case, we have at hand only a lower bound on the lifetime of the $`(\beta \beta )_{0\nu }`$ decaying nuclei. This alone is not enough. However when supplied with the results of neutrino oscillation experiments and the tritium beta decay mass bounds, interesting, but this time neutrino mass scheme dependent conclusions can be drawn. The situation will undoubtedly improve, when the future GENIUS experiment will furnish its first results.
In this paper, we combine the neutrino oscillation data, the tritium beta decay bounds and the present and future GENIUS bounds on the $`(\beta \beta )_{0\nu }`$ lifetime (if their results are not positive, since in the opposite case the situation is clear, as stated above) in order to derive conditions which would define the neutrino as a Dirac particle. To this end we take into account the possibility of having three or four light neutrinos. Therefore our conclusions will be valid in either case, whether the LSND results would eventually be ruled out or confirmed.
The novelty in respect to previous analyses is the consideration of the case of three and four neutrinos together, supplemented with the most recent bounds on the mixing matrix elements and the variation of the minimal neutrino mass $`(m_\nu )_{min}`$ in its full range. Furthermore we do not assume that neutrinos are Majorana particles like in so many other works .
The oscillation data is described in terms of differences of neutrino mass squares $`\delta m_{ij}^2m_i^2m_j^2`$. Define a scheme of neutrino masses as an ordering of these $`\delta m^2`$’s. For three neutrinos there are two such schemes, whereas for four neutrinos there are six. In the latter case, four schemes can be rejected by means of a combined analysis . All four possibilities are shown in Fig. 2 and Fig 3. It should be clear that with the strong ordering of $`\delta m^2`$’s ($`\delta m_{sol}^2\delta m_{atm}^2\delta m_{LSND}^2`$), it is only the largest that actually matters. Therefore for three neutrinos we need only to know $`\delta m_{atm}^23\times 10^3eV^2`$ , whereas for four $`\delta m_{LSND}^21eV^2`$ . The second important information is that the spectrum of the masses of light neutrinos is bounded from above by $`m_\nu <2.5eV`$ . With the additional information on the value of one mass, we recover the complete mass spectrum of neutrinos.
The basic quantity to consider is the effective neutrino mass, measured in $`(\beta \beta )_{0\nu }`$ decay
$$|m_\nu |\left|\underset{i=light}{}U_{ei}^2m_i\right|,$$
(1)
with $`U`$ the mixing matrix which we assume unitary. Current and future bounds on $`|m_\nu |`$ are given in Table I. At present, nothing can be said about the phases of $`U`$, therefore we allow them to have the least favorable values. Our analysis is based on the following statement: if the minimum of $`|m_\nu |`$ with respect to $`U_{ei}`$ within their allowed range, $`(|m_\nu |)_{min}`$, exceeds the experimental bound, then neutrinos are Dirac particles .
Suppose that the solar neutrino oscillations take place mostly between the first and the second neutrino states, then
$$(|m_\nu |)_{min}=\mathrm{min}\left|l^em_as^em_b\right|,$$
(2)
with
$$l^e[l_{}^e,l_+^e],s^e[s_{}^e,s_+^e],$$
(3)
and
$$l_+^e=(1s_+^e),l_{}^e=(1s_+^e)\sqrt{1\mathrm{sin}^22\theta _{sun}},$$
(4)
where in the case of three neutrinos
$$s_+^e=s_{}^e=|U_{e3}|^2,$$
(5)
and in the case of four
$$s_+^e=|U_{e3}|^2+|U_{e4}|^2,s_{}^e=\left||U_{e3}|^2|U_{e4}|^2\right|.$$
(6)
The mass parameters in Eq. 2 can be chosen in two different ways, depending on the ordering of masses of the neutrino states.
Schemes $`A_3`$ and $`A_4`$. The behavior of the minimum of the effective mass is the following
$$(|m_\nu |)_{min}=\{\begin{array}{c}s_{}^e\sqrt{(m_\nu )_{min}^2+\delta m^2}l_+^e(m_\nu )_{min},\\ 0,\\ l_{}^e(m_\nu )_{min}s_+^e\sqrt{(m_\nu )_{min}^2+\delta m^2},\end{array}$$
(7)
with $`\delta m^2\delta m_{atm}^2`$ for three neutrinos and $`\delta m^2\delta m_{LSND}^2`$ for four. The third range occurs only if $`l_{}^e>s_+^e`$. This can be rewritten as
$$\mathrm{sin}^22\theta _{sun}<\frac{12s_+^e}{(1s_+^e)^2},$$
(8)
Obviously these schemes always allow for a Majorana mass. If however the condition Eq. 8 is satisfied, then it is possible to derive a lower bound on the mass spectrum which excludes this possibility. This bound will be larger than the value of $`(m_\nu )_{min}`$ which marks the beginning of the third range in Eq. 7
$$(m_\nu )_{min}=\left(\frac{\delta m^2(s_+^e)^2}{(l_{}^e)^2(s_+^e)^2}\right)^{1/2}.$$
(9)
This number will be small in the case of three neutrinos if $`\mathrm{sin}^22\theta _{sun}`$ is not too close to one, due to the small $`\delta m_{atm}^2`$, but can quickly grow large in the case of four.
Schemes $`B_3`$ and $`B_4`$. At the condition Eq. 8 we have
$$(|m_\nu |)_{min}=l_{}^e\sqrt{(m_\nu )_{min}^2+\delta m^2}s_+^e(m_\nu )_{min}.$$
(10)
Obviously, if the masses of the neutrinos are distributed according to these schemes the neutrino nature can be determined irrespective of the lowest mass, depending solely on the experimental bound on $`|m_\nu |`$
Dependence on the mixing matrix. The condition Eq. 8 plays a crucial role in the problem. The availability of lower bounds on the neutrino mass is tightly connected to it. We therefore have to investigate the size of both sides of the inequality.
The quantity $`s_+^e`$ can be fitted to various experimental data. It turns out that the solar neutrino flux observation is the least constraining since one has a bound of $`s_+^e<.67.73`$ . Remark however that the difference between $`\chi ^2`$ for $`s_+^e=0`$ and the above result is almost irrelevant. Therefore if other experiments constrain the considered quantity to be close to zero, then solar neutrino oscillation should be viewed as a two state phenomenon and 3 neutrino fits to solar data alone seem to be of little value. This will be important subsequently. The currently strongest constraint comes from a combined analysis of atmospheric and reactor data. Namely one has $`s_+^e<.01.02`$ . However a recent fit to the new (830-920 days) data of Super-Kamionkande gives a best value of $`s_+^e.03`$. There has also been some speculation about strengthening this bound by adding some theoretical assumptions on the shape of the mass matrix. It seems that $`s_+^e0`$ is the most preferred value . In Fig. 1, the dependence of the upper bound on $`\mathrm{sin}^22\theta _{sun}`$ is shown as function of $`s_+^e`$. In the range $`s_+^e[0,.04]`$ this bound is between $`.998`$ and $`1`$.
Let us now turn to the experimental bounds on $`\mathrm{sin}^22\theta _{sun}`$. Various fits have been performed for two , three and even four neutrino oscillation scenarios. The most recent which take into account the full data (rates, energy spectrum, day-night asymmetry in the case of the MSW solution, and the seasonal variation in the case of the VO solution) give a best fit value of $`\mathrm{sin}^22\theta _{sun}<1`$. The $`95\%c.l.`$ bounds for the MSW solution are given in Table II. The most important conclusions are that in neither case do $`\mathrm{sin}^22\theta _{sun}`$ exceed the bound given by condition Eq. 8.
For the Vacuum Oscillation solution best fit value we take $`\mathrm{sin}^22\theta _{sun}=.93`$ . The above values have been used to draw Fig. 2 and Fig. 3 for the case $`s_+^e=.01`$.
Discussion. The most stringent of the oscillation solutions when question is of the neutrino nature, is the SMA. Already with the present bound one must adopt the Dirac description in the $`B_4`$ scheme. With negative results of GENIUS the same will hold true of the $`B_3`$ scheme. Currently a bound of $`m_\nu >.22eV`$ can be taken as a delimiter of the two neutrino natures for the $`A_3,A_4`$ and $`B_3`$ schemes.
At the $`95\%c.l.`$ the remaining solutions LMA and LOW define the neutrino nature if $`m_\nu >1.5eV`$, with the present bound and the $`A_3,A_4,B_3`$ schemes and $`m_\nu >1.1eV`$ in $`B_4`$. With the GENIUS I bound this can be shifted down to $`m_\nu <.16eV`$ ($`A_3`$), $`m_\nu <.14eV`$ ($`B_3`$) and $`m_\nu <.22eV`$ ($`A_4`$). Finally GENIUS II would give $`m_\nu <.05eV`$ for the $`A_3`$ and $`m_\nu <.12eV`$ for the $`A_4`$ scheme, but would completely exclude a Majorana neutrino in the $`B_3`$ scheme.
We do not have a $`95\%c.l.`$ bound for the VO solution. However, with the assumption that here also $`\mathrm{sin}^22\theta _{sun}<.98`$ the same conclusions would hold as above.
If one takes the best fit values for the different solutions, all of the bounds can be improved.
An interesting consideration is to take into account the current best fit for the Hot Dark Matter neutrino constituent. One then has $`_{light}m_\nu 2eV`$ . The SMA solution rules out in this case Majorana neutrinos in the $`A_3`$ and $`B_3`$ schemes. With a negative result from GENIUS all solutions share this property in the $`A_3,B_3`$ and $`B_4`$ schemes. Obviously, if we take the previous result $`_{light}m_\nu 45eV`$ , then the constraint is even stronger as already now most of the schemes are ruled out and GENIUS would close the situtation.
Final considerations. Several points must be stressed. First, the bound on $`|m_\nu |`$ depends strongly on the precise determination of nuclear matrix elements. It is acknowledged that they suffer from an imprecision at the level of a factor of 3 . Second, the confidence level that we assumed, $`95\%`$ corresponds to $`2\sigma `$. At $`3\sigma `$ which is $`99\%c.l.`$, the $`\mathrm{sin}^22\theta _{sun}`$ value of one is allowed. This unfortunately ruins the validity of condition Eq. 8 with the same effect on our predictions. Third, the figures have been drawn for $`\delta m_{LSND}^2=1eV^2`$. The experimentally allowed range is $`\delta m_{LSND}^2=.22eV^2`$. Adoption of some other difference of mass squared would require a rescaling of the results for the $`A_4`$ and $`B_4`$ neutrino schemes.
Conclusions. The nature of neutrinos that seemed so evasive is closer and closer to the reach of experiments. Combining oscillation data, tritium beta decay and neutrinoless double beta decay is a powerful method of constraining the parameter space in which Majorana neutrinos are allowed. In this paper we have derived bounds on the masses in several schemes and shown which schemes are or will be excluded out of consideration if neutrinos are to be truly neutral particles. We still have to wait for MiniBoone or SNO, to decide on the existence of the sterile neutrino, and for SNO, Borexino, Kamland to determine the solar neutrino solution and finally for more statistics in all the experiments to constrain more precisely the $`\mathrm{sin}^22\theta _{sun}`$.
Acknowledgments. This work was supported by the Polish Committee for Scientific Research under Grant No. 2P03B05418. J.G. would like to thank the Alexander von Humboldt-Stiftung for fellowship. |
warning/0003/cond-mat0003337.html | ar5iv | text | # HARVESTING THERMAL FLUCTUATIONS: ACTIVATION PROCESS INDUCED BY A NONLINEAR CHAIN IN THERMAL EQUILIBRIUM
## 1 Introduction
The search for mechanisms that may induce the spontaneous localization of vibrational energy in molecular materials has surfaced in a variety of contexts where such localized energy may then trigger other events. These may include switching and other threshold phenomena, chemical reactions, local melting and other deformational effects, and even detonation. In the Kramers problem a particle moving in a bistable potential is used as a model for a chemical process. The trajectory of the particle is associated with the reaction coordinate (RC). One well of the bistable potential represents the “reactant” state, the other the “product” state, and separating them is the “activation barrier.” The bistable potential is connected to a thermal environment, typically through fluctuating and dissipative terms, and every once in a while a large thermal fluctuation causes the particle to surmount the barrier and move from one well to the other. The average rate of occurrence of these events is associated with the reaction rate. This mesoscopic Langevin-type of approach admits of an underlying microscopic description of the thermal environment and its coupling to the bistable system. For instance, the usual Langevin equation with an instantaneous dissipation and Gaussian $`\delta `$-correlated fluctuations can be derived from a picture in which the system is harmonically coupled to an infinite number of harmonic oscillators with a uniform spectrum. A generalized Langevin picture involving dissipative memory terms and correlated fluctuations is associated with a more complex spectrum . It is clear, and has become a topic of considerable interest, that the nature of the environment and its coupling to the bistable system profoundly influence the transition rate.
A different but related set of problems that has attracted intense interest in recent years concerns the spontaneous localization of vibrational energy in periodic nonlinear arrays. The pioneering work of Fermi, Pasta and Ulam demonstrated that a periodic lattice of coupled nonlinear oscillators is not ergodic, and that energy in such a lattice may never be distributed uniformly. A great deal of work has since followed in an attempt to understand how energy is distributed in discrete nonlinear systems . The existence of solitons and more generally of breathers and other energy-focusing mechanisms, and the stationarity or periodic recurrence or even slow relaxation of such spatially localized excitations, are viewed as nonlinear phenomena with important consequences in many physical systems . The search for localization mechanisms that are robust even when the arrays are in a thermal environment has, on the one hand, narrowed the problem (because some localization mechanisms are fragile against thermal fluctuations) but on the other hand broadened it (because new entropy-driven localization mechanisms become possible). Thermal effects may be particularly important in biophysical and biochemical applications at the molecular level .
The interest in the distribution and motion of energy in periodic arrays arises in part because localized energy in these systems may be mobile, in contrast with systems where energy localization occurs through disorder. Localized energy that moves with little or no dispersion may appear at one location on an array and may then be able to move to another where it can be used in a subsequent process. Traditional harmonic models suffer from the fact that dispersion thwarts such a mechanism for energy transfer. There has been a surge of recent activity in an attempt to understand the thermal conductivity of nonlinear chains
The connection between the study of perfect nonlinear arrays and the Kramers problem arises because such arrays may themselves serve as models for a heat bath for other systems connected to them . Albeit in different contexts, “perfect” arrays serving as energy storage and transfer assemblies for chemical or photochemical processes are not uncommon , and literature on the subject goes back for two decades . We thus consider the following variant of the Kramers problem: a bistable system connected to a nonlinear chain, which is in turn connected to a heat bath in the usual Langevin manner (see Fig. 1). The bistable system is only connected to the environment through its embedding in the nonlinear chain, and therefore the ability of the chain to spontaneously localize thermal energy and perhaps to transport it to the location of the bistable system can profoundly affect the transition rate. We investigate the behavior of this model for different types of anharmonic chains and thereby establish the important role of the nature of the environment on these chemical model systems.
In Section 2 we discuss the energy landscape typical of various nonlinear chains in thermal equilibrium. In Section 3 the variant of the Kramers problem wherein a bistable system is connected to each of the different chains is presented. Section 4 details our results for the transition statistics in the bistable system. We compare and contrast the transition statistics in the different chains and compare them to those found in the standard (Markovian) Kramers and generalized Kramers problems. We conclude with a summary and some notes on future directions in Section 5.
## 2 Nonlinear Chains
The simplest nonlinear periodic arrays consist of masses connected by springs that may be harmonic or anharmonic. The masses may also experience a local harmonic or anharmonic potential. In a recent paper we presented a detailed view of the thermal landscape of arrays with local hard (the “$`\varphi ^4`$ model”), harmonic, or soft potentials and harmonic interactions . Here we present the complementary analysis (more interesting, it turns out, in the context of the Kramers problem) of the thermal landscape of masses connected by anharmonic springs (with no local potentials).
The spontaneous localization of energy in any system in thermal equilibrium is simply a reflection of the thermal fluctuations described by statistical mechanics and is unrelated to system dynamics. On the other hand, the way in which these fluctuations dissipate and/or move and disperse, that is, the temporal evolution of thermal fluctuations, is dictated by the system dynamics and, in particular, by the channels connecting the chain to the thermal environment (dissipation) and the masses to one another (intermolecular interactions).
We pose the following questions: 1) How is the energy distributed in an equilibrium nonlinear chain at any given instant of time, and how does this distribution depend on the anharmonicity? Can one talk about spontaneous energy localization in thermal equilibrium, and, if so, what are the mechanisms that lead to it? 2) How do local energy fluctuations in such an equilibrium array relax in a given oscillator? Are there circumstances in the equilibrium system wherein a given oscillator remains at a high level of excitation for a long time? 3) Can local high-energy fluctuations move in some nondispersive fashion along the array? Can an array in thermal equilibrium transmit long-lived high-energy fluctuations from one region of the array to another with little dispersion?
In our earlier work we showed that in harmonically coupled nonlinear chains (“diagonal anharmonicity”) in thermal equilibrium, high-energy fluctuation mobility does not occur beyond that which is observed in a harmonic chain. Herein we show that the situation might be quite different if there is “nondiagonal anharmonicity,” that is, if the interoscillator interactions are anharmonic.
Our model consists of a one-dimensional array of $`N`$ unit-mass sites, each connected by a potential $`V(x_nx_{n\pm 1})`$ to its nearest neighbors that may be harmonic or anharmonic:
$$H=\underset{n=1}{\overset{N}{}}\frac{p_n^2}{2}+\underset{n=1}{\overset{N}{}}V(x_nx_{n1}).$$
(1)
We assume periodic boundary conditions and consider three prototype potentials:
$`V_h(x)`$ $`={\displaystyle \frac{1}{2}}kx^2+{\displaystyle \frac{1}{4}}k^{}x^4`$ $`\mathrm{hard},`$ (2)
$`V_0(x)`$ $`={\displaystyle \frac{1}{2}}kx^2`$ $`\mathrm{harmonic},`$ (3)
$`V_s(x)`$ $`={\displaystyle \frac{k}{k^{}}}\left[|x|{\displaystyle \frac{1}{k^{}}}\mathrm{ln}(1+k^{}|x|)\right]`$ $`\mathrm{soft}.`$ (4)
At small amplitudes the three potentials are harmonic with the same force constant $`k`$. The independent parameters $`k`$ and $`k^{}`$ allow control of the harmonic component and the degree of anharmonicity of the chain. Elsewhere we have argued that the overarching characteristic of anharmonic oscillators is the dependence of frequency on energy. For a harmonic oscillator the frequency is $`\sqrt{k}`$ independent of energy; for a hard oscillator the frequency increases with energy, and for a soft oscillator the frequency decreases with energy.
The set of coupled stochastic equations of motion for the masses is that obtained from the Hamiltonian, Eq. (1), augmented by the usual Langevin prescription for coupling a system to a heat bath at temperature $`T`$:
$$\ddot{x}_n=\frac{}{x_n}\left[V(x_{n+1}x_n)+V(x_nx_{n1})\right]\gamma \dot{x}_n+\eta _n(t),$$
(5)
where a dot represents a derivative with respect to time. The $`\eta _n(t)`$ are mutually uncorrelated, zero-centered, Gaussian, $`\delta `$-correlated fluctuations that satisfy the fluctuation-dissipation relation $`\eta _n(t)\eta _j(t^{})=2\gamma k_BT\delta _{nj}\delta (tt^{})`$. The numerical integration of the stochastic equations for all our simulations is performed using the second order Heun’s method (which is equivalent to a second order Runge Kutta integration) with time step $`\mathrm{\Delta }t=0.005`$. In each simulation the system is initially allowed to relax for enough iterations to insure thermal equilibrium, after which we take our “measurements.”
The equilibrium results to be presented here complement our observations, presented elsewhere, on the way in which these same chains propagate an energy pulse as well as a sustained signal applied at a particular site .
A set of energy landscapes is shown in Fig. 2. Along the horizontal direction in each panel lies a thermalized chain of oscillators; the vertical upward progression shows the evolution of this equilibrium system with time. The gray scale represents the energy, with darker shading reflecting more energetic regions.
Several noteworthy features are evident in the figure. The energy fluctuations are greatest in the soft chain. This feature, seen earlier in chains with local anharmonic potentials , is a consequence of the effect that we have called entropic localization. In the soft chain not only are the thermal fluctuations greater at a given temperature, a result easily obtained from a simple virial analysis, but the free energy is minimized by a nonuniform distribution of energy that populates regions of phase space where the density of states is high. We have argued that this localization mechanism is robust against temperature increases – indeed, it becomes more effective with increasing temperature. A second distinctive feature of the soft chain is the persistence of the energy fluctuations: damping is not particularly effective for soft chains. The only other mechanism for removal of localized energy from a particular location is along the chain. This is clearly not an effective mechanism, a result that is in agreement with our analysis of the propagation of an externally applied pulse in the soft chain . The speed of propagation (in all chains) of a pulse of a given energy is essentially proportional to the average frequency associated with that energy, and in the soft chain this average frequency decreases with increasing energy . Although we do not see an obvious connection between these excitations and solitons at zero temperature (which are not entropic localization mechanisms) , there may be a closer connection with more generalized excitations such as breathers .
In the hard chain (Fermi-Pasta-Ulam chain) the total energy as well as the energy fluctuations are considerably smaller but quite mobile with little dispersion, as can easily be shown on the basis of a straightforward statistical mechanical analysis. In the hard chain the average frequency increases with energy and therefore more energetic pulses tend to travel more rapidly. We have also shown that the dispersion of energy in a hard chain is slow – this is seen here in the integrity of the spontaneously localized pulses over a much longer time than in the harmonic chain.
An important question of course concerns the parameter regimes where the differences illustrated in Fig. 2 are observed. We have chosen potential parameters that insure clear distinctions in the displacement amplitudes associated with the three potentials at the chosen temperature. The only restriction is that the temperature not be “too low,” that is, we avoid the region where all three potentials are essentially harmonic. We have chosen very low damping for the illustration. The soft energy landscape is far less sensitive to damping than the hard array. An increase in damping would readily slow down the motion of the high-energy fluctuations and would shorten their lifetime. Further, while the speed of the energy fluctuation pulses is sensitive to the potential parameters, their lifetime and dispersion properties are less so (as long as one is in the highly anharmonic regime). On the other hand, the persistence of the fluctuations in the soft array is quite sensitive to the harmonic contribution to the potential. All else remaining fixed, the landscapes remain qualitatively similar as temperature increases: the fluctuations in the soft array become even stronger relative to the others, and the pulse speeds in the hard array become even higher.
Suppose now that a bistable “impurity” is embedded in each of these chains, as illustrated in Fig. 1. When sufficient energy reaches the impurity, a transition may occur from one well to the other. The statistical and dynamical properties of these transitions are not obvious, and are explored in the next section.
## 3 Kramers Problem and Statement of our Variant
### 3.1 Traditional Kramers Problem
The original Markovian Kramers problem describes the reaction coordinate $`y`$ evolving in the bistable potential
$$V_b(y)=\frac{V_0}{4}(y^21)^2$$
(6)
according to the usual Langevin prescription
$$\ddot{y}=\frac{dV_b(y)}{dy}\gamma _b\dot{y}+\eta _b(t)$$
(7)
where $`\gamma _b`$ is the dissipation parameter (the subscript for “bistable” distinguishes this from the other dissipation parameters) and $`\eta _b(t)`$ represents Gaussian, zero-centered, $`\delta `$-correlated fluctuations that satisfy the fluctuation-dissipation relation appropriate for temperature $`T`$, $`\eta _b(t)\eta _b(t^{})=2\gamma _bk_BT\delta (tt^{})`$. The rate coefficient $`k_r`$ for transitions from one metastable well to the other is expressed as
$$k_r=\kappa k^{TST},$$
(8)
where $`k^{TST}`$ is the rate obtained from transition state theory for activated crossing, which in our units ($`V_0=1`$, frequencies $`\sqrt{2}`$ at the bottom of the two wells and unit frequency at the top of the barrier) is
$$k^{TST}=\frac{\sqrt{2}}{\pi }e^{1/4k_BT}.$$
(9)
This is the highest possible rate because it assumes no recrossings of the barrier when the particle moves from one well to the other. The “transmission coefficient” $`\kappa <1`$ captures the effects of recrossings. The dependence of $`\kappa `$ on the various parameters of the problem has been the subject of intense study over many years . Its dependence on $`\gamma _b`$ and temperature is exemplified in the simulations shown in Fig. 3. In particular, we note the occurrence of a maximum: as predicted by Kramers, the transmission coefficient at high friction (diffusion-limited regime) decays as $`\gamma _b^1`$ (and is independent of temperature); at low friction (energy-limited regime) Kramers predicted that $`\kappa `$ is proportional to $`\gamma _b/k_BT`$.
An important generalization of the original Kramers problem, the so-called Grote-Hynes problem , reformulates the model in terms of the generalized Langevin equation
$$\ddot{y}=\frac{dV_b(y)}{dy}_0^t𝑑t^{}\mathrm{\Gamma }(tt^{})\dot{y}(t^{})+\eta _b(t)$$
(10)
where $`\mathrm{\Gamma }(tt^{})`$ is a dissipative memory kernel and the fluctuation-dissipation relation is now generalized to $`\eta _b(t)\eta _b(t^{})=k_BT\mathrm{\Gamma }(tt^{})`$. The dissipative memory kernel reflects the dynamics of the thermal environment and is characterized by its own time scales. A frequent choice is an exponential, but other forms that have been used include a Gaussian and a decaying oscillatory memory kernel. The introduction of additional parameters of course changes the behavior of the transmission coefficient.
The main point here is to call attention to the fact that the transmission coefficient has the same value for two different values of the dissipation parameter $`\gamma _b`$, and that therefore one can not conclude whether the system is in one regime (diffusion-limited) or another (energy-limited) simply from the value of the transition rate. One needs to know the trend with changing dissipation parameter, and one requires further information about the dynamics underlying a given transition rate. Not surprisingly, these dynamics turn out to be entirely different in different regimes . The time dependence of the transmission coefficient is a direct reflection of the explicit trajectories of the particle as it transits from one well to the other. A number of investigators have looked at the time dependence of the transmission coefficient in the diffusion-limited and energy-limited regimes, and also at the effect of different types of memory kernels .
Of interest to us here are the different dynamical behaviors in the diffusion-limited regime and the energy-limited regime. In Fig. 4 we show two views of each of two typical trajectories of the reaction coordinate for the Markovian Kramers problem. The transmission coefficients associated with these two trajectories are not too different (see below), but
they correspond to different damping, placing them on opposite sides of the turnover in the $`\kappa `$ vs $`\gamma _b`$ curve. The trajectory in the first panel is in the diffusion-limited regime; the third panel shows an expanded view of a portion of this trajectory. The particle performs rather erratic motion within one well and once in a while it surmounts the barrier. When the particle surmounts the barrier it does not spend much time in the barrier region before being trapped again in one or the other well. The crossing trajectories thus tend to involve only one or a very small number of crossings/recrossings. The trajectory in the second panel, a portion of which is expanded in the fourth panel, is energy-limited. The particle performs a fairly periodic motion within one well. Barrier crossing events tend to retain the particle in the barrier region for several recrossings; a phase space analysis shows that the associated trajectories are rather smooth oscillations from one side to the other of the potential well above the barrier . Correlation functions associated with these trajectories are presented and discussed in Section 4.
### 3.2 Variant of the Kramers Problem
We would like to understand the way in which the very different thermal landscapes described in Section 2 affect the dynamics of a reaction coordinate evolving in a bistable “impurity” embedded in these environments. The connection of the bistable impurity to the thermal environment occurs only through its connection to the chain, that is, we set $`\gamma _b=0`$.
We need to specify how the bistable system interacts with the chain. We insert the impurity along the chain between sites $`i`$ and $`i+1`$ and connect it to each of these two sites (see Fig. 1). It is customary to choose a simple interaction potential with a harmonic dependence $`V_{int}(x,y)(xy)^2`$ for each chain site connected to the impurity. Here $`y`$ is the reaction coordinate and $`x`$ stands for the coordinate of the chain site connected to the impurity. However, this interaction tends to destabilize the bistability in that it causes the neighbors to pull the bistable particle toward the barrier rather than toward its natural metastable states. The interaction thus lowers the barrier of the bistable impurity. Since we do not want to “bias” the problem in this way, we have chosen an interaction that instead tends to favor the already metastable states:
$$V_{int}(x,y)=\frac{k_{int}}{2}\left(\frac{y^21}{2}x\right)^2.$$
(11)
Near the bistable minima (which are shifted by the interaction) the total potential for the reaction coordinate is still harmonic, and near the maximum at $`y=0`$ it is still parabolic. The barrier height is modulated by the motion of the neighbors (somewhat reminiscent of the barrier fluctuations in resonant activation problems). At large values of $`y`$ the interaction hardens the bistable potential.
The equations of evolution then have the following contributions. For a site in the chain not connected to the bistable impurity we have, as before, Eq. (5). For the bistable impurity
$$\ddot{y}=\frac{dV_b(y)}{dy}\frac{V_{int}(x_{i+1},y)}{y}\frac{V_{int}(x_i,y)}{y}.$$
(12)
For the site to the left of the bistable impurity
$$\ddot{x}_i=\frac{V(x_ix_{i1})}{x_i}\frac{V_{int}(x_i,y)}{x_i}\gamma \dot{x}_i+\eta _i(t)$$
(13)
and similarly for $`x_{i+1}`$. Comparisons are made for the same temperature, damping coefficients, and interaction parameter ($`k_{int}=0.1`$ throughout this work) varying only the nature of the chain.
Figure 5 shows trajectories of the bistable impurity embedded in each of the three chains. In the hard chain the trajectory is rather similar to that of a Markovian Kramers particle in the diffusion-limited regime, while in the soft chain it is closer to that of the energy-limited regime. This is a direct reflection of the behavior seen in Fig. 2, that is, of the fact that in the hard chain independent thermal fluctuations created elsewhere along the chain have a good chance of reaching the bistable impurity, causing erratic motion. An occasional large fluctuation causes a transition over the barrier, usually unaccompanied by recrossings: the same energy mobility that brings independent fluctuations to the impurity also makes it easy for the impurity to then lose a particular energy fluctuation back to the chain. In the soft chain, on the other hand, the particle performs fairly periodic motion within one well. Only fluctuations in the sites immediately adjacent to the impurity can excite the impurity; fluctuations originating elsewhere do not travel to the impurity. Strong fluctuations are therefore rarer but more energetic and more persistent, so transition events occur less often. However, once such a fluctuation occurs it tends to remain in that region for a long time; the reaction coordinate therefore recrosses the barrier a large number of times until it eventually loses the excess energy and is trapped again in one of the wells.
A second set of trajectories associated with the same bistable impurity in the same three chains at the same temperature but with a (10-fold) higher dissipation parameter is shown in Fig. 6. Not surprisingly, the trajectories are now more similar to one another, but nevertheless there are still important and revealing differences that will be made evident in our discussion in the next section. Furthermore, a comparison of the two sets will allow important observations concerning the trends associated with increased damping.
In the next section we provide a quantitative characterization of the differences in the trajectories and a comparison of these results with those of the traditional Kramers problem.
## 4 Results for Transition Rates
A useful description of the bistable system in different regimes is provided by the normalized correlation function
$$C(\tau )\frac{y(t+\tau )y(t)}{y^2(t)}$$
(14)
where the brackets indicate an average over $`t`$. Since $`y(t)=0`$, this correlation function decays to zero. When the thermal environment strongly and rapidly changes the particle momentum, the trajectory is erratic and the correlation function decays monotonically and exponentially. The decay time is a measure of the mean time between crossing events from one well to the other, and its inverse can be identified with the transition rate $`k_r`$. If on the other hand the effects of the thermal environment are weak, then the trajectory is determined mainly by the deterministic potential and remains correlated over much longer periods of time.
The correlation functions for the Markovian Kramers trajectories of Fig. 4 are shown in Fig. 7. In the high-dissipation regime the correlation function is monotonic and decays exponentially over essentially all times. This is a reflection of the essentially random motion within each well and between wells (the correlation functions for portions of the trajectory entirely within one well are also monotonically decreasing, albeit not to zero). The slope of the high-$`\gamma _b`$ curve in the right panel leads to a mean time between crossings of $`\tau _c250`$.
The oscillations in the low-$`\gamma _b`$ correlation function reflect mainly the systematic periodic motion of the particle within each well, i. e., of the portions of the trajectory that evolve for a long time near $`y=1`$ or near $`y=1`$. The period of these oscillations for the parameters used here is $`t_{bottom}=\sqrt{2}\pi `$, and this is very nearly the period of the oscillations in the figure. Crossing events from one well to the other are mostly separated by long times and are essentially independent (however, see further discussion below). Hence the logarithmic rendition in the right panel gives a straight line. Its slope leads to a mean time between crossing events of $`\tau _c450`$.
The Kramers correlation functions serve as a point of reference for an interpretation of the correlation functions associated with our variant of the Kramers problem. These are shown in a number of figures starting with Fig. 8, which shows the correlations functions associated with the trajectories in Figs. 5 and 6. We stress that in each panel the $`\gamma `$ and $`k_BT`$ are the same in all cases, as is the coupling of the chain to the bistable system; only the nature of the chain has changed. The first panel shows a correlation function for the harmonic chain that is oscillatory at early times, and quite similar to the energy-limited Markovian Kramers case (see also the corresponding trajectories in Figs. 5 and 4). We conjecture that the harmonic chain provides a thermal environment comparable to the low-damping Markovian Kramers environment. The correlation function associated with the hard chain is similar to the behavior at higher damping in the Kramers case. The correlation function associated with the soft chain also decays in an oscillatory fashion, but in a more complex way than in the energy-limited Markovian Kramers case. The alternation in the amplitudes is a consequence of the presence of sustained bursts of energy that cause a finite fraction of the trajectory to occur above the barrier, leading to many correlated recrossing events. The particle oscillates above the barrier for intervals much longer than in the Markovian Kramers trajectory. The typical oscillation period above the barrier is about twice as long as $`t_{bottom}`$ in our example (detailed discussions of these times can be found in our earlier work ). This effect is already slightly visible in the low-$`\gamma _b`$ Kramers correlation function in Fig. 7, but it is much stronger in the soft chain. To reproduce this behavior in the Kramers model it is necessary to consider the generalized Kramers model with a memory friction: there is clearly an additional memory effect in the soft chain that allows the energy to remain trapped in the region of the bistable impurity for a long time. This is in accord with the notion that transitions in the soft chain are caused by local thermal fluctuations in the nearest neighbors of the impurity. The impurity may periodically exchange energy with these neighbors before the fluctuation eventually dissipates away, and this causes repeated recrossings. This in turn leads to the conclusion that the memory friction in the Kramers model needed to reproduce the soft chain environment would most likely be oscillatory .
The periods of oscillations in the harmonic and soft chains are somewhat different from those of the Kramers curve in Fig. 7 and from each other. This is due to differences in the effective potentials.
The eventual decay of the correlation functions for all three chains is exponential. We find for the times between transition events (single or bursts as appropriate) $`\tau _c101`$ for the hard chain, $`\tau _c453`$ for the harmonic chain, and $`\tau _c1540`$ for the soft chain.
A similar set of correlation functions associated with the higher-damping-parameter trajectories of Fig. 6 is shown in the right panel of Fig. 8. The dynamics of the bistable system in the hard chain with increasing damping does not change in character, whereas the oscillations in the harmonic and soft chains become less pronounced as these systems move toward the diffusion-limited regime. First we note that all the curves become steeper, which translates to a shorter mean time between crossing events and therefore a higher transition rate for all three chains. The specific values we obtain are $`\tau _c72`$ for the hard chain, $`\tau _c170`$ for the harmonic chain, and $`\tau _c215`$ for the soft chain. The decrease in time between crossing events is most pronounced for the soft chain. This is consistent with the notion that the soft chain is in the energy-limited regime where small increases in effective damping cause the greatest increases in the transition rate (see Fig. 3). The harmonic chain lies closer in this sense to the turnover region, and the hard chain even closer yet. The second point is that this apparent trend for the hard chain indicates that it, too, lies on the low-damping side of the turnover in spite of the diffusion-limited aspects of its dynamics. This is the reason for the very small oscillations visible at the earliest times in the hard chain correlation functions. It is apparent that neither the trajectory itself nor even the shape of the correlation function at one value of the damping provides unequivocal information to determine which side of the turnover regime one is on; it is necessary to investigate the trend.
It is interesting to investigate whether our variants of the thermal environment can actually be “pushed” across the turnover point by increasing the damping on the chain. For this purpose we present a series of correlation functions for each of the chains for different values of the damping parameter. The other parameters, including the temperature, remain fixed and equal to the values given earlier.
Figure 9 shows results for the soft and harmonic chains. With increasing damping the early-time oscillations in the correlation function in the soft chain first lose some of the “alternation” feature typical of a long oscillatory dissipative memory kernel and eventually the correlation function loses its oscillatory character altogether. The crossing rate continues to increase as damping increases, so throughout this series one is still on the low damping side of the turnover. The right panel of Fig. 9 shows the correlation functions for the impurity in the harmonic chain. The trend is similar to that of the soft chain but, in all respects, indicative of the fact that the harmonic environment is closer to the turnover region than the soft environment. Thus the oscillations disappear sooner, and the increase in the transition rate with increasing damping is smaller.
Perhaps the most interesting features are seen in Fig. 10. Here we clearly see the very small short-time oscillations, which disappear as damping increases. The transition rate is quite insensitive to damping in the range $`\gamma =0.005`$$`0.1`$ shown in the figure (the line for $`\gamma =0.2`$, not explicitly shown, also falls in the same regime). The turnover value must therefore be within this range. To ascertain if this is so, we also exhibit the correlation function for a considerably larger value of the dissipation parameter, $`\gamma =1.0`$. The slower decay for $`\gamma =1`$ is clear in both panels.
## 5 Conclusions
There has been a dearth of information on the effects on the activation process of nonlinearities in the environment. We have taken an approach here that goes part way, much in the tradition of modeling efforts for a variety of systems interacting with a complex environment: the “immediate surroundings” of the reaction coordinate are described microscopically, while the interaction of this immediate environment with other degrees of freedom is handled phenomenologically.
We find that the dynamics of the activation process in some parameter regimes are profoundly affected by the nature of the chain. If the damping parameter connecting the chain to the heat bath is sufficiently low, a soft chain provides an environment very similar to that of the Grote Hynes model with an oscillatory memory kernel in the energy-limited regime , while a hard chain provides an environment akin to that of the Kramers model in the diffusion-limited regime . This in turn means that in such parameter regimes the hard chain is a more effective mediator of the activation process than is the soft chain.
A number of interesting questions concerning these systems are currently under investigation. One concerns the influence of boundary conditions on the behavior that we have described . A second problem concerns the effect on the reaction coordinate of a pulse or a sustained signal applied somewhere else along the chain. We have showed that the propagation of such a pulse or signal is strongly affected by the nature of the chain , and we expect these differences in turn to affect the response of a bistable impurity to these excitations. Such models are interesting in the context of physical or biophysical situations wherein energy is released at some location (provided perhaps by a chemical reaction or an absorption process at that location), which must then move to another location (that of the bistable impurity) to effect some further chemical process (represented by the activation process). The usual linear chain models are plagued by the excessive dispersion that would make such transmission inefficient. Nonlinearities in the environment may provide the necessary mobility with little attendant dispersion, thus greatly increasing the efficiency of such a process.
## Acknowledgments
R. R. gratefully acknowledges the support of this research by the Ministerio de Educación y Cultura through Postdoctoral Grant No. PF-98-46573147. A. S. acknowledges sabbatical support from DGAPA-UNAM. This work was supported in part by the Engineering Research Program of the Office of Basic Energy Sciences at the U.S. Department of Energy under Grant No. DE-FG03-86ER13606, and in part by the Comisión Interministerial de Ciencia y Tecnología (Spain) Project No. DGICYT PB96-0241. |
warning/0003/math-ph0003038.html | ar5iv | text | # The canonical transformations of the dynamical multiparameter systems as recurrence relations for the models on the grating
## I Introduction
Many equations of the theoretical physics especially of quantum mechanics are linear. In this connection there is a possibility of constructing solutions of these equations searching the canonical transformations of the linear dynamical systems (but this canonical transformations are different from those of quantum theory). These transformations can be formulated in terms of the creation and annihilation operators for some “quantum numbers” of the corresponding quantities that conserve their values. The indicated possibility was considered in , where the method of constructing the recurrence relation for searching for the eigenvalues and eigenfunctions of linear operators is proposed. This method is restricted by the case of one-dimensional systems described by the Lagrangian containing the spectral parameters in a potential part only.
In the present paper the generalization of this method for the case of the multidimensional systems and systems with the Lagrangian containing spectral parameters in a kinetic part too (for example, for the radial equation in hydrogen-like atom theory) is constructed.
The physical meaning of the considered canonical transformations for the classical theory differs from that for the quantum theory. The general point is that these transformations can be considered as translations on the multidimensional grating in the phase space. In the classical theory the parameters of the grating’s knots define the different configurations of considered class of dynamical systems (for example, the set of mechanical oscillators with different masses and spring rigidities). Conditions of the invariance under the canonical transformations along the grating define the admissible grating parameters and, therefore, solutions and a spectrum of oscillations.
In quantum theory the parameters of the knots give the possible states of the same quantum system. In its turn the translation along the grating is defined by the creation and annihilation operators stipulating the transition from one state to another.
The canonical recurrence transformations obtained in this way are the analogies of the known Beclund’s transformations for finite-dimensional dynamical systems. Note that they do not coincide by their form with the standard recurrence relations cited in handbooks. The difference between them is explained by the way of their construction. The main here is that the canonical recurrence relations conserve the symplectic structure of the corresponding dynamical systems in contrast with the standard ones. That is why from the standpoint of the theory of dynamical systems the canonical recurrence relations have a certain advantage. This relations can be reduced to the standard ones by the corresponding normalization of eigenfunctions.
## II A canonical map
Let us consider the sets $`𝒟=\{q,p,H\}`$ and $`\stackrel{~}{𝒟}=\{\stackrel{~}{q},\stackrel{~}{p},\stackrel{~}{H}\}`$ of the $`s`$-dimensional dynamical systems with the generalized coordinates $`q=q(t)\{q^1(t),..,q^s(t)\}`$, $`\stackrel{~}{q}=\stackrel{~}{q}(t)\{\stackrel{~}{q}^1(t),\mathrm{},\stackrel{~}{q}^s(t)\}`$, momenta $`p=p(t)\{p_1(t),\mathrm{},p_s(t)\}`$, $`\stackrel{~}{p}=\stackrel{~}{p}(t)\{\stackrel{~}{p}_1(t),\mathrm{},\stackrel{~}{p}_s(t)\}`$ and Hamiltonians $`H=H(p,q,t)`$, $`\stackrel{~}{H}=\stackrel{~}{H}(\stackrel{~}{p},\stackrel{~}{q},t)`$.
We shall call the mapping $`:𝒟\stackrel{~}{𝒟}`$ the canonical one, if it is defined by the set $`=\{F\}`$ of generating functions $`F=F(q,\stackrel{~}{q},t)`$, so that for an arbitrary system $`(q,p,H)𝒟`$ and for some function $`F`$ there is the system $`(\stackrel{~}{q},\stackrel{~}{p},\stackrel{~}{H})\stackrel{~}{𝒟}`$, for which the relations
$$p_i=\frac{F}{q^i},\stackrel{~}{p}_i=\frac{F}{\stackrel{~}{q}^i},\stackrel{~}{H}=H+\frac{F}{t}(i,j,k,\mathrm{}=1,2,\mathrm{},s),$$
(1)
are realized.
If $`𝒟=\stackrel{~}{𝒟}`$, we shall say that given set of dynamical systems $`𝒟`$ is invariant under the canonical transformation $`F`$.
Let $`𝒟`$ be a parametrized class of dynamical systems described by the Lagrangian
$$L=L(q,\dot{q},t,\lambda ,\mu _a)=\frac{1}{2}\left[g_{ij}\dot{q}^i\dot{q}^j(U_{ij}\lambda g_{ij})q^iq^j\right],$$
(2)
where $`\dot{q}=dq/dt`$, $`g_{ij}=g_{ij}(\mu _a,t)`$, $`U_{ij}=U_{ij}(\mu _a,t)`$ are some functions, $`detg_{ij}0`$, $`\{\lambda ,\mu _a\}D`$, $`D`$ is some set of parameters, $`a=1,\mathrm{},d`$. The ordinary summation convention works here and on.
If we write the Lagrange-Euler equations for the system (2) in the form
$$\widehat{P}_k^lq^kg^{li}\left(\frac{d}{dt}g_{ik}\frac{d}{dt}+U_{ik}\right)q^k=\lambda q^l$$
(3)
we shall make a conclusion that $`\lambda `$ is a spectral parameter of the eigenvalue problem of the $`d`$-parametric set of the linear operators $`\widehat{P}_k^l(\mu _a,t)`$. For the Hamiltonians of the set (2) we have
$$H(p,q,t,\lambda ,\mu _a)=\frac{1}{2}g^{ij}p_ip_j\frac{1}{2}(U_{ij}\lambda g_{ij})q^iq^j,$$
(4)
where
$$p_i=g_{ij}\dot{q}^j,(g^{ik}g_{kj}=\delta _j^i).$$
(5)
The condition of the invariance of the set $`𝒟`$ of dynamical systems under the canonical transformations $``$ is
$$\stackrel{~}{H}(\stackrel{~}{p},\stackrel{~}{q},t,\stackrel{~}{\lambda },\stackrel{~}{\mu }_a)=H(\stackrel{~}{p},\stackrel{~}{q},t,\stackrel{~}{\lambda },\stackrel{~}{\mu }_a),\{\stackrel{~}{\lambda },\stackrel{~}{\mu }_a\}D.$$
(6)
Under sequential actions of the series of the canonical transformations a sequence of the systems of the type (2) and the parameters $`\{\lambda _m,\mu _{n_a}|m,n_a=1,2,\mathrm{}\}D`$ appears. Introduce the collective indices $`K=(m,n_a)`$, $`\stackrel{~}{K}=(\stackrel{~}{m},\stackrel{~}{n}_a)`$ and designate $`(\lambda _m,\mu _{n_a})\sigma _K`$, $`(\lambda _{\stackrel{~}{m}},\mu _{\stackrel{~}{n_a}})\sigma _{\stackrel{~}{K}}`$. Let the corresponding coordinates and momenta be $`q_K,q_{\stackrel{~}{K}}`$ and $`p_K,p_{\stackrel{~}{K}}`$. Further, $`F_{K\stackrel{~}{K}}=F_{K\stackrel{~}{K}}(q_K,q_{\stackrel{~}{K}},t)`$ be the generating function of the canonical transformation $`F`$: $`𝒟𝒟`$, so that $`\{q_K,p_K,H(p_K,q_K,t,\sigma _K)\}`$ $``$ $`\{q_{\stackrel{~}{K}},p_{\stackrel{~}{K}},H(p_{\stackrel{~}{K}},q_{\stackrel{~}{K}},t,\sigma _{\stackrel{~}{K}})\}`$. Then the conditions (1), (6) give the following equation for the generating function
$$\frac{F_{K\stackrel{~}{K}}}{t}+H(\frac{F_{K\stackrel{~}{K}}}{q_K},q_K,\sigma _K,t)=H(\frac{F_{K\stackrel{~}{K}}}{q_{\stackrel{~}{K}}},q_{\stackrel{~}{K}},\sigma _{\stackrel{~}{K}},t).$$
(7)
Further we shall write the indices $`K`$ and $`\stackrel{~}{K}`$ for the generating functions only, and for an arbitrary physical variable $`f`$ we shall write $`f`$ or $`\stackrel{~}{f}`$ instead of $`f_K`$ or $`f_{\stackrel{~}{K}}`$ respectively. Then it is possible to rewrite the condition of the invariance (7) for the system (2) with the Hamiltonian (4) in the following form:
$$2\frac{F_{K\stackrel{~}{K}}}{t}+g^{ij}\frac{F_{K\stackrel{~}{K}}}{q^i}\frac{F_{K\stackrel{~}{K}}}{q^j}\stackrel{~}{g}^{ij}\frac{F_{K\stackrel{~}{K}}}{\stackrel{~}{q}^i}\frac{F_{K\stackrel{~}{K}}}{\stackrel{~}{q}^j}=(U_{ij}\lambda g_{ij})q^iq^j(\stackrel{~}{U}_{ij}\stackrel{~}{\lambda }\stackrel{~}{g}_{ij})\stackrel{~}{q}^i\stackrel{~}{q}^j.$$
(8)
## III A recurrence relations as a canonical map
Let us consider the problem of searching for the solutions of Eq.(8). The canonical map, that converts the linear dynamical systems into linear ones, must be linear. That is why the generating function $`F_{K\stackrel{~}{K}}`$ must be a quadratic function of the generalized coordinates $`q,\stackrel{~}{q}`$. Therefore we search for $`F_{K\stackrel{~}{K}}`$ in the form
$$F_{K\stackrel{~}{K}}=\frac{1}{2}\left(2\gamma _{ij}q^i\stackrel{~}{q}^jb_{ij}q^iq^jc_{ij}\stackrel{~}{q}^i\stackrel{~}{q}^j\right),$$
(9)
where $`\gamma _{ij},b_{ij},c_{ij}`$ are some functions. Substituting Eq.(9) for Eq.(8) and equating the coefficients at $`q^iq^j`$, $`q^i\stackrel{~}{q}^j`$, $`\stackrel{~}{q}^i\stackrel{~}{q}^j`$, we obtain
$`2\dot{\gamma }_{rs}g^{ij}(\gamma _{ir}b_{js}+\gamma _{jr}b_{is})=\stackrel{~}{g}^{ij}(\gamma _{is}c_{jr}+\gamma _{js}b_{ir}),`$ (10)
$`\dot{b}_{rs}+g^{ij}b_{ir}b_{js}+\lambda g_{rs}U_{rs}=\stackrel{~}{g}^{ij}\gamma _{ir}\gamma _{js},`$ (11)
$$\dot{c}_{rs}+g^{ij}\gamma _{ir}\gamma _{js}=\stackrel{~}{g}^{ij}c_{ir}c_{js}\stackrel{~}{U}_{rs}+\stackrel{~}{\lambda }\stackrel{~}{g}_{rs}.$$
(12)
If we find one of the particular solutions $`\gamma _{ij},b_{ij},c_{ij}`$, then, using Eqs.(1), (5), (9), it will be easy to construct the binomial recurrence relations for the eigenfunctions $`q(t)`$ and $`\stackrel{~}{q}(t)`$ of the set of operators $`\widehat{P}_k^l`$
$$\left(b_{ij}+g_{ij}\frac{d}{dt}\right)q^j=\gamma _{ij}\stackrel{~}{q}^j,$$
(13)
$$\left(c_{ij}\stackrel{~}{g}_{ij}\frac{d}{dt}\right)\stackrel{~}{q}^j=\gamma _{ij}q^j.$$
(14)
The obtained relations, increasing and decreasing $`K`$, are the analogies of the well-known Beclund’s transformations for finite-dimensional systems. They are also the generalization of the known relations of the factorization method , . It means, that Eqs.(13), (14) install the correlation between the canonical transformation method and factorization one.
The algebraic trinomial recurrence relations can be obtained considering the sum of the generating functions of the two sequential transformations
$$F_{K\stackrel{~}{K}\stackrel{~}{\stackrel{~}{K}}}=F_{K\stackrel{~}{K}}+F_{\stackrel{~}{K}\stackrel{~}{\stackrel{~}{K}}}.$$
(15)
According to Eq.(1) we have $`F_{K\stackrel{~}{K}\stackrel{~}{\stackrel{~}{K}}}/\stackrel{~}{q}=0`$, from which the linear algebraic recurrence relation
$$\gamma _{ij}q^j(c_{ij}+\stackrel{~}{b}_{ij})\stackrel{~}{q}^j+\stackrel{~}{\gamma }_{ij}\stackrel{~}{\stackrel{~}{q}}^j=0,$$
(16)
follows. Here $`\stackrel{~}{b}_{ij},\stackrel{~}{c}_{ij},\stackrel{~}{\gamma }_{ij}`$ are the coefficients of a quadratic form of the type (9) for the generating function $`F_{\stackrel{~}{K}\stackrel{~}{\stackrel{~}{K}}}`$.
The generating function $`F_{K\stackrel{~}{\stackrel{~}{K}}}`$ of the canonical transformation $`(q,p)(\stackrel{~}{\stackrel{~}{q}},\stackrel{~}{\stackrel{~}{p}})`$, which is a composition of the canonical transformations $`(q,p)(\stackrel{~}{q},\stackrel{~}{p})`$ and $`(\stackrel{~}{q},\stackrel{~}{p})(\stackrel{~}{\stackrel{~}{q}},\stackrel{~}{\stackrel{~}{p}})`$, can be constructed excluding the intermediate state $`\stackrel{~}{q}`$ from Eq.(15) by means of Eq.(16). It makes possible to simplify the procedure of searching for the generating function of the transformation $`K\stackrel{~}{K}`$ decomposing it into more simple steps of the calculating the elementary generating functions of the transformations: $`K=\{m,n_1,\mathrm{}\}\{m+1,n_1,\mathrm{}\}\{m+1,n_1+1,\mathrm{}\}\mathrm{}\stackrel{~}{K}`$, and after this to construct their composition.
From the standpoint of classical mechanics we consider here the canonical transformation theory of an infinite-dimensional system with the action
$$S=\frac{1}{2}𝑑t\underset{K}{}\left\{g_{ij}\dot{q}_K^i\dot{q}_K^j+(U_{ij}\lambda _mg_{ij})q_K^iq_K^j\right\},$$
describing a multicomponent model on a $`(d+1)`$-dimensional grating. The values of the parameters $`\sigma _K=\{\lambda _m,\mu _{n_1},\mathrm{},\mu _{n_d}\}`$ that define grating knots are determined from the condition of an invariance of the dynamical system under grating translations induced by the canonical transformations of the system.
One more interesting possibility connected with the method of the inverse scattering problem is shown below by the examples of more simple dynamical systems.
For the particular case of the set of $`d`$-component three-parametrical dynamical systems $`(s=1)`$ let us denote: $`g_{11}m(\nu ,t)`$, $`U_{11}m(\nu ,t)U(\mu ,t)`$, where $`\mu ,\nu `$ are parameters. Write the Lagrangian (2) in the form
$$L=\frac{m(\nu ,t)}{2}\left[\dot{q}^2+(U(\mu ,t)\lambda )q^2\right].$$
(17)
It is convenient to find the generating function $`F_{K\stackrel{~}{K}}`$ in such a form
$$F_{K\stackrel{~}{K}}=mS,S=\frac{1}{2a}(2\gamma q\stackrel{~}{q}bq^2c\stackrel{~}{q}^2),$$
(18)
where $`\gamma `$ is some unknown constant, $`a,b,c`$ are desired functions of $`t`$. The conditions (1) may be rewritten in the form
$$\dot{q}=\frac{S}{q},\dot{\stackrel{~}{q}}=\frac{m}{\stackrel{~}{m}}\frac{S}{\stackrel{~}{q}},\stackrel{~}{H}=H+\frac{}{t}(mS).$$
(19)
The recurrence relations
$$\left(b+a\frac{d}{dt}\right)q=\gamma \stackrel{~}{q},\left(ca\frac{m}{\stackrel{~}{m}}\frac{d}{dt}\right)\stackrel{~}{q}=\gamma q$$
(20)
and equation for the function $`S`$
$$\frac{2}{m}\frac{}{t}(mS)+\left(\frac{S}{q}\right)^2\frac{m}{\stackrel{~}{m}}\left(\frac{S}{t}\right)^2(\lambda U)q^2\frac{\stackrel{~}{m}}{m}(\stackrel{~}{\lambda }\stackrel{~}{U})\stackrel{~}{q}^2=0.$$
(21)
follow from Eq.(6), (19). Eqs. (10)–(12) for determining $`a,b,c,\gamma `$ acquire the form
$$\dot{a}=\frac{\dot{m}}{m}a+\frac{m}{\stackrel{~}{m}}cb,$$
(22)
$$\frac{m}{\stackrel{~}{m}}\dot{c}+\dot{b}=(\lambda \stackrel{~}{\lambda }+\stackrel{~}{U}U)a,$$
(23)
$$a\dot{b}=(\lambda U)a^2+\frac{m}{\stackrel{~}{m}}(bc\gamma ^2).$$
(24)
The system of equations (22)–(23) is linear with respect to the unknown functions $`a,b,c`$ and does not contain the constant $`\gamma `$, while the equation (24) is quadratic and contains $`\gamma `$. The procedure of the searching for the solutions of the system (22)–(24) consists in the finding a more simple particular solution of the homogeneous linear undetermined system (22)- (23) for which the quadratic equation (24) is satisfied at the some unknown parameters $`\{\stackrel{~}{\lambda },\stackrel{~}{\mu },\stackrel{~}{\nu }\}`$. Then the constant $`\gamma `$ will be determined.
The algebraic trinomial recurrence relations (without derivatives) and their compositions can be found by the way analogous to obtaining Eq.(16) (see also). As a result we shall have
$$\frac{\gamma _{KK+1}}{a_K}q_K\left(\frac{c_{K+1K+1}}{a_K}+\frac{b_{K+1K+1}}{a_{K+1}}\right)q_{K+1}+\frac{\gamma _{K+1K+2}}{a_{K+1}}q_{K+2}=0,$$
(25)
where $`K`$ is the collective index of the totality of the parameters $`\{\lambda _m,\mu _n,\nu _l\}`$, i.e. if $`K=\{m,n,l\}`$ then $`K+1`$ means that one of the indices $`m,n,l`$ is increased by one. The coefficients in Eq.(25) correspond to that of the generating functions $`S_{KK+1}`$ and $`S_{K+1K+2}`$ of the type (18) of the transformations $`q_Kq_{K+1}`$ and $`q_{K+1}q_{K+2}`$.
Writing the similar equations for the sequence of the transformations $`q_{K+1}q_{K+2}q_{K+3}`$ and excluding the coordinate $`q_{K+2}`$ by means of Eq.(25), we obtain the two relations, which can be written in the matrix form
$$Q_{K+2}=B_KQ_K,$$
(26)
where
$$Q_K=\left(\begin{array}{c}q_K\\ q_{K+1}\end{array}\right),Q_{K+2}=\left(\begin{array}{c}q_{K+2}\\ q_{K+3}\end{array}\right),$$
(27)
and $`B_K`$ is a $`2\times 2`$ matrix, which is expressed through the coefficients of the quadratic forms $`\{S_{KK+1},\mathrm{},S_{K+2K+3}\}`$ of the above sequence of the transformations; its explicit form is not indicated here because of its inconvenience.
From the other hand the differential equation (20) can be rewritten in these terms as
$$\frac{dQ_K}{dt}=A_KQ_K,$$
(28)
where
$$A_K=\frac{1}{a_Km_{K+1}}\left(\begin{array}{cc}m_{K+1}b_{KK}& m_{K+1}\gamma _{KK+1}\\ m_K\gamma _{KK+1}& m_Kc_{K+1K+1}\end{array}\right).$$
(29)
The conditions of compatibility of Eqs.(26)–(27)
$$\frac{dB_K}{dt}+B_KA_KA_{K+2}B_K=0$$
(30)
are equations for the coefficients $`a_K,b_K,c_K,\gamma _K`$ of the sequence of the generating functions $`S_{KK+1}`$ and represent zero curvature conditions for the models on the grating . They are the conditions of an applicability of the inverse scattering problem method for integrable models. In our case they are realized as a consequence of the equations of the type (22)–(24).
Thus the regular way of the constructing zero curvature conditions, which can be generalized for nonlinear dynamical integrable systems, follows from our analysis.
## IV The confluent hypergeometric equation
Many equations of quantum mechanics under the separation of variables and after isolation of an angular part lead to the second order ordinary differential equations, which can be reduced to the hypergeometric (or confluent hypergeometric) equation. In this connection it is important to consider the canonical transformation theory of the equations of such a type in general case.
Let us consider construction of the canonical recurrence relations for the confluent hypergeometric equation
$$t\ddot{q}+(\beta t)\dot{q}\alpha q=0,$$
(31)
where $`\alpha ,\beta `$ are some parameters. It is the Lagrange-Euler equation for a dynamical system with the Lagrangian
$$L=\frac{1}{2}t^\beta e^t(\dot{q}^2+\frac{\alpha }{t}q^2).$$
(32)
Comparing Eq.(32) with Eq.(17), we can conclude that
$$\lambda =0,\nu =\beta ,\mu =\alpha ,U(\alpha ,t)=\frac{\alpha }{t},m(\beta ,t)=t^\beta e^t.$$
Therefore the system (22)–(24) has the form
$$\dot{a}=\frac{\dot{m}}{m}a+\frac{m}{\stackrel{~}{m}}cb,\frac{m}{\stackrel{~}{m}}\dot{c}+\dot{b}=\frac{\stackrel{~}{\alpha }\alpha }{t}a$$
(33)
$$a\dot{b}=\frac{\alpha }{t}a^2+\frac{m}{\stackrel{~}{m}}(bc\gamma ^2)$$
(34)
According to the above-mentioned, we shall find the generating functions of the elementary canonical transformations a) $`\{\alpha ,\beta \}\{\stackrel{~}{\alpha },\beta \}`$, b) $`\{\alpha ,\beta \}\{\alpha ,\stackrel{~}{\beta }\}`$ only. Besides note that Eqs.(33)–(34) can be written as equations with rational on $`t`$ coefficients. Therefore it is naturally to find the solution in a class of rational functions.
### A A canonical transformation $`\{\alpha ,\beta \}\{\stackrel{~}{\alpha },\beta \}`$
In this case $`\beta =\stackrel{~}{\beta }`$, $`m=\stackrel{~}{m}`$ and the system (33)–(34) acquires the form
$$\dot{a}=(\frac{\beta }{t}1)a+cb,\dot{c}+\dot{b}=\frac{\stackrel{~}{\alpha }\alpha }{t}a,$$
(35)
$$a\dot{b}=bc\gamma ^2a^2\frac{\alpha }{t}.$$
(36)
We shall obtain the simplest rational solution of the undetermined system (35) by assumption $`a=t`$. From (35) we shall find
$$b=\frac{1}{2}(\stackrel{~}{\alpha }\alpha 1)t+b_0,c=\frac{1}{2}(\stackrel{~}{\alpha }\alpha +1)t+b_0\beta +1,$$
(37)
where $`b_0`$ is the integration constant. Substituting Eq.(37) for Eq.(36) and equating the coefficients at the same powers of $`t`$, one obtains
$$b_0=\alpha ,\stackrel{~}{\alpha }=\alpha +1,\gamma =\sqrt{\alpha (\alpha \beta +1)}$$
(38)
We have restricted ourselves here by the positive value of the radical. As a result we have
$$F_{\alpha ,\beta \alpha +1,\beta }=\frac{1}{2}t^{\beta 1}e^t\left(2\sqrt{\alpha (\alpha \beta +1)}q\stackrel{~}{q}\alpha q^2(t+\alpha \beta +1)\stackrel{~}{q}^2\right)$$
(39)
for the generating function (18). The canonical recurrence transformations (20) have the form
$$\left(\alpha +t\frac{d}{dt}\right)q=\sqrt{\alpha (\alpha \beta +1)}\stackrel{~}{q},\left(t+\alpha \beta +1t\frac{d}{dt}\right)\stackrel{~}{q}=\sqrt{\alpha (\alpha \beta +1)}q.$$
(40)
To determine the connection with the standard recurrence relations and thus with the confluent hypergeometric function $`M(\alpha ,\beta ,t)`$, we shall make the following substitutions
$$q=y(\alpha )\sqrt{N(\alpha )},\stackrel{~}{q}=y(\alpha )\sqrt{N(\alpha +1)},y(\alpha )M(\alpha ,\beta ,t)$$
(41)
in Eq.(40). From the condition of coincidence of the obtained recurrence relations with the standard ones (rather with their consequence)
$$\left(\alpha +t\frac{d}{dt}\right)y(\alpha )=\alpha y(\alpha +1),\left(\alpha \beta +t1t\frac{d}{dt}\right)y(\alpha +1)=(\alpha \beta +1)y(\alpha )$$
(42)
the functional equation for the normalization multiplier $`N(\alpha )`$
$$\alpha N(\alpha )=(\alpha \beta +1)N(\alpha +1)$$
(43)
follows. It can be solved by using the gamma-function’s property $`\mathrm{\Gamma }(z+1)=z\mathrm{\Gamma }(z)`$ from where the substitutions
$$\alpha =\frac{\mathrm{\Gamma }(\alpha +1)}{\mathrm{\Gamma }(\alpha )},\alpha \beta +1=\frac{\mathrm{\Gamma }(\alpha \beta +2)}{\mathrm{\Gamma }(\alpha \beta +1)}.$$
(44)
follow. Using these substitutions, equation (43) can be reduced to the form
$$\frac{\mathrm{\Gamma }(\alpha \beta +1)}{\mathrm{\Gamma }(\alpha )}N(\alpha )=\frac{\mathrm{\Gamma }(\alpha \beta +2)}{\mathrm{\Gamma }(\alpha +1)}N(\alpha +1)=\frac{\mathrm{\Gamma }(\alpha \beta +3)}{\mathrm{\Gamma }(\alpha +2)}N(\alpha +2)=\mathrm{}.$$
(45)
The additional equations arise owing to considering the next elementary steps by the parameter $`\alpha `$. From here it is easy to see that $`\mathrm{\Gamma }(\alpha \beta +1)N(\alpha )/\mathrm{\Gamma }(\alpha )=C_1=\mathrm{const}`$. Therefore
$$N(\alpha )=C_1\frac{\mathrm{\Gamma }(\alpha )}{\mathrm{\Gamma }(\alpha \beta +1)}.$$
Thus, we have obtained the “canonical” solution of the confluent hypergeometrical equation in the form
$$q(t)=C_1M(\alpha ,\beta ,t)\sqrt{\frac{\mathrm{\Gamma }(\alpha )}{\mathrm{\Gamma }(\alpha \beta +1)}}.$$
(46)
### B A canonical transformation $`\{\alpha ,\beta \}\{\alpha ,\stackrel{~}{\beta }\}`$
In this case $`\alpha =\stackrel{~}{\alpha }`$ and the system (33)–(34) can be rewritten in the form
$$t^{\beta \stackrel{~}{\beta }}cb=\dot{a}+(1\frac{\beta }{t})a,t^{\beta \stackrel{~}{\beta }}\dot{c}+\dot{b}=0,$$
(47)
$$a\dot{b}+\frac{\alpha }{t}a^2+t^{\beta \stackrel{~}{\beta }}(\gamma ^2bc)=0.$$
(48)
We shall obtain the simplest rational solution of the inmohomogeneous undetermined system (47) by taking $`a=1`$. Then a particular solution of the obtained nonhomogeneous system has the form
$$c=\frac{\beta }{\stackrel{~}{\beta }\beta +2}t^{\stackrel{~}{\beta }\beta 1},b=\frac{\stackrel{~}{\beta }\beta 1}{\stackrel{~}{\beta }\beta +2}\frac{\beta }{t}.$$
(49)
From the quadratic equation (48) we find
$$\stackrel{~}{\beta }=\beta +1,\gamma =\sqrt{\beta \alpha }.$$
(50)
For the generating function and canonical recurrence transformations we have
$$F_{\alpha ,\beta \alpha ,\beta +1}=\frac{1}{2}t^\beta e^t\left(2\sqrt{\beta \alpha }q\stackrel{~}{q}+q^2+\beta \stackrel{~}{q}^2\right),$$
(51)
$$\left(1+\frac{d}{dt}\right)q=\sqrt{\beta \alpha }\stackrel{~}{q},\left(\beta +t\frac{d}{dt}\right)\stackrel{~}{q}=\sqrt{\beta \alpha }q.$$
(52)
The recurrence relations increasing and decreasing $`\beta `$ that follow from the recurrence relations of the handbook
$$\beta \left(1+\frac{d}{dt}\right)y(\beta )=(\beta \alpha )y(\beta +1),\left(\beta +t\frac{d}{dt}\right)y(\beta +1)=\beta y(\beta ),$$
(53)
where $`y(\beta )M(\alpha ,\beta ,t)`$, can be obtained from Eq.(52) by the above procedure of the constructing of Eq. (42)–(46). As a result one has
$$q(t)=C_2\frac{\mathrm{\Gamma }(\beta )}{\sqrt{\mathrm{\Gamma }(\beta \alpha )}}M(\alpha ,\beta ,t).$$
(54)
## V The hypergeometric equation
As it was mentioned, some “radial equations” of the quantum theory are reduced to the hypergeometric-like equations. Therefore consider a dynamical system described by the hypergeometrical equation
$$t(1t)\ddot{q}((\alpha +\beta +1)t\zeta )\dot{q}\alpha \beta q=0,$$
(55)
where $`\alpha ,\beta ,\zeta `$ are some parameters, as another application of the developed theory. It is the Lagrange-Euler equation of a dynamical system with the Lagrangian
$$L=\frac{1}{2}t^\zeta (1t)^{\alpha +\beta \zeta +1}\left(\dot{q}^2+\frac{\alpha \beta }{t(1t)}q^2\right).$$
(56)
Comparing Eq.(56) with Eqs.(2), (17) we have: $`\lambda =0,\mu _1=\alpha ,\mu _2=\beta ,\mu _3=\zeta ,`$
$$U=\frac{\alpha \beta }{t(1t)},m=t^\zeta (1t)^{\alpha +\beta \zeta +1}.$$
(57)
The system (56) contains the three parameters $`\alpha ,\beta ,\zeta `$. Therefore it is necessary to find the generating functions of three canonical transformations. Due to the symmetry between the parameters $`\alpha `$ and $`\beta `$, it is sufficient to search for the generating functions of the transformations a) $`\{\alpha ,\beta ,\zeta \}\{\stackrel{~}{\alpha },\beta ,\zeta \}`$ and b) $`\{\alpha ,\beta ,\zeta \}\{\alpha ,\beta ,\stackrel{~}{\zeta }\}`$ only.
### A A canonical transformation $`\{\alpha ,\beta ,\zeta \}\{\stackrel{~}{\alpha },\beta ,\zeta \}`$
For the elementary transformation $`\{\alpha ,\beta ,\zeta \}\{\stackrel{~}{\alpha },\beta ,\zeta \}`$ the system (22)–(24) can be rewritten the form
$$(1t)^{\alpha \stackrel{~}{\alpha }}cb=\dot{a}\left(\frac{\zeta }{t}\frac{\alpha +\beta \zeta +1}{1t}\right)a,$$
(58)
$$(1t)^{\alpha \stackrel{~}{\alpha }}\dot{c}+\dot{b}=\frac{\stackrel{~}{\alpha }\alpha }{t(1t)}\beta a,$$
(59)
$$a\dot{b}=\frac{\alpha \beta }{t(1t)}a^2+(1t)^{\alpha \stackrel{~}{\alpha }}(bc\gamma ^2).$$
(60)
Similarly to the first case of the previous example we obtain the simplest rational solution of the homogeneous undetermined system (58)–(59) at $`a=t`$. Then the obtained inhomogeneous system has the following particular solution
$$b=\alpha \frac{\stackrel{~}{\alpha }\alpha 1}{\stackrel{~}{\alpha }\alpha 2}\frac{\alpha +\beta \zeta +1}{1t},c=\left(\beta +\frac{\alpha +\beta \zeta +1}{\stackrel{~}{\alpha }\alpha 2}\frac{1}{1t}\right)(1t)^{\stackrel{~}{\alpha }\alpha }.$$
(61)
In this case the quadratic equation (60) is satisfied at
$$\alpha =\stackrel{~}{\alpha }+1,\gamma =\sqrt{\alpha (\alpha \zeta +1)}.$$
(62)
So for the generating function we have
$$F_{\alpha ,\beta ,\zeta \alpha +1,\beta ,\zeta }=\frac{1}{2}t^{\zeta 1}(1t)^{\alpha +\beta \zeta +1}\left[2\sqrt{\alpha (\alpha \zeta +1)}q\stackrel{~}{q}\alpha q^2(\alpha \zeta +1+\beta t)\stackrel{~}{q}^2\right].$$
(63)
The corresponding recurrence transformations have the form
$`\left(\alpha +t{\displaystyle \frac{d}{dt}}\right)q`$ $`=`$ $`\sqrt{\alpha (\alpha \zeta +1)}\stackrel{~}{q},`$ (64)
$`\left(\alpha \zeta +1\beta tt(1t){\displaystyle \frac{d}{dt}}\right)\stackrel{~}{q}`$ $`=`$ $`\sqrt{\alpha (\alpha \zeta +1)}q.`$ (65)
The standard recurrence transformation
$`\left(\alpha +t{\displaystyle \frac{d}{dt}}\right)y(\alpha )`$ $`=`$ $`\alpha y(\alpha +1),`$ (66)
$`\left(\alpha \zeta +1+\beta tt(1t){\displaystyle \frac{d}{dt}}\right)\stackrel{~}{q}`$ $`=`$ $`\sqrt{\alpha (\alpha \zeta +1)}q,`$ (67)
where $`y(\alpha )M(\alpha ,\beta ,\zeta ,t)`$ — hypergeometric function, can be obtained by making such the substitution
$$q(t)=A_1\sqrt{\frac{\mathrm{\Gamma }(\alpha )}{\mathrm{\Gamma }(\alpha \zeta +1)}}M(\alpha ,\beta ,\zeta ,t)$$
(68)
in Eq.(64).
### B A canonical transformation $`\{\alpha ,\beta ,\zeta \}\{\alpha ,\beta ,\stackrel{~}{\zeta }\}`$
In this case the system (22)–(24) acquires the form
$`t^{\zeta \stackrel{~}{\zeta }}(1t)^{\stackrel{~}{\zeta }\zeta }cb`$ $`=`$ $`\dot{a}\left({\displaystyle \frac{\zeta }{t}}{\displaystyle \frac{\alpha +\beta \zeta +1}{1t}}\right)a,`$ (69)
$`t^{\zeta \stackrel{~}{\zeta }}(1t)^{\stackrel{~}{\zeta }\zeta }\dot{c}+\dot{b}`$ $`=`$ $`0,`$ (70)
$$a\dot{b}=\frac{\alpha \beta }{t(1t)}a^2+t^{\zeta \stackrel{~}{\zeta }}(1t)^{\stackrel{~}{\zeta }\zeta }(b\gamma ^2).$$
(71)
We obtain the simplest solution of the system (69) at $`a=1t`$. Then a particular solution (69) acquires the form
$$b=\frac{\zeta }{\stackrel{~}{\zeta }\zeta 2}\frac{1t}{t}+\frac{\zeta }{t}\alpha \beta ,c=\frac{\zeta }{\stackrel{~}{\zeta }\zeta 2}t^{\stackrel{~}{\zeta }\zeta 1}(1t)^{\zeta \stackrel{~}{\zeta }+1}.$$
(72)
The quadratic equation (71) leads to the following constraints
$$\stackrel{~}{\zeta }=\zeta +1,\gamma =\sqrt{(\zeta \alpha )(\beta \zeta )}.$$
(73)
Therefore for the generating function and the recurrence relations we find
$$F_{\alpha ,\beta ,\zeta \alpha ,\beta ,\zeta +1}=\frac{1}{2}t^\zeta (1t)^{\alpha +\beta \zeta }\left(2\sqrt{(\zeta \alpha )(\beta \zeta )}q\stackrel{~}{q}+(\alpha +\beta \zeta )q^2+\zeta \stackrel{~}{q}^2\right),$$
(74)
$`\left(\zeta \alpha \beta +(1t){\displaystyle \frac{d}{dt}}\right)q`$ $`=`$ $`\sqrt{(\zeta \alpha )(\beta \zeta )}\stackrel{~}{q},`$ (75)
$`\left(\zeta +t{\displaystyle \frac{d}{dt}}\right)\stackrel{~}{q}`$ $`=`$ $`\sqrt{(\zeta \alpha )(\beta \zeta )}q.`$ (76)
Making the substitution
$$q(t)=A_2\sqrt{\frac{\mathrm{\Gamma }(\zeta \alpha )\mathrm{\Gamma }(\zeta \beta )}{\mathrm{\Gamma }(\zeta )}}M(\alpha ,\beta ,\zeta ,t)$$
(77)
we reduce Eq.(75) to the relations which follow from the standard recurrence formulae for the hypergeometric function $`M(\alpha ,\beta ,\zeta ,t)`$
$`\left(\zeta \alpha \beta +(1t){\displaystyle \frac{d}{dt}}\right)y(\zeta )`$ $`=`$ $`{\displaystyle \frac{(\zeta \alpha )(\zeta \beta )}{\zeta }}y(\zeta +1),`$ (78)
$`\left(\zeta +t{\displaystyle \frac{d}{dt}}\right)y(\zeta +1)`$ $`=`$ $`\zeta y(\zeta ),`$ (79)
where $`y(\zeta )M(\alpha ,\beta ,\zeta ,t)`$. |
warning/0003/astro-ph0003084.html | ar5iv | text | # Models for the Magnitude-Distribution of Brightest Cluster Galaxies
## 1 Introduction
Among the most luminous bodies in the universe are the brightest, or first-ranked, galaxies in rich clusters. These galaxies have absolute magnitudes between -21.5 and -23.3 and are among the farthest observable objects. In addition, the magnitudes of these brightest cluster galaxies (BCGs) are highly uniform, with a dispersion of 0.32 magnitudes (Hoessel & Schneider 1985). Their uniformity and large luminosity make BCGs excellent standard candles. The uniformity of BCG magnitudes raises a particularly important question regarding their nature (Peebles 1968; Sandage 1972). Are BCGs simply the brightest of a statistical set of galaxies or do they belong to a special class of objects? If a special class of galaxies exists, do all clusters have special galaxies and are they always first-ranked (Bhavsar 1989)? We investigate these questions using extreme value theory (Fisher & Tippett 1928).
## 2 Extreme Value Theory
The motivation for studying extreme phenomena is practical. Many of the memorable experiences in our lives can be classified as statistical extremes. Examples of maximum extremes are floods, the hottest summer temperatures and the lengths of the longest caterpillars. Examples of minimum extremes are draughts, stock market crashes and the wing-spans of the smallest hummingbirds. Some extremes do not effect our lives and others turn them upside down. The desire to understand these types of phenomena prompts the study of extreme value theory.
Fisher & Tippett (1928) show that the distribution of statistically largest or smallest extremes tends asymptotically to a well-determined and analytic form for a general class of parent distributions. Extremes drawn from sufficiently large and steeply falling parent distributions have this form. One may find the original argument in Fisher & Tippett (1928). Their derivation is reconstructed in greater detail by Bhavsar & Barrow (1985), who apply extreme value theory in an analysis of BCG magnitudes. Fisher & Tippett’s result states that the cumulative distribution of maximum extremes is given by:
$$F(x)=e^{e^{a(xx_0)}}.$$
(1)
This distribution is known as the $`Gumbel`$ distribution. (For smallest extremes, one substitutes $`xx`$.) From $`F`$ we may calculate the differential distribution (or probability density):
$$f(x)=ae^{a(xx_0)e^{a(xx_0)}},$$
(2)
where $`f(x)`$ = $`F^{}(x)`$; $`x_0`$ is the mode of the extremes and $`a>0`$ is a measure of the steepness of fall of the parent distribution. The probability density is normalized to unity. The mean, median and standard deviation of the distribution given in Bhavsar & Barrow (1985) correspond to:
$$<x>=x_0+\frac{0.577}{a};med(x)=x_0+\frac{0.367}{a};\sigma ^2=\frac{\pi ^2}{6a^2},$$
(3)
where $`0.577\mathrm{\Gamma }^{}(1)`$ is Euler’s constant, $`0.367ln(ln(2))`$ and $`\sigma `$ is the standard deviation of the extremes. The standard form for the Gumbel, $`F(x)`$ and $`f(x)`$, is shown in Fig. 1. Note that for BCGs, we will be considering minimum extremes (because more negative magnitudes are brighter) and the curves will be inverted ($`xx`$). Henceforth, we will call $`f(x)`$ the Gumbel distribution.
## 3 Brightest Cluster Galaxies
### 3.1 Past Results
Researchers have described BCGs as special, statistical extremes of a normal population and a mixture of the two (Peebles 1968; Peach 1969; Sandage 1972, 1976; Bhavsar & Barrow 1985; Bhavsar 1989; Postman & Lauer 1995).
The motivation for proposing that BCGs are special is due to the small dispersion observed in BCG magnitudes (Peach 1969; Sandage 1972, 1976). These authors argue that such a small dispersion is not sufficiently explained by the steepness of the luminosity function. In addition, astronomers observe a class of BCGs that are morphologically different, called $`cD`$ galaxies. These galaxies are giant ellipticals and often have features, such as multiple nuclei and large envelopes, that distinguish them from normal galaxies.
On the other hand, Peebles (1968) argues that BCGs are just the extreme tail-end of normal galaxies that form in clusters via some stochastic process. In this case, the brightest galaxy in a given cluster is simply the brightest normal galaxy and, therefore, the distribution of BCG magnitudes is a Gumbel. (It is interesting to note that Peebles, independently of Fisher & Tippett, derived the Gumbel distribution for BCGs for the special case of an exponential luminosity function.)
Bhavsar (1989) contends neither of these scenarios adequately describes the observed distribution of BCG magnitudes and argues for a mixed population. Suppose that a special class of Galaxies exists but that not all clusters have a special galaxy. In clusters with no special galaxy, the BCG is simply the brightest normal galaxy. In a cluster containing at least one special galaxy, either all the normal galaxies are fainter, or the brightest normal galaxy(ies) out-shines the special one(s) and attains the first rank. For these reasons, one might expect both types of galaxies to comprise the BCG population. In what follows, we investigate these assumptions quantitatively by analyzing Lauer & Postman’s (1994) data set and revisiting the one used by Bhavsar (1989).
### 3.2 The Distribution of BCG magnitudes
In the case of one population, the distribution function is straight forward. If BCGs are all drawn from a special class of objects, it has been assumed that BCG magnitudes are normally distributed (Peach 1969; Sandage 1972, 1976; Postman & Lauer 1995). In this case, referred to henceforth as model A, the probability distribution of special galaxies, $`f_{sp}`$, is a Gaussian, $`f_g`$, with mean $`M_g`$, standard deviation $`\sigma `$ and normalization such that the integral over all magnitudes, $`M`$, is unity. The distribution function is as follows:
$$f_{sp}(M)=f_g=\frac{1}{\sigma \sqrt{2\pi }}e^{\frac{(MM_g)^2}{2\sigma ^2}}.$$
(4)
If BCGs are simply the brightest of a normal set of galaxies (Peebles 1968), henceforth referred to as model B, the probability distribution of their magnitudes, $`f_{nor}`$, is a Gumbel, $`f_G`$, given by Equation (2), with $`xM`$ and $`x_0M^{}=M_G+\frac{0.577}{a}`$ (Bhavsar & Barrow 1985):
$$f_{nor}(M)=f_G=ae^{a(MM^{})e^{a(MM^{})}},$$
(5)
where $`M_G`$ is the mean of the extremes and $`a`$ is a measure of the steepness of fall of the parent distribution.
In the case of two populations (Bhavsar 1989), we derive the distribution that $`M`$ should have from the contributions of the two individual populations. Consider $`N`$ clusters of galaxies and suppose that $`n<N`$ have at least one special galaxy. Let the independent magnitude-distribution of normal and special galaxies, respectively, be $`f_{nor}`$ and $`f_{sp}`$. The total magnitude-distribution function, $`f_{tot}`$, is then given by:
$`f_{tot}(M)`$ $`=`$ $`d[f_{sp}{\displaystyle _M^{\mathrm{}}}f_{nor}(M^{})dM^{}+`$ (6)
$`f_{nor}{\displaystyle _M^{\mathrm{}}}f_{sp}(M^{})dM^{}]+`$
$`(1d)f_{nor},`$
where $`d=n/N`$. The first (second) term is the probability of picking a special (normal) galaxy, with absolute magnitude $`M`$, from a cluster containing both populations with the condition that all the normal (special) galaxies are fainter. The third term gives the probability of picking a galaxy, with absolute magnitude $`M`$, in clusters containing only normal galaxies. Equation (6) is true for all well-behaved functions $`f_{nor}`$ and $`f_{sp}`$. If $`f_{nor}`$ and $`f_{sp}`$ are normalized to unity, then so is the resulting total distribution function $`f_{tot}`$. (Note that Equation (6) works, in general, whenever there are two independent populations competing for first rank.)
For BCGs, we consider three different two-population models. The first is the case discussed above with the brightest normal galaxies comprising one population and a special class of galaxies comprising the other. We call this case model C and write the total distribution as $`f_{Gg}`$ (where ‘$`Gg`$’ stands for ‘$`G`$umbel + $`g`$aussian’). To obtain the final form of $`f_{Gg}`$ we note that:
$$I_G=_M^{\mathrm{}}f_G(M^{})𝑑M^{}=F(M),$$
(7)
where $`F(M)`$ is given by Equation (1) with $`xM`$, and $`x_0M_G+\frac{0.577}{a}`$. Second, we note that:
$$I_g=_M^{\mathrm{}}f_g(M^{})𝑑M^{}=(1\pm erf|MM_g|)/2,$$
(8)
where $`erf`$ is the error function. The upper sign is for $`M<M_g`$ and the lower sign is for $`M>M_g`$. Thus, we may rewrite $`f_{Gg}`$ by substituting in $`I_G`$ and $`I_g`$:
$$f_{Gg}(M)=d[f_gI_G+f_GI_g]+(1d)f_G.$$
(9)
Other possible combinations of assigning $`f_G`$ and $`f_g`$ to the two populations result in models D and E. In the case of model D, both distributions are Gaussian and the total distribution function, $`f_{gg}`$, is given by:
$$f_{gg}(M)=d[f_{g2}I_{g1}+f_{g1}I_{g2}]+(1d)f_{g1},$$
(10)
where the notation is self-evident and the two Gaussians are characterized, respectively, by $`M_{g1}`$, $`\sigma _1`$ and $`M_{g2}`$, $`\sigma _2`$. In the case of model E, both distributions are Gumbels ($`f_{sp}`$ is also a Gumbel) and the total distribution function, $`f_{GG}`$, is given by:
$$f_{GG}(M)=d[f_{G2}I_{G1}+f_{G1}I_{G2}]+(1d)f_{G1},$$
(11)
where the two Gumbels are characterized, respectively, by $`M_{G1}`$, $`a_1`$ and $`M_{G2}`$, $`a_2`$. Table 1 summarizes the forms of the five models.
## 4 Modeling the Data
### 4.1 Data Sets
We utilize two data sets from the literature. First, we reanalyze the data used by Bhavsar (1989). This is a 93 member subset of 116 metric BCG visual-intrinsic (VI) magnitudes compiled by Hoessel, Gunn & Thuan (1980), henceforth referred to as “HGT”. These 93 are the data from clusters of richness 0 and 1 only; Bhavsar ignores the rest of the BCGs in order to keep the data set homogeneous. The BCG magnitudes are internally consistent to 0.04 magnitudes, as published in HGT. Second, we analyze the 119 metric BCG magnitudes, taken in the Kron-Cousins $`R_c`$ band, compiled by Lauer & Postman (1994), henceforth referred to as “LP”. The data were corrected for local and possible large scale galactic motions. The 119 LP data are comprised of BGCs from 107 clusters of richness 0 & 1, and 9, 2 and 1 of richness 2, 3 and 4, respectively (Abell, Corwin & Olowin 1989). We find that removing the 12 BCGs from clusters of richness class $``$ 2 does not significantly change the distribution of the LP data. This is consistent with Sandage’s (1976) result that BCG magnitude is independent of cluster-richness. The internal consistency of the set is 0.014 magnitudes, as published in Postman & Lauer (1995). Bhavsar (1989) proposes a two-population model for the HGT data. His maximum-likelihood fit is consistent with the data and has parameter-values consistent with physically measured quantities. Postman & Lauer (1995) conclude that the LP data are consistent with a Gaussian.
There are differences in the data sets that could be the reason for the disagreement between Bhavsar (1989) and Postman & Lauer (1995). The two were obtained in different optical bands. The mean of the HGT data set is 0.2 magnitudes brighter than the mean of the LP data set. The two data sets have 34 galaxies in common. Comparing the subset of 34, we find that the HGT values are, on average, 0.06 $`\pm `$ 0.19 magnitudes brighter than the LP values. A two-sample Kolmogorov-Smirnov (K-S) Test addresses the consistency of the two data sets in describing the same population of objects. The null hypothesis is that the same distribution describes both data sets. We find that the two data sets fail the null hypothesis at the 82% confidence level. Therefore, we do not expect the same parameters or distribution to describe both sets. These discrepancies may need further investigation, but such an analysis is outside the scope of this work. We investigate each data set separately and present our results.
### 4.2 Fitting Method
We consider models A-E discussed above. The two-population distributions have five parameters each: two means, two standard deviations and the fraction, $`d`$, of clusters that contain a special population of galaxies. If there is no population of special galaxies, then $`d`$ = 0. We use maximum-likelihood fitting. The theory behind this method is discussed in Press, et al. (1992). The Maximum-Likelihood fit to a data set of size $`N`$ for a function, $`f`$, are the parameters, $`𝐚`$, that maximize the likelihood function:
$$L=\underset{i=1}{\overset{N}{}}f(x_i;𝐚),$$
(12)
where the $`f(x_i;𝐚)`$ are the values of the probability density, $`f`$, evaluated at each of the $`N`$ data points, $`x_i`$. For a certain $`f`$, one finds the set of parameters that maximizes the product, $`L`$.
## 5 Results
### 5.1 Parameters and Fits
After obtaining parameters from the maximum-likelihood method for models A-E for both data sets, we compute the K-S statistics. We list the results in Tables 2 & 3, respectively. Lower values of the K-S $`D`$-statistic correspond to lower values of rejection probability, $`P`$, and thus denote a better fit. Figs. 2 & 3 illustrate the performance of each of the five models. Note that the distributions use the parameters obtained by the maximum-likelihood method, using $`every`$ data point, and are $`not`$ a fit to the particular histograms.
### 5.2 Comparison with Previous Work
We compare our results with Bhavsar (1989) and Postman & Lauer (1995). Bhavsar’s (1989) two-population model is our model C. He uses maximum-likelihood fitting and his best-fitting parameters are $`M_G`$ = -22.31, $`M_g`$ = -22.79, $`d`$ = 0.63, $`a`$ = 4.01 and $`\sigma `$ = 0.21. Our parameters are in excellent agreement. Minor variation is expected due to differences in fitting techniques. Postman & Lauer (1995) argue against Bhavsar’s two-population model and claim that BCG magnitudes are Gaussian, based on a 26% confidence level.
In agreement with both Bhavsar (1989) and Postman & Lauer (1995), it is clear from Tables 2 & 3 and Figs. 2 & 3 that for both data sets no Gumbel distribution describes the BCG data. This rejects the Gumbel hypothesis (model B) with 85% and 93% confidence levels, respectively, for the HGT and LP sets. For the HGT data, the Gaussian fails at the 53% confidence level, while for the LP data, the rejection confidence is 16%. The difference between our value of 16% and Postman & Lauer’s value arises because our result is for the maximum-likelihood fit Gaussian, while Postman & Lauer’s is for a Gaussian with the same mean and standard deviation as the LP data.
The relatively high rejection-confidence of the one-population models has motivated us to investigate two-population models. The presence of cD galaxies strongly suggests the possibility of another population. Overall, the two-population models fit the data much better than do the Gumbels and as well or better than do the respective Gaussians. The larger number of parameters is taken into account by the statistical estimators when calculating the confidence of rejecting the null hypothesis. Moreover, the parameters are physical quantities that are observationally verifiable (Bhavsar 1989).
Our result that no one model or set of parameters describes both data sets is consistent with the fact that a two-sample K-S Test indicates that the sets are not consistent with one another. Postman & Lauer (1995) have raised questions regarding HGT’s BCG classification and sky subtraction.
### 5.3 Physical Motivation
Researchers have suggested various mechanisms whereby a second population with a brighter average metric magnitude could evolve from the bright normal galaxies. Cannibalism, the process by which large galaxies in the central regions of rich clusters grow at the expense of smaller galaxies (Ostriker & Hausman 1977; Hausman & Ostriker 1978), is one possibility. The existence of giant elliptical and cD galaxies near the centre of approximately half of all rich clusters supports this hypothesis. These galaxies always lie at the tail-end of their cluster-luminosity functions. The occurrence of cannibalism continues to be debated (Merritt 1984).
Motivated by the existence in the literature of strong arguments for such a process, we build a very simple schematic to study its $`statistical`$ effects on the population of first-ranked galaxies. We make two assumptions: (i) at an early epoch the BCGs all belonged to one population and (ii) galaxies from the bright end of this population evolve, resulting in a random boost to their luminosity. We construct a set of $`N`$ galaxies with an exponential luminosity function between absolute magnitudes -22.0 and -23.0. This represents the galaxies at the bright end of cluster luminosity functions that are candidates for a boost. A random number, $`n`$, of these galaxies undergo a random boost between 0.1 and 0.9 magnitudes. We label the boosted subset as $`n_b`$. We choose this range for the following reasons. First, Hausman & Ostriker (1978) show via a simulation that one would expect a large galaxy to gain, on average, 0.5 magnitudes during its first cannibalistic encounter. This is consistent with Aragon-Salamanca, Baugh & Kauffmann (1998), who state that BCGs were approximately 0.5 magnitudes fainter at $`z`$ = 1. Second, we limit ourselves to one encounter because Merritt (1984) argues that the time scale for galactic encounters is too long for cannibalism to be common in the universe. We wish to investigate the magnitude-distribution of the resulting boosted population. These represent the special galaxies mentioned previously. Specifically, this distribution could give us insight into the form of $`f_{sp}`$.
To our surprise, we find that the distribution, $`f_{sp}`$, of $`n_b`$ is a Gumbel! The K-S Test rejects the Gaussian hypothesis at the 98% confidence level. Conversely, the Gumbel distribution, with the same mean and deviation as the data, fits well, with only a 7% confidence level for rejection. We summarize these results in Table and Fig. 4. Thus, the two-population model E (a combination of two Gumbels), which is best-fitting for the newer LP data, has a physical basis.
## 6 Conclusion
For more than thirty years, cosmologists have debated the nature of the magnitude-distribution of brightest cluster galaxies. Peebles (1968) and Sandage (1972, 1976) & Peach (1969) reach markedly different conclusions. More recently, Bhavsar (1989) and Postman & Lauer (1995) differ regarding the population(s) that comprise the first-ranked galaxies. In light of this controversy, we have conducted a new examination of the distribution of BCG magnitudes. We consider the BCGs as $`a`$ $`class`$ $`of`$ $`objects`$ to which we may apply well established results from extreme value theory. We find that there are a number of models that perform well in describing the HGT and LP data sets. Though a Gaussian fits both data sets, the confidence limits warrant further investigation of two-population models.
Tables 2 & 3 clearly show that we should reject the Gumbel (model B) as a fit, i.e., the hypothesis that all BCGs are statistical extremes. The Gaussian (model A) is marginally acceptable but without physical basis. Two-population models, in particular, the three combinations of $`f_G`$ and $`f_g`$, describe the data very well. Tables 2 & 3 show their relative merits. Model E stands out as giving the best overall fit and is motivated by a physical basis. Therefore, it is most likely that there are two populations of BCGs: the extremes of a normal population and a class of atypical galaxies with a brighter average mean.
We thank Marc Postman for sending us the LP data. This research was supported by an ANN grant from the US Department of Education and the Kentucky Space Grant Consortium. |
warning/0003/cond-mat0003393.html | ar5iv | text | # Influence of auto-organization and fluctuation effects on the kinetics of a monomer-monomer catalytic scheme.
## I Introduction
Catalytically activated processes play a significant role in numerous technologies as they serve to produce required products from the species which are non-reactive in normal physical conditions; these chemically stable species may, however, enter into reaction in the presence of some third substance - the catalytic substrate . Despite of the widespread use of such processes, the knowledge of the underlying physics and chemistry still rests largely on phenomenological ideas and prescriptions, and thus remains a challenge for fundamental research .
At the simplest, mean-field level of description, reaction kinetics theory presumes that the reaction rate should be considered as the product of the reactant densities and the rate constant, which is dependent on the nature of the binding forces and on the particles’ dynamics. This rate constant is proportional to the hopping rate if the process is diffusion controlled, or to the reaction frequency, in case when the reaction is kinetically controlled . Hence, a clear cut separation is presumed to exist between the local variables, that could be derived, say, from quantum mechanics, and collective variables, expressed in the most simple way as the product of mean densities of the particles involved.
Recently, one of the most active developments in the field has been the recognition of the substantial importance of the multi-particle effects, spatial fluctuations and self-organization, as opposed to conventional local energetic considerations, which emphasized the purely chemical constraints and focused on the refined descriptions of the elementary reaction act. Subsequently, statistical physics concepts were introduced in order to describe anomalous fluctuation-induced behaviors of non-catalytic chemical reactions and simplest catalytic schemes , as well as such to gain a better understanding of such collective phenomena as wave formation, presence of excitable media or stochastic aggregation in chemical systems (see Refs. and references therein). One conclusion that can be drawn from the statistical physics approach is that fluctuations, either spatial or temporal, may drive the reactive system into a set of new states, that cannot be understood and described in terms of mean-field kinetic equations.
In this paper we discuss in detail the influence of spatial fluctuations, statistical self-organization effects and effects of random diffusive motion of reactants on the kinetics of catalytic reactions, using as a particular example the Langmuir-Hinshelwood type reaction scheme. This reaction process, which is also often referred to as the monomer-monomer catalytic scheme, involves two different kinds of species, $`A`$ and $`B`$, which are deposited (continuously in time with mean intensity $`I`$) onto the catalytic substrate by an external source, diffuse on the substrate and react at encounters forming an inert reaction product $`O`$, $`A+BO`$, which is then removed from the system. Our aims here are twofold. First, we show that this seemingly simple catalytic reaction (which has, in fact, several practical applications (see, e.g. Ref.)) shows quite a rich behavior and represents an ideal illustrative example of the statistical physics effects in the reaction kinetics, which may be generic to more complicated schemes involved in real ”stuff” catalytic processes. Second, we develop a unified analytical description which takes into account explicitly the influence of spatial correlations, dimensionality of space and the way how the particles are introduced into the reaction bath on the time evolution of observables - mean particle densities. Moreover, our approach can be routinely generalized for kinetic description of more complex reaction schemes.
We focus here on several different aspects of the Langmuir-Hinshelwood reaction scheme. In particular, we address the question of how the kinetics depends on the dimensionality $`d`$ of the reactive system. In addition to the standard Langmuir-Hinshelwood model in which the catalytic substrate is a two dimensional flat surface, we analyse kinetics of $`A+B0`$ reactions followed by an external input of reactive species in one- and three-dimensional systems, i.e. the situations appropriate to reactions in capillary geometries and annealing of the radiation damage in solids . We show that the monomer-monomer catalytic reaction proceeds quite differently in low dimensional ($`d=1,2`$) and three-dimensional systems. Besides, we examine how the way of particles’ injection into the system affects the properties of stationary states and also how these stationary states are approached in time. We consider here two different types of external input; in the first one (case I) the $`A`$ and $`B`$ particles are introduced independently of each other at random moments of time and at random positions in space, while in the second case (case II) an $`A`$ and a $`B`$ species are introduced in correlated pairs of a fixed radius, the pairs being injected at random moments of time and at random positions in space. We show that also the way of input does matter significantly and may result in a completely different behavior.
The paper is structured as follows: In Section II we formulate the model, introduce a general analytic approach and, in terms of this approach, derive closed-form equations describing the time evolution of mean particles’ densities and pairwise correlation functions. In Section III we present solution to these equations in one-, two- and three-dimensional systems corresponding to different ways of particles injection to the reaction bath. Finally, we conclude in Section IV with a brief summary of our results and discussion.
## II Definitions and basic equations.
In this section we present kinetic description of the monomer-monomer catalytic scheme involving diffusive particles in terms of a certain analytical approach, which takes explicitly into account the influence of pairwise correlations on the time evolution of mean densities; such a description has been first proposed in Ref., which analysed the effects of fluctuations on the kinetics of $`A+BO`$ reactions and yielded the celebrated $`t^{d/4}`$-law for the decrease of particles mean densities. The subsequent works extended this approach to more general reaction schemes (e.g. reversible and coagulation reactions), three-body and catalytic reactions, and also included the possibility of mutual long-ranged interparticle interactions. Here we will focus on the application of this approach to the analytical description of the $`A+BO`$ reactions kinetics in systems with a continuous, random external input of the reactive species. We will recover some of already known results obtained for the monomer-monomer catalytic scheme involving diffusive monomers and will establish several new ones concerning the long-time relaxation of mean particles density to their steady-state values and the dependence of this steady-state densities on the system parameters.
We start with the formulation of the model. Consider a $`d`$-dimensional reaction bath of volume $`V`$ (we suppose that $`V`$ is sufficiently large such that we can discard different finite-size effects, e.g. hard-core exclusion between particles, the termination of reaction within a finite time interval or saturation (poisoning)) in which $`A`$ and $`B`$ particles are continiously introduced by an external random source. The statistical properties of the source will be defined below. After injection to the system, the $`A`$ and $`B`$ particles begin to diffuse. For simplicitly we assume that their diffusion constants are equal to each other, i.e. $`D_A=D_B=D`$. It will be made clear below that such a description is also appropriate to the case of non-equal diffusivities; the calculations in this case will be only essentially more lengthy . Now, the reaction event is defined as follows: When any two $`A`$ and $`B`$ particles approach each other at a fixed separation $`R`$ (the reaction radius), they may enter into reaction forming (an inert with respect to the reaction) reaction product $`O`$. The recombination upon an encounter of $`A`$ and $`B`$ happens with a finite probability $`p`$ (with probability $`q=1p`$ the particles can be reflected) which defines the constant of an elementary reaction act, $`K`$. This constant describes the intrinsic chemical activities of $`A`$ and $`B`$ molecules and is dependent on the nature of the intra- and intermolecular binding forces. In the following we will suppose that this purely ”chemical” constant is known $`a`$ $`priori`$. Besides, we will assume that $`K`$ is large (i.e. the probability of particles reflection from each other is low) and thus will emphasize the ”statistical physics” effects on the reaction kinetics, rather than the effects of chemical constraints. Consequently, all the factors which are exponentially small with $`K`$ will be neglected here.
Now, we define the statistical properties of the particles generation more precisely. Let $`I_A(r,t)`$ and $`I_B(r,t)`$ be the local, at the point with the radius-vector $`r`$, intensities of the production rates of $`A`$ and $`B`$ particles. We take that the volume average values of the production rates obey
$$\frac{1}{V}_V𝑑rI_A(r,t)=\frac{1}{V}_V𝑑rI_B(r,t)=I,$$
(1)
which means that particles generation is steady in time and the mean production rates $`I`$ of $`A`$ and $`B`$ particles are equal to each other.
Defining the correlations in the production rates $`I_{A,B}(r,t)`$ we will consider here two different situations. In the first one (case I), we suppose that $`A`$ and $`B`$ particles are introduced into the reaction bath statistically independent of each other ; that is, the fluctuations of the sources are correlated neither in space, nor in time. In this case, we have
$`{\displaystyle \frac{1}{V}}{\displaystyle _V}𝑑rI_A(r,t)I_A(r+\lambda ,t+\tau )I^2=I\delta (\lambda )\delta (\tau ),(a)`$ (2)
$`{\displaystyle \frac{1}{V}}{\displaystyle _V}𝑑rI_B(r,t)I_B(r+\lambda ,t+\tau )I^2=I\delta (\lambda )\delta (\tau ),(b)`$ (3)
$`{\displaystyle \frac{1}{V}}{\displaystyle _V}𝑑rI_B(r,t)I_A(r+\lambda ,t+\tau )=\mathrm{\hspace{0.33em}0}(c)`$ (4)
In the second case we suppose that $`A`$ and $`B`$ particles are introduced as correlated $`AB`$ pairs<sup>*</sup><sup>*</sup>*The particles in the pair can, of course, diffuse apart after injection., separated by a fixed distance $`\lambda _g`$ . Such type of external generation may arise in chemical systems in which complex reaction product $`O`$ is continuously forced to break-up by an external radiation (say, laser pulses) into the correlated pairs of the component molecules. Here the radius of pair $`\lambda _g`$ will be mainly determined by the difference of energy ”attributed” to $`O`$ and the energy required to dissociate the reaction product. Another example in which such pairs are produced is the annealing of radiation damage in solids. When the solid is irradiated, atoms are knocked out of their places in the lattice to become interstitials and leave behind a vacancy, and then the vacancies and interstitials diffuse and recombine. In case II, different $`AB`$ pairs are statistically uncorrelated and occur, with an average intensity $`I`$, at random positions in the reaction bath. Then, the fluctuations of the sources obey :
$`{\displaystyle \frac{1}{V}}{\displaystyle _V}𝑑rI_A(r,t)I_A(r+\lambda ,t+\tau )I^2=I\delta (\lambda )\delta (\tau ),(a)`$ (5)
$`{\displaystyle \frac{1}{V}}{\displaystyle _V}𝑑rI_B(r,t)I_B(r+\lambda ,t+\tau )I^2=I\delta (\lambda )\delta (\tau ),(b)`$ (6)
$`{\displaystyle \frac{1}{V}}{\displaystyle _V}𝑑rI_B(r,t)I_A(r+\lambda ,t+\tau )=\gamma _d(\lambda _g)I\delta (|\lambda |\lambda _g)(c)`$ (7)
In Eq.(3.c), the parameter $`\gamma _d(\lambda _g)`$ is the normalization factor which arises because of different possible angular orientations of a given $`AB`$ pair in a $`d`$-dimensional continuum; the value of $`\gamma _d(\lambda _g)`$ depends on the dimensionality of the reaction bath and for $`d=1,2`$ and $`3`$ equals respectively $`1`$, $`(2\pi \lambda _g)^1`$ and $`(4\pi \lambda _g^2)^1`$.
Let $`C_A(r,t)`$ and $`C_B(r,t)`$ denote the local densities of $`A`$ and $`B`$ particles at point with radius-vector $`r`$ at time $`t`$. The time evolution of local densities due to the diffusion of species, their reaction and an external production can be described by the following reaction-diffusion equations :
$`\dot{C}_A(r,t)=`$ $``$ $`\gamma _d(R)K{\displaystyle _V}𝑑r^{}\delta (|rr^{}|R)C_A(r,t)C_B(r^{},t)+`$ (8)
$`+`$ $`D\mathrm{\Delta }_rC_A(r,t)+I_A(r,t),`$ (9)
$`\dot{C}_B(r,t)=`$ $``$ $`\gamma _d(R)K{\displaystyle _V}𝑑r^{}\delta (|rr^{}|R)C_B(r,t)C_A(r^{},t)+`$ (10)
$`+`$ $`D\mathrm{\Delta }_rC_B(r,t)+I_B(r,t),`$ (11)
where the symbol $`\mathrm{\Delta }_r`$ denotes the $`d`$-dimensional Laplace operator acting on the spatial variable $`r`$, and the integration with the delta-function $`\delta (|rr^{}|R)`$ accounts for all possible orientations of an $`AB`$ pair, at which an elementary reaction act can take place.
Now, an experimentally accessible property is not, however, the local density, but rather its volume averaged value
$$C(t)=\frac{1}{V}_V𝑑rC_{A,B}(r,t)$$
(12)
To find an equation which governs the time evolution of $`C(t)`$, let us first represent the local densities in the form
$$C_{A,B}(r,t)=C(t)+\delta C_{A,B}(r,t),$$
(13)
where $`\delta C_{A,B}(r,t)`$ will denote local deviations of particles’ densities from their mean values. By definition,
$$\frac{1}{V}_V𝑑r\delta C_{A,B}(r,t)=\mathrm{\hspace{0.33em}0}.$$
(14)
Then, substituting Eq.(7) into Eqs.(4) and (5) and taking the volume average, we get the following equation
$$\dot{C}(t)=K\left[C^2(t)+G_{AB}(|\lambda |=R,t)\right]+I,$$
(15)
in which $`G_{AB}(\lambda ,t)`$ stands for the pairwise, central correlation function of the form
$$G_{AB}(\lambda ,t)=\frac{1}{V}_V_V𝑑r𝑑r^{}\delta (rr^{}\lambda )\delta C_A(r,t)\delta C_B(r^{},t),$$
(16)
the variable $`\lambda `$ being a $`d`$-dimensional correlation parameter.
Therefore, Eq.(9) shows that the time evolution of the mean particle density is ostensibly coupled to the evolution of the pairwise correlations in the reactive system. Neglecting these correlations, i.e. setting $`G_{AB}(\lambda ,t)=0`$, which is equivalent to the traditional, ”mean-field” assumption that the particles’ spatial distribution is uniform, we obtain the customary, text-book ”law of mass action”. Such an approximation predicts a linear in time growth of mean density at relatively short times, i.e.
$$C(t)It,$$
(17)
and, in the large-$`t`$ limit, an exponentially fast relaxation to the equilibrium density $`C(t=\mathrm{})=(I/K)^{1/2}`$, i.e.
$$C(t)(\frac{I}{K})^{1/2}\left[\mathrm{\hspace{0.33em}1}\mathrm{exp}(2(IK)^{1/2}t)+\mathrm{}\right]$$
(18)
The short-time behavior as in Eq.(11) is quite reasonable and describes the regime in which the particles are merely added into the (initially empty) system by the external source and the reaction between them is negligible, i.e. the regime in which particle density remains very small. As for the analytical prediction in Eq.(12), one may question its validity on intuitive grounds. First, in the system under consideration diffusive motion of particle is the only mechanism to bring particles together and let them react. This motion is essentially $`d`$-dependent and should evidently entail $`d`$-dimensional behaviors, whilst Eq.(12) is independent of the dimensionality of the reaction bath. Second, it shows that with an increase of the chemical reaction constant $`K`$ the equilibrium density tends to zero, which is apparently an artificial behavior. Below we will show that the actual behavior of $`C(t)`$ as $`t\mathrm{}`$ is very different. This turns out to depend essentially on the dimensionality of the reactive system and also on the way how the particles are injected into the system. We show analytically that in the case I, when $`A`$ and $`B`$ particles are introduced into the system statistically independ of each other, in low dimensional systems (i.e. $`d=1,2`$) there occurs a strong reaction-induced spatial organization of particles. Here the absolute value of the pairwise correlation function $`G_{AB}(R,t)`$ (which is negative) grows in time, which induces an unlimited (in absence of hard-core exclusion) growth of mean particles’ densities $`C(t)`$. Therefore, in the case I in low dimensional systems the steady state does not exist, in a striking contradiction to the prediction of Eq.(12). Similar results have been obtained previously in Refs.. In the case I, in three dimensional systems the steady state density is well-defined but, however, turns out to be different from that predicted by Eq.(12). In particular, it is dependent on particles’ diffusion constant, which reflects the pathological behavior of correlations in the reactive system . Actually, we show that correlations of fluctuations in particles local densities are essentially long-ranged and obey, in the steady-state, an $`\mathrm{a}\mathrm{l}\mathrm{g}\mathrm{e}\mathrm{b}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{c}`$ law $`G_{AB}(|\lambda |,t)1/|\lambda |`$. Finally, we proceed to show that in three dimensional systems with uncorrelated generation of particles the long-time relaxation of mean particles density to its steady-state value is described by a power-law with characteristic exponent $`1/2`$, in contrast to the exponential dependence predicted by Eq.(12). Such a behavior has been first conjectured in Ref. and here will be deduced analytically. For the case II, we show that steady-state density exists in all dimensions . However, its value is also different from that predicted by Eq.(12) and depends, for instance, on the radius of the generated pairs, $`\lambda _g`$. The approach to the steady state in this case is also not exponential in time and is described by a universal algebraic law, reminiscent of the long-time approach to the equilibrium in reversible chemical reactions .
We turn back to Eq.(9) and continue our analysis of the binary reaction kinetics taking into account the influence of pairwise correlations on the time evolution of $`C(t)`$. Omitting the details of the derivation, which can be found in Refs., we write down the system of equations for the time evolution of the pairwise correlation functions
$`\dot{G}_{AB}(\lambda ,t)=`$ $``$ $`KC(t)\left[2G_{AB}(\lambda ,t)+G_{AA}(\lambda ,t)+G_{BB}(\lambda ,t)\right]+`$ (19)
$`+`$ $`\mathrm{\hspace{0.33em}2}D\mathrm{\Delta }_\lambda G_{AB}(\lambda ,t)+I_{AB}(\lambda )+T_{AB}(\lambda ),`$ (20)
$`\dot{G}_{AA}(\lambda ,t)=`$ $``$ $`\mathrm{\hspace{0.33em}2}KC(t)\left[G_{AB}(\lambda ,t)+G_{AA}(\lambda ,t)\right]+`$ (21)
$`+`$ $`\mathrm{\hspace{0.33em}2}D\mathrm{\Delta }_\lambda G_{AA}(\lambda ,t)+I\delta (\lambda )+T_{AA}(\lambda ),`$ (22)
$`\dot{G}_{BB}(\lambda ,t)=`$ $``$ $`\mathrm{\hspace{0.33em}2}KC(t)\left[G_{AB}(\lambda ,t)+G_{BB}(\lambda ,t)\right]+`$ (23)
$`+`$ $`\mathrm{\hspace{0.33em}2}D\mathrm{\Delta }_\lambda G_{BB}(\lambda ,t)+I\delta (\lambda )+T_{BB}(\lambda ),`$ (24)
where $`\mathrm{\Delta }_\lambda `$ denotes the Laplace operator acting on the $`d`$-dimensional variable $`\lambda `$, the symbol $`I_{AB}(\lambda )`$ in Eq.(13) describes the correlations in the production rates of $`A`$ and $`B`$ particles; it is equal to zero in the case I (uncorrelated generation of particles) and to
$$I_{AB}(\lambda )=\gamma _d(\lambda _g)I\delta (|\lambda |\lambda _g)$$
(25)
in case II, when particles are introduced into the system in correlated $`AB`$ pairs. Finally, in Eqs.(13) to (15) the terms $`T_{ij}`$ denote the correlation functions of the third order.
The time evolution of the pairwise correlations is coupled to the evolution of the third-order correlations, which, in turn, depends on the correlations of the fourth order. Thus the non-linearity of the reaction-diffusion Eqs.(4) and (5) entails an infininite hierarchy of equations for correlation functions and one has to resort to some approximate methods.
The most commonly used method invokes truncation of this hierarchy approximating the third-order correlation functions in terms of $`C(t)`$ and $`G_{ij}(\lambda ,t)`$ . Such an approach, as it was first noticed in Ref., results in the Smoluchowski-type approximate results with improved numerical coefficients and is appropriate for the description of the single-species reactions $`A+AO`$, but not for the description of reactions involving two different types of particles. The point is that such an approximation misses an important conservation law, which is specific for $`A+BO`$ reactions. Namely, the reaction process conserves the difference $`Z(r,t)`$ of local densities, $`Z(r,t)=C_A(r,t)C_B(r,t)`$, which changes only due to the diffusion of particles and thus is a pure diffusive mode of the system. Conservation of $`Z(r,t)`$ entails, in turn, the conservation of the combination of pairwise correlation functions, $`S_{}(\lambda ,t)=G_{AA}(\lambda ,t)+G_{BB}(\lambda ,t)2G_{AB}(\lambda ,t)`$, which is also a pure diffusive modeOne may readily verify that $`T_{AA}+T_{BB}2T_{AB}`$ is exactly equal to zero.. Consequently, only such truncation scheme should correctly describe the behavior of the binary reaction, which does not violate the important conservation laws .
The most simple truncation scheme, which preserves the conservation laws, has been first proposed in Ref.. In this scheme the third-order correlation functions, i.e. $`T_{ij}`$, were set equal to zero. Such a truncation, as it was shown in Refs. and is equivalent to the assumption that fields $`\delta C_{A,B}(\lambda ,t)`$ have Gaussian distribution. Then, the fourth-order correlation functions automatically decouple into the product of pairwise correlation functions and the third-order correlations are equal to zero. Such an approach leads to, for instance, the correct long-time decay law of the particles mean densities, i.e. the $`t^{d/4}`$-law, but fails to reproduce correctly the intermediate time behaviors; at intermediate times this approach predicts essentially the same behavior as the formal kinetic ”law of mass action” and thus disregards the effects of particles’ diffusion at the intermediate-time stage.
This shortcomming has been revisited and improved in Refs., in which it has been shown that correlation functions of the third order are small only in the limit $`|\lambda |>R`$, while in the domain $`|\lambda |R`$ they are singular and this singularity has an impact on the behavior of the pairwise correlation at the intermediate times. In a discrete-space picture, essential at scales $`|\lambda |R`$, the third-order correlation functions have been computed explicitly ,
$$T_{AA}T_{BB}T_{AB}\gamma _d(R)\dot{C}(t)\delta (|\lambda |R).$$
(26)
Substituting Eq.(17) into Eqs.(13) to (15) one gets then a closed with respect to $`C(t)`$ and $`G_{ij}(\lambda ,t)`$ system of equations. To solve them, it is expedient to represent the pairwise correlations in the form
$$G_{ij}(\lambda ,t)=\widehat{G}_{ij}(\lambda ,t)+g_{ij}(\lambda ,t),$$
(27)
where $`\widehat{G}_{ij}(\lambda ,t)`$ denotes a ”singular” part of the pairwise correlation functions, which accounts merely for the behavior of the third-order correlations, and $`g_{ij}(\lambda ,t)`$ \- the ”fluctuational” part, which accounts for the fluctuation spectrum of the external source and fluctuations stemming out of reaction and diffusive processes.
The ”singular” part of the pairwise correlation functions has been determined in Ref.. In particular, the leading at sufficiently large times behavior of $`\widehat{G}_{ij}(|\lambda |=R,t)`$ have been found to be as follows:
$$\widehat{G}_{ij}(R,t)\dot{C}(t)(\pi t/8D)^{1/2},$$
(28)
in one-dimensional,
$$\widehat{G}_{ij}(R,t)\dot{C}(t)\frac{ln(Dt/R^2)}{8\pi D},$$
(29)
and
$$\widehat{G}_{ij}(R,t)\dot{C}(t)\mathrm{\hspace{0.33em}8}\pi DR,$$
(30)
in two-, and three-dimensional systems, respectively.
Now, inserting Eqs.(19) to (21) to Eq.(9) we obtain the following equation for the time evolution of particles mean density
$$\dot{C}(t)=\frac{KK_S(d)}{K+K_S(d)}\left[C^2(t)+g_{AB}(|\lambda |=R,t)\right]+\frac{I}{1+K/K_S(d)}$$
(31)
where $`K_S(d)`$ obeys, as $`t\mathrm{}`$,
$$K_S(d=1)(\frac{8D}{\pi t})^{1/2},$$
(32)
$$K_S(d=2)\frac{8\pi D}{ln(Dt/R^2)},$$
(33)
and
$$K_S(d=3)\mathrm{\hspace{0.33em}8}\pi DR$$
(34)
One may readily notice that in three-dimensions the $`K_S(d)`$, Eq.(25), coincides exactly with the so-called ”diffusive” Smoluchowski constant; a reaction constant which has been first calculated by Smoluchowski in his approximate description of the effects of diffusion on the chemical reactions kinetics. This constant accounts for, heuristically, the ”resistivity” of random, diffusive transport of particles with respect to reaction. Employing the Smoluchowski method, the analogues of such a constant have been obtained in Ref. for one- and two-dimensional systems. Remarkably, our results in Eqs.(23) and (24) coincide with those obtained in Refs. and . We note also that the prefactor before the brackets in Eq.(22), i.e. the ratio $`K_{app}=KK_S(d)/(K+K_S(d))`$, is the so-called effective or apparent reaction constant, which was first derived for three-dimensional systems in Ref.. Therefore, an account of the ”singular” part of the third-order correlation function and subsequent extraction of the ”singular” part in the pairwise correlators leads us to the results equivalent to those obtained in terms of the Smoluchowski approach.
Hence, Eq.(12), in which one sets $`g_{AB}(|\lambda |=R,t)=0`$ and $`K_S(d)=\mathrm{}`$ reduces to the formal kinetic ”law of mass action”, while setting $`g_{AB}(|\lambda |=R,t)=0`$ and using $`K_S(d)`$ as in Eqs.(23) to (25), one obtains the effective kinetic equation of the Smoluchowski-type approach. Below, we proceed to show that taking into account the time evolution of the pairwise correlations, i.e. the term $`g_{AB}(|\lambda |=R,t)`$, one arrives at completely different physical behavior as compared to the ones predicted by the formal kinetic and Smoluchowski approaches.
Finally, we obtain the following system of equations, obeyed by the ”fluctuational” part of the pairwise correlation functions. It reads
$`\dot{g}_{AB}(\lambda ,t)=`$ $``$ $`KC(t)\left[2g_{AB}(\lambda ,t)+g_{AA}(\lambda ,t)+g_{BB}(\lambda ,t)\right]+`$ (35)
$`+`$ $`\mathrm{\hspace{0.33em}2}D\mathrm{\Delta }_\lambda g_{AB}(\lambda ,t)+I_{AB}(\lambda ),`$ (36)
$`\dot{g}_{AA}(\lambda ,t)=\mathrm{\hspace{0.33em}2}KC(t)\left[g_{AB}(\lambda ,t)+g_{AA}(\lambda ,t)\right]+\mathrm{\hspace{0.33em}2}D\mathrm{\Delta }_\lambda g_{AA}(\lambda ,t)+I\delta (\lambda ),`$ (37)
$`\dot{g}_{BB}(\lambda ,t)=\mathrm{\hspace{0.33em}2}KC(t)\left[g_{AB}(\lambda ,t)+g_{BB}(\lambda ,t)\right]+\mathrm{\hspace{0.33em}2}D\mathrm{\Delta }_\lambda g_{BB}(\lambda ,t)+I\delta (\lambda )`$ (38)
Equations (26) to (28), accompanied by Eq.(22), are now closed with respect to mean particles’ densities and pairwise correlations, and suffice the computation of the time evolution of the monomer-monomer reaction scheme.
## III Kinetics of the monomer-monomer reaction scheme.
Below we will analyse solutions of Eqs.(22) to (28) in systems of different dimensionalities and with different types of external particle generation. The derivation of results in case of one-dimensional systems will be presented in detail. The steps involved for such a derivation in higher dimensions are essentially the same and here we will merely discuss the results.
### A Low dimensional systems.
Let us start with the case of one-dimensional systems in which an external source produces uncorrelated $`A`$ and $`B`$ particles.
We note first that the system of equations (26) to (28) possesses two integrable combinations
$$S_{}(\lambda ,t)=\mathrm{\hspace{0.33em}2}g_{AB}(\lambda ,t)g_{AA}(\lambda ,t)g_{BB}(\lambda ,t),$$
(39)
which is related to the conserved property $`Z(r,t)`$, and
$$S_+(\lambda ,t)=\mathrm{\hspace{0.33em}2}g_{AB}(\lambda ,t)+g_{AA}(\lambda ,t)+g_{BB}(\lambda ,t)$$
(40)
These integrable combinations obey
$$\dot{S}_{}(\lambda ,t)=\mathrm{\hspace{0.33em}2}D\mathrm{\Delta }_\lambda S_{}(\lambda ,t)\mathrm{\hspace{0.33em}2}I\delta (\lambda ),$$
(41)
which is thus the pure diffusive mode of the system, not affected by the reaction, and
$$\dot{S}_+(\lambda ,t)=\mathrm{\hspace{0.33em}2}D\mathrm{\Delta }_\lambda S_+(\lambda ,t)\mathrm{\hspace{0.33em}4}KC(t)S_+(\lambda ,t)+\mathrm{\hspace{0.33em}2}I\delta (\lambda )$$
(42)
The desired property, i.e. the correlation function $`g_{AB}(\lambda ,t)`$ which enters the Eq.(22), may be then expressed in terms of these integrable combinations as
$$g_{AB}(\lambda ,t)=\frac{1}{4}(S_{}(\lambda ,t)+S_+(\lambda ,t))$$
(43)
Consider now the solutions to Eqs.(31) and (32) in one-dimensional systems. Applying the Fourier transformation over the variable $`\lambda `$,
$$S_\pm (\omega ,t)=\frac{1}{\sqrt{2\pi }}_{\mathrm{}}^{\mathrm{}}𝑑\lambda \mathrm{exp}(i\omega \lambda )S_\pm (\lambda ,t)$$
(44)
to Eqs.(31) and (32), and assuming that at $`t=0`$ no $`A`$ and $`B`$ particles were present in the system, one readily gets that the Fourier-images of the integrable combinations follow
$`S_{}(\omega ,t)`$ $`=`$ $`I\sqrt{{\displaystyle \frac{2}{\pi }}}{\displaystyle _0^t}𝑑\tau exp(2D\tau \omega ^2)=`$ (45)
$`=`$ $`{\displaystyle \frac{I}{D\omega ^2\sqrt{2\pi }}}(1exp(2D\tau \omega ^2)),`$ (46)
and
$$S_+(\omega ,t)=I\sqrt{\frac{2}{\pi }}_0^t𝑑\tau exp(2D\tau \omega ^2\mathrm{\hspace{0.33em}4}K_\tau ^t𝑑\tau ^{}C(\tau ^{}))$$
(47)
Now we notice that in the extreme situation, when reaction act occurs at any encounter of any $`A`$ and $`B`$ particle (i.e. when $`K=\mathrm{}`$) the second integrable combination $`S_+(\omega ,t)`$ vanishes since the integral $`_\tau ^t𝑑\tau ^{}C(\tau ^{})`$ is obviously positively defined. One can show, however, that even for the finite $`K`$s the influence of $`S_+(\omega ,t)`$ on the pairwise correlation function, Eq.(33), is not essential at large times and the dominant contribution to $`g_{AB}`$ comes from $`S_{}(\omega ,t)`$.
We note that setting $`t=\mathrm{}`$ in Eq.(35) we obtain that $`S_{}(\omega ,\mathrm{})`$ has a steady-state spectrum of the form $`S_{}(\omega ,\mathrm{})1/\omega ^2`$, i.e. the spectrum which has an essential singularity when $`\omega 0`$. Such a singular behavior of the fluctuation spectrum of the pairwise correlations in systems with binary reactions followed by an external uncorrelated production of the reactive species has been first obtained, in terms of a different than ours approach, in Refs. and . The authors concluded thus that the steady state of such a system is highly anomalous; since such a singularity is not integrable in low dimensional systems, the steady state values of the integrable combination $`S_{}(\lambda ,t)`$ and thus of the correlation function $`g_{AB}`$ are infinitely large, which means that as time evolves the system progressively coarses into the domains containing particles of only one type.
Consider now how the integrable combination $`S_{}(\lambda ,t)`$ and the correlation function $`g_{AB}`$ grow in time. Taking the inverse Fourier transformation of the first line in Eq.(35) we get
$$S_{}(\omega ,t)=\frac{I}{\sqrt{2\pi D}}_0^t\frac{d\tau }{\sqrt{\tau }}\mathrm{exp}(\lambda ^2/8D\tau )$$
(48)
The integrand in Eq.(37) is a bell-shaped function with its maximum at point $`\tau =\lambda ^2/8D`$. For bounded $`\lambda `$, the bulk contribution to the integral comes from the algebraic tail $`1/\sqrt{\tau }`$ and consequently, the leading at $`tR^2/8D`$ behavior of the integrable combination follows
$$S_{}(|\lambda |=R,t)=I\sqrt{\frac{2t}{\pi D}}$$
(49)
Accordingly, the absolute value of the ”fluctuational” part of the pairwise correlation function grows in time as
$$g_{AB}(|\lambda |=R,t)=I\sqrt{\frac{t}{8\pi D}}$$
(50)
Inserting the just derived expression into the Eq.(22) and noticing that the correct asymptotical behavior of the mean density obtains when the terms in brackets compensate each other, i.e., when
$$C(t)=\sqrt{g_{AB}(|\lambda |=R,t)},$$
(51)
we find that
$$C(t)=I^{1/2}(\frac{t}{8\pi D})^{1/4},$$
(52)
i.e. in one-dimensional systems with random uncorrelated generation of the reactive species the mean particle density grows sublinearly in time as time progresses. The behavior as in Eq.(41) has been also obtained earlier in Refs..
Consider now how the situation will be changed in the case II, when $`A`$ and $`B`$ particles are introduced into the reactive bath as correlated pairs. In this case one readily gets that the Fourier-image of the integrable combination $`S_{}(\lambda ,t)`$ obeys the following equation
$$\dot{S}_{}(\omega ,t)=2D\omega ^2S_{}(\omega ,t)\mathrm{\hspace{0.33em}2}I(1cos(\omega \lambda _g))$$
(53)
whose solution will read
$$S_{}(\omega ,t)=\frac{I(1cos(\omega \lambda _g))}{D\omega ^2\sqrt{2\pi }}(1\mathrm{exp}(2Dt\omega ^2))$$
(54)
One may readily notice a very important feature of Eq.(43); in a striking contrast to the case I, here the steady-state spectrum is not singular in the limit $`\omega 0`$, but tends to a constant value
$$S_{}(\omega 0,t=\mathrm{})=\frac{I\lambda _g^2}{\sqrt{8\pi }D},$$
(55)
which means that $`S_{}(\lambda ,\mathrm{})`$ and hence, $`g_{AB}(\lambda ,\mathrm{})`$ are bounded in systems of any dimensionality, and thus the well-defined steady-state mean density $`C(t=\mathrm{})`$ also exists. We notice, however, that the steady-state pairwise correlation function is proportional to $`\lambda _g^2`$ and thus may increase indefinitely with growth of $`\lambda _g`$. This unbounded growth is, of course, quite consistent with the result in Eq.(39), since the limit $`\lambda _g\mathrm{}`$ corresponds to the case of uncorrelated generation of particles.
Now, the inverse Fourier transformation gives us
$`S_{}(\lambda ,t)`$ $`=`$ $`{\displaystyle \frac{I}{\sqrt{2\pi D}}}{\displaystyle _0^t}{\displaystyle \frac{d\tau }{\sqrt{\tau }}}\{\mathrm{exp}({\displaystyle \frac{\lambda ^2}{8D\tau }})`$ (56)
$``$ $`{\displaystyle \frac{1}{2}}\mathrm{exp}({\displaystyle \frac{(\lambda \lambda _g)^2}{8D\tau }}){\displaystyle \frac{1}{2}}\mathrm{exp}({\displaystyle \frac{(\lambda +\lambda _g)^2}{8D\tau }})\}=`$ (57)
$`=`$ $`{\displaystyle \frac{I\lambda }{4D\sqrt{\pi }}}\{\mathrm{\Gamma }(1/2,{\displaystyle \frac{\lambda ^2}{8Dt}}){\displaystyle \frac{1}{2}}\mathrm{\Gamma }(1/2,{\displaystyle \frac{(\lambda _g\lambda )^2}{8Dt}})`$ (58)
$``$ $`{\displaystyle \frac{1}{2}}\mathrm{\Gamma }(1/2,{\displaystyle \frac{(\lambda _g+\lambda )^2}{8Dt}})\},`$ (59)
where $`\mathrm{\Gamma }(\alpha ,x)`$ denotes the incomplete Gamma function .
Consider now the asymptotic behavior of the pairwise correlation function, Eq.(33), for different values of parameters $`\lambda `$ and $`\lambda _g`$, and different values of time $`t`$.
One readily gets from Eq.(45) that at sufficiently short times, when $`\lambda \lambda _g8Dt`$, the pairwise correlation function obeys
$$g_{AB}(\lambda ,\lambda _g,t)=\frac{3I\lambda _g^2(8Dt)^{3/2}}{8\sqrt{\pi }D\lambda ^4}\mathrm{exp}(\frac{\lambda ^2}{8Dt}),$$
(60)
which shows that correlations drop off as a Gaussian function at large scales.
Now, at short scales, such that $`\lambda \lambda _g`$ and $`\lambda 8Dt`$, and when $`\lambda _g`$ is sufficiently large, $`\lambda _g8Dt`$, we obtain that $`g_{AB}(\lambda ,\lambda _g,t)`$ obeys Eq.(39), which is not a surprising result since at such scales the correlations in the particles’ injection should be irrelevant.
Within the opposite limit, when $`8Dt\lambda \lambda _g`$ the correlator follows
$$g_{AB}(\lambda ,\lambda _g,t)\frac{I(8Dt)^{3/2}}{16\sqrt{\pi }D\lambda ^2}\mathrm{exp}(\frac{\lambda ^2}{8Dt}),$$
(61)
which is reminiscent of the behavior in Eq.(46).
Finally, in the limit when both $`\lambda 8Dt`$ and $`\lambda _g8Dt`$ (and $`\lambda _g>\lambda `$), i.e. in the limit of very long times, we find the following asymptotic expansion
$$g_{AB}(\lambda ,\lambda _g,t)\frac{I(\lambda _g\lambda )}{16D}\left[1\frac{\lambda _g^2}{\lambda \sqrt{\pi Dt}}\mathrm{exp}(\frac{\lambda ^2}{8Dt})+𝒪(1/t)\right],\lambda _g\lambda ,$$
(62)
where the symbol $`𝒪(1/t)`$ signifies that the correction terms decay with time as $`1/t`$.
Equation (48) suffices to derive the large-$`t`$ evolution of the mean particle density in the case of generation by correlated pairs, which reads
$$C(t)\sqrt{\frac{I(\lambda _gR)}{16D}}\left[1\frac{\lambda _g^2}{2R\sqrt{\pi Dt}}+𝒪(1/t)\right],\lambda _gR$$
(63)
Equations (49) reveals two surprising features; first, the steady-state density turns out to be dependent both on the diffusion constant and on the radius of pairs, generated by the source. Such a dependence is, of course, inconsistent with the predictions of the formal kinetic approach, based on the text-book ”law of mass action”. Second, the approach of particles’ densities to their steady-state values obeys a power-law dependence, in a striking contrast to the exponential one predicted both by the formal kinetic and the Smoluchowski approach.
To close this subsection let us briefly consider the behavior of solutions of the reaction-diffusion equations (22), (26) to (28) in two-dimensional systems.
In the case I, we have from Eqs.(31) and (33) that as $`t\mathrm{}`$ the pairwise correlation function grows (by absolute value) as
$$g_{AB}(|\lambda |=R,t)Iln(Dt/R^2),DtR^2,$$
(64)
and consequently, we get from Eq.(40) that in this case at large times the mean particle density exhibits logarithmically slow growth ,
$$C(t)\sqrt{Iln(Dt/R^2)}$$
(65)
Now, in the case II, we obtain that the steady-state exists and $`g_{AB}(|\lambda |=R,\mathrm{})`$ behaves as
$$g_{AB}(|\lambda |=R,\mathrm{})Iln(\lambda _g),$$
(66)
and hence, the steady-state density turns to be a slowly growing function of the radius of the generated pairs,
$$C(t=\mathrm{})\sqrt{Iln(\lambda _g)}$$
(67)
Some analysis shows also that such a steady-state is approached via an algebraic law
$$C(t)C(t=\mathrm{})(Dt)^1,$$
(68)
in contrast to the exponential in time dependence predicted by mean-field descriptions.
### B Three-dimensional systems.
As we have already mentioned, in the case I the steady-state fluctuation spectrum is characterized by an essential singularity of the type $`1/\omega ^2`$ as $`\omega 0`$. In three-dimensional systems such a singularity is integrable, which insures that the steady-state correlations exist and vanish as $`\lambda \mathrm{}`$. Consequently, the steady-state mean particle density $`C(t=\mathrm{})`$ should exist in 3D also in this case. Let us analyse now the form of these correlations. Solving Eq.(31) in 3D we find (up to the correction terms which are exponentially small with $`K`$) that
$$g_{AB}(\lambda ,t=\mathrm{})\frac{I}{8\pi D\lambda },$$
(69)
i.e. $`AB`$ correlations vanish with the distance between particles $`\lambda `$ as $`1/\lambda `$, which shows that in the monomer-monomer catalytic scheme taking place in three dimensional systems the correlations in the steady-state show a quasi-long-range order decaying as the first inverse power of the interparticle distance.
Now, substituting Eq.(55) into the Eq.(22) we find the following expression for the steady-state density in 3D,
$$C(t=\mathrm{})=\sqrt{(\frac{1}{8\pi DR}+\frac{1}{K})I},$$
(70)
which shows that $`AB`$ correlations lead here to an effective renormalization of the reaction constant in the steady-state, i.e. $`C(t=\mathrm{})`$ has the form $`C(t=\mathrm{})=\sqrt{I/K_{app}}`$, where $`K_{app}`$ is the mentioned above apparent reaction constant .
Consider now how such a steady-state is approached at long times. Expanding the solution of Eq.(31) near the steady-state, we have that pairwise correlations approach the steady-state as a power-law,
$$g_{AB}(|\lambda |=R,t)\frac{I}{8\pi DR}\left[1\frac{R}{\sqrt{\pi Dt}}+O(1/t)\right],$$
(71)
which yields, in turn, a power-law relaxation of the mean particle density to the steady-state
$$C(t)C(t=\mathrm{})(Dt)^{1/2}$$
(72)
Therefore, in contrast to low dimensional systems, in three dimensional systems with random uncorrelated generation of the reactive species the essential singularity in the fluctuation spectrum is integrable, correlations vanish with the distance between particles and the steady-state mean particle density exists. However, the steady-state density is different from that predicted by the mean-field ”law of mass action” and shows, in particular, dependence on the particles’ diffusivity $`D`$. Besides, Eq.(58) reveals that here approach to the steady-state is described by a power-law with the characteristic exponent $`1/2`$, which is essentially non-mean-field behavior.
Finally, for the case II we find the following results for the correlation function and mean density. In the steady-state the $`AB`$ correlations are equal to zero for $`\lambda \lambda _g`$ (again, apart from the exponentially small with $`K`$ terms). In the domain $`\lambda <\lambda _g`$ the correlations exist and are described by
$$g_{AB}(\lambda ,t=\mathrm{})\frac{I}{8\pi D\lambda }(1\frac{\lambda }{\lambda _g}),$$
(73)
which reduces to the result in Eq.(55) when $`\lambda _g=\mathrm{}`$. In constrast to the behavior as in Eq.(55), however, the correlations vanish at finite values of the correlation parameter $`\lambda `$.
Now, Eq.(59) yields for the steady-state mean-particle density
$$C(t=\mathrm{})=\sqrt{I(\frac{1}{K}+\frac{1R/\lambda _g}{8\pi DR})},$$
(74)
which is less than the steady-state density in the case I, Eq.(56), due to a factor $`1R/\lambda _g`$, which renormalizes the Smoluchowski constant. Consequently, for $`\lambda _g>R`$ apparent rate constant here takes the form
$$K_{app}=\frac{8\pi DRK}{8\pi DR+(1R/\lambda _g)K}$$
(75)
We find also that such a steady-state is approached via a power-law,
$$C(t)C(t=\mathrm{})(Dt)^{3/2}$$
(76)
which is faster than the approach described in Eq.(58), but still very different from the exponential behavior predicted by mean-field analysis.
## IV Conclusion.
To summarize, we have shown that both the cases I and II the fluctuations effects dominate the kinetics of the monomer-monomer catalytic scheme involving diffusive particles and induce essential departures from the predictions of the mean-field approaches. In the case I, the effects of fluctuation are especially pronounced in low dimensional systems - the steady-state does not exist and mean particle density grow indefinitely in time, in absence of hard-core exclusion between particles. In three dimensions the steady-state exists, but is characterized by very strong interparticle correlations, which, in turn, have a strong impact on the value of the steady-state mean particle density. The steady-state density is different from that predicted by mean-field ”law of mass action”. The approach to this steady-state is described by an anomalous power-law with the characteristic exponent $`1/2`$, which stems from the presence of essential singularity in the steady-state fluctuation spectrum. In the case II, the steady-state fluctuation spectrum and the steady-state mean particle density exist in any dimension, but show an anomalous, non-mean-fields dependence on the particles’ diffusivity and the radius of pairs, generated by the source. Approach to the steady-state follows a universal power-law with the characteristic exponent $`d/2`$, which resembles, apart from the dependence of the prefactors on the system parameters (e.g. constant of the backward reaction) the long-time approach to the equilibrium in reversible chemical reactions . The origin of such a behavior is that the fluctuation spectrum in the steady-state is flat at small values of the wave-vector, i.e. the essential singularity in the steady-state spectrum of fluctuations is screened.
Figure Caption.
Fig.1. Langmuir-Hinshelwood reaction on a two-dimensional catalytic substrate. Black and grey spheres denote particles of $`A`$ and $`B`$ species, respectively; $`(1)`$ describes the situation in which an $`A`$ and a $`B`$ appear within the reactive distance from each other and may enter into reaction. |
warning/0003/hep-th0003229.html | ar5iv | text | # Untitled Document
Vacuum Energy and Casimir Force in a Presence of Skin-depth Dependent Boundary Condition
S.L. Lebedev <sup>1</sup><sup>1</sup>1 E-mail: lsl@chuvsu.ru Permanent adress: 428000 Cheboksary, Chuvash I.Ya. Yakovlev State Pedagogical University.
> The vacuum energy-momentum tensor (EMT) and the vacuum energy corresponding to massive scalar field on $`\mathrm{}_t\times [0,l]\times \mathrm{}^{D2}`$ space-time with boundary condition involving a dimensional parameter ($`\delta `$) are found. The dependent on the cavity size $`l`$ Casimir energy $`\stackrel{~}{E}_C`$ is a uniquely determinable function of mass $`m`$, size $`l`$ and ”skin-depth” $`\delta `$. This energy includes the ”bulk” and the surface (potential energy) contributions. The latter dominates when $`l\delta `$. Taking the surface potential energy into account is crucial for the coincidence between the derivative $`\stackrel{~}{E}_C/l`$ and the $`ll`$-component of the vacuum EMT. Casimir energy $`\stackrel{~}{E}_C`$ and the bulk contribution to it are interconnected through Legendre transformation, in which the quantity $`\delta ^1`$ is conjugate to the vacuum surface energy multiplied by $`\delta `$. The surface singularities of the vacuum EMT do not depend on $`l`$ and, for even $`D`$, $`\delta =0`$ or $`\mathrm{}`$, possess finite interpretation. The corresponding vacuum energy is finite and retains known analytical dependence on the dimension $`D`$.
1. Introduction
The determination of vacuum energy of fields confined to finite volumes is the basic part in the calculations of Casimir force as well as of some bag characteristics of hadrons . At the same time the renormalization approach appears to be not unique: the counterterms necessary to make the vacuum energy finite can depend on the cavity size so that the regularization dependence of Casimir energy appears . Concrete calculations for the scalar case and Dirichlet or Neumann boundary condition on a spherical boundary have been performed in the recent papers . General consideration shows however that at least for these boundary conditions the surface singularities of the vacuum EMT (and, hence, the aforementioned counterterms) can not depend on the size of the cavity. The detection of such dependences in the papers listed above is a consequence of a coincidence between the local (curvature) and global (size) parameters determining the spherical boundary.
We consider the vacuum characteristics of massive scalar field defined on the domain $`\mathrm{}_t\times [0,l]\times \mathrm{}^{D2}`$. In the absence of curvature we introduce ”skin-depth” parameter $`\delta `$ in boundary condition <sup>2</sup><sup>2</sup>2We use the system of units where $`\mathrm{}=c=1`$. The signature of metric is $`(+,,,\mathrm{})`$. $`x^\mu =(t,x_1,𝐱_{})`$. Brackets $`[,]_+`$ mean anticommutator.
$$_1\phi (t,0,𝐱_{})=\frac{1}{\delta }\phi (t,0,𝐱_{}),\phi (t,l,𝐱_{})=0$$
$`(1)`$
This boundary condition implies the presence of the surface potential energy
$$\frac{1}{2\delta }d^{D2}𝐱_{}\phi ^2(t,0,𝐱_{})$$
$`(2)`$
in the total hamiltonian of the model. The latter is, as usual, space integral of 00-component of the energy-momentum tensor
$$\stackrel{~}{T}_{\mu \nu }=\frac{1}{2}[_\mu \phi ,_\nu \phi ]_+\frac{1}{2}g_{\mu \nu }(\phi )^2+\frac{1}{2}g_{\mu \nu }m^2\phi ^2+\xi ^\lambda (g_{\mu \nu }[\phi ,_\lambda \phi ]_+g_{\mu \lambda }[\phi ,_\nu \phi ]_+)$$
$`(3)`$
where the last divergence term should be taken with coefficient $`\xi =1/4`$. This is the only value of $`\xi `$ guaranteeing the diagonality of hamiltonian and its conservation .
2. Energy-momentum tensor and energy of vacuum
The normalized solutions of Klein-Gordon-Fock equation have the form:
$$\stackrel{~}{\phi }_k(x)=(2\omega )^{\frac{1}{2}}(2\pi )^{1\frac{D}{2}}\mathrm{exp}[i\omega t+i\mathrm{𝐪𝐱}_{}]\psi _k(x_1),$$
$`(4)`$
where $`\omega =\sqrt{m^2+𝐪^\mathrm{𝟐}+k^2}`$, and the descrete set of functions
$$\psi _k(x)=N_k\mathrm{sin}k(xl),N_k=\left[\frac{l}{2}\left(1\frac{\mathrm{sin}2kl}{2kl}\right)\right]^{1/2}$$
$`(5)`$
represents the resonator modes. Wave number $`k=z/\delta (\stackrel{~}{l}l/\delta )`$ should be determined from the spectral equation
$$\mathrm{\Delta }(z)z^1tgz\stackrel{~}{l}+1=0,$$
$`(6)`$
following from boundary condition (1). The basis of functions (4) is used to define creation-annihilation operators and vacuum state in a standard manner <sup>3</sup><sup>3</sup>3It should be noted, however, that canonical quantization of the system (3) for $`\xi =1/4`$ as a system with a higher derivatives, should account for its degeneracy ..
With the help of Cauchy theorem applied to meromorphic function $`\mathrm{\Delta }(z)`$, it is possible to transform the sums over transcendential roots of eq.(6) into corresponding integrals. The scheme of calculation fits in the formula
$$\underset{z_{n\delta }>0}{}\frac{f(z_{n\delta })}{1{\displaystyle \frac{\mathrm{sin}2z_{n\delta }\stackrel{~}{l}}{2z_{n\delta }\stackrel{~}{l}}}}=\frac{\stackrel{~}{l}}{\pi }\underset{0}{\overset{\mathrm{}}{}}f(z)𝑑z\frac{\stackrel{~}{l}}{2\pi i}\underset{0}{\overset{\mathrm{}}{}}\frac{(1t)[f(it)f(it)]}{sht\stackrel{~}{l}+tcht\stackrel{~}{l}}e^{t\stackrel{~}{l}}𝑑t,$$
$`(7)`$
where $`f(z)`$ is a function analitical in the right half-plane, and $`z_{n\delta }(n=1,2,\mathrm{})`$ are the (real) roots of eq.(6). With the use of eq.(7) one could attach the integral representation to the unrenormalized vacuum EMT. The renormalization is reduced to the subtraction of Minkowskian space contribution (D-regularization supposed):
$$T_{\mu \nu }_M=\frac{\mathrm{\Gamma }(\frac{1D}{2})}{\sqrt{\pi }(4\pi )^{D/2}}_0^{\mathrm{}}𝑑kM^{D3}diag\left[\begin{array}{c}M^2\\ k^2(1D)\\ M^2\end{array}\right],$$
$`(8)`$
$`M(m^2+k^2)^{1/2}`$. The final answer i.e. renormalized EMT of vacuum, looks like
$$\stackrel{~}{T}_{\mu \nu }_{\delta ,l}=\frac{\mathrm{K}_D}{\delta }\underset{\mu }{\overset{\mathrm{}}{}}𝑑t(t^2\mu ^2)^{\frac{D1}{2}}\left\{\frac{(1t)e^{t\stackrel{~}{l}}}{sht\stackrel{~}{l}+tcht\stackrel{~}{l}}diag\left[\begin{array}{c}ch2tx^{}1\\ \frac{t^2(1D)}{t^2\mu ^2}\\ 1ch2tx^{}\end{array}\right]+e^{2tx^{}}diag\left[\begin{array}{c}1\\ 0\\ I\end{array}\right]\right\}.$$
$`(9).`$
$`(D2)`$-dimensional diagonal of $`\stackrel{~}{T}_{\mu \nu }_{\delta ,l}`$ corresponding to $`𝐱_{}`$ contains expressions of the same type. The following notations were used in eq.(9):
$$\mathrm{K}_D=\frac{\delta ^{1D}}{2(4\pi )^{\frac{D1}{2}}\mathrm{\Gamma }\left(\frac{1+D}{2}\right)},\mu =m\delta ,x^{}=(lx_1)/\delta .$$
$`(10)`$
The energy of vacuum could be determined directly, without resort to eq.(9), as a sum of half-frequences interpreted e.g. with the help of zeta-regularization method . At the same time, Green function method used to obtain vacuum energy should rely on the tensor $`\stackrel{~}{T}_{\mu \nu }`$ (3) , but not its first three terms
$$T_{\mu \nu }=\frac{1}{2}[_\mu \phi ,_\nu \phi ]_+\frac{1}{2}g_{\mu \nu }(\phi )^2+\frac{1}{2}g_{\mu \nu }m^2\phi ^2.$$
$`(11)`$
$`11`$-component of tensor (3) gives Casimir pressure $`P=\stackrel{~}{T}_{11}_{\delta ,l}`$ which, because of translational symmetry, coincides with $`T_{11}_{\delta ,l}`$ (i.e. the divergence term in (3) does not affect the pressure). The role of divergence term is evident from the fact that value $`\xi =1/4`$ is the only one at which equality
$$\frac{\stackrel{~}{E}_{vac}}{l}=\stackrel{~}{T}_{11}_{\delta ,l}$$
$`(12)`$
holds. Notice that conservation of the total energy and equality (12) both rely on the value $`\xi =1/4`$.
In proving the eq.(12) it is essential that vacuum energy per unit area of the boundary
$$\stackrel{~}{E}_{vac}=_{x_0}^{ly_0}𝑑x_1\stackrel{~}{T}_{00}_{\delta ,l}=\stackrel{~}{E}_C(l,\delta )+\stackrel{~}{E}_{w1}(x_0,\delta )+\stackrel{~}{E}_{w2}(y_0,0)+\mathrm{},$$
$`(13)`$
where dots denote terms vanishing at $`x_0,y_00`$ as well as at $`l\mathrm{}`$. Only finite part of $`\stackrel{~}{E}_{vac}`$, i.e. Casimir energy
$$\stackrel{~}{E}_C(l,\delta )=\mathrm{K}_D\underset{\mu }{\overset{\mathrm{}}{}}𝑑t\frac{(t^2\mu ^2)^{\frac{D1}{2}}e^{t\stackrel{~}{l}}}{sht\stackrel{~}{l}+tcht\stackrel{~}{l}}\left[\stackrel{~}{l}(t1)\frac{1}{t+1}\right],$$
$`(14)`$
depends on the size of the domain and vanishes when $`l\mathrm{}`$. Boundary divergences are present in r.h.s. of eq.(13) in the form of the energies of the walls, $`\stackrel{~}{E}_{w1,2}`$. For example,
$$\stackrel{~}{E}_{w1}(x_0,\delta )=\mathrm{K}_D\mu ^{D1}\underset{1}{\overset{\mathrm{}}{}}\frac{d\xi }{2\xi }(\xi ^21)^{\frac{D1}{2}}\frac{1\mu \xi }{1+\mu \xi }e^{2mx_0\xi }.$$
$`(15)`$
The specific representation of energies $`\stackrel{~}{E}_{w1,2}`$ depends on regularization employed, but Casimir energy is independent of the latter. Notice that the terms discarded in the r.h.s. of eq.(13) have the property of ”double vanishing” (with respect to limits of $`x_0,y_0`$ and $`l`$) so that energies $`\stackrel{~}{E}_{w1,2}`$ of the walls are determined uniquely within the given regularization scheme.
By a complete analogy with $`\stackrel{~}{E}_{vac}`$, one can find its bulk ($`E_{vac}`$) and surface ($`\mathrm{\Pi }_{vac}=\stackrel{~}{E}_{vac}E_{vac}`$) components, the first dertermined with the help of density $`T_{00}`$ (11). For either of these components the expansion of the form (13) exists giving rize to $`E_C(l,\delta )`$ and $`\mathrm{\Pi }_C(l,\delta )`$ correspondingly. For example,
$$\mathrm{\Pi }_C(l,\delta )=\mathrm{K}_D(1D)\underset{\mu }{\overset{\mathrm{}}{}}𝑑t\frac{t^2(t^2\mu ^2)^{\frac{D3}{2}}e^{t\stackrel{~}{l}}}{(t+1)(sht\stackrel{~}{l}+tcht\stackrel{~}{l})},$$
$`(16)`$
and $`E_C=\stackrel{~}{E}_C\mathrm{\Pi }_C`$. Since $`\mathrm{\Pi }_C(l,\delta )0`$, derivative $`E_C/l`$ does not coincide with the Casimir pressure (12). Nevertheless,
$$\delta \frac{\stackrel{~}{E}_C}{\delta }=\mathrm{\Pi }_C,$$
$`(17)`$
and, with the notations $`\lambda =\delta ^1,f=\mathrm{\Pi }_C/\lambda `$, one finds that $`\stackrel{~}{E}_C`$ and $`E_C`$ are interconnected through Legendre transformation:
$$\stackrel{~}{E}_C(l,\lambda )=E_C(l,f(l,\lambda ))+\lambda f(l,\lambda ),$$
$`(18)`$
so that
$$\left(\frac{E_C}{l}\right)_f=\left(\frac{\stackrel{~}{E}_C}{l}\right)_\lambda ,\mathrm{and}\left(\frac{E_C}{f}\right)_l=\lambda .$$
$`(19)`$
Now energy $`E_C`$ determines the Casimir pressure but under specific condition of constancy of the quantity $`f`$ being one half the ”Casimir” part of $`\phi (0)^2`$, see (2).
3. Asymptotic properties
For the purpose of comparison with electrodynamics, here we consider a massless case. The behaviour of integral (14) in Dirichlet ($`\delta l`$) and Neumann ($`\delta l`$) regimes is displayed by the following expansions ($`m=0`$):
$$\stackrel{~}{E}_C(l,\delta )=\mathrm{K}_D\frac{\zeta (D)\mathrm{\Gamma }(D)}{(2\stackrel{~}{l})^{D1}}\left[1+C_{D1}^1\stackrel{~}{l}^1C_D^2\stackrel{~}{l}^2+C_{D+1}^3\left(1+\frac{\zeta (D+2)}{2\zeta (D)}\right)\stackrel{~}{l}^3\right],$$
$`(20)`$
$$\stackrel{~}{E}_C(l,\delta )=\mathrm{K}_D\stackrel{~}{l}^{D+1}\{\begin{array}{c}A_D(D1)A_{D2}\stackrel{~}{l},D=3,4,\mathrm{},\hfill \\ \\ A_2+\stackrel{~}{l}ln(4\gamma _E\stackrel{~}{l}/\pi )\stackrel{~}{l},D=2.\hfill \end{array}$$
$`(21)`$
Here $`C_n^m`$ denotes binomial coefficient, $`A_D=2^{1D}(12^{1D})\mathrm{\Gamma }(D)\zeta (D)`$, $`ln\gamma _E=0.577`$, and eq.(21) contains the leading corrections only. The expansion of pressure $`\stackrel{~}{T}_{11}_{\delta ,l}`$ corresponding to (20) has the form ($`\delta l)`$:
$$\begin{array}{c}\stackrel{~}{T}_{11}_{\delta ,l}=\frac{(D1)\mathrm{\Gamma }(D)\zeta (D)}{(4\pi )^{\frac{D1}{2}}\mathrm{\Gamma }\left(\frac{1+D}{2}\right)}(2l)^D\times \\ \\ \times \left[1+C_D^1\stackrel{~}{l}^1C_{D+1}^2\stackrel{~}{l}^2+C_{D+2}^3\left(1+\frac{\zeta (D+2)}{2\zeta (D)}\right)\stackrel{~}{l}^3+\mathrm{}\right],\end{array}$$
$`(22)`$
and for Neumann regime (21)
$$\stackrel{~}{T}_{11}_{\delta ,l}=\frac{(D1)l^D}{2(4\pi )^{\frac{D1}{2}}\mathrm{\Gamma }\left(\frac{D+1}{2}\right)}\left[A_D(D2)A_{D2}\stackrel{~}{l}+\mathrm{}\right].$$
$`(23)`$
The four terms in the square brackets of eq.(22) (term with $`\zeta `$-functions excluding) could be obtained through the Taylor series for $`(l+\delta )^D`$. Thus, eq.(22) demonstrates the role of $`\delta `$ as penetration depth . Unlike Dirichlet case (22), the correction term in the brackets of eq.(23) emerges due to surface energy $`\mathrm{\Pi }_C`$ only. Numerical analysis of the formulas (14) and (16) taken at $`m=0`$ shows the dominance of the surface contribution $`\mathrm{\Pi }_C`$ over the bulk one ($`E_C`$) in the region $`l(1÷4)\delta `$. The difference in signs between those quantities is responsible for the shift of the position of the minimum from $`l4.5\delta `$ (for $`E_C`$) to $`l1.8\delta `$ (for $`\stackrel{~}{E}_C`$). At that time, the qualitative behaviours of energies $`E_C`$ and $`\stackrel{~}{E}_C`$ as functions of $`l`$ are alike showing a typical van-der-vaals character.
At $`D=4`$ formula (22) takes the form
$$\stackrel{~}{T}_{11}_{\delta ,l}=\frac{\pi ^2}{480l^4}\left[1+4\frac{\delta }{l}10\frac{\delta ^2}{l^2}+20\left(1+\frac{9\pi ^2}{185}\right)\frac{\delta ^3}{l^3}+\mathrm{}\right],$$
$`(24)`$
and should be compared with the corresponding expression in the case of electromagnetic field confined to the region between impedance walls ($`\delta `$\- skin-depth parameter):
$$P=\frac{\pi ^2}{240l^4}\left[1+\frac{16}{3}\frac{\delta }{l}24\frac{\delta ^2}{l^2}+\frac{640}{7}\left(1+\frac{9\pi ^2}{740}\right)\frac{\delta ^3}{l^3}+\mathrm{}\right].$$
$`(25)`$
Formula (25) except the last term $`\delta ^3/l^3`$ in the square brackets, was extracted from . The corresponding coefficients of expansions (25) and (24) are related to each other in ratios 1, 1.33, 2.4, 3.46, hence showing the growing effect of spin when going deeper into the boundary.
4. Interpretation of surface singularities at $`\delta =0`$ and $`\mathrm{}`$
The surface singularities of the vacuum EMT is a stumbling-block problem for any field theory expected to establish the connection between its local properties and observational predictions . It is shown below that the method of dimensional regularization could be applied not only to interpret the singular sums like $`E_{vac}=\underset{\nu }{}\frac{1}{2}\omega _\nu `$, but gives a reasonable finite answer for the energy density of vacuum as well. The method works for even $`D,\delta =+0`$ <sup>4</sup><sup>4</sup>4The singular nature of Dirichlet limit (contrary to Neumann one) is explained in . or $`\mathrm{}`$ and leads to an energy expression consistent with the one of $`\zeta `$-regularization method.
Boundary divergences are represented in r.h.s. of eq.(13) by the energies of the walls ($`\stackrel{~}{E}_{w1,2}`$). Correspondingly, we consider 00-component of a tensor (9) in the limit $`l\mathrm{},\delta =+0`$ (alternative case $`\delta =\mathrm{}`$ differs from (26) only in sign):
$$\stackrel{~}{T}_{00}_{0,\mathrm{}}=\left(\frac{m^2}{2\pi }\right)^{{\scriptscriptstyle \frac{D}{2}}}\{\begin{array}{c}\rho ^{D/2}K_{D/2}(\rho ),0<\rho \underset{}{<}1,x/l0;\\ \\ (\rho ^{})^{D/2}K_{D/2}(\rho ^{}),0<\rho ^{}\underset{}{<}1,x^{}/l0\end{array}$$
$`(26a,b)`$
($`\rho =2mx,\rho ^{}=2mx^{}=2m(lx)`$). Exploiting recurrence relations between modified Bessel functions $`K_\nu (\rho )`$ , let us transform energy density (26a):
$$\stackrel{~}{T}_{00}_{0,\mathrm{}}\stackrel{~}{T}_{00}_{0,\mathrm{}}\frac{d}{d\rho }f_D(\rho )=\left(\frac{m^2}{2\pi }\right)^{{\scriptscriptstyle \frac{D}{2}}}\frac{\rho ^{1ϵ}K_{ϵ1}(\rho )}{(1D)(3D)\mathrm{}(12ϵ)}.$$
$`(27)`$
Here $`ϵ\frac{D}{2}n0,n=1,2,\mathrm{}`$, and function
$$f_D(\rho )=\left(\frac{m^2}{2\pi }\right)^{{\scriptscriptstyle \frac{D}{2}}}\left[\frac{\rho ^{1\frac{D}{2}}K_{\frac{D}{2}}(\rho )}{1D}+\frac{\rho ^{2\frac{D}{2}}K_{\frac{D}{2}1}(\rho )}{(1D)(3D)}+\mathrm{}+\frac{\rho ^{1ϵ}K_ϵ(\rho )}{(1D)(3D)\mathrm{}(12ϵ)}\right]$$
$`(28)`$
possesses following properties: $`i)`$ for $`\mathrm{}\mathrm{e}D<0`$ ($`\mathrm{}\mathrm{e}ϵ<n`$)
$$f_D(0)=f_D(\mathrm{})=0,$$
$`(29)`$
so that divergence addition in the l.h.s. of eq.(27) does not affect the vacuum energy of the wall
$$\stackrel{~}{E}_{w1}(0,0)=\left(\frac{m^2}{2\pi }\right)^{{\scriptscriptstyle \frac{D}{2}}}_0^{\mathrm{}}𝑑x\rho ^{D/2}K_{D/2}(\rho )=\frac{m^{D1}\mathrm{\Gamma }\left(\frac{1D}{2}\right)}{8(4\pi )^{\frac{D1}{2}}}.$$
$`(30)`$
(the vacuum energy of Neumann wall $`\stackrel{~}{E}_{w1}(0,\mathrm{})=\stackrel{~}{E}_{w1}(0,0)=E_{w1}(0,0)`$);
$`ii)`$ a term $`f_D^{^{}}(\rho )`$ in the l.h.s. of eq.(27) taken at ”physical” values of $`D=2,4,6,\mathrm{}`$, acts as a counterterm eliminating all nonintegrable singularities of density (26a);
$`iii)`$ this counterterm preserves an exponential decreasing behaviour of the modified density (27) at $`x>\mathrm{}/mc`$. The latter is not still uniquely defined: one can add to $`f_D(\rho )`$ any regular function having the property (29) (see ).
Thus, at $`\delta =+0`$ the total energy of vacuum includes Casimir part
$$\stackrel{~}{E}_C(l,0)=\frac{m^Dl}{(4\pi )^{\frac{D1}{2}}\mathrm{\Gamma }(\frac{D+1}{2})}\underset{1}{\overset{\mathrm{}}{}}𝑑\xi \frac{(\xi ^21)^{\frac{D1}{2}}}{e^{2\xi ml}1}$$
$`(31)`$
and doubled topological (according to ) energy (30) caused by the presence of the edge of the manifold. At $`\delta =\mathrm{}`$ the total energy of vacuum reduces to $`\stackrel{~}{E}_C(l,\mathrm{})`$ because the energies of Neumann ($`x=0`$) and Dirichlet ($`x=l`$) walls cancell out. Energy $`\stackrel{~}{E}_C(l,\mathrm{})=E_C(l,\mathrm{})`$. The integral representation for it can be obtained from (31) by substitution ”$`+`$” for ”$``$” in the denominator of the integrand and by multiplying the whole expression by $`(1)`$.
Dimensional regularization method does not result in satisfactory expressions for the vacuum energy when $`0<\delta <\mathrm{}`$. As it may be seen from eq.(15) (taken at $`x_0=0`$), for even $`D`$ the corrections of order $`\mu ,\mu ^2,\mathrm{}`$ to the energy (30) cannot be interpreted in the way like (30). On the other hand, for odd $`D3`$ the energy (30) is infinite. Now, the depending on $`\mu `$ ”corrections” find finite interpretation, but being summed up, they lead to logarithmic divergence in the Neumann limit $`\mu \mathrm{}`$.
The analytical structure of the surface singularities of the vacuum EMT (26) with respect to dimension $`D`$ is just that property which makes dimensional regularization applicable to define finite density (27). In view of the absence of translational symmetry it seems quite natural for density like (27) to exist. At that time finite renormalization of the type suggested in eliminates the energy of the walls from $`\stackrel{~}{E}_{vac}=\stackrel{~}{E}_C(l,0)+2\stackrel{~}{E}_w`$. This procedure needs in explaination how to interpret the expressions like (27). It should be mentioned in addition that the energy density (27) (but not (26)) vanishes when $`m=0`$. This disappearance was assumed to be characteristic of conformally invariant models .
5. Conclusion
Below some comments on literature related to the present topic are given. The massless case for the scalar model at hand (dimension $`D=2,3`$) was considered in works and formulas consistent with (9) ( $`D=3,m=0`$, ) and (14) ( $`D=2,m=0`$, ) were derived. The work exploits an idea of replacement of the boundary with a singular non-locally regularized potential, that occured to be equivalent to taking potential energy (2) into account.
Formula (2.18) from corresponds to our formula (31). Doubled energy (30) appears in as well, but it was not associated there with the vacuum energy of half-space. Our formula (9) at $`\delta =+0`$ (or, equivalently, (31) and (12)) gives the expression
$$\begin{array}{c}\stackrel{~}{T}_{11}_{0,l}=\frac{m^D(1D)}{(4\pi )^{\frac{D1}{2}}\mathrm{\Gamma }\left(\frac{1+D}{2}\right)}\underset{1}{\overset{\mathrm{}}{}}\frac{\xi ^2(\xi ^21)^{\frac{D3}{2}}}{e^{2ml\xi }1}𝑑\xi |_{ml1}=\\ \\ =\frac{m^D}{(4\pi ml)^{\frac{D1}{2}}}e^{2ml}\left[1+\frac{(D1)(D+5)}{16ml}+\mathrm{}\right],\end{array}$$
$`(32)`$
which does not coincide with formula (2.13) from . The latter, being taken at zero temperature, includes physically unacceptable dependence of pressure on (arbitrary) renormalization parameter. Such dependence, as it was noted above, is characteristic of curved boundary under the condition of coincidence between the parameters determining its curvature and size.
The author would like to acknowledge useful discussions with A.I. Nikishov and V.I. Ritus and financial support from RFFR (grants 95-02-04219-a and 96-15-96463).
References |
warning/0003/cond-mat0003239.html | ar5iv | text | # Evolution of the universality class in slightly diluted (1>𝑝>0.8) Ising systems.
## Abstract
The crossover of a pure (undiluted) Ising system (spin per site probability $`p=1`$) to a diluted Ising system (spin per site probability $`p<0.8`$) is studied by means of Monte Carlo calculations with $`p`$ ranging between 1 and 0.8 at intervals of 0.025. The evolution of the self averaging is analyzed by direct determination of the normalized square widths $`R_M`$ and $`R_\chi `$ as a function of $`p`$. We find a monotonous and smooth evolution from the pure to the randomly diluted universality class. The $`p`$-dependent transition is found to be indepent of size ($`L`$). This property is very convenient for extrapolation towards the randomly diluted universality class avoiding complications resulting from finite size effects.
Systems with quenched randomness have been studied intensively for several decades . One of the first results was establishing the so called Harris criterion , which predicts that a weak dilution does not change the critical behavior’s character near second order phase transitions for systems of dimension $`d`$ with specific heat exponent lower than zero (the so called P systems), $`\alpha _{pure}<0\nu _{pure}>2/d`$ due to the hyperscaling relationship, in the undiluted case. This criterion has been confirmed by several renormalization group (RG) analyses , and by scaling analysis . It was shown to hold also in strongly diluted systems by Chayes et al. . For $`\alpha _{pure}>0`$ (the so called R systems), for example the Ising 3D case, the system fixed point flows from that of a pure (undiluted) fixed point towards a new stable fixed point at which $`\alpha _{random}<0`$ for diluted systems.
Recently Ballesteros et al. have used the Monte Carlo approach to study diluted Ising systems in two, three and four dimensions . The existence of a new universality class for the randomly diluted Ising system (RDIS) (different from that of the pure Ising model, and $`p`$-independent being $`p`$ the spin per site probability) is proved using an infinite volume extrapolation technique based upon the leading correction to scaling. The critical exponents obtained this way agree with the experimental critical exponents for a random disposition of vacancies in diluted magnetic systems .
The crossover from the pure Ising system $`p=1`$ to the randomly diluted system may occur for very large values of the average density of occupied sites ($`1>p>0.8`$), i.e. systems with a very small amount of vacancies. In this region, the specific heat critical exponent must flow from a value grater than zero, $`\alpha _{pure}>0`$ for $`p=1`$, to a value smaller than zero, $`\alpha _{random}<0`$ for $`p=0.8`$. It means that in principle is possible to expect the existence of a critical density $`p_c`$ at which its value is equal to zero. The $`p_c`$ value has been found to be around $`0.9`$ .
In principle it is not clear whether this crossover should occur smoothly or whether the crossover should take place sharply at a critical value $`p_c,`$ separating the two distinct universality classes. This is a crucial question to establish whether the slightly diluted systems should be considered as pure (basically undiluted), as randomly diluted systems, or, on the contrary, as intermediate states between both extreme classes. There is an intrinsic difficulty in detecting the evolution of critical exponents from pure to diluted random Ising systems due to the fact that they are very similar (see Table I). Following Ballesteros et al. we find $`\alpha _{random}=0.051,\beta _{random}=0.3546,\gamma _{random}=1.342`$ in comparison with the pure undiluted values: $`\alpha _{pure}=0.11,\beta _{pure}=0.3267,\gamma _{pure}=1.237`$ (incidentally, this does not happen if the disorder is long range correlated ). That is why it is useful to study some other universal quantity which clearly indicates the difference between the pure and the random universality classes.
For a random hypercubic sample of linear dimension $`L`$ and number of sites $`N=L^d`$, any observable singular property $`X`$ presents different values for different random realizations corresponding to the same average dilution. This means that X behaves as a stochastic variable with average $`[X]`$, variance $`(\mathrm{\Delta }X)^2`$ and a normalized square width $`R_X=(\mathrm{\Delta }X)^2/[X]^2`$. This quantity allows us to determine properly the evolution from the pure to the randomly diluted system by an investigation of its self averaging behavior. A system is said to exhibit self averaging (SA) if $`R_X0`$ as $`L\mathrm{}`$. If the system is away from criticality, $`L>>\xi `$ (being $`\xi `$ the correlation length) the central limit theorem indicates that strong SA must be expected. However, the self-averaging behavior of a ferromagnet at criticality (where $`\xi >>L)`$ is not so obvious. This point has been studied recently. Wiseman and Domany (WD) have investigated the self-averaging of diluted ferromagnets at criticality by means of finite-size scaling calculations, concluding weak SA for both the P and R cases. In contrast Aharony and Harris (AH), using a renormalization group analysis in $`d=4\epsilon `$ dimensions, proved the expectation of a rigorous absence of self-averaging in critically random ferromagnets . More recently, Monte Carlo simulations where used to check this lack of self-averaging in critically disordered magnetic systems . The absence of self-averaging was confirmed. The source of the discrepancy with previous scaling analysis by WD was attributed to the particular size ($`L`$) dependence of the distribution of pseudocritical temperatures used in their work.
In the present work we study the evolution of the normalized square width from $`p=1`$ to $`p=0.8`$ at small steps $`\mathrm{\Delta }p=0.025,`$ in an effort to characterize in detail the evolution of the normalized square width from $`R_M=R\chi =0`$ to the zone where lack of self-averaging $`(R_M0`$ and $`R\chi 0)`$ appears. We will determine whether there is some sharp critical value $`p_c`$ separating both universality classes, or there is a smooth evolution. We have performed Monte Carlo calculations using the Wolff single cluster algorithm in diluted three dimensional Ising systems at criticality for different values of the site occupation spin probability $`1>p>0.8.`$ In order to obtain good enough statistics in our determination of the normalized square width for magnetization and susceptibility, we have used 500 samples for the sizes $`L=20,40,60.`$ The magnetization and susceptibility of each sample was determined using 50.000 MCS leaving the previous 100.000 MCS for thermalization. The critical temperature for each dilution was taken by interpolation between the data reported by Heuer et al. and Ballesteros et al. . We may note that there are no much data in the literature about the critical temperature in this region of slightly diluted systems. A ”nearly-linear” extrapolation of the data for $`T_c`$ vs. $`p`$ from $`p=1`$ to $`p=0.8`$ seems clear from Fig.1. To check this we have calculated the critical temperature for several values of $`p`$ by means of statistics on the Binder Cumulant, and we have found that the data lie over the interpolation functions previously considered.
We can build histograms with the values obtained for susceptibility or magnetization at criticality. In the case of very high $`p`$ values, corresponding to very-low dilution, the width of these histograms is very small, indicating the existence of proper self-averaging. However, for somewhat lower values of $`p`$, the system starts its crossover to the randomly diluted behavior and the width of the histograms begins to increase, indicating that lack of self-averaging is taken place. Fig.2a and b show the evolution of these histograms for the case of the susceptibility and for $`L=60`$. Note that the width of the histograms increases monotonously as $`p`$ decreases, indicating a smooth flow towards the random diluted universality class. We will see this point more clearly studying the value of the normalized square width.
The results obtained for the normalized square width of the magnetization and the susceptibility are presented in Fig3 and Fig4 respectively. Note how in both cases we find a smooth evolution indicating that the crossover from the pure fixed point to the randomly diluted fixed point takes place smoothly and continuously and that there is no apparent critical value of $`p_c`$ acting as a boundary between the two regimens.
The value of the normalized square width for a given $`p`$, can be strongly affected by finite size effects. In order to obtain a value of $`R_M`$ or $`R_\chi `$ independent of $`p`$ for $`p<0.8`$ it is necessary to consider very high values of $`L`$ or to use the so called infinite volume extrapolation . However, for small dilution $`(1>p>0.8)`$ the values of the normalized square width are nearly unaffected by finite size effects , but they are dependent of $`p`$. That is the reason why in Fig3 and Fig4 all the data seem to collapse over the same curve. This does not happen for $`p<0.8`$ where finite size effects clearly appear. To show this, we present in Fig5 data for the susceptibility together with data by Ballesteros et al. for different values of $`L`$ and for values of $`p<0.8.`$ Note that the tendency of the data for $`L\mathrm{}`$, seems to be towards $`R_\chi (p=0.8)`$. It means that the effect of the finite size is to introduce a apparent increase in the value of the normalized square width which should not exist for the sample with $`L=\mathrm{}`$. If we consider the data in the $`p`$-dependent zone, that is $`1>p>0.8`$, where there is small $`L`$ dependence, we can make and extrapolation to $`pp_p`$ (being $`p_p`$ the probability for which the system suffers percolation: $`p_p0.31`$) that is not going to be affected by finite size effects. In our case we have used a hyperbolic tangent to fit our data, $`R_\chi (L,1p)=R_\chi (\mathrm{})tanh[const(1p)]`$, leaving free the universal value of the normalized square width $`R_\chi (\mathrm{})`$ and the slope constant ($`const)`$. Results are shown in Fig5. For $`L=40,60`$ we find a p-independent universal value $`R_\chi (\mathrm{})0.155`$, very close to previously reported values : $`0.150(7)`$ . For $`L=20,`$ on the other hand, finite size effects are important even in the region $`(1>p>0.8)`$, and the extrapolation gives a somewhat higher value $`R_\chi (\mathrm{})0.19`$.
In conclusion, we have presented Monte Carlo data for diluted Ising systems in the region where crossover to the diluted random universality class takes place $`(1>p>0.8).`$ The evolution of the normalized square width for the magnetization $`R_M`$ and the susceptibility $`R_\chi `$ indicates a smooth transition with no critical probability $`p_c`$ (corresponding to a well defined boundary between the pure and the randomly diluted universality classes). The transition zone studied is $`p`$-dependent but $`L`$-independent. This result is very convenient for extrapolation to the universal value $`R_\chi (\mathrm{})`$ which is independent of $`L.`$
We acknowledge financial support from CGCyT through grant PB96-0037.
Figure Captions
Fig1: Critical temperature vs. spin per site probability in the slightly diluted zone. Our data is compared with the extrapolation performed for data by Heuer et al. and Ballesteros et al. .
Fig2: Normalized histograms for the susceptibility values obtained at criticality. The values of the spin per site probability considered are (a) 0.8,0.825,0.85, 0.875 and (b) 0.9,0.925,0.95,0.975.
Fig3: Normalized square width for the magnetization vs. $`1p`$, for values of $`L=20,40,60`$. Dotted line is just a guide for the eye.
Fig4: Normalized square width for the susceptibility vs. $`1p`$, for values of $`L=20,40,60`$. Dotted line is just a guide for the eye.
Fig5: Normalized square width for the susceptibility vs. $`1p`$ for values of $`L=20,40,60`$ (circles), together with the data reported in . The two continuous lines indicate the universal value of $`R_\chi `$ for the randomly diluted Ising system reported in . The hyperbolic extrapolation of the data is indicated by a segmented line for $`L=40,60`$ and by a doted line for $`L=20`$. |
warning/0003/hep-ph0003126.html | ar5iv | text | # ON THE EFFECTIVE LIGHT-CONE QCD-HAMILTONIAN
## 1 Introduction
The bound-state problem in a field theory particularly gauge field theory is a very difficult many-body problem and has not been solved thus far. But perhaps an exact solution is not necessary for reconciling the so succesful constituent quark models with a covariant theory such as QCD and for being able to compute observables like the structure functions and distribution amplitudes from an underlying theory. Perhaps it suffices to have approximate or QCD-inspired solutions which are not wrong from the outset. The present work is still ongoing and contributes to these questions. Perhaps one begins to understand the essential point.
## 2 A QCD-inspired effective interaction
The light-cone approach to the bound state problem aims at diagonalizing the invariant mass-squared operator.$`^\mathrm{?}`$ Its matrix elements are obtained directly from the gauge field Lagrangian in the light-cone gauge and are explicitly tabulated.$`^\mathrm{?}`$ The many-body aspect of the problem was reduced recently, by the method of iterated resolvents,$`^\mathrm{?}`$ to the problem of phrasing an effective Hamiltonian which acts only in the Fock space of one quark and one anti-quark. The reduction is not exact but rests on certain simplifying assumptions which can be checked only a posteriori. Quite cautiously, one should therefore speak of a ‘QCD-inspired’ effective Hamiltonian. Important is that both the effective and the full Hamiltonian have in principle the same eigenvalue spectrum and that the Fock space is systematically reduced to the $`q\overline{q}`$-space. Important is also that the Fock-space is not truncated, that all Lagrangian symmetries are preserved, and that the higher Fock-space amplitudes can be retrieved systematically from the $`q\overline{q}`$-projection once that is available.
The effective interaction has several contributions which are illustrated in Fig. 1. The present work is restricted to flavor-off-diagonal mesons, to mesons where quark and anti-quark have different flavors. Then the effective interaction due to the annihilation of two gluons is kinematically suppressed, and the final (one-body) integral equation in the $`q\overline{q}`$-space is governed by an effective one gluon exchange, i.e.
$`M^2x,\stackrel{}{k}_{};\lambda _1,\lambda _2|\psi =\left[{\displaystyle \frac{\overline{m}_1^2+\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}}{x}}+{\displaystyle \frac{\overline{m}_2^2+\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}}{1x}}\right]x,\stackrel{}{k}_{};\lambda _1,\lambda _2|\psi `$
$`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{\lambda _q^{},\lambda _2^{}}{}}{\displaystyle }{\displaystyle \frac{dx^{}d^2\stackrel{}{k}_{}^{}}{\sqrt{x(1x)x^{}(1x^{})}}}R(x^{},\stackrel{}{k}_{}^{};\mathrm{\Lambda })\times `$
$`\times {\displaystyle \frac{4}{3}}{\displaystyle \frac{\overline{\alpha }(Q;\mathrm{\Lambda })}{Q^2}}\lambda _1,\lambda _2|S|\lambda _1^{},\lambda _2^{}x^{},\stackrel{}{k}_{}^{};\lambda _1^{},\lambda _2^{}|\psi .`$ (1)
Here, $`M^2`$ is the invariant-mass squared eigenvalue and $`x,\stackrel{}{k}_{};\lambda _1,\lambda _2|\psi `$ the associated eigenfunction. It is the probability amplitude for finding a $`q\overline{q}`$-Fock state in which the quark has longitudinal momentum fraction $`x`$ and transversal momentum $`\stackrel{}{k}_{}`$ and the anti-quark correspondingly $`1x`$ and $`\stackrel{}{k}_{}`$, and where the respective helicities are $`\lambda _1`$ and $`\lambda _2`$. The mean four-momentum transfers of the quarks and the spinor factor are respectively defined by
$`Q^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[(k_1k_1^{})^2+(k_2k_2^{})^2\right],`$ (2)
$`S=\lambda _1,\lambda _2|S|\lambda _1^{},\lambda _2^{}`$ $`=`$ $`\left[\overline{u}(k_1,\lambda _1)\gamma ^\mu u(k_1^{},\lambda _1^{})\right]\left[\overline{v}(k_2^{},\lambda _2^{})\gamma _\mu v(k_2,\lambda _2)\right].`$ (3)
The effective quark masses $`\overline{m}_{1,2}`$ and the effective coupling constant $`\overline{\alpha }(Q;\mathrm{\Lambda })`$ depend both, in general, on a regularization scale $`\mathrm{\Lambda }`$. The regulator function $`R(x^{},\stackrel{}{k}_{}^{};\mathrm{\Lambda })`$ restricts the range of integration and in general is a function of the same $`\mathrm{\Lambda }`$, see also below.
The current work is restricted to the lowest order of approximation (LOA) where $`\overline{m}=m`$ and $`\overline{\alpha }(Q;\mathrm{\Lambda })=\alpha `$. The expressions for the next-to-lowest order (NLO) can be found elsewhere.$`^\mathrm{?}`$ One has a fair confidence into the validity of Eq.(1), since the alternative method of Hamiltonian flow $`^\mathrm{?}`$ leads to the very same equation as in LOA.$`^\mathrm{?}`$ The same equation had also been obtained prior to that,$`^{\mathrm{?},\mathrm{?}}`$ with however less stringent arguments. The work of Wilson et al. $`^\mathrm{?}`$ had the same emphasis, but that it did not lead to the same formulas, see also Refs.$`^{\mathrm{?},\mathrm{?}}`$.
## 3 The crucial point in a simple model
Why is a regularization necessary at all?
In light-cone parametrization, the quark is at rest relative to the anti-quark when $`\stackrel{}{k}_{}=0`$ and $`x=\overline{x}\overline{m}_1/(\overline{m}_1+\overline{m}_2)`$. For very small deviations from these ‘equilibrium values’ the spinor factor is diagonal in the helicities, $`\lambda _1,\lambda _2|S|\lambda _1^{},\lambda _2^{}4\overline{m}_1\overline{m}_2\delta _{\lambda _1,\lambda _1^{}}\delta _{\lambda _2,\lambda _2^{}}`$, as can be veryfied from an explicit presentation of $`S`$.$`^\mathrm{?}`$ For very large deviations, particularly for $`\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}`$, holds $`Q^2\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}`$ and $`|S|2\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}`$. The ratio is then a simple dimensionless number, $`S/Q^2=2`$, independent of $`\stackrel{}{k}_{}^{}`$. Both extremes are combined here by
$`{\displaystyle \frac{|S|}{Q}}^2={\displaystyle \frac{4\overline{m}_1\overline{m}_2}{Q^2}}+2,`$ (4)
which is considered as a model of the interaction in singlet states, dropping the less important spin-orbit interaction.
Unfortunately, I have not realized earlier $`^{\mathrm{?},\mathrm{?}}`$ that the innocent and finite number ‘2’ in Eq.(4) has the same orign as the familiar divergencies in gauge field theory. The latter arise, typically, when bilinar products of the elementatry Dirac interaction like in $`S`$ divided by some energy denominator like $`Q^2`$ are integrated over all phase space.
It was clear from the outset that the many-body Hamiltonian has to be regularized.$`^\mathrm{?}`$ Since fancy methods like dimensional regularization are not applicable, Fock-space regularization was considered to be sufficient.$`^\mathrm{?}`$ In practice, it has lead to difficulties and was was replaced by vertex regularization,$`^\mathrm{?}`$ where the elementary Dirac interaction at a vertex was endorsed with kind of a form factor $`F(\mathrm{\Lambda })`$, i.e.
$`\left[\overline{u}(1)\gamma ^\mu u(2)\right]ϵ_\mu (3)\left[\overline{u}(1)\gamma ^\mu u(2)\right]ϵ_\mu (3)F(k_1,k_2,k_3;\mathrm{\Lambda }).`$ (5)
Requiring that the off-shell mass of the two scattered particles $`M_0^2=(k_2+k_3)^2`$ is limited, $`F`$ becomes the step function $`F(k_1,k_2,k_3;\mathrm{\Lambda })=\mathrm{\Theta }(M_0^2\mathrm{\Lambda }^2)`$. In the absolute squares of the effective interaction, as in Eq.(1), it appears as a regulator function R as a consequence of regulating the theory from the outset.
## 4 Rewriting the integral equation in conventional momenta
Unpleasant is that $`x`$ and $`\stackrel{}{k}_{}`$ have different ranges $`(0x1,\mathrm{}k_x\mathrm{})`$. Instead of $`x`$ a $`k_z`$ $`(\mathrm{}k_z\mathrm{})`$ can be introduced by the transform $`^\mathrm{?}`$
$`x(k_z)`$ $`=`$ $`{\displaystyle \frac{E_1+k_z}{E_1+E_2}},\text{with }E_{1,2}=\sqrt{\overline{m}_{1,2}^{\mathrm{\hspace{0.17em}2}}+\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}+k_z^2}.`$ (6)
The Jacobian of the transformation can be cast into the form
$`{\displaystyle \frac{dx}{x(1x)}}`$ $`=`$ $`{\displaystyle \frac{1}{A(k)}}{\displaystyle \frac{dk_z}{m_r}},\text{with }A(k)={\displaystyle \frac{1}{m_r}}{\displaystyle \frac{E_1E_2}{E_1+E_2}},`$ (7)
$`\text{and}{\displaystyle \frac{1}{m_r}}`$ $`=`$ $`{\displaystyle \frac{1}{\overline{m}_1}}+{\displaystyle \frac{1}{\overline{m}_2}},m_s=\overline{m}_1+\overline{m}_2.`$ (8)
In the same convention the kinetic energy becomes
$`T(k)`$ $`=`$ $`{\displaystyle \frac{\overline{m}_1^2+\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}}{x}}+{\displaystyle \frac{\overline{m}_2^2+\stackrel{}{k}_{}^{\mathrm{\hspace{0.17em}2}}}{1x}}m_s^2C(k)\stackrel{}{k}^2,`$ (9)
$`\text{with}C(k)`$ $`=`$ $`(E_1+\overline{m}_1+E_2+\overline{m}_2)\left({\displaystyle \frac{1}{E_1+\overline{m}_1}}+{\displaystyle \frac{1}{E_2+\overline{m}_2}}\right),`$ (10)
without approximation. If one substitutes the wave function according to
$`\psi (x,\stackrel{}{k}_{})`$ $`=`$ $`\sqrt{{\displaystyle \frac{A(x,\stackrel{}{k}_{})}{x(1x)}}}\varphi (x,\stackrel{}{k}_{})`$ (11)
and drops explicit reference to the helicities, one arrives at an equation,
$`\left[M^2m_s^2C(k)\stackrel{}{k}^2\right]\varphi (\stackrel{}{k})={\displaystyle \frac{1}{4\pi ^2m_r}}{\displaystyle \frac{d^3\stackrel{}{k^{}}R}{\sqrt{A(\stackrel{}{k})A(\stackrel{}{k}^{})}}\frac{4}{3}\frac{\overline{\alpha }S}{Q^2}\varphi (\stackrel{}{k^{}})},`$ (12)
which is identical with Eq.(1), but which looks like one with usual 3-momenta.
## 5 Interpretation in configuration space
The main reason for introducing conventional 3-momenta is interpretation. It is more transparent in configuration space. Since the various factors $`A(\stackrel{}{k})`$ with their square-roots prevent straight-forward Fourier transforms, the non-relativistic approximation ($`\stackrel{}{k}^{\mathrm{\hspace{0.17em}2}}\overline{m}_{1,2}^2`$) is applied consistently. Using
$`A(k)=1,C(k)={\displaystyle \frac{m_s}{m_r}},Q^2=(\stackrel{}{k}\stackrel{}{k}^{})^2,`$ (13)
the model interaction of Eq.(4), and in addition the regulator $`R=1`$, Eq.(12) leads to a local Schrödinger equation
$`\left[M^2m_s^2{\displaystyle \frac{m_s}{m_r}}\stackrel{}{k}^2\right]\varphi (\stackrel{}{k})`$ $`=`$ $`{\displaystyle \frac{4\alpha }{3\pi ^2}}{\displaystyle d^3\stackrel{}{k^{}}\left(\frac{m_s}{Q^2}+\frac{1}{2m_r}\right)\varphi (\stackrel{}{k^{}})},`$ (14)
$`\left[M^2m_s^2{\displaystyle \frac{m_s}{m_r}}\stackrel{}{p}^{\mathrm{\hspace{0.17em}2}}\right]\psi (\stackrel{}{r})`$ $`=`$ $`2m_sV(r)\psi (\stackrel{}{r}),\text{ with }\stackrel{}{p}i\stackrel{}{}.`$ (15)
The potential $`V(r)`$ bears great similarity with $`V_{\text{hf-s}}(r)`$, the hyperfine interaction in the singlet channel found for hydrogen in all textbooks, i.e.
$`V(r)`$ $`=`$ $`{\displaystyle \frac{4}{3}}\alpha \left({\displaystyle \frac{1}{r}}+2{\displaystyle \frac{\pi }{m_rm_s}}\delta ^{(3)}(\stackrel{}{r})\right),`$ (16)
$`V_{\text{hf-s}}(r)`$ $`=`$ $`\alpha \left({\displaystyle \frac{1}{r}}+g_p{\displaystyle \frac{\pi }{m_em_p}}\delta ^{(3)}(\stackrel{}{r})\right).`$ (17)
The strange ‘2’ in Eq.(4) finds its explanation as the gyromagnetic ratio for a fermion, with $`g_p=2`$. But Eq.(16) has no solution! A Dirac-delta function is no proper function and must be regulated, for instance by a Yukawa-potential
$`V(r)`$ $`=`$ $`{\displaystyle \frac{4\alpha }{3}}\left({\displaystyle \frac{1}{r}}+{\displaystyle \frac{\mu ^2}{m_rm_s}}{\displaystyle \frac{e^{\mu r}}{r}}\right).`$ (18)
Important is that the delta ($`d^3\stackrel{}{r}\delta ^{(3)}(\stackrel{}{r})=1`$) and the Yukawa have the same strength $`d^3\stackrel{}{r}[\mu ^2\text{exp}(\mu r)/(2\pi r)]=1`$. Transforming back to momentum space gives
$`\left[M^2m_s^2{\displaystyle \frac{m_s}{m_r}}\stackrel{}{k}^2\right]\varphi (\stackrel{}{k})={\displaystyle \frac{4\alpha }{3\pi ^2}}{\displaystyle d^3\stackrel{}{k^{}}\left(\frac{m_s}{Q^2}+\frac{1}{2m_r}\frac{\mu ^2}{\mu ^2+Q^2}\right)\varphi (\stackrel{}{k^{}})}.`$ (19)
Obviously, replacing in configuration space the Dirac-delta function with a Yukawa potential corresponds to introducing in momentum space the regulator function $`R=\mu ^2/(\mu ^2+Q^2)`$.
The regularization of a Dirac-delta function is an old theme of nuclear physics in the context of pairing theory.$`^\mathrm{?}`$ It was the point of orign for the similarity transform,$`^{\mathrm{?},\mathrm{?}}`$ and was investigated recently,$`^\mathrm{?}`$ again.
## 6 Renormalization
The eigen values of the integral equation Eq.(19) depend now on a regularization scale $`\mu `$. In line with the current understanding of field theory one replaces $`\alpha `$ and $`m`$ by functions of $`\mu `$, such that every eigenvalue is stationary against small variations of $`\alpha `$, $`m`$ and $`\mu `$, i.e. $`dM_n^2/d\mu =0`$! The general procedure, however, is not possible here, since $`\alpha `$ and $`\mu `$ appear in Eq.(19) in the typical combination $`\alpha \mu ^2/(\mu ^2+Q^2)`$. One therefore proceeds by looking for a function $`\alpha _\mu `$ such, that the calculated mass of the pion, for example, agrees with the empirical value. The so obtained function $`\alpha _\mu `$ is considered universal. The actual value of $`\mu `$ can then be fixed by a second requirement.
## 7 The pion
For carrying out, in practice, the programme of Sec. 6 one needs an efficient tool for solving Eq.(19). Such one has been developed recently,$`^\mathrm{?}`$ restricted however for simplicity to spherically symmetric s-states. Fixing the up and the down mass to $`\overline{m}_u=\overline{m}_d=1.16`$ (the unit of mass is chosen here as 350 MeV$`/c^2`$), the calculated pion mass is required to agree with the experimental value $`^\mathrm{?}`$ to within eight digits. For every value of $`\mu `$ one obtains thus an $`\alpha _\mu `$ which is displayed in Fig. 3. The even greater surprise came when investigating the spectrum of the pion along the so obtained $`\alpha _\mu `$, as given in Fig. 3. The lowest state, the $`\pi ^+`$, stays nailed fixed to the empirical value, while the second and the third (as well as the higher ones) have an extremum at $`\mu 3.8`$. Such an extremum was not really expected, but in view of the renormalization group is useful: $`\mu `$ determines itsself from the solution!
For suffiently large values of $`\mu `$, say for $`\mu 10`$, see Fig. 3, the coupling constant decreases strongly and the spectrum becomes more and more the familiar Bohr spectrum with a small hyperfine shift. Obviously, it can be calculated perturbatively, indeed, for sufficiently small $`\alpha `$.
Having fixed now the regularization scale $`\mu \mu _0=3.8`$ gives the coupling constant $`\alpha \alpha _{\mu _0}=0.6904`$, see also Fig. 3. The wave function of the pion is plotted in Fig. 5. By a fit to an approximate expression $`\mathrm{\Phi }_a(p)=10/(1+p^2/p_a^2)`$, a width $`p_a=1.471`$ is obtained which is much larger than $`p_c=\frac{4}{3}\alpha m_r=0.5339`$, the Bohr momentum of the Coulomb wave function with the same $`\alpha `$. Having the wave function, one could, in principle, calculate the form factor and the distribution amplitude according to the exact expressions.$`^\mathrm{?}`$ This is however not a trivial undertaking, and, instead of, the above approximate expression was Fourier transformed to $`\psi (\stackrel{}{r})e^{p_ar}`$. The calculated root-mean-sqare radius of the pion becomes then
$$<r^2>^{\frac{1}{2}}=\frac{\sqrt{3}}{p_a}=0.663\text{ fm}.$$
(20)
The good agreement with the experimental value $`<r^2>_\pi ^{\frac{1}{2}}=0.657`$ fm$`^\mathrm{?}`$ is actually the result of fitting the quark masses $`\overline{m}_u=\overline{m}_u=406`$ MeV, see above. This exhausts all freedom in choosing the parameters of Eq.(19).
One should emphasize that the Yukawa potential in Eq.(19) at the scale $`\mu _0`$ pulls down essentially only one state. The others are left rather unperturbed at their Bohr values, see Fig. 3. The analogy with the pairing model in early nuclear physics is more than obvious.$`^\mathrm{?}`$ In a future and more rigorous solution of the full Eq.(12) one expects that the excited states can be disentangled into almost degenerate singlets and triplets, which in turn can be interpreted as an excitation of the pseudo-scalar pion, or the ground state of a vector meson, respectively. In any case, the first excited state of the simple model correlates very well with the mass of the $`\rho ^+`$ and the other vector mesons, see below.
## 8 On the question of confinement
The question of confinement is closely related to the potential $`V(r)`$, but the relation is subtle. One of the great advantages of light-cone quantization is the additivity of the free part and the interaction. Because of the dimensions of invariant mass squares the relation of potential and kinetic energy is somewhat hidden. In analogy to a classical Hamiltonian, the invariant mass-squared operator $`M^2`$ in Eq.(19) is interpreted as the square of a position and momentum dependent mass
$`M(r,p)=m_s\sqrt{1+{\displaystyle \frac{p^{\mathrm{\hspace{0.17em}2}}}{m_sm_r}}+{\displaystyle \frac{2}{m_s}}V(r)}.`$ (21)
For $`V(r)m_s`$ and $`p^{\mathrm{\hspace{0.17em}2}}m_sm_r`$ it could be expanded as
| $`M(r,p)`$ | $``$ | $`m_s`$ | + | $`{\displaystyle \frac{p^{\mathrm{\hspace{0.17em}2}}}{2m_r}}`$ | + | $`V(r)`$, |
| --- | --- | --- | --- | --- | --- | --- |
| | $``$ | rest mass | + | kinetic energy | + | potential energy, |
These conditions, however, can not be satisfied everywhere. I therefore propose to introduce a relativistic potential energy by $`W(r)M(r,p=0)`$. This function is plotted in Fig. 5. It vanishes at $`r_0=0.5194`$ fm and the classical turning point is $`r_t=0.5300`$ fm, in agreement with Eq.(20). Since the sum of the quark masses is 812 MeV, the ionization threshold occurs at 542 MeV: In order to liberate the quarks in the pion, one has to invest more than three times the rest energy of a pion.
## 9 The meson masses
The remaining parameters of the theory, the masses of the strange, charm and bottom quark can then be determined by reproducing the masses of the pseudo-scalar mesons $`K,^{}`$ $`D^0`$ and $`B,^{}`$ respectively. This gives $`\overline{m}_s=508`$, $`\overline{m}_c=1666`$ and $`\overline{m}_b=5054\text{ MeV}`$. The remaining off-diagonal pseudo-scalar mesons are then calculated straightforwardly and compiled in Table 1. In view of the simplicity of the model, the agreement with the empirical values is remarkable, indeed, and at least as good as for any other model. The strong correlation of the first excited state with the vector mesons was mentioned.
## 10 Summary and some conclusions
The very difficult problem of solving the many-body Hamiltonian for QCD is replaced here by the problem of solving an effective QCD-inspired Hamiltonian. It is obtained from the QCD-Lagrangian in the light-cone gauge by the method of iterated resolvents or the Hamiltonian flow equations in the lowest non-trivial approximation. In order to avoid the numerical complexities of the full solution in light-cone coordinates, this effective interaction was stripped-off of almost all ingredients except the Coulomb and the hyperfine interaction, which are analyzed in a non-relativistic treatment with only 3-momenta. It is argued why the hyperfine interaction in the singlet channel needs to be regulated like the usual divergencies in gauge field theory.
It was courageous to apply such a simple model to the mystery particle of QCD, to the pion, but it turns out that both the pion’s mass and size are obtained without vacuum condensates of any kind, just by adjusting the canonical parameters of a gauge field theory, the coupling constant and the quark masses. The necessary regularization constant determines itself by the theory, in line with the current interpretation of the renormalization group. For the pion, as well as for the other mesons, one gets thus for the first time a QCD-inspired light-cone wave function by which one can compute the form factor by the available exact formulas, as well as the higher Fock space amplitudes.
The present approach exposes screening rather than confinement. The removal energy of a quark from a meson is higher than the pion’s mass. A bare quark is thus unlikely to be observed. One wonders whether this is not much more a physical picture than the strict linear confinement obtained by lattice gauge theory, the only available alternative approach for a non-perturbative description. But like the light-cone approach, the latter is burdened with its own pecularities among them the far extrapolation down to the pion.
In fact the present model runs short in several aspects. That the regulator function should apply to the whole kernel, and not only to one piece of the interaction, is a technical detail, which can be easily corrected in future work. More importent is whether the two-gluon annihilation interaction, which was omitted here on perpose, can account for the flavor diagonal mesons and the so important aspects of isospin.
Astounding is also that the present model for QCD differs from QED only by the color factor $`\frac{4}{3}`$ in the coupling constant. Can it be true that the essential difference between these two theories with their so distinctly different phenomenology is manifest only as large and small coupling?
Depite all of that I conclude that the light-cone approach to hadronic physics may now have a fair chance to compute the hadronic spectra and to predict the hadronic wave functions in line with a covariant theory.
## References |
warning/0003/hep-th0003242.html | ar5iv | text | # On the Landau-Ginzburg description of Boundary CFTs and special Lagrangian submanifolds
## 1 Introduction
A complete microscopic description of D-branes wrapped on supersymmetric cycles is available in the the cases where these cycles are submanifolds in flat spaces like tori. The description can also be fairly reliably extended to spaces where the techniques of conformal field theories constructed from purely free fields can be easily applied, as in the case of orbifolds. However it is only recently that the case of D-branes living in non-trivial curved spaces and wrapped on supersymmetric cycles in these spaces have begun to be investigated systematically from a microscopic viewpoint. Following Ooguri et. al., who specified the boundary conditions on the worldsheet $`N=2`$ supersymmetry generators and explained their geometric significance, further efforts have concentrated on extending the boundary conformal field theory description of D-branes to the case of Calabi-Yau (CY) manifolds. Calabi-Yau manifolds in three complex dimensions have been the subject of special attention in view of their importance of these manifolds for string compactification. (See ref. for a nice summary. For earlier work that dealt with similar issues without however explicitly describing D-branes, see ref. .)
In the closed string case, string propagation on Calabi-Yau manifolds can be described by a variety of techniques depending on which region of the space of complex structure and Kähler moduli of the CY manifold one wishes to concentrate on. At the so-called Gepner point in the moduli space of some CY manifolds, explicit descriptions are available in terms of the tensor product of $`N=2`$ conformal field theories. This point can also be described by using the Landau-Ginzburg (LG) description of these conformal field theories. The LG description provides a link between the abstract geometrical structure encoded in the CFT and a more explicit description in terms of the co-ordinates of the algebraic geometric picture of the CY manifold. The LG description can be used also for CY manifolds that may not have a corresponding Gepner construction. More generally, the LG models may be viewed as the description appropriate to a particular region in the enlarged moduli space of Calabi-Yau vacua.
For the study of D-branes one can use the corresponding extensions of these descriptions to world-sheets with boundary. In the case of the Gepner construction, one may use the boundary conformal field theory techniques due to Cardy, to provide an explicit construction of boundary states associated to D-branes. However to make the geometric picture of D-branes more explicit, one may, in simple cases, work with a functional integral description of such theories with an explicit Lagrangian involving free bosons and free fermions. For more complicated examples of CY manifolds one would like to extend the LG description to world-sheets with boundary.
Substantial progress has been achieved in the application of the methods of boundary conformal field theory to the case of D-branes wrapped on supersymmetric cycles in the CY. Following on the work of Recknagel and Schomerus that used the Gepner model construction for the description of the boundary states relevant to D-brane constructions on supersymmetric cycles on Calabi-Yau manifolds, the specific case of D-branes on the quintic Calabi-Yau manifold was studied in detail in the work of Brunner et.al.. Among other results, a particularly important one (and relevant to the results of this paper as we shall explain below) was their use of the identification of the Witten index in the open string sector between two boundaries with the intersection matrix between the corresponding D-branes to study systematically some properties of D-branes with both A-type and B-type boundary conditions (in the notation of Ooguri et.al ). Subsequent papers have utilised these techniques to particularly study B-type boundary states in other Calabi-Yau manifolds.
Despite this impressive progress, several important puzzles and open questions remain. It would take us too far afield to list these but there are two that are the underlying theme of the present paper. What precisely is the geometric interpretation of the large number of D-brane like boundary states that are to be found in the boundary CFTs arising from Gepner type constructions? Secondly, are there more general geometric constructions that may or may not be realised in the boundary conformal field theory approach? While the answer to the second question is generally yes, we still need to explicitly investigate such constructions. We note that these questions need to be clarified further separately in the case of A-type and B-type boundary conditions. For A-type boundary conditions one may appeal to the modified geometric hypothesis of ref. . According to this hypothesis we may expect that the masses and charges for branes with A-type boundary conditions conditions computed in the “large volume” limit continue to hold in the “small volume” limit also. Thus the A-type boundary conditions can in principle be computed in a suitable description that keeps explicit track of the geometry associated with the corresponding Gepner construction (modulo some caveats that we shall discuss later). In the case of B-type branes this is not expected to be true. Following the method developed by Brunner et. al. the data at the Gepner point have to be monodromy transformed to the large volume limit (by using the monodromy transformations computed in the mirror CY) to obtain the corresponding interpretation of these branes. In the case of B-type boundary conditions the corresponding charges in the large volume descriptions have been obtained of all the boundary states obtained in several examples including the quintic. However a full geometric or physical understanding is still lacking, particularly with regard to the description at the Gepner point. In the case of the A-type boundary conditions only one class of boundary states have been tentatively identified with the corresponding geometric construction.
In this paper, as a first step in trying to answer these questions, we will investigate in detail general classes of A-type boundary conditions from a more geometric viewpoint. This will lead us to not only investigate boundary conditions related to the Recknagel-Schomerus construction but also more general constructions that clearly go beyond the Recknagel-Schomerus class.
In an earlier paper, the correspondence between boundary states in boundary CFT and boundary conditions in LG models was studied. This correspondence was explicitly illustrated in the case of the supersymmetric one-cycles of the two-torus, using the common discrete symmetries of the boundary conformal field theory and the boundary LG theory. A general class of linear boundary conditions in the LG models was also described. These are relevant to both the case of D-branes wrapped on the middle-dimensional cycles of a CY as well as the case of even-dimensional D-branes wrapped on holomorphic sub-manifolds of a CY. However an explicit identification of the boundary states of the CFT with those from the LG theory was still lacking.
In this paper we will begin by considering linear A-type boundary conditions in LG models, of the form discussed in our earlier paper. By explicitly performing the open-string Witten index calculation in the LG model and comparing it to the boundary conformal field theory calculation we will definitively show the equivalence of this class of linear boundary conditions with the $`L=k/2`$ class of boundary states in the boundary CFT.
We will then turn to more general, generically non-linear, A-type boundary conditions and show the consistency conditions that are required to ensure that these describe supersymmetric middle-dimensional cycles in Calabi-Yau manifolds. It is well known that in the case of $`N=2`$ world-sheet supersymmetry several interesting features of the conformal limit are seen even when the theory is perturbed away from this limit. Hence models with $`N=2`$ world-sheet supersymmetry even away from the conformal limit are of interest. In our discussion therefore we will consider A-type boundary conditions in Landau-Ginzburg descriptions of minimal models, both without and with perturbations by relevant operators. We extend this discussion to more general cases.
A summary of the main results of the paper is as follows:
1. We compute the open-string Witten index in the LG model and provide evidence that the linear class of boundary conditions in the minimal model correspond to the $`L=k/2`$ boundary states in the minimal model.
2. We show that A-type $`N=2`$ supersymmetry is preserved if the submanifold is Lagrangian. The complete set of boundary conditions associated with this Lagrangian submanifold are presented. The analysis is used to provide a microscopic description of the special Lagrangian submanifolds in $`^n`$ due to Harvey and Lawson.
3. For the cases with a superpotential $`W`$ (which describe hypersurfaces in $`^n`$), we show that one needs to have the boundary conditions have vanishing Poisson bracket with $`(W\overline{W})`$. Thus, these submanifolds are necessarily pre-images of straight lines in the $`W`$-plane.
4. For a single minimal model, we find non-linear boundary conditions by perturbations of the quasi-homogeneous potential by relevant operators. The boundary conditions correspond to straight lines in the $`W`$-plane passing through the minima of the perturbed potential.
5. For the case of the quintic CY threefold, we use these methods to provide an explicit microscopic description of the $`T^3`$ special Lagrangian sub-manifold in the infinite complex structure limit.
Observations closely related to items 3 and 4 above from a slightly different viewpoint<sup>1</sup><sup>1</sup>1These results have been reported by C. Vafa in a recent conference talk. We thank C. Vafa for bringing this to our attention. appear in the work of Hori, Iqbal and Vafa. Some of these results have been reported in a recent paper by Hori and Vafa which appeared while this manuscript was under preparation.)
The organisation of the paper is as follows: In section 2, we discuss the case of LG models with boundary and list a linear class of boundary conditions obtained in . In section 3, we review the construction of A-type boundary states in a single minimal model using Cardy’s prescription. In section 4, we carry out the open-string Witten index computation and provide a map from the boundary conditions in the LG model to a class of boundary states in the corresponding CFT. This is illustrated for the case of a single minimal model and for the Gepner model associated with the quintic. In section 5, we consider general boundary conditions consistent with A-type $`N=2`$ worldsheet supersymmetry. Using this microscopic description, we obtain conditions under which the boundary conditions describe a supersymmetric cycle (special Lagrangian). We apply the methods to some simple examples. We conclude in section 6 with some remarks.
## 2 Landau-Ginzburg theories with boundary
### 2.1 Notation and Conventions
We work in $`N=2`$ superspace with coordinates $`x^m`$, $`\theta ^\alpha `$, $`\overline{\theta }^{\dot{\alpha }}`$ ($`m=0,1`$, $`\alpha ,\dot{\alpha }=+,)`$. Left movers are specified by the index $``$ and right movers by the index $`+`$. The worldsheet has Lorentzian signature (metric=Diag$`(1,+1)`$) and has a boundary at $`x^1=0`$ and is topologically a half-plane.
The Lagrangian for a $`N=2`$ supersymmetric Landau-Ginzburg theory is constructed from chiral superfields ($`y^m=x^m+i\theta ^\alpha \sigma _{\alpha \dot{\alpha }}^m\overline{\theta }^{\dot{\alpha }}`$)
$$\mathrm{\Phi }(x,\theta )=\varphi (y)+\sqrt{2}\theta ^\alpha \psi _\alpha (y)+\theta ^\alpha \theta _\alpha F(y)$$
and anti-chiral superfields. The Lagrangian for $`n`$ chiral superfields $`\mathrm{\Phi }_i`$ is given by
$$S=d^2xd^4\theta K(\mathrm{\Phi },\overline{\mathrm{\Phi }})d^2x𝑑\theta ^+𝑑\theta ^{}W(\mathrm{\Phi })d^2x𝑑\overline{\theta }^+𝑑\overline{\theta }^{}\overline{W}(\overline{\mathrm{\Phi }}),$$
(1)
where $`K`$ is the Kähler potential and $`W`$ is the holomorphic superpotential. We will choose the Kähler potential to be $`K=_i\overline{\mathrm{\Phi }}_i\mathrm{\Phi }_i`$. In the conformal case, the superpotential is taken to be quasi-homogeneous: $`W(\lambda ^{n_i}\mathrm{\Phi }_i)=\lambda ^dW(\mathrm{\Phi }_i)`$, where $`n_i`$ are some integers which are related to the charges of the superfields $`\mathrm{\Phi }_i`$. The Lagrangian takes the following form after the auxiliary fields $`F_i`$ are eliminated<sup>2</sup><sup>2</sup>2In addition, we have symmetrised the action of the derivatives occuring in the Kinetic energy term for the fermions as is done when one is considering worldsheets with boundary.
$`S={\displaystyle }d^2x(_m\overline{\varphi }_i^m\varphi _i+i\overline{\psi }_i(\stackrel{}{_0}+\stackrel{}{_1})\psi _i+i\overline{\psi }_{+i}(\stackrel{}{_0}\stackrel{}{_1})\psi _{+i}`$
$`|{\displaystyle \frac{W}{\varphi _i}}|^2{\displaystyle \frac{^2W}{\varphi _i\varphi _j}}\psi _i\psi _{+j}{\displaystyle \frac{^2\overline{W}}{\overline{\varphi }_i\overline{\varphi }_j}}\overline{\psi }_{+i}\overline{\psi }_j),`$ (2)
where $`A\stackrel{}{_i}B\frac{1}{2}[A(_iB)(_iA)B]`$.
The Lagrangian is invariant under the supersymmetry transformations parametrised by $`ϵ_\alpha `$ and $`\overline{ϵ}_{\dot{\alpha }}`$. The transformations of the fields are given by
$`\delta \varphi _i`$ $`=`$ $`\sqrt{2}(ϵ_{}\psi _{+i}+ϵ_+\psi _i)`$
$`\delta \psi _{+i}`$ $`=`$ $`i\sqrt{2}(_0+_1)\varphi _i\overline{ϵ}_{}+\sqrt{2}ϵ_+{\displaystyle \frac{\overline{W}}{\overline{\varphi }_i}}`$ (3)
$`\delta \psi _i`$ $`=`$ $`i\sqrt{2}(_0_1)\varphi _i\overline{ϵ}_++\sqrt{2}ϵ_{}{\displaystyle \frac{\overline{W}}{\overline{\varphi }_i}}`$
We will be interested in considering the case when boundary conditions preserve part of the supersymmetry. Further, the boundary conditions should cancel the ordinary variations of the action modulo the bulk equations of motion. The ordinary variation of the action gives rise to the following boundary terms
$`\delta _{\mathrm{ord}}S={\displaystyle 𝑑x^0}`$ $`([(_1\overline{\varphi }_i)\delta \varphi +\delta \overline{\varphi }_i(_1\varphi _i)]|_{x^1=0}`$
$`+`$ $`{\displaystyle \frac{i}{2}}[\delta \overline{\psi }_i\psi _i\overline{\psi }_i\delta \psi _i\delta \overline{\psi }_{+i}\psi _{+i}+\overline{\psi }_{+i}\delta \psi _{+i}]|_{x^1=0})`$ (4)
There are two inequivalent sets of boundary conditions which preserve different linear combinations of the left and right $`N=2`$ supersymmetries.
A-type boundary conditions: These are boundary conditions such that the unbroken $`N=2`$ supersymmetry is generated by
$$ϵ_+=\eta \overline{ϵ}_{},$$
(5)
and the complex conjugate equation and $`\eta =\pm 1`$ corresponds to the choice of spin-structure on the worldsheet.
B-type boundary conditions: These are boundary conditions such that the unbroken $`N=2`$ supersymmetry is generated by
$$ϵ_+=\eta ϵ_{},$$
(6)
and the complex conjugate equation and $`\eta =\pm 1`$ corresponds to the choice of spin-structure on the worldsheet. The two boundary conditions are related by the mirror automorphism of the $`N=2`$ supersymmetry algebra under which the left-moving $`U(1)`$ current changes sign.
### 2.2 A-type boundary conditions
Under A-type boundary conditions, the unbroken $`N=2`$ supersymmetry is given by the condition
$$ϵ_+=\eta \overline{ϵ}_{},$$
(7)
where $`\eta =\pm 1`$. In an earlier paper, it was shown that the following conditions<sup>3</sup><sup>3</sup>3The boundary conditions have been adapted to the notation used in this paper. preserve $`N=2`$ supersymmetry and that the boundary terms in ordinary variations (eqn. (4)) of the Lagrangian vanish.
$`(\psi _{+i}A_{ij}\eta \overline{\psi }_j)|_{x^1=0}=0,`$
$`_1(\varphi _i+A_{ij}\overline{\varphi }_j)|_{x^1=0}=0,`$
$`_0(\varphi _iA_{ij}\overline{\varphi }_j)|_{x^1=0}=0,`$ (8)
$`\left(A_{ij}{\displaystyle \frac{W}{\varphi _j}}{\displaystyle \frac{\overline{W}}{\overline{\varphi }_i}}\right)|_{x^1=0}=0,`$
where $`A`$ is a symmetric matrix satisfying $`AA^{}=1`$.
For the $`k`$-th minimal model, the LG description has a superpotential given by $`W=\varphi ^{k+2}/(k+2)`$, the condition involving the superpotential becomes
$$A^{k+2}=1,$$
(9)
which is a condition on the parameter $`A`$ appearing in the boundary condition. Thus, $`A`$ can be any $`(k+2)`$-th root of unity. Hence there are $`(k+2)`$ different boundary conditions which are consistent with $`N=2`$ supersymmetry.
Under the action of the generator $`g`$ of the group $`Z_{k+2}`$, one can easily check that $`AA\mathrm{exp}(4\pi i/(k+2))`$. Suppose we choose to label the different $`(k+2)`$ roots of unity by
$$A_m=\mathrm{exp}(2\pi m/k+2).$$
Then under the action of $`g`$, $`A_mA_{m+2}`$. This suggests that the $`m`$ label here can be associated with the $`M`$ labels of the boundary states constructed in the corresponding minimal model. For odd $`k`$, the allowed values of $`A`$ form a $`(k+2)`$ dimensional orbit while for even $`k=2n`$, one obtains two $`(n+1)`$ dimensional orbits of the $`Z_{n+1}`$ subgroup of the $`Z_{k+2}`$.
### 2.3 B-type boundary conditions
Under B-type boundary conditions, the unbroken $`N=2`$ supersymmetry is given by the condition
$$ϵ_+=\eta ϵ_{},$$
(10)
where $`\eta =\pm 1`$. The following linear boundary conditions were constructed in the LG model
$`(\psi _{+i}+\eta B_{i}^{}{}_{}{}^{j}\psi _j)|_{x=0}=0,`$
$`_1(\varphi _i+B_{i}^{}{}_{}{}^{j}\varphi _j)|_{x=0}=0,`$
$`_0(\varphi _iB_{i}^{}{}_{}{}^{j}\varphi _j)|_{x=0}=0,`$
$`\left({\displaystyle \frac{W}{\varphi _i}}+B_{i}^{}{}_{}{}^{j}{\displaystyle \frac{W}{\varphi _j}}\right)|_{x=0}=0,`$ (11)
where the boundary condition is specified by a hermitian matrix $`B`$ which satisfies $`B^2=1`$. Since $`B`$ squares to one, its eigenvalues are $`\pm 1`$. An eigenvector of $`B`$ with eigenvalue of $`+1`$ corresponds to a Neumann boundary condition and $`1`$ corresponds to a Dirichlet boundary condition. Associated with every eigenvector with eigenvalue $`+1`$, there is a non-trivial condition involving the superpotential which is given by the last of the above boundary conditions.
For a LG model with a single chiral superfield such as the minimal model, the consistency condition involving the superpotential does not permit the imposition of a Neumann boundary condition on the scalar field. Thus, one can only impose Dirichlet boundary conditions on the scalar. We will not consider B-type boundary states further in this paper.
## 3 Boundary States in the $`N=2`$ minimal models
### 3.1 Notation and Conventions
The $`k`$-th $`N=2`$ minimal models has central charge $`c=3k/(k+2)`$. The primary fields of the model are labelled by three integers $`(l,m,s)`$ with
$`l`$ $`=`$ $`0,\mathrm{},k,`$
$`m`$ $`=`$ $`(k+1),k,\mathrm{},(k+2)\mathrm{mod}(2k+4),`$
$`s`$ $`=`$ $`1,0,1,2\mathrm{mod}4,`$
subject to the constraint that $`l+m+s`$ is even. In addition there is a field identification given by
$$(l,m,s)(kl,m+k+2,s+2).$$
Even $`s`$ refers to the NS sector and odd $`s`$ refers to the R sector fields.
A complete set of labels for the minimal model (using the field identification mentioned above) are given by
$$l=0,\mathrm{},k/2,m=(k+1),k,\mathrm{},(k+2)\mathrm{and}s=1,0,1,2,$$
where $`k/2`$ is the largest integer less than or equal to $`k/2`$ and $`(l+m+s)`$ is even.
Another equivalent set of labels is given by
$$l=0,\mathrm{},k,m=(k+1),k,\mathrm{},(k+2)\mathrm{and}s=0,1,$$
where again we have the condition that $`(l+m+s)`$ must be even. The dimension $`h`$ and $`U(1)`$ charge $`q`$ of the fields are given by
$`h_{l,m,s}`$ $`=`$ $`{\displaystyle \frac{l(l+2)m^2}{4(k+2)}}+{\displaystyle \frac{s^2}{8}}`$
$`q_{l,m,s}`$ $`=`$ $`{\displaystyle \frac{m}{k+2}}{\displaystyle \frac{s}{2}}\mathrm{mod}2`$ (12)
The $`k`$-th minimal model has a $`_{k+2}\times _2`$ discrete symmetry. The action of the discrete symmetry on the fields is given by
$`g\mathrm{\Phi }_{l,m,s}=e^{\frac{2\pi im}{k+2}}\mathrm{\Phi }_{l,m,s},`$ (13)
$`h\mathrm{\Phi }_{l,m,s}=()^s\mathrm{\Phi }_{l,m,s},`$ (14)
where $`g`$ and $`h`$ generate the $`_{k+2}`$ and $`_2`$ respectively. We will be interested in the action of $`\mathrm{exp}(i\pi J_0)`$ on a bulk state:
$$\mathrm{exp}(i\pi J_0)|l,m,s=\mathrm{exp}(i\pi [\frac{m}{k+2}\frac{s}{2}])|l,m,s.$$
(15)
Note that this is not necessarily equal to $`()^{F_L}`$ when one is considering a single minimal model (as opposed to a Gepner construction involving integer $`U(1)`$ charges). However, one can see that product $`()^{F_L}\mathrm{exp}(i\pi J_0)`$ commutes with all generators of the $`N=2`$ supersymmetry algebra. In the minimal model, this product on general grounds should be given by $`f(g,h)`$, where $`f`$ is some function of the the discrete symmetries which commute with $`N=2`$ supersymmetry. Thus, we will also be interested in defining the action of $`()^{F_L}`$ on the states $`|l,m,s`$. We will require that $`()^{F_L}`$ gives $`\pm 1`$ acting on the NS sector states. For odd $`k`$, we find
$$()^{F_L}|l,m,s=\mathrm{exp}(i\pi [m+\frac{s}{2}])|l,m,s.$$
(16)
The above assignment is consistent with the identification $`|l,m,s|kl,m+k+2,s+2`$ of states. Thus, for odd $`k`$, $`f(g,h)=g^{(k+3)/2}`$.
For a given representation $`p`$ of the $`N=2`$ algebra, the character is defined as
$$\chi _p(q,z,u)=e^{2i\pi u}\mathrm{Tr}_p[e^{2i\pi zJ_0}q^{(L_0\frac{c}{24})}]$$
(17)
where $`q=\mathrm{exp}(2i\pi \tau )`$ and $`u`$ is an arbitrary phase. The trace runs over the representation denoted by $`p`$. The characters of the $`N=2`$ minimal models are defined in terms of the Jacobi theta functions $`\theta _{n,m}(\tau ,z,u)`$ and characters of a related parafermionic theory $`C_m^l(\tau )`$ as:
$$\chi _{l,m}^{(s)}(q,z,u)=\underset{j\mathrm{mod}k}{}C_{m+4js}^l(\tau )\theta _{2m+(4js)(k+2),2k(k+2)}(\tau ,2kz,u).$$
(18)
The characters $`\chi _{l,m}^{(s)}`$ have the property that they are invariant under $`ss+4`$ and $`mm+2(k+2)`$ and are zero if $`(l+m+s)`$ is odd. By using the properties of the theta functions, the modular transformation of the minimal model characters is found to be
$$\chi _{l,m}^{(s)}(\widehat{q},0,0)=C\underset{l^{},m^{},s^{}}{}\mathrm{sin}(l,l^{})_k\mathrm{exp}\left(\frac{i\pi mm^{}}{k+2}\right)\mathrm{exp}(\frac{i\pi ss^{}}{2})\chi _{l^{},m^{}}^{(s^{})}(q,0,0)$$
(19)
where $`\widehat{q}=\mathrm{exp}(2i\pi /\tau )`$; $`(l,l^{})_k\left(\frac{\pi (l+1)(l^{}+1)}{k+2}\right)`$ and $`C=1/\sqrt{2}(k+2)`$.
### 3.2 A-type Boundary States in the $`N=2`$ minimal models
We will consider A-series which has a diagonal partition function. For A-type boundary conditions, there are Ishibashi states for each possible value of $`(l,m,s)`$. We will label these states by $`|l,m,s`$. Using Cardy’s prescription, we can construct the boundary states
$$|L,M,S=\sqrt{C}\underset{l,m,s}{}\frac{\mathrm{sin}(L,l)_k}{\left[\mathrm{sin}(l,0)_k\right]^{\frac{1}{2}}}\mathrm{exp}\left(\frac{i\pi Mm}{k+2}\right)\mathrm{exp}(\frac{i\pi Ss}{2})|l,m,s$$
(20)
where we have used upper case letters to represent the boundary state and lower case letters for the Ishibashi states. One can check that the boundary states $`|L,M,S`$ and $`|L,M,S+2`$ differ only in the sign occurring in front of the RR-sector (i.e., odd $`s`$ ) Ishibashi states. Thus, it suffices to study only the $`S=0,1`$ states.
The field identification $`(l,m,s)(kl,m+k+2,s+2)`$ in the bulk minimal model extends to the boundary states as $`|L,M,S|kL,M+k+2,S+2`$. The annulus amplitude $`𝒜_{L,M,S}(q)`$ (with modular parameter $`q`$) which is given by the modular transform of the cylinder amplitude $`0,0,0|\widehat{q}^{L_0c/24}|L,M,S`$, is given by
$$𝒜_{L,M,S}(q)=\chi _{L,M}^{(S)}(q).$$
(21)
Note that this vanishes when $`(L+M+S)`$ is odd. Thus, we impose the additional condition that $`(L+M+S)`$ be even.
The full set of boundary states that we obtain are specified by the following values of $`(L,M,S)`$:
$$L=0,\mathrm{},k/2,M=(k+1),k,\mathrm{},(k+2)\mathrm{and}S=0,2,$$
In this labelling convention, we will sometimes loosely refer to the $`S=2`$ state as an antibrane (in analogy with the situation in the full Gepner construction) since the $`S=2`$ boundary state differs from the $`S=0`$ state (for identical values of $`L,M`$) by an overall sign in front of the RR Ishibashi states. This set of labels takes care of the identification of boundary states mentioned earlier except for the case when $`k`$ is even and $`L=k/2`$. For this case, the antibrane corresponding to the boundary state $`|k/2,M,0`$ is $`|k/2,M+k+2,0`$.
Under the discrete symmetries of the minimal model $`_{k+2}\times _2`$, the boundary states transform as
$`g|L,M,S=|L,M+2,S`$ (22)
$`h|L,M,S=|L,M,S+2`$ (23)
Thus all A-type boundary states can be classified into orbits of the discrete symmetry. When $`k`$ is odd, there are $`k/2=(k1)/2`$ orbits of length $`(k+2)`$ after taking into account the identification of the boundary states mentioned earlier. For even $`k`$, when $`l<k/2`$, then the states are in orbits of length $`(k+2)`$. However, when $`l=k/2`$, since $`l=kl`$, the orbit length is shorter and equals $`(k+2)/2`$ (provided one ignores the distinction between the $`S=0`$ and $`S=2`$ states).
The characters of the full $`N=2`$ supersymmetry algebra are given by the combination $`(\chi _{lm}^{(s)}+\chi _{lm}^{(s+2)})`$. It is thus of interest to construct boundary states corresponding to these characters. In this regard consider
$$|L,M,\pm \frac{1}{\sqrt{2}}(|L,M,S\pm |L,M,S+2).$$
(24)
From the earlier discussion, it is clear that the states $`|L,M,+`$ involve Ishibashi states from the NSNS sector and $`|L,M,`$ involve Ishibashi states from the RR sector. These states also are more natural in the construction of boundary states in the Gepner model since the tensor product of boundary states $`_i|L_i,M_i,+`$ incorporates the condition that NSNS states of each sub-theory (labelled by $`i`$) are tensored to each other and the tensor product of boundary states $`_i|L_i,M_i,`$ works similarly for the RR states. The annulus amplitude $`𝒜_{L,M,\pm }(q)`$ (with modular parameter $`q`$) which is given by the modular transform of the cylinder amplitude $`0,0,+|\widehat{q}^{L_0c/24}|L,M,\pm `$, is given by
$$𝒜_{L,M,\pm }(q)=\chi _{L,M}^{(S)}(q)\pm \chi _{L,M}^{(S+2)}(q),$$
(25)
where $`S=L+M`$ mod $`2`$. Under the discrete symmetries of the minimal model $`_{k+2}\times _2`$, the boundary states transform as
$`g|L,M,\pm `$ $`=`$ $`|L,M+2,\pm `$ (26)
$`h|L,M,\pm `$ $`=`$ $`\pm |L,M,\pm `$ (27)
As before, all states except the case when $`k`$ is even and $`L=k/2`$, the states can be arranged in $`_{k+2}`$ orbits. However, when $`k`$ is even and $`L=k/2=n`$, the orbit length is shorter. One has
$$g^{n+1}|L,M,\pm =\pm |L,M,\pm ,$$
Thus, they have orbit length $`(n+1)=(k+2)/2`$.
## 4 Computing the open-string Witten index
We have so far constructed A-type boundary conditions in the Landau-Ginzburg model and constructed A-type boundary states in the corresponding minimal model. However, since the number of boundary conditions is smaller than the number of states, we would like to identify to which boundary states to which they correspond. This is not easy to do even in simple cases such as the Ising model with boundary. A useful tool in this regard is to classify boundary conditions and boundary states in terms of discrete symmetries such as the $`_2`$ in the Ising model. In the present problem, as we have already seen, there is a $`_{k+2}`$ discrete group which organises the boundary conditions and boundary states into orbits. It turns out that this alone is sufficient to provide an identification for the even $`k`$ minimal model. The LG boundary conditions form two orbits of the $`_{(k+2)/2}`$ subgroup of $`_{k+2}`$. This uniquely identifies them with the $`L=k/2`$ boundary states.
This is not the case for odd $`k`$ where the LG boundary conditions form a single $`Z_{k+2}`$ orbit which is true of all boundary states in the corresponding minimal model. In order to make the identification, we will use a open-string Witten index computation (due to Douglas and Fiol). In the context of Calabi-Yau threefolds, this index computes the intersection matrix between three cycles. We will compute the index in both the LG as well as boundary CFT and show that the LG boundary conditions correspond to the $`L=k/2`$ boundary states.
Let $`|B`$ and $`|B^{}`$ be two boundary states. The Witten index is defined as
$$\stackrel{~}{}_{BB^{}}={}_{RR}{}^{}B^{}|()^{F_L}\stackrel{~}{q}^{(L_0c/24)}|B_{RR}^{},$$
(28)
where $`|B_{RR}`$ refers to the RR part of the boundary state. In the open-string channel, this counts the number of Ramond ground states of the Hamiltonian $`H_{BB^{}}`$:
$$\stackrel{~}{}_{BB^{}}=\mathrm{Tr}_{BB^{}}\left[()^Fq^{(L_0c/24)}\right].$$
(29)
As discussed by Witten, the operator $`()^{F_L}`$ can be replaced by $`\mathrm{exp}(i\pi J_0^L)`$ where $`J_0^L`$ is the zero-mode of the left-moving $`U(1)`$ current. Thus, we will be computing the following object
$$_{BB^{}}={}_{RR}{}^{}B^{}|e^{i\pi J_0^L}\widehat{q}^{(L_0c/24)}|B_{RR}^{}.$$
(30)
In the open-string channel, this will be given by
$$_{BB^{}}=\mathrm{Tr}_{BB^{}}\left[e^{i\pi J_0}q^{(L_0c/24)}\right],$$
(31)
where $`J_0`$ is the charge associated with the unbroken U(1).
For the level $`k`$ ($`k`$ odd) minimal model, one can study the action of $`()^{F_L}`$ and $`\mathrm{exp}(i\pi J_0^L)`$ on the boundary states. For A-type boundary states, we can see that
$`\mathrm{exp}(i\pi J_0^L)|L,M,S`$ $`=`$ $`|L,M+1,S+1`$ (32)
$`()^{F_L}|L,M,S`$ $`=`$ $`|L,M+k+2,S1.`$ (33)
Thus, we see that for A-type boundary states (and odd $`k`$)
$$\mathrm{exp}(i\pi J_0^L)=g^{\frac{k+3}{2}}h()^{F_L}.$$
(34)
### 4.1 Boundary Minimal Model Calculation
We will now compute the following in the boundary minimal model
$$_{L,M,0;L^{},M^{},0}{}_{RR}{}^{}L^{},M^{},0|\mathrm{exp}(i\pi J_0)\widehat{q}^{(L_0c/24)}|L,M,0_{RR}^{}.$$
(35)
This calculation is identical to the one in the appendix of ref. tailored to the case of a single minimal model. We reproduce it here for completeness. Using the expression for the boundary states constructed using Cardy’s prescription, we get
$$_{L,M,0;L^{},M^{},0}=C\underset{l,m,s}{\overset{R}{}}\frac{\mathrm{sin}(L,l)_k\mathrm{sin}(L^{},l)_k}{\mathrm{sin}(l,0)_k}\mathrm{exp}\left(\frac{i\pi (MM^{}+1)m}{k+2}\right)e^{i\pi s/2}\chi _{lm}^s(\widehat{q}),$$
(36)
where the $`R`$ in the summation refers to the restriction to the Ramond sector (i.e, $`s=\pm 1`$). On transforming to the open-string channel by an S-transformation, one obtains
$`_{L,M,0;L^{},M^{},0}`$ $`=`$ $`C^2{\displaystyle \underset{l,m,s}{\overset{R}{}}}{\displaystyle \underset{l^{},m^{},s^{}}{}}{\displaystyle \frac{\mathrm{sin}(L,l)_k\mathrm{sin}(L^{},l)_k\mathrm{sin}(l.l^{})_k}{\mathrm{sin}(l,0)_k}}e^{\left(\frac{i\pi \mu m}{k+2}\right)}e^{i\pi s(1+s^{})/2}\chi _{l^{}m^{}}^s^{}(q),`$ (37)
$`=`$ $`2C^2{\displaystyle \underset{l,m}{\overset{R}{}}}{\displaystyle \underset{l^{},m^{}}{\overset{R}{}}}{\displaystyle \frac{\mathrm{sin}(L,l)_k\mathrm{sin}(L^{},l)_k\mathrm{sin}(l.l^{})_k}{\mathrm{sin}(l,0)_k}}e^{\left(\frac{i\pi \mu m}{k+2}\right)}I_l^{}^m^{}(q),`$
where $`\mu MM^{}+m^{}+1`$ and
$$I_l^m(q)\chi _{l,m}^{(1)}(q)\chi _{l,m}^{(1)}(q)=\delta _{m,l+1}\delta _{m,l1}.$$
(See ref. for the above relation.) In the above, we have carried out the $`s`$ and $`s^{}`$ summations and hence the restriction $`R`$ now implies that $`(l+m)`$ and $`(l^{}+m^{})`$ must be odd. On carrying out the summation over $`m`$, we get
$$_{L,M,0;L^{},M^{},0}=2C^2(k+2)\underset{l=0}{\overset{k}{}}\underset{l^{},m^{}}{\overset{R}{}}\frac{\mathrm{sin}(L,l)_k\mathrm{sin}(L^{},l)_k\mathrm{sin}(l.l^{})_k}{\mathrm{sin}(l,0)_k}\delta _\mu ^{(k+2)}()^{\frac{\mu (l+1)}{(k+2)}}I_l^{}^m^{}(q),$$
(38)
where $`\delta _{\mu ,0}^{(k+2)}`$ is the periodic delta function of period $`(k+2)`$ i.e, it is non-vanishing for $`\mu =0`$ mod $`(k+2)`$. We can now carry out the summation over $`l`$. We then obtain
$$_{L,M,0;L^{},M^{},0}=C^2(k+2)^2\underset{l^{},m^{}}{\overset{R}{}}N_{LL^{}}^l^{}\delta _{MM^{}+m^{}+1}^{(2k+4)}I_l^{}^m^{}(q),$$
(39)
where $`N_{LL^{}}^l^{}`$ is the $`SU(2)`$ level $`k`$ fusion coefficient and $`\delta _M^{(2k+4)}`$ is the periodic delta function with period $`(2k+4)`$. On carrying out the summations over $`l^{}`$ and $`m^{}`$ we obtain (after substituting for $`C`$)
$$_{L,M,0;L^{},M^{},0}=N_{L,L^{}}^{MM^{}},$$
(40)
where we have continued the top index $`M`$ of the $`SU(2)`$ fusion coefficient $`N_{L,L^{}}^M`$ to values mod $`(2k+4)`$ following the work of Brunner et. al.. The continuation is given by
$$N_{L,L^{}}^{l2}=N_{L,L^{}}^l\mathrm{and}N_{L,L^{}}^1=N_{L,L^{}}^{k+1}=0,$$
where $`l=0,\mathrm{},k`$. Thus the intersection number is given by the appropriate fusion coefficient. Following ref. , we can write the fusion coefficient $`N_{L,L^{}}^M`$ as a matrix in the index $`M`$. This matrix can be represented as a polynomial in $`g`$, the generator of $`_{k+2}`$. Using this presentation of the fusion coefficient, the Witten index for odd $`k`$ and $`L=L^{}=(k1)/2`$ is given by
$$^{mm}=\left(1+g+\mathrm{}+g^{(k1)/2}g^1g^2\mathrm{}g^{(k1)/2}\right),$$
(41)
where we use the superscript $`mm`$ to denote that this is a minimal model computation.
### 4.2 Boundary Landau-Ginzburg Calculation
We will now compute the Witten index in the LG model. The worldsheet is assumed to have the topology of an annulus (of width $`\pi `$). We will impose A-type boundary conditions at the two ends of the strip i.e., at $`x^1=0`$ and $`x^1=\pi `$. We Wick rotate the time coordinate to Euclidean space and make it periodic. At $`x^1=\pi `$. we impose the condition
$`_1\mathrm{Re}\varphi `$ $`=`$ $`\mathrm{Im}\varphi =0,`$
$`\psi _+`$ $`=`$ $`\overline{\psi }_{},`$ (42)
This corresponds to the choice $`A=1`$ in the notation of the earlier section. At $`x^1=0`$, we impose
$`_1\mathrm{Re}\left(\mathrm{exp}({\displaystyle \frac{i\pi m}{k+2}})\varphi \right)`$ $`=`$ $`\mathrm{Im}\left(\mathrm{exp}({\displaystyle \frac{i\pi m}{k+2}})\varphi \right)=0,`$
$`\psi _+`$ $`=`$ $`\mathrm{exp}({\displaystyle \frac{2i\pi m}{k+2}})\overline{\psi }_{},`$ (43)
This corresponds to the choice $`A_m=\mathrm{exp}(\frac{2i\pi m}{k+2})`$.
We will use the doubling trick to convert the annulus into a torus. The doubling for fermions is done by introducing the extended fermion $`\mathrm{\Psi }(x^1,x^0)`$.
$$\mathrm{\Psi }(x^1,t)=\{\begin{array}{ccc}\psi _{}(x^1,t)& \mathrm{for}& 0\sigma \pi \\ \overline{\psi }_+(2\pi x^1,t)& \mathrm{for}& \pi x^12\pi \end{array}$$
(44)
This automatically imposes the condition $`\psi _{}=\overline{\psi }_+`$ at the boundary at $`x^1=\pi `$. The boundary condition at $`x^1=0`$ becomes the periodicity on the extended fermion $`\mathrm{\Psi }`$. The bosonic fields can also be doubled in a similar fashion.
The counting of the Ramond ground states for the above situation can now be seen to be identical to the counting of Ramond ground states in the twisted sector of a certain orbifold: it is the $`m`$th twisted sector of the orbifolding of the $`k`$th minimal model by $`Z_{k+2}`$. This computation has been carried out by Vafa and we quote his result. For $`m0`$, there is precisely one ground state. This observation more or less uniquely identifies the boundary condition with the $`l=k/2`$ boundary states. From eqn. (41), where we have given the Witten index for the $`l=k/2`$ boundary states: one can clearly see that there is typically one Ramond ground state in all sectors except in one case. For $`m=0`$, i.e., for the case where one has identical boundary conditions on both ends of the annulus, one is dealing with the untwisted sector. In this sector, free field methods cannot be used. However, one has $`(k+1)`$ Ramond ground states. However, none of them satisfy the $`J_L=J_R`$ boundary condition in the open-string channel. Thus, all Ramond ground states are projected out and the Witten index is zero. (If $`k`$ were even, there is one Ramond ground state with vanishing left and right $`U(1)`$ charges and hence the Witten index is one in this case.)
In order to completely carry out the full LG computation, we need to suitably assign a fermion number to the ground state. Let us assign fermion number $`()^m`$ to the Ramond ground state<sup>4</sup><sup>4</sup>4The interchange of boundary conditions at $`x^1=0,\pi `$ can be represented by $`A_mA_m`$. This choice makes the Witten index antisymmetric under the exchange. of the boundary condition given by $`A_m`$, for $`m=1,\mathrm{},k+1`$. Using the conventions of Brunner et al., we can rewrite the above result as (for odd $`k`$)
$$^{LG}=\left(g+\mathrm{}+g^{(k+1)/2}g^{(k+3)/2}\mathrm{}g^{k+1}\right),$$
(45)
where we use the superscript $`LG`$ to indicate that the Witten index was computed in the LG model. Note that
$$^{mm}=g^{(k+1)/2}^{LG},$$
(46)
This difference can be understood as follows: The computation in the LG model is a Witten index computation while the minimal model computation is one where $`\mathrm{exp}(i\pi J_0)`$ replaces $`()^F`$. Eqn.(34) provides the relation between the two operations (on the boundary states). Thus one interprets the $`\mathrm{exp}(i\pi J_0)`$ to correspond to an additional time-twisting by $`g^{(k+3)/2}`$ in the Witten index computation done in the doubled theory. Following the method used in closed string LG orbifolds as in ref. , this can be seen to be be equivalent to an additional space-twisting by $`g^{(k+3)/2}`$ which provides the required shift of $`g^{(k+1)/2}`$ in the calculation. The minus sign comes from the action of $`h`$, which maps branes to anti-branes and thus changes the sign in the Witten index computation. Thus, this identifies the LG boundary conditions with the $`L=(k1)/2`$ boundary states for odd $`k`$.
### 4.3 Landau-Ginzburg Orbifolds
The work of Greene, Vafa and Warner showed the relationship between certain LG orbifolds and Gepner models. For example, the Gepner model description of the Calabi-Yau threefold, the quintic, at a special point in its moduli is given by the tensor product of five copies of $`k=3`$ minimal models subject to certain projections. The LG description involves five chiral superfields $`\mathrm{\Phi }_i`$ with superpotential
$$W(\mathrm{\Phi })=\mathrm{\Phi }_1^5+\mathrm{\Phi }_2^5+\mathrm{}+\mathrm{\Phi }_5^5.$$
Further, the orbifolding corresponds to the identification $`\varphi _i\alpha \varphi _i`$, for all $`i=1,\mathrm{},5`$ ($`\alpha `$ is a non-trivial fifth root of unity). It was argued by Greene, Vafa and Warner, that this condition is equivalent to the integer $`U(1)`$ projection in the corresponding Gepner model.
In the boundary LG, we find the following set of A-type boundary conditions given by the matrix $`A(\{m_i\})=\mathrm{Diag}(\alpha ^{m_1},\mathrm{},\alpha ^{m_5})`$. The equivalence relation mentioned earlier implies that the two matrices given by parameters $`\{m_i\}`$ and $`\{m_i+2\}`$ are equivalent. As before, we would like to calculate the Witten index which is equivalent to the intersection matrix for these cycles.
Before orbifolding it is clear that the intersection matrix is simply the product of the intersection matrix of the individual theories. The further orbifolding by the diagonal $`_5`$(which results in the integer U(1) charge projection) is implemented as a projection (and hence time-twisting ) in the closed-string channel. This will therefore show up as a sum over space-twisted sectors in the open-string channel. Hence in the computation of the Witten index in the doubled torus the final result is as follows:
$$=\underset{\nu =0}{\overset{4}{}}\underset{i=1}{\overset{5}{}}N_{1,1}^{m_im_i^{}4+2\nu }$$
(47)
where we $`\{m_i\}`$ and $`\{m_i^{}\}`$ are the boundary conditions at the two ends of the annulus and $`N_{1,1}^M`$ is the $`SU(2)_3`$ fusion coefficent. The end result can be written in the compact notation of ref. as
$$=\underset{i=1}{\overset{5}{}}(g_i+g_i^2g_i^3g_i^4)$$
(48)
subject to the condition that $`g_1g_2g_3g_4g_5=1`$. The result is as written in ref. and is consistent with the $`L=1`$ assignment of boundary state labels for each individual minimal model.
The LG calculation, especially the part involving summing over twisted sectors, is quite similar to the spacetime intersection matrix calculation (see section 2 of ref. ). In the spacetime calculation, the intersection calculation involves summing over patches which becomes different twisted sectors in the LG orbifold computation. For a single minimal model, we needed to explain the shift between the $`()^F`$ and $`\mathrm{exp}(i\pi J_0)`$ computations. However, for the LG orbifold, the two results are identical due to the condition $`g_1g_2g_3g_4g_5=1`$. Finally, the similarity with the spacetime intersection calculation suggests that the expectation from the modified geometric hypothesis that the central charges of the A-branes should be the same at different points in the Kähler moduli space is true.
The methods used here clearly apply to more general examples involving the linear boundary conditions in LG models as discussed in section 2. We will later see more general conditions where the Witten index computation is quite difficult.
## 5 Non-linear boundary conditions in LG models
As we have seen, the boundary LG description seems to provide fewer boundary conditions than the corresponding boundary minimal model. This situation holds for more general examples such as the one involving LG description of Calabi-Yau manifolds. The class of boundary conditions considered in correspond to the linear class. We shall now try to generalise these conditions and see if we can obtain new conditions.
### 5.1 LG models with a single chiral superfield
We shall first consider the simplest case of an LG model involving a single chiral superfield $`\mathrm{\Phi }`$. The most general boundary condition is given by
$$F(\varphi ,\overline{\varphi })=0,$$
(49)
where $`F(\varphi ,\overline{\varphi })`$ is a real function. We will have to impose additional conditions such that A-type $`N=2`$ supersymmetry $`ϵ_+=\eta \overline{ϵ}_{}`$ is preserved and all boundary terms (eqn. (4)) which appear in the ordinary variation of the Lagrangian vanish. In order to achieve this, we will first consider all new conditions generated under the unbroken A-type $`N=2`$ supersymmetry.
The first supersymmetric variation leads to the following condition:
$$\frac{F}{\varphi }\psi _++\eta \frac{F}{\overline{\varphi }}\overline{\psi }_{}=0.$$
(50)
The supersymmetric variation of the above equation leads to the following additional conditions:
$`\left[{\displaystyle \frac{F}{\varphi }}_1\varphi {\displaystyle \frac{F}{\overline{\varphi }}}_1\overline{\varphi }\right]i\left[{\displaystyle \frac{F}{\varphi }}{\displaystyle \frac{F}{\overline{\varphi }}}\right]^{\frac{1}{2}}K\psi _{}\overline{\psi }_{}`$ $`=`$ $`0`$ (51)
$`\{F,W(\varphi )\overline{W}(\overline{\varphi })\}_{PB}`$ $`=`$ $`0,`$ (52)
where $`K`$ is the extrinsic curvature of the curve $`F=0`$ in the complex $`\varphi `$-plane given by
$$K=\left[\frac{F}{\varphi }\frac{F}{\overline{\varphi }}\right]^{\frac{3}{2}}\left[\left(\frac{F}{\overline{\varphi }}\right)^2\frac{^2F}{\varphi ^2}2\left|\frac{F}{\varphi }\right|^2\frac{^2F}{\varphi \overline{\varphi }}+\left(\frac{F}{\varphi }\right)^2\frac{^2F}{\overline{\varphi }^2}\right]$$
One can check that the boundary terms in the ordinary variation vanish under these boundary conditions. The linear cases discussed in section 2 correspond to the case when $`K=0`$ (since the boundary curves are straight lines in the $`\varphi `$-plane) and are clearly seen to be special case of the more general boundary condition $`F=0`$.
The vanishing of the Poisson bracket $`\{F,W(\varphi )\overline{W}(\overline{\varphi })\}_{PB}`$ imposes an important restriction on the possible boundary curves in the $`\varphi `$-plane. Since in a two-dimensional phase space, there can be at most one constant of motion, the only possible boundary condition is
$$F=W(\varphi )\overline{W}(\overline{\varphi })ic,$$
(53)
where $`c`$ is a real constant. These correspond to straight lines in the W-plane which are parallel to the real $`W`$ axis. For the case when $`W=\varphi ^{k+2}/(k+2)`$, the pre-image of $`F=0`$ in the $`\varphi `$-plane will generically have $`(k+2)`$ components. When $`c=0`$, these $`(k+2)`$ pre-images are precisely the $`(k+2)`$ linear boundary conditions that we have already obtained!
We can now discuss as to how the other boundary conditions will appear in the LG model. In this regard, we would like to make the following observations: (i) The superpotential $`W`$ has $`(k+1)`$ degenerate minima at $`\varphi =0`$. (ii) We will require that all the curves $`F=0`$ should pass through the minima which fixes the constant $`c=0`$. (iii) The minima can be made non-degenerate by deforming the potential. A possible deformation is to add $`\lambda \varphi `$ to the superpotential. This leads to non-degenerate minima located at the $`(k+1)`$ roots of $`\lambda `$. By a suitable rescaling, we can set $`\lambda =1`$. We will require that the only allowed values of the constant $`c`$ are such that $`F=0`$ passes through one of the minima.
Thus, we propose that the boundary states for $`L=0,\mathrm{},k`$ correspond to the boundary conditions in the LG model given by the pre-images of the the straight lines in the W-plane:
$$F_L(\varphi ,\overline{\varphi })=W(\varphi )\overline{W}(\overline{\varphi })ic_L=0,$$
where $`c_L=2\mathrm{I}\mathrm{m}W(\varphi _L)`$, where $`\varphi _L`$ are the minima of the bosonic potential. Each $`F_L`$ will have $`(k+2)`$ components which will be asymptotic to the $`k+2`$ lines obtained in the linear class of boundary conditions. This presumably should enable us to associate them with the $`M`$ label of the boundary states. Thus, the boundary states correspond to real algebraic curves in the $`\varphi `$ plane whose image in the $`W`$ plane are straight lines parallel to the Re$`W`$ axis. In the degenerate case, it is easy to see that all $`c_L`$ are coincident. Since we are as yet unable to compute the Witten index in these non-linear situations, the identification cannot be made more precise.
### 5.2 The general case
We will now consider the general case of a LG model with $`n`$ chiral superfields and arbitrary superpotential. We will impose $`n`$ independent conditions
$$F_a(\varphi ,\overline{\varphi })=0,$$
(54)
where $`F_a`$ are real functions. We will use the indices $`i,j,\mathrm{}`$ to denote the superfields and the indices $`a,b,c,\mathrm{}`$ to indicate the boundary conditions. Let $`\mathrm{\Sigma }`$ denote the sub-manifold in $`^n`$ (with complex coordinates $`\varphi _i`$ and $`\overline{\varphi }`$) obtained by imposing these conditions. We will in addition require that the functions be compatible:
$$\{F_a(\varphi ,\overline{\varphi }),F_b(\varphi ,\overline{\varphi })\}_{PB}=0.$$
(55)
We will assume that for all point on $`\mathrm{\Sigma }`$, the normals $`\stackrel{}{n}_a(_iF_a,\overline{}_iF_a)`$ span the normal bundle $`𝒩\mathrm{\Sigma }`$. The vanishing of the Poisson bracket can be rewritten as
$$\stackrel{}{n}_a\stackrel{}{t}_b=0$$
(56)
where $`\stackrel{}{t}_b(_iF_a,\overline{}_iF_a)`$ are tangent vectors to the curve $`F_b=0`$. It follows that they span the tangent bundle $`T\mathrm{\Sigma }`$. Thus, $`\mathrm{\Sigma }`$ is a Lagrangian submanifold of $`^n`$ by construction. The induced metric (first fundamental form) on $`\mathrm{\Sigma }`$ is given by
$$h_{ab}=\stackrel{}{t}_a\stackrel{}{t}_b=\stackrel{}{n}_a\stackrel{}{n}_b.$$
(57)
Let $`h^{ab}`$ denote the inverse of the metric.
The first supersymmetric variation of the boundary conditions leads to
$$\frac{F_a}{\varphi _i}\psi _{+i}+\eta \frac{F_a}{\overline{\varphi }_i}\overline{\psi }_i=0,$$
(58)
where the complex conjugate conditions are implicitly assumed. Defining
$$\chi _{\pm a}\frac{F_a}{\varphi _i}\psi _{\pm i},$$
the above condition takes the simple form
$$\chi _{+a}+\eta \overline{\chi }_a=0.$$
(59)
Further supersymmetric variation of the above condition gives rise to the following terms after imposing $`ϵ_+=\eta \overline{ϵ}_{}`$
$`{\displaystyle \frac{^2F_a}{\varphi _i\varphi _j}}`$ $`\left[\sqrt{2}ϵ_{}\psi _{+j}+\sqrt{2}\eta \overline{ϵ}_{}\psi _j\right]\psi _{+i}`$
$`+{\displaystyle \frac{^2F_a}{\varphi _i\overline{\varphi }_j}}`$ $`\left[\sqrt{2}\overline{ϵ}_{}\overline{\psi }_{+j}+\sqrt{2}\eta ϵ_{}\overline{\psi }_j\right]\psi _{+i}`$
$`+{\displaystyle \frac{F_a}{\varphi _i}}`$ $`\left[i\sqrt{2}(_0+_1)\varphi _i\overline{ϵ}_{}+\sqrt{2}\eta \overline{ϵ}_{}{\displaystyle \frac{\overline{W}}{\overline{\varphi }_i}}\right]`$ (60)
$`+\eta {\displaystyle \frac{^2F_a}{\overline{\varphi }_i\overline{\varphi }_j}}`$ $`\left[\sqrt{2}\overline{ϵ}_{}\overline{\psi }_{+j}+\sqrt{2}\eta ϵ_{}\overline{\psi }_j\right]\overline{\psi }_i`$
$`+\eta {\displaystyle \frac{^2F_a}{\overline{\varphi }_i\varphi _j}}`$ $`\left[\sqrt{2}ϵ_{}\psi _{+j}+\sqrt{2}\eta \overline{ϵ}_{}\psi _j\right]\overline{\psi }_i`$
$`+\eta {\displaystyle \frac{F_a}{\overline{\varphi }_i}}`$ $`\left[i\sqrt{2}(_0_1)\overline{\varphi }_i\eta \overline{ϵ}_{}+\sqrt{2}\overline{ϵ}_{}{\displaystyle \frac{W}{\varphi _i}}\right]=0.`$
In the above, the terms involving $`_0\varphi `$ can be seen to vanish using $`_0F=0`$. The remaining terms can be rearranged in an elegant fashion using the extrinsic curvature tensor $`K_{abc}`$ as defined in the appendix. After eliminating $`\chi _{+a}`$ and $`\overline{\chi }_{+a}`$ respectively in terms of $`\overline{\chi }_a`$ and $`\chi _a`$, we obtain
$`ϵ_{}`$ $`K_{abc}\chi _{}^b\chi _{}^c`$
$`+\overline{ϵ}_{}`$ $`\left(+i({\displaystyle \frac{F_a}{\varphi _i}}_1\varphi _i{\displaystyle \frac{F_a}{\overline{\varphi }_i}}_1\overline{\varphi }_i)+K_{abc}\chi _{}^b\overline{\chi }_{}^c\right)`$ (61)
$`+\eta \overline{ϵ}_{}`$ $`\left({\displaystyle \frac{F_a}{\varphi _i}}{\displaystyle \frac{\overline{W}}{\overline{\varphi }_i}}+{\displaystyle \frac{F_a}{\overline{\varphi }_i}}{\displaystyle \frac{W}{\varphi _i}}\right)=0`$
where $`\chi _{}^a=h^{ab}\chi _b`$. In the above terms, it can be seen that the term multiplying $`ϵ_{}`$ cancels since a symmetric object multiplies an antisymmetric object. The terms multiplying $`\overline{ϵ}_{}`$ lead to two new conditions rather than a single one. There are two ways to understand this: First, the terms involving the superpotential are multiplied with an $`\eta `$ and if we insist on a single condition, the bosonic boundary condition ends up depending on the spin structure. In addition, the vanishing of the boundary terms in the ordinary variation also requires two conditions. The two conditions are
$`\left(\left[{\displaystyle \frac{F_a}{\varphi _i}}_1\varphi _i{\displaystyle \frac{F_a}{\overline{\varphi }_i}}_1\overline{\varphi }_i\right]iK_{abc}\chi _{}^b\overline{\chi }_{}^c\right)=0`$ (62)
$`\{F_a(\varphi ,\overline{\varphi }),W(\mathrm{\Phi })\overline{W}(\overline{\varphi })\}_{PB}=0`$ (63)
We note that the demonstration of the cancellation of the boundary terms of the ordinary and supersymmetric variation of the action is tedious but straightforward. The full set of boundary conditions obtained by the the requirement of unbroken $`N=2`$ supersymmetry of the A-type is equivalent to requiring that the submanifold $`\mathrm{\Sigma }`$ be Lagrangian. For the case without a superpotential, this corresponds to the microscopic(worldsheet) realisation of situations considered by Harvey and Lawson. The new feature that we obtain is that in the presence of a superpotential, there is an additional condition that the real conditions $`F_a`$ have a vanishing Poisson bracket with $`(W\overline{W})`$. This suggests that one must necessarily choose one of the conditions to be $`F=(W\overline{W})ic`$ where $`c`$ is a real constant. This can be seen as a consequence of the fact that in a phase space of real dimension $`2n`$, there can only $`n`$ independent commuting constants of motion.
### 5.3 Spacetime supersymmetry and the special Lagrangian condition
The special Lagrangian condition<sup>5</sup><sup>5</sup>5This was derived for the first time using spacetime supersymmetry by ref. . Of the subsequent literature on this approach, the one closest in spirit to our microscopic viewpoint is ref. . which is necessary for spacetime supersymmetric D-brane configuration appears in our microscopic description as follows. (We will first discuss the case when there is no superpotential. This is the case where the spacetime is $`^n`$). Let $`\mathrm{{\rm Y}}`$ and $`\overline{\mathrm{{\rm Y}}}`$ respectively be the holomorphic $`(n,0)`$ form and anti-holomorphic $`(0,n)`$ form on $`^n`$. In the microscopic description, we can choose
$`\mathrm{{\rm Y}}`$ $``$ $`\psi _1\psi _2\mathrm{}\psi _n`$ (64)
$`\overline{\mathrm{{\rm Y}}}`$ $``$ $`\overline{\psi }_{+1}\overline{\psi }_{+2}\mathrm{}\overline{\psi }_{+n},`$ (65)
This choice is dictated by the fact that under the A-twist, $`\psi _i`$ become $`(1,0)`$ forms and $`\overline{\psi }_{+i}`$ become $`(0,1)`$ forms on $`^n`$. One can also see (by bosonising the fermions, for example) that $`\mathrm{{\rm Y}}`$ generates spectral flow in the left-moving $`N=2`$ under which the NS and R sectors get mapped to each other. $`\overline{\mathrm{{\rm Y}}}`$ has a similar action in the right moving sector except that it is a spectral flow of opposite sign. The requirements for having boundary conditions on the worldsheet which preserve spacetime supersymmetry are:
* The boundary conditions must preserve a global $`N=2`$ worldsheet supersymmetry.
* The boundary conditions must preserve a linear combination of the two spectral flow generators.
In our earlier considerations, we have ensured that the first part has been satisfied. The second condition can be stated as follows:
$$\mathrm{{\rm Y}}=\eta ^n\mathrm{exp}(i\alpha )\overline{\mathrm{{\rm Y}}},$$
(66)
for some constant $`\alpha `$. Under the general A-type boundary conditions discussed above, one can see that (using eqn. (58))
$`\mathrm{\Delta }\mathrm{{\rm Y}}`$ $``$ $`\mathrm{\Delta }\psi _1\psi _2\mathrm{}\psi _n`$
$`=`$ $`()^n\overline{\mathrm{\Delta }}\overline{\psi }_{+1}\overline{\psi }_{+2}\mathrm{}\overline{\psi }_{+n}`$
$`=`$ $`()^n\overline{\mathrm{\Delta }}\overline{\mathrm{{\rm Y}}},`$
where $`\mathrm{\Delta }\mathrm{Det}\frac{F_a}{\varphi _i}`$. The special Lagrangian condition can now be restated as
$$\mathrm{\Delta }=()^n\mathrm{exp}(i\alpha )\overline{\mathrm{\Delta }}.$$
(68)
The $`()^n`$ can always be absorbed into the phase $`\alpha `$ and thus is not crucial. Hence the general boundary conditions which preserve spacetime supersymmetry have to satisfy eqn. (68).
So far we have discussed special Lagrangian condition for the case when the superpotential was zero, i.e., the target space was $`^n`$. In the presence of a superpotential, we have seen that it is necessary to choose one of the conditions, say
$$F_1=W(\varphi )\overline{W}(\overline{\varphi }).$$
Further, let us assume that the superpotential is homogeneous and that $`\varphi _i`$ are homogeneous coordinates on $`^{n1}`$<sup>6</sup><sup>6</sup>6The generalisation to the case of hypersurfaces in weighted projective spaces is obvious. We shall restrict to projective spaces for simplicity.. In order that the boundary conditions carry over to $`^{n1}`$, we will require that the $`F_a`$ be homogeneous under real scalings: $`\varphi _i\lambda \varphi _i`$, for real $`\lambda 0`$. Clearly, this is satisfied by $`F_1=W(\varphi )\overline{W}(\overline{\varphi })`$.
Suppose, we have chosen $`F_a`$ which satisfy the conditions mentioned in the previous paragraph and that the special Lagrangian condition in $`^n`$ given in eqn. (68) is also satisfied. We will now show that this implies that one obtains a special Lagrangian submanifold $`\mathrm{\Sigma }`$ of the Calabi-Yau manifold described by the equation $`W=0`$ in $`^{n1}`$. The global holomorphic $`(n2,0)`$ form on the Calabi-Yau manifold is given by
$$\mathrm{\Omega }=_\gamma \frac{ϵ^{i_1\mathrm{}i_n}\varphi _{i_1}d\varphi _{i_2}\mathrm{}d\varphi _{i_n}}{W(\varphi )},$$
(69)
where $`\gamma `$ is a curve in $`^{n1}`$ enclosing $`W=0`$. On $`\mathrm{\Sigma }`$, the $`F_a`$ satisfy
$$dF_a=\frac{F_a}{\varphi _i}d\varphi _i+\frac{F_a}{\overline{\varphi }_i}d\overline{\varphi }_i=0$$
(70)
Further, the homogeneity condition on the boundary conditions $`F_a`$ can be written as
$$\varphi _i\frac{F_a}{\varphi _i}+\overline{\varphi }_i\frac{F_a}{\overline{\varphi }_i}=d_aF_a,$$
(71)
where $`d_a`$ is the degree of $`F_a`$. Using the above two relations and the fact that we choose $`W(\varphi )=\overline{W}(\overline{\varphi })`$ as one of our boundary conditions, one can see that on $`\mathrm{\Sigma }`$
$$ϵ^{i_1\mathrm{}i_n}\varphi _{i_1}d\varphi _{i_2}\mathrm{}d\varphi _{i_n}|_\mathrm{\Sigma }=\frac{\overline{\mathrm{\Delta }}}{\mathrm{\Delta }}ϵ^{j_1\mathrm{}j_n}\overline{\varphi }_{j_1}d\overline{\varphi }_{j_2}\mathrm{}d\overline{\varphi }_{j_n}|_\mathrm{\Sigma }$$
(72)
This implies that
$`\mathrm{\Omega }|_\mathrm{\Sigma }`$ $`=`$ $`()^n{\displaystyle \frac{\overline{\mathrm{\Delta }}}{\mathrm{\Delta }}}\overline{\mathrm{\Omega }}|_\mathrm{\Sigma }`$ (73)
$`=`$ $`\mathrm{exp}(i\alpha )\overline{\mathrm{\Omega }}|_\mathrm{\Sigma }`$
which is the special Lagrangian condition on the Calabi-Yau manifold. Note that this has not been derived using any spacetime inputs but rather from the worldsheet analysis of the LG model with boundary.
An important question to consider is whether the homogeneity condition, eqn. (71), which certainly appears natural, is too restrictive. One possibility is to allow for the condition
$$\left(\varphi _i\frac{F_a}{\varphi _i}+\mathrm{exp}(i\theta )\overline{\varphi }_i\frac{F_a}{\overline{\varphi }_i}\right)|_\mathrm{\Sigma }=0\mathrm{mod}W,$$
(74)
where $`\theta `$ is a constant. The mod $`W`$ degree of freedom reflects the fact that the integral in eqn. (69) has support only at the zeros of $`W`$. We will use this weaker condition shortly in an example. With this weaker condition, one obtains
$$\mathrm{\Omega }|_\mathrm{\Sigma }=\mathrm{exp}(i\alpha +i\theta )\overline{\mathrm{\Omega }}|_\mathrm{\Sigma }.$$
(75)
### 5.4 Examples
We first illustrate the case without a superpotential by using the classic example of Harvey and Lawson. The construction provides special Lagrangian submanifolds with topology $`^+\times T^{n1}`$ on $`^n`$. We will also show that this leads naturally to a $`T^{n1}`$ Lagrangian fibration of $`^{n1}`$.
The conditions of Harvey and Lawson can be implemented as boundary conditions in our worldsheet theory:
$`F_1`$ $`=`$ $`\{\begin{array}{c}\mathrm{Re}(\varphi _1\mathrm{}\varphi _n)c_1\mathrm{for}\mathrm{even}n\\ \mathrm{Im}(\varphi _1\mathrm{}\varphi _n)c_1\mathrm{for}\mathrm{odd}n\end{array}`$ (76)
$`F_a`$ $`=`$ $`|\varphi _1|^2|\varphi _a|^2c_a\mathrm{for}2an`$ (77)
where $`c_a`$ are some real constants. Linear combinations of the $`c_a`$ correspond to the radii of the circles of $`T^{n1}`$. When $`c_1=0`$, the $`^+`$ corresponds to the value of $`|\varphi _1|^2`$ and the $`T^{n1}`$ corresponds to the $`(n1)`$ phases left unfixed by the condition $`F_1=0`$.
In order to extend these boundary conditions to $`^{n1}`$, the condition of homogeneity on the $`F_a`$ implies that all the constants $`c_a`$ must be necessarily set to zero. In this limit, the Lagrangian submanifold appears to be singular at the origin. This however is not a point in $`^{n1}`$. Thus, the submanifold is non-singular. Further, the (real) scaling degree of freedom in the homogeneous coordinates $`\varphi _i`$ of $`^{n1}`$ eats up the $`^+`$ degree of freedom leaving us with a $`T^{n1}`$. In the inhomogeneous coordinates of $`^{n1}`$, where we set $`\varphi _1=1`$, the radii of all circles is set to unity. Thus the Lagrangian submanifold, $`T^{n1}`$ in $`^{n1}`$ is obtained at a specific point in its moduli space. We will momentarily see how to generalise this.
As pointed out by Strominger, Yau and Zaslow, the existence of a mirror partner for a Calabi-Yau threefold implies that the Calabi-Yau manifold admits a $`T^3`$ fibration. Consider the mirror quintic given by the equation in $`^4`$:
$$W=\varphi _1^5+\mathrm{}+\varphi _5^55\psi \varphi _1\varphi _2\varphi _3\varphi _4\varphi _5=0.$$
(78)
The large complex structure limit corresponds to $`|\psi |\mathrm{}`$. In the infinite limit, the quintic breaks up into five $`^3`$ given by setting $`\varphi _i=0`$ for $`i=1,\mathrm{},5`$. While this is a degenerate limit, the $`T^3`$ special Lagrangian fibre is seen easily by using the earlier construction for $`^3`$. It has been argued by Strominger et. al., that this $`T^3`$ will be special Lagrangian in the neighbourhood of the infinite $`|\psi |`$ limit(see ref. for a discussion).
Since it in general rather hard to construct special Lagrangian submanifolds, it is of interest to see how the above example fits into our construction. In the infinite complex structure limit, it is interesting to note that $`F^1`$ chosen by Harvey and Lawson is indeed equal to $`(W\overline{W})`$! It immediately follows therefore that if we also choose the conditions
$$F_a=|\varphi _1|^2|\varphi _a|^2c_a\mathrm{for}2a5$$
(79)
then we obtain a supersymmetric cycle. We have introduced four constants $`c_a`$ which break the homogeneity condition. However, one can check that the weaker condition mentioned earlier holds:
$$\left(\varphi _i\frac{F_a}{\varphi _i}\overline{\varphi }_i\frac{F_a}{\overline{\varphi }_i}\right)|_\mathrm{\Sigma }=0\mathrm{mod}W,$$
(80)
A calculation shows that except for $`F_1=(W\overline{W})`$, the above condition holds identically (without the mod $`W`$ condition). It is of interest to count the number of independent parameters. First, it may seem that we have four angles and thus a $`T^4`$. However, since $`W=0`$ necessarily requires one of the $`\varphi _i`$ to identically vanish, the angle associated with the vanishing $`\varphi _i`$ does not exist. Further, projectivisation leaves only three independent real variables coming from the four $`c_a`$. These presumably correspond to the moduli associated with the $`T^3`$. The example discussed above is not something specific to the quintic but can be extended to a larger class of CY threefolds. See for instance, ref. .
## 6 Conclusions and Outlook
It is clear that the methods that we have outlined in this paper have obvious generalizations. First, the methods can be extended to LG models that are associated to CY hypersurfaces in weighted projective spaces. Second, since we are working in the LG model we can extend our techniques to hypersurfaces that are described by more general potentials than those of the Fermat type. Thirdly, the techniques could equally well be applied in the case of non-quasi-homogeneous potentials relevant to massive $`N=2`$ supersymmetric theories. There would however be some difference in the geometric interpretation of the boundary conditions in these cases.
The special Lagrangian submanifolds considered here, described by a set of real equations $`F_a=0`$ in some ambient $`^n`$ can be thought of as a real algebraic variety. This is in line with the theory in the bulk corresponding to strings propagating on a complex algebraic variety. It would be interesting to see how other structures that were seen in the bulk, like the operator ring for instance, carry over to the boundary theory. It is not clear however what the potential role, if any at all, of anholonomic boundary conditions, which of course are allowed in general. While inequalities together with conditions of the form $`F_a=0`$ are in general relevant to what are known to mathematicians as semi-algebraic real sets. It would be interesting to see if they have a role in the context of special Lagrangian sub-manifolds in CY.
One of the themes of this paper was to obtain boundary conditions corresponding to all the boundary states in a single minimal model. While the non-linear boundary conditions we have constructed have suggested a possible way out, the story is far from complete. It is of interest to be able to compute the Witten index for the non-linear case in order to be sure of the identification proposed in this paper. Solving this problem is of interest in making a clear geometric identification of the Recknagel-Schomerus class of boundary states. In particular, we do not yet have a clear geometric identification of all the $`Lk/2`$ states in the boundary CFT. We may add that even in the linear class of boundary conditions we have not yet explored the cases where the matrix A is non-diagonal. This should help us in examining boundary states in the Gepner construction that do not belong to the Recknagel-Schomerus class of states. Such states would naturally arise from the possible fixed-point resolutions of the modular transformation matrix of the full Gepner CFT.
It is of interest to extend our analysis to the case of B-type boundary conditions. However, the LG description of B-branes will be rather different from the large volume CY description since the geometry and charge of the B-branes are not expected to remain invariant. Nevertheless, it may be possible to track some B-branes from the large volume CY limit to the LG phase without encountering lines of marginal stability. Assuming that this is possible, then one might be able to calculate the worldvolume superpotential directly in the LG model.
As we discussed in section 2, it is not possible to impose Neumann boundary conditions on all fields in the LG model. It is also not possible to impose Neumann boundary conditions on the LG field of a single minimal model. This strongly suggests that all the states of the Recknagel-Schomerus class must arise from Dirichlet-type boundary conditions in the LG and suitable modifications thereof. This shows that for example, a D6-brane wrapping a CY will look rather different in the LG limit. From the work of Brunner et. al. we know that the corresponding state exists at the Gepner point in the moduli space. It would be of interest to describe this state in the LG formalism. Given the identification of the linear LG boundary conditions of A-type with the $`L=1`$ boundary states of the A-type boundary CFT, the case of B-type Dirichlet boundary conditions on all LG fields described in section 2, most likely are the $`\{L_i=1,\mathrm{for}\mathrm{all}i\}`$ B-type states.
The B-type states may also be studied by using mirror symmetry on A-type states in the LG model. In the LG models, the orbifolding technique provides a simple method of constructing the mirror CY and should also provide a corresponding method for the construction of A-type boundary states in the mirror.
Given the close interplay of both complex and Kähler moduli in the description of D-branes on CY threefolds (a natural consequence of spacetime $`N=2`$ supersymmetry being broken to $`N=1`$ by the D-branes), the linear sigma model(LSM) description is better suited in some ways for a microscopic description on CY threefolds. For example, one can show that the D6-brane (all Neumann boundary conditions) in the CY limit starts looking like an all Dirichlet boundary condition as one goes to “small volumes” and to the LG phase. In the neighbourhood of vanishing Kähler parameter (for the quintic), the CY as seen by the D-brane appears to be in a non-commutative phase. These issues will be discussed in a forthcoming paper. Related remarks appear in the work of Hori and Vafa. The transitions discussed by Joyce also seem well suited for an LSM description.
Acknowledgments We would like to thank the following for useful discussions: R. Balakrishnan, V. Balakrishnan, G. Date, S. Lakshmi Bala, D. S. Nagaraj, K. Paranjape and especially T. Sarkar.
## Appendix A Extrinsic curvature of Lagrangian submanifolds
We will be considering a Lagrangian submanifold $`\mathrm{\Sigma }`$ of $`^n`$ which is implicitly specified by $`n`$ independent real functions
$$F_a(\varphi ,\overline{\varphi })=0,$$
where $`\varphi _i`$ are complex coordinates on $`^n`$. The Lagrangian condition implies that the Poisson bracket of the $`n`$ functions vanish. Further, the normals $`n_a^i=(_iF_a,\overline{}_iF_a)`$ span the normal bundle $`𝒩\mathrm{\Sigma }`$ and the tangents $`t_a^i=(_iF_a,\overline{}_iF_a)`$ span the tangent bundle $`T\mathrm{\Sigma }`$ and $`T^n=𝒩\mathrm{\Sigma }T\mathrm{\Sigma }`$. The vanishing Poisson bracket ensures that $`\stackrel{}{n}_a\stackrel{}{t}_b=0`$.
The tangential derivatives $`D_at_a^i_i`$ satisfy $`[D_a,D_b]=0`$ by virtue of the vanishing of the Poisson bracket $`\{F_a,F_b\}_{PB}=0`$. Thus, locally on $`\mathrm{\Sigma }`$, there exists a coordinate system $`\sigma _a`$ such that $`/\sigma _a=D_a`$. The induced metric(first fundamental form) in this coordinate system is given by
$$h_{ab}=\stackrel{}{t}_a\stackrel{}{t}_b=\stackrel{}{n}_a\stackrel{}{n}_b.$$
(81)
The extrinsic curvature tensor (second fundamental form) $`\stackrel{}{K}_{ab}`$ is defined as follows<sup>7</sup><sup>7</sup>7We follow the lectures of F. David in defining the extrinsic curvature tensor after providing the required generalisations.
$$(t_a^i_i)t_b^j=(t_b^i_i)t_a^j=K_{ab}^j+\mathrm{\Gamma }_{ab}^ct_c^j,$$
(82)
where $`\mathrm{\Gamma }_{ab}^c`$ is the Christoffel connection with respect to the induced metric on $`\mathrm{\Sigma }`$. Thus, $`\stackrel{}{K}_{ab}`$ is normal to the surface $`\mathrm{\Sigma }`$ (since the second term projects out the tangential component of $`(t_a^i_i)t_b^j`$). Since, $`\stackrel{}{n}_a`$ span the $`𝒩\mathrm{\Sigma }`$, we can decompose $`\stackrel{}{K}_{ab}`$ into
$$K_{abc}\stackrel{}{K}_{ab}\stackrel{}{n}_c.$$
(83)
One can explicitly calculate $`K_{abc}`$ as defined above and we obtain
$$K_{abc}=\left[\frac{F_c}{\varphi _i}\frac{F_b}{\varphi _j}\frac{^2F_a}{\varphi _i\varphi _j}\frac{F_c}{\overline{\varphi }_i}\frac{F_b}{\varphi _j}\frac{^2F_a}{\varphi _i\overline{\varphi }_j}\frac{F_c}{\varphi _j}\frac{F_b}{\overline{\varphi }_i}\frac{^2F_a}{\varphi _i\overline{\varphi }_j}+\frac{F_c}{\overline{\varphi }_i}\frac{F_b}{\overline{\varphi }_j}\frac{^2F_a}{\overline{\varphi }_i\overline{\varphi }_j}\right]$$
(84)
One can verify, that $`K_{abc}`$ is a completely symmetric tensor. The symmetry under the exchange $`ab`$ is the usual symmetry property of the extrinsic curvature tensor. However, for Lagrangian submanifolds, one has the isomorphism between the normal bundle and the tangent bundle which enables one to make it fully symmetric. Further, if $`\mathrm{\Sigma }`$ is special Lagrangian, then the trace of the extrinsic curvature tensor with respect to the induced metric vanishes. |
warning/0003/nlin0003027.html | ar5iv | text | # The Scalings of Scalar Structure Functions in a Velocity Field with Coherent Vortical Structures
## I Introduction
The theory of turbulent passive scalars has received much attention recently. The mixing of a scalar field $`\theta `$ in a velocity field $`𝐯`$ is governed by the advection-diffusion equation
$$_t\theta (𝐱,t)+𝐯(𝐱,t)\theta (𝐱,t)=\kappa ^2\theta (𝐱,t)+f(𝐱,t)$$
(1)
where $`\kappa `$ denotes the molecular diffusivity of the scalar $`\theta `$ and $`f(𝐱,t)`$ is an external source ( forcing) driving the scalar. The mixing of a scalar field $`\theta `$ is characterised by its structure functions $`S_p[\theta (𝐱+𝐫)\theta (𝐱)]^p=(\delta \theta (𝐫))^p`$ for any number $`p`$. If we want to find $`S_p`$ then we need models of the velocity field $`𝐯`$.
The model which has attracted much attention recently is the Kraichnan model where the velocity field $`𝐯`$ is considered to be incompressible, statistically isotropic, white-noise in time($`\delta `$-correlated) and Gaussian. Furthermore it has homogeneous increments with power law spatial correlations
$`[v_i(𝐫,t)v_i(0,0)][v_j(𝐫,t)v_j(0,0)]=`$ (2)
$`2\delta (t)r^h\left[(h+d1)\delta _{ij}h{\displaystyle \frac{r_ir_j}{r^2}}\right]`$ (3)
(4)
where the scaling exponent $`h]0,2[`$ and $`d`$ is the dimension of space so that $`i,j=1,2,\mathrm{},d`$. The above tensorial structure of the velocity field is in conformity with incompressibility. The Kraichnan model also assumes a forcing $`f(𝐱,t)`$ in (1) that is an independent Gaussian random field with zero mean. The forcing is white in time and its covariance is assumed to be a real, smooth, positive-definite function with rapid decay in space so that the forcing is homogeneous, isotropic and takes place on the integral scale $`L`$. The generic scaling behaviour of the structure function $`S_pr^{\zeta _p}`$, $`rL`$, of passive scalars in the Kraichnan model was established in. The scaling exponents of this formalism are of the form $`\zeta _p=\zeta _p(d,h,p)`$ where $`h[0,2]`$ is the H$`\ddot{o}`$lder exponent in (4). In the context of this model Balkovsky and Lebedev and Chertkov used the instantonic formalism in a $`d`$\- dimensional space to find the scaling exponents for large $`p`$. It was also shown in the instantonic formalism that $`lim_p\mathrm{}\zeta _p\frac{d(2h)^2}{8h}`$ which is independent of $`p`$. The scaling exponents were also calculated using other techniques in the limits $`h0`$, $`d\mathrm{}`$ and $`p\mathrm{}`$, and a $`2h`$ expansion of $`\zeta _p`$ was proposed in . It was found that $`\zeta _p`$ does depend on $`p`$ for small values of $`p`$ in the Kraichnan model.
Our work lies in the opposite extreme of the Kraichnan model. We work in the regime where we have a persistent vortical velocity field frozen in time in two dimensions. The important differences between this model and the Kraichnan model are in the structure of the velocity field infinitely correlated in time in this model but delta correlated in time in the Kraichnan model; and vortical in space in this model, but Gaussian in the Kraichnan model. In this model, the velocity field is that of spatially distributed noninteracting two dimensional vortices with compact structure. We consider the spatial distribution of vortices to be dilute in that they are far from each other and therefore maintain their structure and spatial position for an indefinite period. We also consider this distribution to be homogeneous and isotropic and the velocity field to be incompressible, that is $`𝐯=0`$. The model of the velocity field considered here is an artificial model of planar homogeneous turbulence where the emphasis is on the coherent vortex aspect of the flow. In order of presentation, the first aim of this model is to demonstrate that in the case of the unforced scalar ($`f=0`$ in 1) we can quantify the statistics of the turbulent scalar field in terms of the scalar’s spiral geometry generated by the coherent vortical structures in the flow(sections III and IV). The second aim is to derive the Batchelor $`k^1`$ power spectrum and all the corresponding structure functions for the scalar field in the case where the scalar is forced and in a statistically steady state (section V). Such a spectrum has been recently observed by Jullien et. al., in a 2-D turbulent flow with well defined, albeit short lived, coherent vortical structures.
In the next section we discuss the scenario of a decaying scalar field in an isolated vortex. In section III, we define structure functions and calculate the spectrum of higher order correlation functions for a single spiral created by a single vortex and decaying by the action of molecular diffusion. In section IV we generalise our analysis to many non-interacting vortices and calculate the scalings of the structure functions of the decaying scalar field. In section V we calculate the generalised power spectra and the corresponding structure functions of statistically stationary scalar spirals by applying the time-average operation approach of Lundgren. In section VI we discuss the phenomenology behind the $`k^1`$ scalings of the generalised power spectra. Section VII contains conclusions, discussion and the obtainment of the $`(k\mathrm{ln}k)^1`$ scalings of the generalised power spectra in smooth chaotic flows.
## II Passive scalar in a planar vortex
The advection of a decaying passive scalar field by a single planar vortex has been studied by Flohr and Vassilicos. We use the formulation used in namely:
$$_t\theta +\mathrm{\Omega }(r)_\varphi \theta =\kappa ^2\theta $$
(5)
where $`_t={\displaystyle \frac{}{t}}`$ and $`_\varphi ={\displaystyle \frac{}{\varphi }}`$ and $`\mathrm{\Omega }(r)=\mathrm{\Omega }_0\left(\frac{r}{L}\right)^s`$ and $`L`$ is the maximum distance of the scalar interface from the centre of the vortex. This equation describes the advection and diffusion of a scalar field $`\theta `$ in the azimuthal plane $`𝐫=(r,\varphi )`$ by a steady vortex with azimuthal velocity component $`u_\varphi (r)=r\mathrm{\Omega }(r)=L\mathrm{\Omega }_0\left(\frac{r}{L}\right)^{1s}`$ and vanishing radial and axial velocity components. Direct Numerical Simulations and experiments in the laboratory have demonstrated the existence and importance of coherent vortices in two-dimensional turbulence and in two-component turbulence in stably-stratified flow with and without rotation of the reference frame . Note that axial velocity fields of the form $`u_\varphi (r)=L\mathrm{\Omega }_0\left(\frac{r}{L}\right)^{1s}`$ have been used in and that their large wavenumber energy spectrum has the form $`E(k)k^{5+2s}`$ for $`1/2<s<2`$ with the appropriate large scale bound. We choose $`s1`$ to ensure that $`u_\varphi (r)`$ does not increase with increasing $`r`$ and $`s<2`$ to ensure that the energy spectrum is steeper than $`k^1`$. The initial scalar field $`\theta _0=\theta (𝐱,t=0)`$ is characterised by a regular interface between $`\theta _00`$ and $`\theta _0=0`$ with minimal distance $`r_0`$ and maximal $`L`$ from the rotation axis. By regular structure we mean that the interface has no irregularities on scales smaller than $`L`$. Nothing else needs to be specified about the initial scalar field $`\theta _0(𝐱)`$. Such a patchy initial condition where all the non-zero scalar is confined within a regular interface mimics well initial conditions in certain laboratory experiments where scalar is injected in the flow in the form of blobs(e.g. ).
As time proceeds, the patch winds around the vortex and builds up a spiral structure and decays due to diffusion. The characteristic time $`\mathrm{\Omega }_0^1`$ is the inverse angular velocity of the vortex at $`L`$. This defines a Péclet number $`Pe=\mathrm{\Omega }_0L^2\kappa ^1`$. We non-dimensionalise equation (5) by using the following transformations
$`L^1rr,\mathrm{\Omega }_0tt,\mathrm{\Omega }_0^1\mathrm{\Omega }(r)\mathrm{\Omega }(r),L^2^2^2,`$
and equation (5) takes the form
$$_t\theta +\mathrm{\Omega }(r)_\varphi \theta =\frac{1}{Pe}^2\theta .$$
(6)
In the non dimensionalised variables we have $`\mathrm{\Omega }(r)=r^s`$, and $`r_0`$ represents $`r_0/L`$ since $`L=1`$. Considering finite diffusivity $`\kappa `$, the form of the general solution of the evolution equation(6) for any initial field $`\theta _0`$ in the limit of large $`t`$, ie. $`t1`$, is the following
$`\theta (𝐫,t)`$ $`=`$ $`{\displaystyle \underset{n}{}}f_n(r,t)e^{in(\varphi \mathrm{\Omega }(r)t)}`$ (7)
$`f_n(r,t)`$ $`=`$ $`f_n(r,0)e^{[\frac{1}{3}n^2\mathrm{\Omega }^^2Pe^1t^3]}`$ (8)
where $`r=𝐫`$ and $`\varphi `$ is the azimuthal angle and $`n`$ is an integer. $`\mathrm{\Omega }^{}`$ is the derivative of $`\mathrm{\Omega }`$ with respect to $`r`$. The angular Fourier coefficients $`f_n(r,t)`$ are time dependent and the initial condition is fully specified by $`f_n(r,0)`$. We do not go in to the details of the solution of (5) which can be found in .
## III Structure Functions of Passive Decaying Scalar in one Vortex
In this work we concentrate in finding the scaling properties of the structure functions of the scalar field in a planar turbulence consisting of coherent vortices. The two point equal time $`p`$-th order structure function is defined as follows
$`𝒮_p(r,t)`$ $`=`$ $`\overline{[\theta (𝐱+𝐫,t)\theta (𝐱,t)]^p}`$ (10)
$`=`$ $`\overline{[\delta \theta (𝐫,t)]^p}.`$ (11)
$`S_p`$ depends only on the magnitude of the distance between two points, when the ensemble average is taken over an isotropic and homogeneous distribution of the scalar field. The overbar denotes ensemble averaging and the brackets imply space averaging ($`\mathrm{}d^2𝐱`$).
Let us first calculate $`[\delta \theta (𝐫,t)]^p`$ for one 2-D vortex. We use the binomial expansion as follows
$`[\theta (𝐱+𝐫,t)\theta (𝐱,t)]^p`$ (13)
$`=\theta ^p(𝐱+𝐫,t)+(1)^p\theta ^p(𝐱,t)`$ (14)
$`+{\displaystyle \underset{q=1}{\overset{p1}{}}}C_{qp}(1)^q\theta ^{pq}(𝐱+𝐫,t)\theta ^q(𝐱,t)`$ (15)
(16)
where $`C_{qp}`$ is the binomial coefficient of the expansion. In order to calculate $`[\delta \theta (𝐫,t)]^p`$ we first determine
$$\theta ^q(𝐱,t)\theta ^{pq}(𝐱+𝐫,t)=B_{pq}(𝐫,t)$$
(17)
which is the $`q`$-th term in the binomial expansion of $`(\delta \theta (𝐫))^p`$ as shown in (16). Now we write the above as
$$\frac{1}{L_A^2}d^2𝐱\theta ^q(𝐱,t)\theta ^{pq}(𝐱+𝐫,t)=B_{pq}(𝐫,t)$$
(18)
where $`L_A`$ is a large scale over which the spatial average may be calculated. The Fourier transform of equation (18) is given by
$$\widehat{F}_{pq}(𝐤,t)=\frac{1}{2\pi }e^{i𝐤𝐫}B_{pq}(𝐫,t)d^2𝐫.$$
(19)
Substituting (18) in (19) and after some standard manipulations we get
$`\widehat{F}_{pq}(𝐤,t)=`$ $`{\displaystyle \frac{1}{2\pi L_A^2}}{\displaystyle \theta ^q(𝐱,t)e^{i𝐤𝐱}d^2𝐱}`$ (21)
$`\times {\displaystyle }\theta ^{pq}(𝐱^{},t)e^{i𝐤𝐱^{}}d^2𝐱^{}.`$
Now if we integrate (LABEL:Ft2) over a circular shell in $`k`$ space we get
$$F_{pq}(k,t)=_0^{2\pi }𝑑A(k)\widehat{F}_{pq}(𝐤,t)$$
(23)
where $`dA(k)kd\varphi _k`$, $`k=|𝐤|`$ and $`\varphi _k`$ is the angle of $`𝐤`$. $`F_{pq}(k,t)`$ could be called the generalised power spectrum of the scalar field in Fourier space. Substituting (LABEL:soln1) and (LABEL:Ft2) in (23) we get the following
$`F_{pq}(k,t)`$ (24)
$`={\displaystyle \frac{1}{(2\pi L_A^2)}}{\displaystyle 𝑑A(k)}`$ (25)
$`\times {\displaystyle }d^2𝐱e^{i𝐤𝐱}\left\{{\displaystyle \underset{n}{}}f_n(x,t)e^{[in(\varphi \mathrm{\Omega }(x)t)]}\right\}^q`$ (26)
$`\times {\displaystyle }d^2𝐱^{}e^{i𝐤𝐱^{}}\left\{{\displaystyle \underset{m}{}}f_m(x^{},t)e^{[im(\varphi ^{}\mathrm{\Omega }(x^{})t)]}\right\}^{pq}`$ (27)
(28)
where $`x=|𝐱|`$ and $`x^{}=|𝐱^{}|`$. After some standard manipulations (28) leads to
$`F_{pq}(k,t)`$ (34)
$`={\displaystyle \frac{1}{(2\pi L_A^2)}}{\displaystyle k𝑑\varphi _k𝑑xxJ_n(kx)2\pi (i)^ne^{in\varphi _k}}`$
$`\times {\displaystyle \underset{n,n_1,n_2\mathrm{}n_{q1}}{}}f_{n_1}f_{n_2}\mathrm{}f_{nn_1\mathrm{}n_{q1}}e^{in\mathrm{\Omega }(x)t}`$
$`\times {\displaystyle }dx^{}x^{}e^{im\varphi _k}J_m^{}(kx^{})2\pi (i)^m`$
$`\times {\displaystyle \underset{m,m_1,m_2\mathrm{}m_{pq1}}{}}f_{m_1}f_{m_2}\mathrm{}f_{mm_1\mathrm{}m_{pq1}}`$
$`\times e^{im\mathrm{\Omega }(x^{})t}`$
(35)
where we have changed summation variables in accordance with the following conditions
$`n=n_1+n_2+n_3\mathrm{}n_q`$ (36)
$`\mathrm{and}`$ $`m=m_1+m_2+m_3+\mathrm{}m_{pq}.`$ (37)
All the $`f_n`$’s and $`f_m`$’s are functions of time $`t`$ and of $`x`$ and $`x^{}`$ respectively. $`J_n(kx)`$ is the Bessel function which has been substituted in place of the integral representation
$$_0^{2\pi }e^{in\varphi }e^{ikxcos(\varphi \varphi _k)}𝑑\varphi =2\pi (i)^ne^{in\varphi _k}J_n(kx).$$
(39)
After integrating (35) with respect to $`\varphi _k`$ and summing over $`m`$ we get the following relation
$`F_{pq}(k,t)`$ (45)
$`={\displaystyle \frac{1}{(2\pi L_A^2)}}{\displaystyle 𝑑xkxJ_n(kx)2\pi (i)^n}`$
$`\times {\displaystyle \underset{n,n_1,n_2\mathrm{}n_{q1}}{}}f_{n_1}f_{n_2}\mathrm{}f_{nn_1\mathrm{}n_{q1}}e^{in\mathrm{\Omega }(x)t}`$
$`\times {\displaystyle }dx^{}x^{}J_n^{}(kx^{})2\pi (i)^n`$
$`\times {\displaystyle \underset{m_1,m_2\mathrm{}m_{pq1}}{}}f_{m_1}f_{m_2}\mathrm{}f_{nm_1\mathrm{}m_{pq1}}`$
$`\times e^{in\mathrm{\Omega }(x^{})t}.`$
(46)
We now have two cases to study (see figure 1). In case I the vortex centre is outside the scalar patch and its nearest distance to the scalar interface is $`r_0`$. In Case II the centre of the vortex is inside the scalar patch and its nearest distance to the scalar interface is $`r_0`$. In view of the above we can write (46) as
$`F_{pq}(k,t)`$ (47)
$`={\displaystyle \frac{1}{(2\pi L_A^2)}}\left[{\displaystyle _0^{r_0}}+{\displaystyle _{r_0}^{\mathrm{}}}\right]dxkxJ_n(kx)2\pi (i)^n`$ (48)
$`\times {\displaystyle \underset{n,n_1,n_2\mathrm{}n_{q1}}{}}f_{n_1}f_{n_2}\mathrm{}f_{nn_1\mathrm{}n_{q1}}e^{in\mathrm{\Omega }(x)t}`$ (49)
$`\times \left[{\displaystyle _0^{r_0}}+{\displaystyle _{r_0}^{\mathrm{}}}\right]dx^{}x^{}J_n^{}(kx^{})2\pi (i)^n`$ (50)
$`\times {\displaystyle \underset{m_1,m_2\mathrm{}m_{pq1}}{}}f_{m_1}f_{m_2}\mathrm{}f_{nm_1\mathrm{}m_{pq1}}`$ (51)
$`\times e^{in\mathrm{\Omega }(x^{})t}.`$ (52)
(53)
In (53) we have four terms of the form as shown below
$`{\displaystyle _0^{r_0}}𝑑x{\displaystyle _0^{r_0}}𝑑x^{}(\mathrm{})+{\displaystyle _0^{r_0}}𝑑x{\displaystyle _{r_0}^{\mathrm{}}}𝑑x^{}(\mathrm{})`$ (54)
$`+{\displaystyle _{r_0}^{\mathrm{}}}𝑑x{\displaystyle _0^{r_0}}𝑑x^{}(\mathrm{})+{\displaystyle _{r_0}^{\mathrm{}}}𝑑x{\displaystyle _{r_0}^{\mathrm{}}}𝑑x^{}(\mathrm{})`$ (55)
(56)
with the integrands denoted by $`(\mathrm{})`$ being the same as (53) for all the terms of the above. For case I, terms containing $`_0^{r_0}(\mathrm{})`$ are zero since $`f_n=0`$ in the regime $`0<x<r_0`$ for all n, even for $`n=0`$, since there is no scalar patch in the region $`0<x<r_0`$. Therefore contributions only come from the term
$`{\displaystyle _{r_0}^{\mathrm{}}}𝑑xx{\displaystyle _{r_0}^{\mathrm{}}}𝑑x^{}x^{}(\mathrm{}).`$
For case II we can legitimately replace $`\theta `$ by $`\theta f_0`$ and get the same result as in case I. Hence the only contributing term is the following
$`F_{pq}(k,t)`$ (57)
$`={\displaystyle \frac{1}{(2\pi L_A^2)}}{\displaystyle _{r_0}^{\mathrm{}}}𝑑xkxJ_n(kx)2\pi (i)^n`$ (58)
$`\times {\displaystyle \underset{n,n_1,n_2\mathrm{}n_{q1}}{}}f_{n_1}f_{n_2}\mathrm{}f_{nn_1\mathrm{}n_{q1}}e^{in\mathrm{\Omega }(x)t}`$ (59)
$`\times {\displaystyle _{r_0}^{\mathrm{}}}dx^{}x^{}J_n^{}(kx^{})2\pi (i)^n`$ (60)
$`\times {\displaystyle \underset{m_1,m_2\mathrm{}m_{pq1}}{}}f_{m_1}f_{m_2}\mathrm{}f_{nm_1\mathrm{}m_{pq1}}`$ (61)
$`\times e^{in\mathrm{\Omega }(x^{})t}.`$ (62)
It is because $`f_0`$ decays with a time scale which is much larger than the decay time of the non-zero harmonics that $`f_0`$ is considered to be a constant and therfore subtracted away from the $`\theta `$ in case II. The same reasoning can be applied for case I. Hence all the $`n_i`$’s and $`m_i`$’s are non-zero in the above equation and in the rest of the paper .
To take into account the fact that diffusion gradually wipes out the spiral structure of the scalar field near the vortex centre (a fact not taken in to account in where the spiral structure is assumed to exist wholly intact until finally destroyed by viscosity), we follow Flohr and Vassilicos and define a critical radius $`\rho `$ which gives a measure of this diffused region. This critical radius is defined in the limit $`Pe\mathrm{}`$ for times $`tPe^{1/3}`$ which are such that
$`f_n(r,t)=f_n(r,0)forr\rho `$
but
$`|f_n(r,t)||f_n(r,0)|forr\rho see(\text{LABEL:soln1}).`$
Hence, we set $`\frac{1}{3}n^2\mathrm{\Omega }^2(\rho )Pe^1t^3=1`$ which implies
$`\rho (t)=\left[{\displaystyle \frac{1}{3}}n^2s^2Pe^1t^3\right]^{\frac{1}{2(s+1)}}.`$
This critical radius is time-dependent and grows with time. It can be thought of as a diffusive length scale over which the harmonics in $`n`$ have diffused and the spiral structure has been smeared out.
In view of the above, the integrals in (62) can be further divided as
$`{\displaystyle _{r_0}^{\mathrm{}}}𝑑x={\displaystyle _{r_0}^\rho }𝑑x+{\displaystyle _\rho ^{\mathrm{}}}𝑑x.`$
The only significantly non-zero contribution comes from the range $`\rho <x<1`$ in the integrals and similarly for $`x^{}`$. Hence we get the following
$`F_{pq}(k,t)`$ (63)
$`={\displaystyle \frac{1}{(2\pi L_A^2)}}{\displaystyle _\rho ^{\mathrm{}}}𝑑xkxJ_n(kx)2\pi (i)^n`$ (64)
$`\times {\displaystyle \underset{n,n_1,n_2\mathrm{}n_{q1}}{}}f_{n_1}f_{n_2}\mathrm{}f_{nn_1\mathrm{}n_{q1}}e^{in\mathrm{\Omega }(x)t}`$ (65)
$`\times {\displaystyle _\rho ^{\mathrm{}}}dx^{}x^{}J_n^{}(kx^{})2\pi (i)^n`$ (66)
$`\times {\displaystyle \underset{m_1,m_2\mathrm{}m_{pq1}}{}}f_{m_1}f_{m_2}\mathrm{}f_{nm_1\mathrm{}m_{pq1}}`$ (67)
$`\times e^{in\mathrm{\Omega }(x^{})t}`$ (68)
Now for large $`kx`$, ie. $`kx1`$, we can use the asymptotic expansion for the Bessel function
$$J_n(kx)\left(\frac{1}{2\pi kx}\right)^{\frac{1}{2}}\left[(i)^{n+\frac{1}{2}}e^{ikx}+(i)^{n+\frac{1}{2}}e^{ikx}\right].$$
(69)
This is appropriate for our analysis if the Fourier modes are to resolve at the very least the distance $`r_0`$ from the centre of the vortex to the scalar patch interface i.e $`1<kr_0`$. After substituting (69) in (68) we use the method of stationary phase to evaluate the integrals where the phase is given by
$$\mathrm{\Phi }=kxn\mathrm{\Omega }(x)t.$$
(70)
The approximation for a general integral of this type is known to be
$`I(x)`$ $`=`$ $`{\displaystyle _a^b}f(t)exp[ix\mathrm{\Psi }(t)]𝑑t`$ (71)
$``$ $`\sqrt{{\displaystyle \frac{\pi }{2x\mathrm{\Psi }^{\prime \prime }(t^{})}}}f(t^{})exp[ix\mathrm{\Psi }(t^{})\pm \pi /4]`$ (72)
where $`t^{}`$ is the t where the derivative of the phase is zero. The condition of stationary phase gives
$$0=\mathrm{\Phi }^{}=kn\mathrm{\Omega }^{}(x_n)t$$
(74)
which picks out points $`x_n`$ where the contribution to the integral is maximum. Finally what we get is
$`F_{pq}(k,t)`$ (78)
$`{\displaystyle \underset{n,n_1\mathrm{}n_{q1}}{}}{\displaystyle \frac{2\pi k}{n|\mathrm{\Omega }^{\prime \prime }(x_n)|t}}`$
$`\times \left({\displaystyle \frac{x_n}{2\pi k}}\right)f_{n_1}f_{n_2}\mathrm{}f_{nn_1n_2\mathrm{}n_{q1}}`$
$`\times {\displaystyle \underset{m_1,m_2\mathrm{}m_{pq1}}{}}f_{m_1}f_{m_2}\mathrm{}f_{nm_1m_2\mathrm{}m_{pq1}}`$
where $`f_n=f_n(x_n,0)`$ and similarly for $`f_m`$. Now from the condition of stationary phase (74) we can find
$$x_n=\left(\frac{snt}{k}\right)^{\frac{1}{s+1}}$$
(80)
where we have used $`\mathrm{\Omega }(r)=r^s`$. The stationary phase contributes only when $`\rho <x_n<1`$ because the spiral structure only exists in that range of distances from the centre of the vortex. The relation between the fractal co-dimension (Kolmogorov capacity) $`D`$ of the scalar spiral and the power law of the decay of the azimuthal velocity of the vortex is,
$$D=\frac{s}{s+1}.$$
(81)
This $`D`$ is such that $`1/2<D<2/3`$ because $`1<s<2`$ and gives a measure of the space-filling property of the spiral. Hence after doing the summations in (LABEL:Ft9) we can show that the power spectrum $`F_{pq}(k,t)`$ scales like
$`F_{pq}(k,t)`$ (82)
$`k^{(32D)}t^{2(1D)}[\mathrm{const}+\mathrm{higher}\mathrm{order}\mathrm{terms}]`$ (83)
(84)
in the limit $`Pe\mathrm{}`$ and in the range $`t<k<\sqrt{\frac{Pe}{t}}`$ for times $`1tPe^{1/3}`$ which is the range of times for which the scalar patch has a well-defined spiral structure in the range of wavenumbers $`t<k<\sqrt{\frac{Pe}{t}}`$(obtained from $`\rho <x_n<1`$). The higher order terms are functions of $`k/t`$, and decay faster than $`(k/t)^1`$ in the range $`t<k<\sqrt{\frac{Pe}{t}}`$, and can therefore be neglected.
We notice that as time runs forward the spiral range $`\left(t<k<\sqrt{\frac{Pe}{t}}\right)`$ shifts to higher values of $`k`$ which is solely due to the vortex continuously wrapping the scalar field in to finner and finner spirals thus generating scales which have higher wavenumbers. This range also shrinks as it shifts to higher values of $`k`$ because of the action of diffusion. In the next section we generalise our results to the case of multiple spirals generated by a dilute collection of noninteracting vortices which may be representative of a turbulent velocity field with coherent vortical structure, perhaps obtained in the experiments of Jullien et al..
## IV Structure functions of passive decaying scalar in a flow consisting of many identical non-interacting vortices
All the analysis in this section is carried out in dimensionalised variables so we invert the transformations of section II. Let us consider many non interacting vortices randomly distributed over 2-D space and sufficiently far apart so that we can safely describe the velocity field in terms of compact vortical structures characterised by
$`\mathrm{\Omega }(x)=\mathrm{\Omega }_0\left({\displaystyle \frac{x}{L}}\right)^s\mathrm{if}{\displaystyle \frac{x}{L}}1`$
$`\mathrm{\Omega }(x)=0\mathrm{if}{\displaystyle \frac{x}{L}}>1`$
where the $`x`$’s are measured from the centre of each vortex at $`𝐱_m`$ for all $`m`$ and
$$min|𝐱_m𝐱_n|L;\mathrm{for}\mathrm{all}\mathrm{m}\mathrm{and}\mathrm{n}.$$
(85)
For the calculation of the generalised power spectrum and structure functions we need only to consider the scalar patches within a distance $`L`$ of each vortex because these scalar patches acquire a spiral structure and thereby dominate the scaling of the structure functions. The scalar field at distances larger than $`L`$ from all vortices contributes an $`𝒪(r/L)`$ term to the structure function for $`rL`$ because the interfacial structure of the scalar field far from the vortices remain regular. As we show in this section, this term is negligible in the $`r/L1`$ limit compared to the $`r`$-dependence of the structure functions caused by the scalar spiral structures around the vortices. It is therefore sufficient to consider that $`\theta (𝐱,t)`$ consists only of the local scalar fields $`\theta _m(𝐱𝐱_m,t)`$ in the vicinity of vortices labelled $`m`$ and write
$`\theta (𝐱,t)={\displaystyle \underset{m}{}}\theta _m(𝐱𝐱_m,t).`$ (86)
Every scalar spiral in the right hand side of (86) is localised within a distance $`L`$ of $`𝐱_m`$ and the condition (85) ensures that they do not overlap each other.
Now we can generalise (LABEL:Ft2) to include the effect of many non-interacting vortices with non-overlapping scalar spirals as shown below
$`\widehat{F}_{pq}(𝐤,t)`$ $`=`$ $`{\displaystyle \frac{1}{2\pi L_A^2}}{\displaystyle \underset{m}{}}{\displaystyle \theta _m^q(𝐱,t)e^{i𝐤𝐱}d^2𝐱}`$ (88)
$`\times {\displaystyle }\theta _m^{pq}(𝐱^{},t)e^{i𝐤𝐱^{}}d^2𝐱^{}.`$
Since the integrals are independent of the $`m`$’s we can take them out of the sum. Hence we get the same result as in (84) multiplied by the number of vortices per unit area, that is
$$F_{pq}(k,t)(kL)^{(32D)}(\mathrm{\Omega }_0t)^{2(1D)}\underset{m}{}\frac{1}{2\pi L_A^2}.$$
(89)
This asymptotic relation is valid when $`\mathrm{\Omega }_0t<kL<\sqrt{\frac{Pe}{\mathrm{\Omega }_0t}}`$ which is found from the condition $`\rho <x_n<l`$ in dimensionalised form and $`1\mathrm{\Omega }_0tPe^{1/3}`$ in the limit $`Pe\mathrm{}`$.
Assuming the distribution of the scalar spirals over the two dimensional space to be homogeneous and isotropic, the power spectrum $`F_{pq}(k,t)=2\pi k\overline{\widehat{F}_{pq}(𝐤,t)}`$ where the bar implies ensemble averaging over the distribution of many spirals (because vortices are non-interacting and spirals are therefore statistically independent from each other, it does make sense for the average over space already included in the definition of $`\widehat{F}_{pq}(𝐤,t)`$ to be taken over an idealised space where there is only one spiral, and for the ensemble average to be taken over the distribution of many spirals). From (16) we can show that for a homogenous distribution of scalar spirals the odd order structure functions vanish. Only the even order structure functions do not vanish, that is for $`p=even`$. Hence from(LABEL:stfunc)
$`𝒮_p(r,t)`$ $`=`$ $`\overline{[\theta (𝐱+𝐫,t)\theta (𝐱,t)]^p}`$ (90)
$`=`$ $`\overline{[\delta \theta (𝐫,t)]^p}`$ (91)
$``$ $`{\displaystyle \left(2\underset{q=1}{\overset{p1}{}}C_{qp}e^{i𝐤𝐫}\right)\overline{\widehat{F}_{pq}(𝐤,t)}k𝑑k𝑑\varphi }`$ (92)
$``$ $`{\displaystyle \left(1e^{i𝐤𝐫}\right)F_{pq}(k,t)𝑑k𝑑\varphi }`$ (93)
$``$ $`{\displaystyle (1J_0(kr))F_{pq}(k,t)𝑑k}`$ (94)
where $`J_0(kr)`$ is same as (39) with $`n=0`$. After substituting (89) in (LABEL:stfunc2) and integrating we find
$`𝒮_p(r,t)\left({\displaystyle \frac{r}{L}}\right)^{2(1D)}(\mathrm{\Omega }_0t)^{2(1D)}`$ (96)
$`\mathrm{in}\mathrm{the}\mathrm{ranges}{\displaystyle \frac{1}{\mathrm{\Omega }_0t}}>{\displaystyle \frac{r}{L}}>\sqrt{{\displaystyle \frac{\mathrm{\Omega }_0t}{Pe}}}`$ (97)
$`\mathrm{and}1\mathrm{\Omega }_0tPe^{1/3}.`$ (98)
## V Structure functions of statistically stationary passive scalar
To achieve a statistically stationary passive scalar field we may imagine that, as scalar patches take spiral shapes and decay, more scalar patches are introduced in to the flow as may well happen in an experimental setup in the laboratory. This procedure soon leads to a situation where many scalar spirals coexist in the flow all in different stages of their evolution. Assuming the rate of regular injection of the scalar blobs to balance exactly the rate of scalar dissipation, we can expect to have a statistically stationary scalar field. In this case, the averaging over many spirals in different stages of their evolution (which is involved in the calculation of the generalised power spectra and structure functions) may be assumed, in the spirit of Lundgren, to be equivalent to averaging over the life-time of a single spiral. Hence, to obtain the generalised power spectra of the statistically stationary scalar we average the previous section’s results over time in the range $`1<\mathrm{\Omega }_0t<Pe^{1/3}`$, which represents the life time of the spirals. The spiral structure lies in the range $`\rho <x_n<L`$ which implies $`\mathrm{\Omega }_0t<kL<\sqrt{\frac{Pe}{\mathrm{\Omega }_0t}}`$. This spiral range of wavenumbers together with the time range of the spiral gives the range of $`\mathrm{\Omega }_0t`$ over which we can average for a given value of $`kL`$. This leads to a time averaged $`F_{pq}(k,t)`$ which takes the form
$`F_{pq}(k)`$ $``$ $`(kL)^1`$ (99)
$`\mathrm{where}`$ $`1<kL<Pe^{1/3},`$ (100)
$`F_{pq}(k)(kL)^{(76D)}Pe^{2(1D)}`$ (101)
$`\mathrm{where}`$ $`Pe^{1/3}<kL<Pe^{1/2}.`$ (102)
(100) and (102) are the result of averaging (89) over the time ranges $`1<\mathrm{\Omega }_0t<kL`$ and $`1<\mathrm{\Omega }_0t<\frac{Pe}{(kL)^2}`$ respectively. These time ranges are determined by the respective wavenumber ranges in (100) and (102). Finally (100) and (102) give the following structure functions
$`𝒮_p(r)constant+\mathrm{ln}\left({\displaystyle \frac{r}{LPe^{1/3}}}\right)`$ (103)
$`\mathrm{where}LPe^{1/3}<r<L,`$ (104)
$`𝒮_p(r)`$ $``$ $`\left({\displaystyle \frac{r}{LPe^{1/3}}}\right)^{6(1D)}`$ (105)
$`\mathrm{where}`$ $`LPe^{1/2}<r<LPe^{1/3}.`$ (106)
In (104) and (106) $`𝒮_p(r)`$ is a time average of $`𝒮_p(r,t)`$ in (98). Note that $`\zeta _p=0`$ with a logarithmic correction in the range $`LPe^{1/3}<r<L`$ and that $`\zeta _p=6(1D)]2,3[`$ in the dissipative range $`LPe^{1/2}<r<LPe^{1/3}`$.
The structure functions $`𝒮_p(r)`$ are not time-dependent and may be interpreted as characterising a scalar field in a statistically steady state achieved with an external large-scale source of scalar (scalar forcing). This scalar forcing may consist of regularly placing in the flow scalar blobs with large-scale smooth interfacial structure. In the spirit of Birkhoff’s Ergodic theorem we should expect Lundgren’s time average assumption to be relevant for the calculation of structure functions when the scalar field is statistically stationary.
## VI Phenomenology
In this section we extract the phenomenology underlying the calculations and results of the previous sections and show that the essential ingredient of this phenomenology are the locality of scalar inter-scale transfer (108) and the Lundgren time-averaging operation. Indeed, as we show in this section, (100) and (102) can be retrieved by a simple phenomenological argument based on these two ingredients.
Let us return to the time-dependent wind-up of scalar spirals. As time proceeds, ie. as $`\mathrm{\Omega }_0t\mathrm{\Omega }_0(t+\delta t)`$, then $`kLkL+L\delta k`$ because of the differential rotation(which amounts to local shear) in every steady vortex and the entire scalar spectrum is shifted towards higher wavenumbers with time (see Figure 2). That is to say that the shearing advection to which the scalar patches are subjected in every steady vortex is such that the generalised power spectra obey
$`F_{pq}(kL+\delta kL,\mathrm{\Omega }_0t+\mathrm{\Omega }_0\delta t)d(kL+\delta kL)`$ (107)
$`=F_{pq}(kL,\mathrm{\Omega }_0t)d(kL)`$ (108)
The amount of scalar variance in the wavenumber band $`d(kL)`$ around wavenumber $`kL`$ is simply transported to wavenumber band $`d(kL+\delta kL)`$ around wavenumber $`kL+\delta kL`$ after an incremental time duration $`\mathrm{\Omega }_0\delta t`$ (see Figure 2). As shown in, the distance $`l`$ between consecutive coils of the scalar spiral in every vortex at a distance $`r`$ from the centre scales as
$$ł\frac{L}{\mathrm{\Omega }_0t}\left(\frac{r}{L}\right)^{1+1/s}.$$
(109)
Letting time vary by a small amount $`\delta t`$, the distance between two coils changes by
$`{\displaystyle \frac{\delta l}{l}}{\displaystyle \frac{\delta t}{t}}`$
the minus sign indicating that $`l`$ is decreasing. Identifying $`k`$ with $`\frac{2\pi }{l}`$ for the purpose of equation (108) so that $`\frac{\delta k}{k}=\frac{\delta l}{l}`$, it follows that (108) becomes
$`F_{pq}[kL\left(1+{\displaystyle \frac{\delta t}{t}}\right),\mathrm{\Omega }_0t\left(1+{\displaystyle \frac{\delta t}{t}}\right)]dkL\left(1+{\displaystyle \frac{\delta t}{t}}\right)`$ (110)
$`=F_{pq}(kL,\mathrm{\Omega }_0t)d(kL).`$ (111)
The solution of this equation is
$$F_{pq}(kL,\mathrm{\Omega }_0t)=L(kL)^1_{pq}\left(\frac{\mathrm{\Omega }_0t}{kL}\right)$$
(112)
where $`_{pq}`$ are arbitrary dimensionless functions. As indicated in figure 2 this form of the generalised spectra is valid in the limit $`Pe\mathrm{}`$ and in the wavenumber range $`\mathrm{\Omega }_0tkL\sqrt{\frac{Pe}{\mathrm{\Omega }_0t}}`$ and time range $`1\mathrm{\Omega }_0tPe^{1/3}`$. Note that $`1\mathrm{\Omega }_0tPe^{1/3}\sqrt{\frac{Pe}{\mathrm{\Omega }_0t}}Pe^{1/2}`$. The inverse of $`\mathrm{\Omega }_0t`$ represents the decaying outer length-scale of the spiral range and the inverse of $`\sqrt{\frac{Pe}{\mathrm{\Omega }_0t}}`$ represents the growing micro-scale of diffusive attrition. A wavenumber in the range $`1kLPe^{1/3}`$ during the time-period $`1\mathrm{\Omega }_0tPe^{1/3}`$ does not have the time to be affected by diffusive attrition and only receives scalar variance activity from lower wavenumbers until $`\mathrm{\Omega }_0t=kL`$. We therefore refer to $`1kLPe^{1/3}`$ as the advective wavenumber range. The time averaged generalised power spectra in this range are given by
$$F_{pq}(kL)=\frac{1}{kL1}_1^{kL}d(\mathrm{\Omega }_0t)L(kL)^1_{pq}\left(\frac{\mathrm{\Omega }_0t}{kL}\right)$$
(113)
and $`_{pq}`$ must be increasing functions of $`\frac{\mathrm{\Omega }_0t}{kL}`$ because the differential rotation’s shearing process causes the power spectra to shift from small to large wavenumbers (see figure 2). Hence we retrieve (100), i.e.
$`F_{pq}(kL)(kL)^1`$
in the advective range $`1kLPe^{1/3}`$ which is well defined in the limit $`Pe\mathrm{}`$.
In the advective-diffusive range $`Pe^{1/3}kLPe^{1/2}`$ a wavenumber $`kL`$ experiences the advection process from $`\mathrm{\Omega }_0t=1`$ until $`kL=\sqrt{\frac{Pe}{\mathrm{\Omega }_0t}}`$ when molecular diffusion sets in. Hence the time averaged generalised power spectra are given by
$$F_{pq}(kL)=\frac{1}{\frac{Pe}{kL^2}1}_1^{\frac{Pe}{kL^2}}d(\mathrm{\Omega }_0t)L(kL)^1_{pq}\left(\frac{\mathrm{\Omega }_0t}{kL}\right)$$
(114)
in the advective diffusive range and using $`_{pq}\left(\frac{\mathrm{\Omega }_0t}{kL}\right)\left(\frac{\mathrm{\Omega }_0t}{kL}\right)^{2(1D)}`$ (see (89)) we retrieve (102), i.e.
$`F_{pq}(kL)Pe^{2(1D)}(kL)^{7+6D}`$
in the advective-diffusive range $`Pe^{1/3}kLPe^{1/2}`$ which is well defined in the limit $`Pe\mathrm{}`$. Note that the diffusive micro-length-scale $`LPe^{1/2}`$ is the Taylor microscale of the scalar field (first introduced by Corrsin in 1951). Note also that $`76D]3,4[`$ and that the experimental results of seem to show a steeper power-law wavenumber spectrum at wavenumbers larger than where the $`k^1`$ spectrum is observed.
## VII Conclusions and Discussion
In a two-dimensional isotropic and homogeneous collection of non-interacting compact and time-independent singular vortices with a large-wavenumber energy spectrum $`E(k)k^\alpha `$ with $`1<\alpha 3`$, the structure functions of an advected and freely decaying scalar field have the following scaling behaviour in the limit where $`Pe\mathrm{}`$
$$𝒮_p(r,t)\left(\frac{r}{L}\mathrm{\Omega }_0t\right)^{2(1D)}$$
(115)
where $`\sqrt{\frac{\mathrm{\Omega }_0t}{Pe}}\frac{r}{L}\frac{1}{\mathrm{\Omega }_0t}`$ and $`1\mathrm{\Omega }_0tPe^{1/3}`$, and the fractal co-dimension $`D`$ of the scalar interfaces is such that $`1/2D<2/3`$.
By applying the Lundgren time-average assumption we obtain predictions for the structure functions of a statistically stationary scalar field in the same 2-D velocity field and the same limit $`Pe\mathrm{}`$
$$𝒮_p(r)constant+\mathrm{ln}\left(\frac{r}{LPe^{1/3}}\right)$$
(116)
in the range $`LPe^{1/3}rL`$ and
$$𝒮_p(r)\left(\frac{r}{LPe^{1/3}}\right)^{6(1D)}$$
(117)
in the range $`LPe^{1/2}rLPe^{1/3}`$. The logarithmic term in (116) corresponds to $`k^1`$ generalised power spectra. It may be worth mentioning that the 2-D velocity fields of Holzer and Siggia where they observe a well-defined $`k^1`$ scalar power spectrum are replete with spiral scalar structures.
Predictions of $`k^1`$ scalar power spectra in the limit $`Pe\mathrm{}`$ have been made for scalar fields in smooth (ie. at least everywhere continuous and differentiable ) homogeneous and isotropic random velocity field with arbitrary dimensionality and time dependence by Chertkov et. al. who generalised and unified the results of Batchelor and Kraichnan . Experimental investigations of the high Péclet number $`k^1`$ scalar power spectrum have been inconclusive in 3-D turbulent flows even though Prasad and Sreenivasan have claimed such a spectrum in a 3-D wake . However $`k^1`$ scalar power spectra have been observed at high Péclet numbers in numerical simulations of scalar fields in 2-D and 3-D chaotic flows and in 2-D velocity fields obeying the stochastically forced Euler equation restricted to a narrow band of small (integral scale) wavenumbers . More recently, $`k^1`$ scalar power spectra have been observed at $`Pe=10^7`$ in 2-D or quasi 2-D statistically stationary turbulent flows in the same range where the velocity field’s energy spectrum is $`k^3`$ by Jullien et al. who have also observed logarithmic scalings in that range for all order structure functions (similarly to (116), but without the ability to establish the $`LPe^{1/3}`$ scaling factor and range). The theory of Chertkov et al. does not apply to this experiment because homogeneous and isotropic random velocity fields which are everywhere continuous and differentiable have energy spectra steeper than $`k^4`$ (see Appendix). The present paper’s theory, however, applies when the energy spectrum scales as $`k^\alpha `$ with $`1<\alpha 3`$ but is limited to time-independent velocity fields. Nevertheless the phenomenology developed in section VI also holds for time-dependent velocity fields and we now apply and generalise it to spatially smooth chaotic flows (and also, by the way, to frozen straining velocity field structures).
The starting point of our phenomenology is the locality of transfer (108). Pedrizzetti and Vassilicos have shown that inter-scale transfer in 2-D compact vortices is indeed local at a given scale when velocity gradients do not vary much in physical space over that scale. This is the case in the 2-D axisymmetric vortices considered in this paper but also in spatially smooth velocity fields. In a spatially smooth chaotic flow the distance $`l`$ between successive folds of the scalar interface decays exponentially as determined by the largest positive Lyapunov exponent $`\lambda `$, ie. $`l(t)e^{\lambda t}`$ which implies $`\frac{\delta l}{l}\lambda \delta t`$. Applying the locality of transfer property we get
$`F_{pq}(k(1+\lambda \delta t),t+\delta t)dk(1+\lambda \delta t)`$ (118)
$`=F_{pq}(k,t)dk`$ (119)
the solution of which is
$$F_{pq}(k,t)=k^1_{pq}\left(\frac{e^{\lambda t}}{k}\right).$$
(120)
This form of the generalised spectra is valid for $`Pe\mathrm{}`$ and as long as $`1<e^{\lambda t}<k`$, so that applying Lundgren’s time-average operation from $`t=0`$ to $`t=\lambda ^1\mathrm{ln}k`$ gives
$`F_{pq}(k)`$ $`=`$ $`k^1{\displaystyle \frac{\lambda }{\mathrm{ln}k}}{\displaystyle _0^{\frac{\mathrm{ln}k}{\lambda }}}_{pq}\left({\displaystyle \frac{e^{\lambda t}}{k}}\right)𝑑t`$ (121)
$``$ $`[k\mathrm{ln}k]^1`$ (122)
because $`_{pq}`$ is an increasing function of $`\frac{e^{\lambda t}}{k}`$. Power spectra of scalar fields in spatially smooth chaotic flows are believed to scale as $`k^1`$ in the limit $`Pe\mathrm{}`$ but our theory predicts $`(k\mathrm{ln}k)^1`$. This is a small correction to the spectrum but an exponentially large correction to the scalar variance in the limit $`Pe\mathrm{}`$.
## VIII Acknowledgments
MAIK and JCV acknowledge support from EPSRC grant GR/K50320 and from EC TMR Research network on intermittency in turbulent systems. MAIK also wishes to thank the Cambridge Commonwealth Trust, Wolfson College, Cambridge and DAMTP for financial support while this work was being completed. JCV acknowledges support from the Royal Society.
## IX Appendix
For a statistically homogeneous velocity field with velocity components $`u_i(𝐱)`$ we can define a correlation function
$`R_{ij}(𝐫)=\overline{u_i(𝐱)u_j(𝐱+𝐫)}`$
and its Fourier transform
$`\mathrm{\Phi }_{ij}(𝐤)={\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle 𝑑𝐫R_{ij}(𝐫)e^{i𝐤𝐫}}.`$
Incompressibility ($`𝐮=0`$) and statistical isotropy of a two-component or 2-D velocity field $`u_i(𝐱)`$, $`i=1,2`$, imply
$`\mathrm{\Phi }_{ij}(𝐤)=\left(\delta _{ij}{\displaystyle \frac{k_ik_j}{k^2}}\right){\displaystyle \frac{E(k)}{\pi k}}`$
where the average kinetic energy per unit mass of the velocity field is $`E=_0^{\mathrm{}}𝑑kE(k),k|𝐤|`$; $`E(k)`$ is the energy spectrum of the velocity field.
One dimensional energy spectra are defined as follows
$`\varphi _{ij}(k_1)={\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^{\mathrm{}}}R_{ij}(r_1,0)e^{ik_1r_1}𝑑r_1`$
which, because of isotropy, are completely characterised by a single component, say $`\varphi _{11}(k_1)`$.
From the above and a few standard manipulations one can get
$`{\displaystyle _1^{\mathrm{}}}𝑑x{\displaystyle \frac{\sqrt{x^21}}{\pi x^2}}{\displaystyle \frac{E(xk_1)}{\varphi _{11}(k_1)}}={\displaystyle \frac{1}{2}}`$
and in a range of wavenumbers where both $`E(k)`$ and $`\varphi _{11}(k_1)`$ are monotonically decreasing functions of $`k`$ and $`k_1`$ respectively, it follows that $`E(k)k^p`$ if and only if $`\varphi _{11}(k_1)k_1^p`$.
The pivotal assumption in is that the velocity field $`u_i(𝐱)`$ is Taylor-expandable up to at least first derivative terms everywhere in physical space. In particular, this means that $`u_1(x_1,0)`$ is differentiable with respect to $`x_1`$ everywhere on the $`x_1`$ axis. If the first derivative of $`u_1(x_1,0)`$ with respect to $`x_1`$ is also continuous everywhere along the $`x_1`$ axis then the Fourier transform $`\widehat{u}_1(k_1)`$ of $`u_1(x_1,0)`$ must decay faster than $`k_1^2`$ . If, however, the first derivative of $`u_1(x_1,0)`$ is not everywhere continuous, then it is discontinuous either on a set of well-separated points or on a more pathological set of points which accumulate (and are therefore not well-separated) in a fractal-like or in a spiral-like manner . In the case where discontinuities in the derivative field of $`u_1(x_1,0)`$ are well-separated, $`\widehat{u}_1(k_1)`$ decays as $`k_1^2`$ because the Fourier transform of well-separated discontinuities between which the field is continuous decays as $`k_1^1`$ and the Fourier transform of the derivative of $`u_1(x_1,0)`$ is equal to $`ik_1\widehat{u}_1(k_1)`$. In the other case where discontinuities are not well-separated and accumulate, the decay of $`\widehat{u}_1(k_1)`$ can be anywhere between $`k_1^1`$ and $`k_1^2`$ , but in this case $`u_1(x_1,0)`$ cannot be considered to be differentiable at those points where discontinuities of its derivative accumulate.
In conclusion, the differentiability of the velocity field everywhere in physical space implies that $`\widehat{u}_1(k_1)`$ must decay at least as $`k_1^2`$ and therefore $`\varphi _{11}(k_1)|\widehat{u}_1(k_1)|^2𝒪(k_1^4)`$ which in turn implies $`E(k)𝒪(k^4)`$.
The condition $`E(k)𝒪(k^3)`$ stated in the conclusion of guarantees that the strain field is large-scale but not that the velocity field is differentiable. The spectral condition required to use the pivotal assumption of differentiability in should in fact be $`E(k)𝒪(k^4)`$. |
warning/0003/astro-ph0003367.html | ar5iv | text | # Chandra, GLAST, and the Galactic Center
## 1 Introduction
In roughly spherical accretion flows, be it Bondi (1952) or advection-dominated accretion flows (ADAFs; Rees et al. 1982, Narayan & Yi 1994), the protons have temperatures comparable to their gravitational potential energy. Close to the black hole, a significant number of protons are energetic enough to exceed the threshold for the production of pions in proton-proton collisions. Neutral pions quickly decay to produce gamma-rays. It has long been recognized that this is a plausible source of gamma-ray emission from spherical accretion flows (Shvartsman 1971; Dahlbacka, Chapline, & Weaver 1974; Colpi, Maraschi, & Treves 1986; Mahadevan, Narayan, & Krolik 1997).
In this paper we place the expected gamma-ray emission from spherical accretion flows on firmer observational ground by relating it to the more readily observable x-ray emission produced by thermal bremsstrahlung. A simple and relatively universal relationship between the two fluxes exists because both are produced by two-body processes. This is discussed in the next section (§2). We then apply these considerations to Sgr A\*, the supermassive black hole at the center of our galaxy (§3). In §4 we briefly summarize our results.
## 2 The Gamma-ray to X-ray Luminosity Ratio
We first give a simple calculation of the gamma-ray to x-ray luminosity ratio assuming self-similar scalings for the density and temperature of the flow and a thermal distribution of protons. We then discuss the uncertainties introduced by these approximations.
We take the temperature and number density of the flow to be
$$\theta _p=\theta _0r^1\mathrm{and}n=n_0r^a,$$
(1)
where $`\theta _p=kT_p/m_pc^2`$ is the dimensionless proton temperature, $`r`$ is the radius in the flow in units of the Schwarzschild radius ($`R_S`$), and $`n_0`$ is the normalization of the density, which depends on, e.g., the black hole mass, the accretion rate, and the viscosity parameter $`\alpha `$. For $`\theta _0=0.15`$, the proton temperature profile is that of non-relativistic Bondi accretion with an adiabatic index of $`\gamma =5/3`$;<sup>2</sup><sup>2</sup>2Relativistic corrections are small. comparable maximal temperatures and identical radial scalings occur in relativistic Bondi accretion (Shapiro 1973) and ADAFs (Narayan & Yi 1995). The electron temperature profile is rather uncertain; fortunately we will only need the electron temperature at large radii, r
>103
>𝑟superscript103r\mathrel{\mathchoice{\vbox{\offinterlineskip\halign{$\m@th\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}}10^{3}, where the flow is well approximated as one temperature. In equation (1) we allow the radial density profile to differ from the canonical Bondi value of $`a=3/2`$; recent work on ADAFs has shown that much smaller values, e.g., $`a=1/2`$, may be appropriate (see §2.2).
The number of $`100`$ Mev gamma-rays produced per second and per cm<sup>3</sup> is given by $`n^2R(\theta _p)`$, where $`R(\theta _p)`$ is the reaction coefficient for thermal protons of temperature $`\theta _p`$. At $`\theta _p0.15`$, $`R(\theta _p)`$ can be approximated by $`R(\theta _p)R_0(\theta _p/0.15)^3`$, where $`R_02\times 10^{17}`$ cm<sup>3</sup> s<sup>-1</sup> (see Fig. 3 of Dermer 1986); for θp
<0.05
<subscript𝜃𝑝0.05\theta_{p}\mathrel{\mathchoice{\vbox{\offinterlineskip\halign{$\m@th\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr} }}}0.05, $`R(\theta _p)`$ decreases much more rapidly than $`\theta _p^3`$. Integrating over the flow, the photon luminosity in $`100`$ Mev gamma-rays is
$$N_\gamma =4\pi R_S^3n_0^2_1^{\mathrm{}}\left(\frac{dr}{r}\right)R(\theta )r^{32a}\frac{2\pi }{a}R_S^3n_0^2R_0.$$
(2)
Spherical accretion flows produce x-rays by thermal bremsstrahlung and by Comptonizing synchrotron photons. We are interested in very low luminosity systems where the former is expected to dominate. Thermal bremsstrahlung emission can also be expressed using a reaction coefficient. The number of x-rays of frequency $`\nu `$ produced per second and per cm<sup>3</sup> is given by $`n^2R_\nu (\theta _e)`$, where $`R_\nu (\theta _e)\beta \theta _e^{1/2}\mathrm{exp}[h\nu /kT_e]`$ and $`\beta 1.3\times 10^{16}`$ cm<sup>3</sup> s<sup>-1</sup> (e.g., Rybicki & Lightman 1979). Integrating over the flow, the photon luminosity in x-rays of frequency $`\nu `$ is
$$N_X=4\pi R_S^3n_0^2\beta _1^{\mathrm{}}\left(\frac{dr}{r}\right)r^{32a}\theta _e^{1/2}\mathrm{exp}[h\nu /kT_e].$$
(3)
Equation (3) shows that at a frequency $`\nu `$ the x-ray emission is dominated by the largest radius which satisfies kTe
>hν
>𝑘subscript𝑇𝑒𝜈kT_{e}\mathrel{\mathchoice{\vbox{\offinterlineskip\halign{$\m@th\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}}h\nu. This is because $`r^3n^2T_e^{1/2}`$ increases with increasing radius. We focus on x-ray emission at $`10`$ keV which is dominated by emission from $`r10^310^4`$. At these radii the flow is quite accurately approximated as one-temperature so we can substitute $`T_e=T_p`$ ($`\theta _e=m_p\theta _p/m_e`$) into equation (3) and perform the integral
$$N_X\frac{4\pi }{3.52a}R_S^3n_0^2\beta ^{}\theta _0^{1/2}r_\nu ^{3.52a},$$
(4)
where $`r_\nu =\theta _0/\theta _\nu `$, $`\theta _\nu =h\nu /m_pc^2`$, and $`\beta ^{}=\beta (m_e/m_p)^{1/2}3\times 10^{18}`$ cm<sup>3</sup> s<sup>-1</sup>.
Combining equations (2) and (4), the ratio of the gamma-ray luminosity at energy $`E_\gamma 100`$ MeV to the x-ray luminosity at energy $`E_X`$ is given by
$$\frac{L_\gamma }{L_X}\left(\frac{E_\gamma }{E_X}\right)\left(\frac{3.52a}{a}\right)r_\nu ^{2a3.5}30\left(\frac{10\mathrm{keV}}{E_X}\right)^{1/2},$$
(5)
where the last approximation takes $`a=3/2`$.
Equation (5) shows that, for the self-similar analysis of this subsection, the gamma-ray to x-ray luminosity ratio of the flow depends only on the radial density profile. Since both pion decay and bremsstrahlung involve two-body processes the luminosity ratio from any spherical shell depends only on the local temperature(s). The radial density profile enters because pions are only produced in interesting numbers very close to the black hole while the x-ray luminosity primarily originates from rather large radii. For $`a=3/2`$, equation (5) predicts $`L_\gamma 30L_X`$, an observationally interesting number (§3), while for $`a=1/2`$, the predicted gamma-ray luminosity is certainly undetectable, $`L_\gamma 10^5L_X`$.
### 2.1 Uncertainties in Bondi-like models ($`a=3/2`$)
Several models of quasi-spherical accretion (Bondi, ADAF) predict a nearly free-fall radial velocity and, as a consequence, $`a=3/2`$. In this case, the primary uncertainty in the $`L_\gamma /L_X`$ estimate of the previous section is the proton temperature: $`\theta _p`$ could be smaller than $`0.1`$ near the black hole. A priori this is quite worrying because of the strong temperature dependence of the pion reaction rate.
We do not believe that this uncertainty poses a serious threat to the estimate of equation (5). Relativistic corrections to the classical Bondi solution are small (Shapiro 1973). As we explain below, the corrections due to rotation in the ADAF solution are relatively small as well.
In principle, ADAF models can have low proton temperatures near the event horizon (e.g., θp
<0.03
<subscript𝜃𝑝0.03\theta_{p}\mathrel{\mathchoice{\vbox{\offinterlineskip\halign{$\m@th\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr} }}}0.03). The low-temperature solutions, however, require small values of the dimensionless viscosity $`\alpha `$, while numerical simulations and theoretical arguments (see §2.2) show that canonical ADAF models are only realizable if $`\alpha `$ is relatively large, roughly α
>0.1
>𝛼0.1\alpha\mathrel{\mathchoice{\vbox{\offinterlineskip\halign{$\m@th\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}}0.1.
For large $`\alpha `$ equation (1) is a reasonable approximation of even general relativistic calculations of the structure of ADAFs (Gammie & Popham 1998, Popham & Gammie 1998; hereafter GP).<sup>3</sup><sup>3</sup>3This is not true for small $`\alpha `$ (e.g., Narayan, Kato, & Honma 1997). For non-rotating black holes and α
>0.1
>𝛼0.1\alpha\mathrel{\mathchoice{\vbox{\offinterlineskip\halign{$\m@th\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}}0.1, for example, our temperature profiles match those of GP very well;<sup>4</sup><sup>4</sup>4GP consider several adiabatic indices for the flow; we compare only with $`\gamma 5/3`$, appropriate for a flow dominated by the energy density of the nearly non-relativistic protons. they find maximal temperatures of $`\theta _p0.1`$, consistent with our value. For rapidly spinning black holes, their temperatures are yet higher, reaching $`\theta _p0.3`$. In addition, the density profile given by equation (1) is a reasonable approximation of the global calculations for large $`\alpha `$. Self-similar solutions predict radial velocities $`\alpha c_s`$, where $`c_s`$ is the sound speed of the gas. At small radii, however, the accreting gas must pass through a sonic point on its way into the black hole. For large $`\alpha `$ the “natural” radial velocity of the flow is of order the sound speed, so little deviation from self-similarity is required to match onto the sonic transition.
To test the estimate of equation (5) we calculated the expected gamma-ray to x-ray luminosity ratio using several of GP’s models and found generally good agreement. For a non-rotating black hole and an accretion flow with $`\alpha 0.3`$, for example, the more detailed calculation yields $`L_\gamma 10L_X`$ for $`E_X=10`$ keV, in reasonable agreement with equation (5).
It is also important to emphasize that gamma-ray emission from pion decay is unlikely to be as sensitive to temperature as suggested by the simple thermal model we have considered. The collisionless plasmas of interest should efficiently accelerate protons to relativistic energies; in this case the total gamma-ray luminosity varies only linearly with changes in the thermal energy of the protons since there are always a substantial number of protons above the pion production threshold (see Mahadevan et al. 1997).<sup>5</sup><sup>5</sup>5Gruzinov & Quataert (1999) describe a proton heating model which yields very little proton acceleration. However, if shocks or reconnection events occur in the accretion flow, a fraction of protons should be accelerated.
Equation (5) predicts a detectable gamma-ray flux only if $`a3/2`$. The above considerations suggest that equation (5) should be a good approximation in this limit.
### 2.2 Non Bondi-like accretion models ($`a<3/2`$)
Modern theories and numerical simulations of quasi-spherical accretion flows suggest that the mean infall velocity can deviate substantially from the free-fall value, resulting in $`a<3/2`$. There are three scenarios: convection-dominated accretion flows (CDAFs), winds, and turbulent heat conduction.
CDAF: Two independent groups have performed numerical simulations of quasi-spherical non-radiating accretion flows with small values of the viscosity parameter (Stone, Pringle, & Begelman 1999; Igumenshchev & Abramowicz 1999). They both find $`a=1/2`$ rather than $`a=3/2`$. Narayan, Igumenshchev, & Abramowicz (2000) and Quataert & Gruzinov (2000) have explained this in terms of a CDAF. In such a flow angular momentum is efficiently transported inwards by strong radial convection. This nearly cancels the outward transport by magnetic fields, leading to a substantially suppressed accretion rate and a much flatter radial density profile.
Winds: For large α
>0.1
>𝛼0.1\alpha\mathrel{\mathchoice{\vbox{\offinterlineskip\halign{$\m@th\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}{\vbox{\offinterlineskip\halign{$\m@th\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr} }}}0.1, CDAFs do not appear to be found in numerical simulations (Igumenshchev & Abramowicz 1999); this is because the infall time of the gas is shorter than the convective turnover time, so convection is less dynamically important. For large $`\alpha `$, however, $`a`$ may differ from $`3/2`$ for a different physical reason; strong outflows may drive away most of the accreting mass (Blandford & Begelman 1999; Igumenshchev & Abramowicz 1999).
Turbulent heat conduction: Conduction preheats the infalling gas, reducing the accretion rate and flattening the density profile (Gruzinov 1999).
## 3 Application to the Galactic Center
Chandra observations of the Galactic Center detect a point source coincident with the non-thermal radio source Sgr A\* to within $`0.5^{\prime \prime }10^5R_S`$. Its $`0.110`$ keV luminosity is $`L_X4\times 10^{33}`$ ergs s<sup>-1</sup> (Baganoff et al. 2000). It is very plausible that this represents the first x-ray detection of the supermassive black hole at the center of our galaxy.
It is natural to interpret Chandra’s detection as thermal bremsstrahlung from large radii in the accretion flow. As shown in §2, such emission would arise from $`r10^4`$; the density required to match the observed luminosity is then $`4\times 10^3`$ cm<sup>-3</sup>. The corresponding accretion rate is $`10^5`$ $`M_{}`$ yr<sup>-1</sup>, if the radial velocity of the gas is of order the sound speed. This is in good agreement with estimates based on the mass-losing stars in the central parsec of the Galactic Center (e.g., Coker & Melia 1997; Quataert, Narayan, & Reid 1999). The bremsstrahlung interpretation predicts the absence of short timescale variability in the observed x-rays (since the emission arises from large radii). It can also be tested by looking for x-ray line emission in deeper Chandra exposures (Narayan & Raymond 1999).
EGRET observations of the Galactic Center region detect a source (2EG 1746-2852) with $`L_\gamma 10^{36}`$ ergs s<sup>-1</sup> and a power law spectrum extending from $`100`$ MeV to $`10`$ GeV (Merck et al. 1996); it appears to be point like within the $`1^o`$ resolution of the instrument. Mahadevan et al. (1997) interpreted this emission as arising from an ADAF around the black hole at the Galactic Center. In their more comprehensive models of Sgr A\*, however, Narayan et al. (1998) were unable to produce gamma-ray emission at the required levels and satisfy other observational constraints. Moreover, the observed spectrum of 2EG 1746-2852 looks very similar to that expected from cosmic rays colliding with a dense cloud of molecular hydrogen.
Our calculation in §2 predicts the expected gamma-ray luminosity from the accretion flow given the x-ray luminosity in thermal bremsstrahlung. We believe that the Chandra observations of the Galactic Center provide this thermal bremsstrahlung luminosity. With this normalization, equation (5) predicts $`L_\gamma 10^{35}`$ ergs s<sup>-1</sup> for $`a=3/2`$. This is well above the $`2\times 10^{33}`$ ergs s<sup>-1</sup> detection threshold of the forthcoming Gamma-ray Large Area Space Telescope (GLAST).<sup>6</sup><sup>6</sup>6see http://glast.gsfc.nasa.gov/SRD In addition, GLAST’s angular resolution is expected to be significantly better than that of EGRET (for, among other things, the express purpose of identifying unidentified EGRET sources). GLAST will therefore likely have the capability of distinguishing a gamma-ray counterpart of Sgr A\* (if it indeed exists) from 2EG 1746-2852.
## 4 Discussion
We have argued that the ratio of the $`100`$ MeV gamma-ray luminosity to the $`10`$ keV x-ray luminosity of a spherical accretion flow depends primarily on its radial density profile (where $`nr^a`$). In particular, for canonical spherical accretion flow models with $`a=3/2`$, $`L_\gamma 30L_X`$; for $`a<3/2`$, $`L_\gamma L_X`$.
Our analysis predicts the $`100`$ MeV gamma-ray luminosity expected from the accretion flow onto the supermassive black hole at the center of our galaxy (§3): $`L_\gamma 10^{35}`$ ergs s<sup>-1</sup> if $`a=3/2`$ while $`L_\gamma 10^{29}`$ ergs s<sup>-1</sup> if $`a=1/2`$. For $`a=3/2`$, this estimate is nearly two orders of magnitude above the detection threshold of the GLAST telescope. We expect, however, that no gamma-rays will be observed coincident with the black hole, supporting theoretical suggestions (see §2.2) that the density profile in spherical accretion flows is significantly flatter than the canonical $`r^{3/2}`$ profile.
###### Acknowledgements.
We thank Charles Gammie and Bob Popham for the use of their global models of ADAFs, Fred Baganoff for information about Chandra observations of the Galactic Center, and John Bahcall for discussions and comments. EQ is supported by NASA through Chandra Fellowship PF9-10008, awarded by the Chandra X–ray Center, which is operated by the Smithsonian Astrophysical Observatory for NASA under contract NAS 8-39073. AG was supported by the W. M. Keck Foundation and NSF PHY-9513835. |
warning/0003/hep-th0003034.html | ar5iv | text | # Real-normalized Whitham hierarchies and the WDVV equations
## Introduction
In the beginning of the 90’s, while studying deformations of $`2D`$ topological field theories (TFT), E. Witten, R. Dijkgraaf, E. Verlinde, and H. Verlinde (\[DVV91b, DVV91a, Wit90\]) wrote down the following overdetermined system of non-linear PDEs for a function $`(t)=(t_0,t_1,\mathrm{})`$:
(WDVV)
$$_{\alpha \beta \lambda }(_{0\lambda \mu })^1_{\mu \gamma \delta }=_{\delta \beta \lambda }(_{0\lambda \mu })^1_{\mu \gamma \alpha },$$
where $`_\alpha ={\displaystyle \frac{}{t_\alpha }}`$, and the matrix $`\eta _{\alpha \beta }=_{0\alpha \beta }`$ is constant and non-degenerate. These equations are now called the WDVV equations. They appear in the following way \[Dij98\].
One can show that the structure of a $`2D`$ TFT is equivalent to a structure of a *Frobenius algebra* $`A`$, i. e., a commutative associative algebra with a unit and a symmetric non-degenerate bilinear form $`\eta `$ such that $`\eta (ab,d)=\eta (a,bd)`$. Let $`\{\varphi _\alpha \}`$ be a basis of $`A`$ ($`\varphi _\alpha `$ correspond to primary fields in TFT) and let $`\eta _{\alpha \beta }=\eta (\varphi _\alpha ,\varphi _\beta )=\varphi _\alpha \varphi _\beta _0`$, $`c_{\alpha \beta \gamma }=\eta (\varphi _\alpha ,\varphi _\beta \varphi _\gamma )=\varphi _\alpha \varphi _\beta \varphi _\gamma _0`$. Deformations of a TFT structure considered in \[DVV91b, DVV91a, Wit90\] correspond to potential deformations of a Frobenius algebra, i. e., there should exist a function $`(t)`$, called the *WDVV potential*, such that $`\eta _{\alpha \beta }=_{0\alpha \beta }(t)`$ and $`c_{\alpha \beta \gamma }(t)=_{\alpha \beta \gamma }(t)`$. Then the WDVV equations are just the associativity conditions for the deformed algebra structure.
It is now clear that the theory of the WDVV equations and its differential-geometric counterpart, the theory of Frobenius manifolds, are related to a whole spectrum of applications ranging from classical differential geometry (Darboux-Egoroff metrics, $`n`$-orthogonal curvilinear coordinate systems, deformations of singularities) to quantum cohomology, Gromov-Witten invariants, integrable hierarchies, and the Seiberg-Witten equations.
Solutions of the WDVV equations that can be obtained from the theory of Whitham hierarchies correspond to the topological Landau-Ginzburg theories and minimal models. For $`A_n`$ minimal models, a Frobenius algebra structure is defined on the space of degree $`n2`$ polynomials in $`p`$ with the help of the superpotential $`W(p)={\displaystyle \frac{p^n}{n}}`$ by
$`uv`$ $`=u(p)v(p)modW^{}(p),`$
$`uv`$ $`=\mathrm{res}_{\mathrm{}}{\displaystyle \frac{u(p)v(p)}{W^{}(p)}}dp.`$
A Frobenius algebra structure is then deformed by deforming the superpotential $`W`$,
$$W(p|a)=\frac{p^n}{n}+a_{n2}p^{n2}+\mathrm{}+a_1p+a_0,$$
where we used the notation $`W(p|a)`$ to separate the variable $`p`$ from the deformation parameters $`a_{n2},\mathrm{},a_0`$. Let $`𝔨(p)`$ be an $`n^{\text{th}}`$ root of $`W(p|a)`$ in the neighborhood of infinity,
$$𝔨^n(p)=nW(p)=p^n+\mathrm{},𝔨(p)=p+O(p^1).$$
Then the deformed basis of primary fields $`\varphi _\alpha `$ is given by
(1)
$$\varphi _\alpha (p|t)=\frac{1}{\alpha +1}\frac{d\mathrm{\Omega }_{\alpha +1}(p|t)}{dp}=p^\alpha +\mathrm{},$$
where $`\mathrm{\Omega }_\alpha (p)=\left[\begin{array}{c}𝔨^\alpha (p)\end{array}\right]_+`$ is a principal (i.e., polynomial) part of the Laurent expansion of $`𝔨^\alpha `$ in the neighborhood of infinity. The flat coordinates $`t_\alpha `$ on the space of the deformation parameters are given by $`t_\alpha =\mathrm{res}_{\mathrm{}}𝔨^{(\alpha +1)}pdW`$, and $`\varphi _\alpha (p|t)=_{t_\alpha }W(p|t)`$. Then
$`\eta _{\alpha \beta }`$ $`=\varphi _\alpha ,\varphi _\beta =\mathrm{res}_{\mathrm{}}{\displaystyle \frac{\varphi _a\varphi _\beta }{W^{}(p)}}dp=\delta _{\alpha +\beta ,n2},`$
$`c_{\alpha \beta \gamma }(t)`$ $`=(\text{coeff}){\displaystyle \underset{q_s|dW(q_s)=0}{}}\mathrm{res}_{q_s}{\displaystyle \frac{d\mathrm{\Omega }_{\alpha +1}d\mathrm{\Omega }_{\beta +1}d\mathrm{\Omega }_{\gamma +1}}{dpdW}}.`$
In \[DVV91b\] it was shown that $`c_{\alpha \beta \gamma }(t)`$ satisfy the integrability conditions and therefore $`c_{\alpha \beta \gamma }(t)=_{\alpha \beta \gamma }(t)`$. The closed expression for this WDVV potential $`(t)`$ was identified by I. Krichever \[Kri92\] with the logarithm of the $`\tau `$-function of a certain reduction of the genus zero Whitham hierarchy.
Whitham equations (or modulation equations) appear in the theory of perturbations of exact algebro-geometric solutions of soliton equations and describe “slow drift” on the moduli space of algebro-geometric data. These equations can be defined with the help of certain differentials $`d\mathrm{\Omega }_i`$ on the universal curve, each differential generating a corresponding Whitham flow on the moduli. A large class of solutions of the Whitham hierarchy is given by the so-called algebraic orbits. Such solutions depend only on finitely many parameters and can be constructed as follows. One starts with a finite-dimensional moduli space, called the universal configuration space, that consists of a curve $`\mathrm{\Gamma }`$, punctures $`P_\alpha `$, and a pair of Abelian integrals $`E`$ and $`Q`$. Then one defines special coordinates on this space (Whitham times), picks a leaf defined by fixing some of them, and maps this leaf to the moduli space of algebro-geometric data in such a way that coordinate lines go to Whitham flows. All information about such algebraic orbit can be encoded in a single function $`\tau (t)`$ that depends only on the moduli. This function is called the $`\tau `$-function of an algebraic orbit.
In the genus zero case Abelian integrals become polynomials, and if we choose an algebraic orbit with $`Q=p`$, then $`E`$ can be identified (up to normalization) with the superpotential $`W`$, $`\mathrm{\Omega }_i`$ define a basis of the corresponding Frobenius algebra, Whitham times $`t_i`$ give flat coordinates on the orbit, and $`F(t)=\mathrm{ln}\tau (t)`$ is a WDVV potential.
This approach can be generalized to moduli spaces of curves of higher genus. One new feature of a higher genus case is a more complicated topology of the moduli space. As a result, the hierarchy has to be extended to include certain (multivalued) differentials $`d\mathrm{\Omega }_A`$ that generate additional Whitham flows. Another important issue is a choice of the normalization. Namely, in the genus zero case, differentials $`d\mathrm{\Omega }_i`$ are completely determined by their expansions in the neighborhoods of marked points $`P_\alpha `$. In the higher genus case, these conditions define $`d\mathrm{\Omega }_i`$ only up to a holomorphic differential. This ambiguity is fixed by introducing a normalization condition. There are two main choices — real normalization, which is defined by
$$\mathrm{}\left[_c𝑑\mathrm{\Omega }_i\right]=0cH_1(\mathrm{\Gamma }_g,),$$
and complex normalization (or normalization w.r.t. $`a`$-cycles). Complex normalization requires making a choice of a canonical basis $``$ of cycles in the homology of $`\mathrm{\Gamma }_g`$ and is then defined by
$$_{a_k}𝑑\mathrm{\Omega }_i=0a_k.$$
Whitham equations were originally derived in \[Kri88\] for real-normalized differentials. However, after the relationship between Whitham equations and WDVV equations was found in \[Kri92\] and \[Dub92\], the focus shifted to the complex normalization condition \[Kri94\]. There were two main reasons for this. First, in the complex-normalized case the $`\tau `$-function is holomorphic, which is important for string theory applications. Second, the derivation of the expression for the $`\tau `$-function relied on the Riemann bilinear identities. Corresponding identities for real normalized differentials are technically more complicated. However, complex normalization has certain disadvantages. In particular, it is well-defined only on the extended moduli space that incorporates the choice of a canonical basis $``$ into moduli data.
In this paper we develop a real-normalized version of the above approach. In this case, Whitham hierarchy can be defined on the usual moduli space of curves with some extra algebro-geometric data. First, we define a real leaf in the universal configuration space, introduce Whitham coordinates on this leaf and map it into the moduli space in such a way that the resulting differentials are real-normalized. Then we prove the real-normalized version of the Riemann bilinear identities (generalized to multivalued differentials). Using these identities we find the formula for the $`\tau `$-function of an algebraic orbit and prove the following theorem for $`F(T)=\mathrm{ln}\tau (T)`$.
###### Theorem.
The third derivatives of $`F(T)`$ are given by the following formula
$$_{T_AT_BT_C}F(T)=\mathrm{}\left[\underset{q_s|dE(q_s)=0}{}\mathrm{res}_{q_s}\frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_Bd\mathrm{\Omega }_C}{dEdQ}\right]$$
This theorem then implies that if we consider a special class of algebraic orbits by choosing $`dQ`$ to be a real-normalized differential with a pole of order two at a puncture $`P_1`$, then Whitham flows define potential deformations of a Frobenius algebra structure with observables corresponding to $`{\displaystyle \frac{d\mathrm{\Omega }_A}{dQ}}`$, and we have the following theorem.
###### Theorem.
The logarithm of a $`\tau `$-function of a (reduced) real-normalized genus $`g`$ Whitham hierarchy is a solution to the WDVV equation.
### Acknowledgments
The author is grateful to I. Krichever for suggesting this problem and for constant attention to this work. The author also thanks Yu. Volvovski for many useful discussions.
## Whitham equations
Whitham equations first appeared in the theory of perturbations of exact algebro-geometric solutions of soliton equations. Such solutions are defined by linear flows on the Jacobian $`\mathrm{Jac}(\mathrm{\Gamma }_g)`$ of an auxiliary algebraic curve $`\mathrm{\Gamma }_g`$ and are expressed in terms of theta functions. Perturbing these solutions by the so-called non-linear WKB (or Whitham averaging \[Whi74\]) method (\[FFM80, DM82, DN83\]) results in a “slow drift” on the moduli $``$ of algebro-geometric data. Equations describing this drift are called *Whitham equations*.
Although Whitham equations are equations on the moduli, they can be conveniently written with the help of certain Abelian differentials $`d\mathrm{\Omega }_A(P|I)`$, $`P\mathrm{\Gamma }_g`$ defined on the universal curve $`𝒩_g^1`$,
$$\begin{array}{ccc}\mathrm{\Gamma }_g& & 𝒩_g^1\\ & & & & \\ & & .\end{array}$$
Each of the differentials $`d\mathrm{\Omega }_A`$ is coupled with a corresponding *Whitham time $`T_A`$* defining $`A^{\text{th}}`$ Whitham flow on $``$. Then, after making a special choice of connection on $`𝒩_g^1`$, Whitham equations can be written in the following implicit form,
$$_Ad\mathrm{\Omega }_B=_Bd\mathrm{\Omega }_A.$$
This form of Whitham equations was first observed by Flashka, Forest, and McLaughlin \[FFM80\] for the KdV equation and hyperelliptic spectral curves. It was later justified by Krichever \[Kri88\] in a more general setting of $`(2+1)`$ equations and general spectral curves.
## Whitham hierarchies
It turns out that one can construct a whole hierarchy of Whitham equations. Following Krichever, we use algebro-geometric approach to construct Whitham hierarchies (Hamiltonian approach to the theory of Whitham equations was developed in \[DN83\]).
We begin with a local definition. Let $`T=\{T_A\}_{A𝒜}`$ be a collection of (real or complex) times, indexed by some set $`𝒜`$, let $`zD`$, and let $`\{\mathrm{\Omega }_A(z|T)\}_{A𝒜}`$ be a collection of functions, meromorphic in $`z`$, each $`\mathrm{\Omega }_A`$ is coupled with the corresponding time $`T_A`$. The functions $`\mathrm{\Omega }_A(z|T)`$ should be thought of as pull-backs via Whitham times of Abelian integrals $`\mathrm{\Omega }_A=𝑑\mathrm{\Omega }_A`$ from a leaf spanned by Whitham flows in $``$, with $`z`$ corresponding to a local coordinate along the fiber. Define a $`1`$-form $`\omega `$ by
$$\omega =\underset{A}{}\mathrm{\Omega }_A(z|T)dT_A.$$
Then, by definition, *Whitham hierarchy* is given by the generating equation
(2)
$$\delta \omega \delta \omega =0,$$
where
$$\delta \omega =\underset{A}{}_B\mathrm{\Omega }_AdT_BdT_A+\frac{d\mathrm{\Omega }_A}{dz}dzdT_A.$$
This is equivalent to the following two equations,
(3)
$$\underset{\{A,B,C\}}{}\epsilon ^{\{A,B,C\}}\left(\frac{d\mathrm{\Omega }_A}{dz}\right)(_B\mathrm{\Omega }_C)=0,\underset{\{A,B,C,D\}}{}\epsilon ^{\{A,B,C,D\}}(_A\mathrm{\Omega }_B)(_C\mathrm{\Omega }_D)=0,$$
where we sum over all possible permutations of a fixed collection of indexes and $`\epsilon `$ is a sign of a permutation.
Usually we have one marked index $`A_0𝒜`$ with
$$X=T_{A_0},p(z|T)=\mathrm{\Omega }_{A_0}(z|T).$$
Then, as long as $`{\displaystyle \frac{dp}{dz}}0`$, we can make a change of coordinates from $`(z,T)`$ to $`(p,T)`$. In these coordinates, equation (3) written for $`A_0`$, $`A`$, $`B`$ takes the form
(4)
$$_A\mathrm{\Omega }_B(p|T)_B\mathrm{\Omega }_A(p|T)+\{\mathrm{\Omega }_A,\mathrm{\Omega }_B\}(p|T)=0,$$
where
$$\{\mathrm{\Omega }_A,\mathrm{\Omega }_B\}(p|T)=_X\mathrm{\Omega }_A\frac{d\mathrm{\Omega }_B}{dp}_X\mathrm{\Omega }_B\frac{d\mathrm{\Omega }_A}{dp}$$
is the usual Poisson bracket. Equations in this form are called *zero-curvature equations*. They can be interpreted as a compatibility conditions for the following Hamiltonian system,
$$_AE(p|T)=\{E,\mathrm{\Omega }_A\}(p|T).$$
Then, if zero-curvature equations are satisfied, there exists (locally) a solution $`E=E(p|T)`$ of this system and, as long as $`{\displaystyle \frac{dE}{dp}}0`$, we can again change coordinates from $`(p,T)`$ to $`(E,T)`$. In these coordinates, the above system takes the form
$$_Ap(E|T)=_X\mathrm{\Omega }_A(E|T)$$
and its compatibility conditions can be written as
$$_A\mathrm{\Omega }_B(E|T)=_B\mathrm{\Omega }_A(E|T).$$
Therefore, there exists (locally) a function $`S=S(E|T)`$ such that $`\mathrm{\Omega }_A(E|T)=_AS(E|T)`$. This function $`S(E|T)`$ is called a *prepotential of the Whitham hierarchy*.
## Algebraic orbits
One can use the above formalism to construct certain exact solutions of Whitham equations. These solutions depend on finitely many parameters and are obtained as follows \[Kri94\]. First we construct Whitham flows on finite-dimensional submanifolds, called *algebraic orbits*, of $``$ (note that $``$ is usually infinite-dimensional) together with Whitham equations that they satisfy. Then these equations are extended to the whole $``$. By making a correct choice of algebraic data, one can obtain usual Whitham equations of the soliton theory.
We illustrate this idea in the special case of the so-called *dispersionless Lax equations*. In this case we take algebraic curve of genus zero, $`\mathrm{\Gamma }P^1`$, with a single puncture $`P_1=\mathrm{}`$. The moduli space is equivalent to a moduli space of local coordinates around $`P_1`$,
$$=\widehat{}_{0,1}=\{\mathrm{\Gamma }P^1;P_1=\mathrm{};z(P)\}\{\nu _s,s=1,\mathrm{}\},$$
where
$$z^1(p)=𝔨(p)=p+\nu _1p^1+\nu _2p^2+\mathrm{},$$
and $`p`$ is the standard coordinate on $``$. Note that in this case $`\mathrm{\Omega }_1=p`$, i.e., the marked index is $`i=1`$. Then, by definition, the dispersionless Lax hierarchy (or dispersionless KP hierarchy) is a set of evolution equations on $`\nu _s=\nu _s(T)`$:
$$_i𝔨(p|T)=\{𝔨,(𝔨^i)_+\}=\{𝔨,\mathrm{\Omega }_i\},\text{where }_i=\frac{}{T_i}.$$
This hierarchy can be thought of as a quasi-classical limit of the usual KP hierarchy, and algebraic orbits correspond to $`n^{\text{th}}`$ order reductions of the KP hierarchy, i.e., nKdV hierarchies. Namely, let
$$E(p)=p^n+u_{n2}p^{n2}+\mathrm{}+u_0$$
and define $`𝔨(p)`$ by the condition $`𝔨(p)^n=E(p)`$. Then $`\nu _s=\nu _s(u_0,\mathrm{},u_{n2})`$ and we obtain a finite-dimensional leaf in $``$. Corresponding evolution equations on $`u_i=u_i(T)`$ can be written as
$`_iE(p|T)`$ $`=\{E,(E^{\frac{i}{n}})_+\}`$ $`\text{in }(p,T)\text{ coordinates},`$
$`_ip(E|T)`$ $`=_X\mathrm{\Omega }_i(E|T)`$ in $`(E,T)`$ coordinates.
Then the solution $`E=E(p|T)`$ (i.e., $`u_i=u_i(T)`$) can be obtained as follows \[Kri88, Kri92\].
###### Theorem.
Let
(5)
$$S(p|T)=\underset{i}{}T_i\mathrm{\Omega }_i(p|u)=\underset{i}{}T_i𝔨^i+O(𝔨^1).$$
Require that
(6)
$$\frac{dS}{dp}(q_s)=0\text{ for all }q_s\text{ such that }\frac{dE}{dp}(q_s)=0.$$
Equation (6) is equivalent to a collection of equations $`F_k(u,T)=0`$ that implicitly define $`u_i=u_i(T)`$.
Then $`u(T)`$ is a solution of the dispersionless Lax equations.
The theorem follows from the fact that $`S`$ defined above is a *global* prepotential, i.e.,
$$_iS(E|T)=\mathrm{\Omega }_i(E|T)=\{\begin{array}{cc}𝔨^i+O(𝔨^1)\text{ near }P_1\hfill & \\ \text{has no other poles}\hfill & \end{array}.$$
From the definition of $`S`$ it is clear that the first condition is satisfied. The role of the condition (6) is to ensure that $`_iS(E|T)`$ is holomorphic on $`\mathrm{\Gamma }P_1`$.
Note that the condition (6) can be also written in the form $`dS=QdE`$ for some polynomial $`Q(p|T)`$. Conversely, we can recover $`T_i`$ from $`S`$ and $`𝔨`$ by the formula
$$T_i=\frac{1}{i}\mathrm{res}_{P_1}𝔨^idS.$$
This observation motivates the following alternative approach. Consider the so-called *universal configuration space*
$$_0(n,m)=\left\{\mathrm{\Gamma }=P^1;P_1=\mathrm{};[z]_n;E,Q\right\}\{u_0,\mathrm{},u_{n2};b_0,\mathrm{},b_m\},$$
where
$$E(p)=p^n+u_{n2}p^{n2}+\mathrm{}+u_0,Q(p)=b_mp^m+\mathrm{}+b_0,$$
$`[z]_n`$ is an $`n`$-jet of a local coordinate near $`P_1`$ and we require that $`E=z^n+O(z)`$ near $`P_1`$. By definition, put
(7)
$$dS=QdE,T_i=\frac{1}{i}\mathrm{res}_{P_1}z^idS.$$
Then
$$_idS(E|T)=d\mathrm{\Omega }_i(E|T),$$
condition (7) is equivalent to a collection of equations $`T_k=T_k(u,b)`$, which can be inverted, $`u_i=u_i(T)`$, $`b_j=b_j(T)`$, and therefore we obtain a map from $`_{0,1}`$ to an algebraic orbit in $`\widehat{}_0`$.
## $`\tau `$-functions and solutions to the WDVV equations
For each algebraic orbit constructed above corresponds a so-called $`\tau `$-function. By definition,
$$\mathrm{ln}\tau (T)=F(T),$$
where
(8)
$$F(T)=\frac{1}{2}\mathrm{res}_{\mathrm{}}\left(\underset{i}{}T_i𝔨^i\right)dS,$$
and $`dS`$ is a (differential of) the prepotential of the corresponding algebraic orbit. Then we have the following theorem (Krichever \[Kri92\]).
###### Theorem.
The derivatives of $`F(T)`$ are given by the following formulas:
$`_iF(T)`$ $`=\mathrm{res}_{\mathrm{}}𝔨^idS`$
$`_{ij}F(T)`$ $`=\mathrm{res}_{\mathrm{}}𝔨^jd\mathrm{\Omega }_i=\mathrm{res}_{\mathrm{}}𝔨^jd\mathrm{\Omega }_i`$
$`_{ijk}F(T)`$ $`={\displaystyle \underset{q_s|dE(q_s)=0}{}}\mathrm{res}_{q_s}{\displaystyle \frac{d\mathrm{\Omega }_id\mathrm{\Omega }_jd\mathrm{\Omega }_k}{dQdE}},\text{where }Q={\displaystyle \frac{dS}{dE}}.`$
###### Corollary.
If we choose
$$Q(p|T)=p,E(p|T_1,\mathrm{},T_{n+1})=p^n+u_{n2}(T)p^{n2}+\mathrm{}+u_0(T),$$
then $`F(T)`$ is a WDVV potential for $`A_{n1}`$ Landau-Ginzburg model defined by a superpotential
$$W(p|t_0,\mathrm{},t_{n2})=\frac{1}{n}E(p|t_0,\frac{t_1}{2},\mathrm{},\frac{t_{n2}}{n},0,\frac{1}{n+1}).$$
## Higher genus case
The above approach can be generalized to algebraic curves of higher genus. This question was considered by Dubrovin \[Dub92\] for Hurwitz spaces and by Krichever \[Kri94\] in general. In the higher genus case, the “universal” moduli space
$$\widehat{}_{g,N}=\left\{\mathrm{\Gamma }_g;P_\alpha ;𝔨_\alpha ^1(P)=z_\alpha (P)\right\}$$
consists of the following algebro-geometric data:
* smooth algebraic curve $`\mathrm{\Gamma }_g`$ of genus $`g`$,
* collection of marked points $`P_\alpha \mathrm{\Gamma }_g,\alpha =1,\mathrm{},N`$,
* local coordinates $`z_\alpha (P)`$ in the neighborhoods of $`P_\alpha `$, $`z_\alpha (P_\alpha )=0`$.
For simplicity, we concentrate exclusively on a single puncture case, $`N=1`$.
To construct a realization of a Whitham hierarchy on $`\widehat{}_{g,1}`$, one can proceed as follows (see \[Kri94\]). Instead of polynomials $`\mathrm{\Omega }_i`$ one has to consider Abelian differentials of the second kind $`d\mathrm{\Omega }_i`$ on $`\mathrm{\Gamma }_g`$ with prescribed behavior near the puncture,
(9)
$$d\mathrm{\Omega }_i=d\left(𝔨^i+O(𝔨^1)\right)\text{near }P_1\text{.}$$
However, condition (9) specifies $`d\mathrm{\Omega }_i`$ only up to a holomorphic differential. To define $`d\mathrm{\Omega }_i`$ uniquely, one has to impose certain normalization conditions. There are two choices — *real normalization* and *complex normalization* (or normalization w.r.t. *$`a`$-cycles*). Real normalization is defined by the condition
(10)
$$\mathrm{}\left[_c𝑑\mathrm{\Omega }_i\right]=0cH_1(\mathrm{\Gamma }_g,).$$
To define complex normalization, one has to first choose a canonical homology basis for $`\mathrm{\Gamma }_g`$,
$$=\left\{a_1,\mathrm{},a_g;b_1,\mathrm{},b_g\right|a_i,b_jH_1(\mathrm{\Gamma }_g,),a_ia_j=b_ib_j=0,a_ib_j=\delta _{ij}\}.$$
As a result, it is necessary to consider the extended moduli space
$$\widehat{}_{g,1}^{}=\{\mathrm{\Gamma }_g;P_1;z(P);\},$$
which is a covering of $`\widehat{}_{g,1}`$. The differentials $`d\mathrm{\Omega }_i`$ are then uniquely normalized by the condition
(11)
$$_{a_k}𝑑\mathrm{\Omega }_i=0a_k.$$
As before, the marked index is $`i=1`$ and $`p(P)=^P𝑑\mathrm{\Omega }_1`$.
## Universal configuration space
In order to construct solutions corresponding to algebraic orbits, it is convenient to use the universal configuration space approach. Following \[KP97, KP98\], define the universal configuration space for Whitham hierarchies,
$$_g(n,m)=\{\mathrm{\Gamma }_g;P_1;[z]_n;E,Q\},$$
to be the moduli space of the following data:
* genus $`g`$ Riemann surface $`\mathrm{\Gamma }_g`$,
* marked point $`P_1\mathrm{\Gamma }_g`$,
* an $`n`$-jet $`[z]_n`$ of local coordinates $`z(P)`$ in the neighborhoods of $`P_1`$, $`z(P_1)=0`$,
* an Abelian integral $`E`$ with pole of order $`n`$ at $`P_1`$ such that
(12)
$$dEd(z^n+O(z))\text{near }P_1\text{}z[z]_n\text{ },$$
* an Abelian integral $`Q`$ with pole of order $`m`$ at $`P_1`$.
More precisely, by Abelian integrals we mean pairs $`E=(dE,P_0^E)`$, $`Q=(dQ,P_0^Q)`$, where $`dE`$, $`dQ`$ are meromorphic differentials of the second kind on $`\mathrm{\Gamma }_g`$, holomorphic on $`\mathrm{\Gamma }_gP_1`$, and with poles of order $`n+1`$ and $`m+1`$ at $`P_1`$, and
$$E(P)=_{P_0^E}^P𝑑E,Q(P)=_{P_0^Q}^P𝑑Q.$$
Alternatively, one can choose a local coordinate $`z[z]_n`$ and define $`E`$ and $`Q`$ to be pairs $`E=(dE,c_E)`$, $`Q=(dQ,c_Q)`$, where near $`P_1`$,
$$Ez^n+c_E+O(z),Qc_m^Qz^m+\mathrm{}+c_1^Qz^1+c_Q+O(z).$$
Note that after $`E`$ is chosen, there is a preferred local coordinate $`z_{}[z]_n`$ defined by
$$E=z_{}^n\text{near }P_1.$$
The moduli space $`_g(n,m)`$ is a complex manifold of dimension $`5g+n+m`$ with at most orbifold singularities. We also need to consider smaller $`(4g+n1)`$-dimensional moduli space
$$_g(n)=\{\mathrm{\Gamma }_g;P_1;[z]_n;E\},$$
as well as the corresponding extended moduli spaces $`_g^{}(n,m)`$ and $`_g^{}(n)`$.
## Construction of algebraic orbits
A general approach for constructing algebraic orbits in the higher genus case is the following. First, choose a leaf $`𝒱(n,m)`$ in $`_g^{}(n,m)`$. The precise definition of such leaf depends on the choice of the normalization. Then introduce special coordinates $`T_A`$, called *Whitham times*, on $`𝒱(n,m)`$, and define the prepotential $`S`$ of the leaf by the formula
$$dS=QdE.$$
*Whitham differentials* $`d\mathrm{\Omega }_A`$ are then obtained from $`dS`$ by
$$_{T_A}dS(E|T)=d\mathrm{\Omega }_A(E|T),$$
where we use a connection on $`𝒩_g^1`$ given by choosing $`E=\text{const}`$ to be horizontal sections. Note that $`dS=_AT_Ad\mathrm{\Omega }_A`$. The leaf $`𝒱(n,m)`$ is then mapped to the corresponding algebraic orbit $`𝒪(n)`$ by the following sequence of maps:
(13)
$$\begin{array}{ccccc}𝒱(n,m)& & 𝒱(n)& & 𝒪(n)\\ & & & & \\ & & 𝒱^{}(n)& & 𝒪^{}(n)\\ \{\mathrm{\Gamma }_g;P_1;[z]_n;E,Q\}& & \{\mathrm{\Gamma }_g;P_1;[z]_n;E\}& & \{\mathrm{\Gamma }_g;P_1;z_{}\},\end{array}$$
where $`𝒱^{}(n)`$ is a leaf of a subfibration of $`𝒱(n)`$ defined by fixing the periods of $`dE`$. Then, in the coordinates $`(E,T)`$ on the “universal curve” $`𝒩_{g^1}`$ over $`𝒪^{}(n)`$, the differential $`dS(E|T)=Q(E|T)dE`$ satisfies the equations $`_AdS(E|T)=d\mathrm{\Omega }_A(E|T)`$. Therefore, $`dS`$ is a prepotential, $`_x\mathrm{\Omega }_A(E|T)=_{T_A}p(E|T)`$, and we obtain solutions to Whitham equations. Note that Whitham coordinates on $`𝒱(n,m)`$ go to Whitham flows on the algebraic orbit.
For each Whitham derivative $`_{T_A}`$ we define a dual “integral” operator $`_{T_A}`$ in such a way that
(14)
$$_{T_A}_{T_B}dS=_{T_A}𝑑\mathrm{\Omega }_B=_{T_B}𝑑\mathrm{\Omega }_A=_{T_B}_{T_A}dST_A,T_B.$$
The exact definition of such operators $`_{T_A}`$ can be rather non-trivial and usually includes integration with some weight over a cycle and, maybe, some correction terms. The above identities can be thought of as a generalization of Riemann bilinear identities for Whitham differentials, and they are proved along the same lines. The $`\tau `$-function of an algebraic orbit is then defined by $`\tau (T)=e^{F(T)}`$, and
(15)
$$F(T)=\frac{1}{2}\underset{T_A}{}T_A_{T_A}𝑑S.$$
It encodes all information about the differentials $`d\mathrm{\Omega }_A`$ and the curve $`\mathrm{\Gamma }_g`$, and its first and second derivatives are given by
$`_{T_A}F(T)`$ $`={\displaystyle \frac{1}{2}}{\displaystyle _{T_A}}𝑑S+{\displaystyle \frac{1}{2}}{\displaystyle \underset{T_B}{}}T_B{\displaystyle _{T_B}}𝑑\mathrm{\Omega }_A`$
$`={\displaystyle \frac{1}{2}}{\displaystyle _{T_A}}𝑑S+{\displaystyle \frac{1}{2}}{\displaystyle \underset{T_B}{}}T_B{\displaystyle _{T_A}}𝑑\mathrm{\Omega }_B={\displaystyle _{T_A}}𝑑S,`$
$`_{T_AT_B}F(T)`$ $`={\displaystyle _{T_A}}𝑑\mathrm{\Omega }_B={\displaystyle _{T_B}}𝑑\mathrm{\Omega }_A.`$
Note that the equality of mixed partials of $`F(T)`$ corresponds to equation $`(\text{14})`$. Third derivatives of $`F(T)`$ are then given by the residue type formula.
## Complex-normalized case
In this case, one chooses
$`t_k`$ $`={\displaystyle \frac{1}{k}}\mathrm{res}_{P_1}z^kQdE,k=1,\mathrm{},n+m,`$
$`t_{h,k}`$ $`={\displaystyle _{a_k^{}}}Q𝑑E,t_{E,a_k}={\displaystyle _{a_k}}𝑑E,t_{Q,a_k}={\displaystyle _{a_k}}𝑑Q,t_{E,b_k}={\displaystyle _{b_k}}𝑑E,t_{Q,b_k}={\displaystyle _{b_k}}𝑑Q,`$
as coordinates on $`_g(n,m)`$. The notation $`a_k^{}`$ indicates that the integral has to be taken over the right side of the cycle. A *complex* $`(3g+n+m)`$-dimensional $`𝒱_{}(n,m)_g^{}(n,m)`$ is defined by imposing the $`a`$-normalization condition on the Abelian differentials $`dE`$ and $`dQ`$,
$$_{a_k}𝑑E=_{a_k}𝑑Q=0a_k.$$
Note that the projection of $`𝒱_{}(n,m)`$ to $`_g(n,m)`$ is not well-defined. Whitham coordinates on the leaf are then given by
$`T_k`$ $`={\displaystyle \frac{1}{k}}\mathrm{res}_{P_1}z^kdS,`$ $`k`$ $`=1,\mathrm{},n+m,`$
$`T_{h,k}`$ $`={\displaystyle _{a_k}}𝑑S,T_{E,k}={\displaystyle _{b_k}}𝑑Q,T_{Q,k}={\displaystyle _{b_k}}𝑑E,`$ $`k`$ $`=1,\mathrm{},g`$
Corresponding Whitham differentials are divided into the following four groups.
* $`_{T_k}dS(E|T)=d\mathrm{\Omega }_k(E|T)`$ are the usual complex-normalized Whitham differentials with prescribed behavior near the puncture,
$$d\mathrm{\Omega }_k=d\left(𝔨^k+O(𝔨^1)\right)\text{ near }P_1,_{a_l}𝑑\mathrm{\Omega }_k=0.$$
* $`_{T_{h,k}}dS(E|T)=d\mathrm{\Omega }_{h,k}(E|T)`$ are holomorphic differentials that are dual to the canonical basis of cycles,
$$_{a_i}𝑑\mathrm{\Omega }_{h,k}=\delta _{ik}.$$
* $`_{T_{E,k}}dS(E|T)=d\mathrm{\Omega }_{E,k}(E|T)`$ are $`a`$-normalized and holomorphic everywhere on $`\mathrm{\Gamma }_g`$ except for the $`a`$-cycles, where they have jumps,
$$d\mathrm{\Omega }_{E,k}(P_{a_l}^+P_{a_l}^{})=\delta _{kl}dE(P_{a_l}),_{a_i}𝑑\mathrm{\Omega }_{E,k}=0.$$
* $`_{T_{Q,k}}dS(E|T)=d\mathrm{\Omega }_{Q,k}(E|T)`$ are $`a`$-normalized and holomorphic everywhere on $`\mathrm{\Gamma }_g`$ except for the $`a`$-cycles, where they have jumps,
$$d\mathrm{\Omega }_{Q,k}(P_{a_l}^+P_{a_l}^{})=\delta _{kl}dQ(P_{a_l}),_{a_i}𝑑\mathrm{\Omega }_{Q,k}=0.$$
The duals $`_{T_A}`$ are then given by
$$_{T_i}𝑑\mathrm{\Omega }=\mathrm{res}_{P_1}𝔨^id\mathrm{\Omega },_{T_{h,k}}𝑑\mathrm{\Omega }=\frac{1}{2\pi \sqrt{1}}\left[_{[b_k]}𝑑\mathrm{\Omega }\right],_{T_{E,k}}𝑑\mathrm{\Omega }=\frac{1}{2\pi \sqrt{1}}\left[_{a_k^+}E𝑑\mathrm{\Omega }\right],$$
where the notation $`_{[b_k]}𝑑\mathrm{\Omega }`$ indicates that a certain correction terms have to be added to make the integral independent on small cycle deformations. In particular,
$$_{[b_k]}𝑑S=_{b_k}𝑑S+T_{E,k}E(a_k^+b_k).$$
This remark becomes very important in the real-normalized case and is discussed at length in the next section.
The third derivatives of $`F(T)`$ are given by the following theorem (\[Kri94\]).
###### Theorem.
$$_{T_AT_BT_C}F(T)=\underset{q_s|dE(q_s)=0}{}\mathrm{res}_{q_s}\frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_Bd\mathrm{\Omega }_C}{dEdQ}.$$
###### Remark.
The residue type formula for the third derivatives of $`F(T)`$ implies that if we consider the reduced hierarchy with $`dQ=dp`$, then $`F(T)`$ is a solution for the WDVV equation.
## Real-normalized Whitham hierarchies
### Real leaf and Whitham times
We begin the study of the real-normalized case by introducing the following real-analytic coordinate system on $`_g(n,m)`$:
$$t_k^𝔯=\mathrm{}\left[\frac{1}{k}\mathrm{res}_{P_1}z^kQdE\right],t_k^𝔦=\mathrm{}\left[\frac{1}{k}\mathrm{res}_{P_1}z^kQdE\right],k=1,\mathrm{},n+m,$$
and
$`t_{E,a_k}^𝔯`$ $`=\mathrm{}\left[{\displaystyle _{a_k}}𝑑E\right],`$ $`t_{E,a_k}^𝔦`$ $`=\mathrm{}\left[{\displaystyle _{a_k}}𝑑E\right],`$ $`t_{Q,a_k}^𝔯`$ $`=\mathrm{}\left[{\displaystyle _{a_k}}𝑑Q\right],`$ $`t_{Q,a_k}^𝔦`$ $`=\mathrm{}\left[{\displaystyle _{a_k}}𝑑Q\right],`$
$`t_{E,b_k}^𝔯`$ $`=\mathrm{}\left[{\displaystyle _{b_k}}𝑑E\right],`$ $`t_{E,b_k}^𝔦`$ $`=\mathrm{}\left[{\displaystyle _{b_k}}𝑑E\right],`$ $`t_{Q,b_k}^𝔯`$ $`=\mathrm{}\left[{\displaystyle _{b_k}}𝑑Q\right],`$ $`t_{Q,b_k}^𝔦`$ $`=\mathrm{}\left[{\displaystyle _{b_k}}𝑑Q\right],`$
$`t_{h,a_k}^𝔦`$ $`=\mathrm{}\left[{\displaystyle _{a_k^{}}}Q𝑑E\right],`$ $`t_{h,b_k}^𝔦`$ $`=\mathrm{}\left[{\displaystyle _{b_k^+}}Q𝑑E\right],`$ $`k`$ $`=1,\mathrm{},g.`$
Note that all coordinates except $`t_{h,a_k}^𝔦`$, $`t_{h,b_k}^𝔦`$ are just the real and imaginary parts of the corresponding complex analytic coordinates. A *real leaf* $`𝒱_{}(n,m)_g(n,m)`$ is defined to be a zero set of the coordinates $`t_{E,a_k}^𝔯,t_{E,b_k}^𝔯,t_{Q,a_k}^𝔯,t_{Q,a_k}^𝔯`$. Alternatively, $`𝒱_{}(n,m)`$ can be defined by the following $`4g`$ basis-independent equations,
(16)
$$\mathrm{}\left[_c𝑑E\right]=0,\mathrm{}\left[_c𝑑Q\right]=0,cH_1(\mathrm{\Gamma }_g,).$$
Therefore, $`𝒱_{}(n,m)`$ is well-defined as a real-analytic submanifold of both $`_g^{}(n,m)`$ and $`_g(n,m)`$.
Before introducing Whitham coordinates on $`𝒱_{}(n,m)`$, we have to make the following important observation. In what follows we work with multivalued differentials that usually have the form $`fd\omega `$, where $`f`$ is an Abelian integral and the differential $`d\omega `$ can again be multivalued. These differentials are well-defined only if we cut the surface $`\mathrm{\Gamma }_g`$. Thus, we always assume that some canonical basis $``$ of $`\mathrm{\Gamma }_g`$ is chosen, and we cut our surface along the representatives $`a_k`$, $`b_k`$ of cycles in $``$. Sometimes these cuts are not enough and we need one extra cut $`\gamma `$. Let us introduce the following notation.
###### Notation.
By $`\widehat{\mathrm{\Gamma }}_g`$ we denote the normal polygon obtained from $`\mathrm{\Gamma }_g`$ by cutting along the $`a`$ and $`b`$ cycles. For each cycle $`a_l`$ we denote by $`a_l^+`$ its *left* and by $`a_l^{}`$ its *right* sides. Same notation applies to $`b`$-cycles. The point of intersection of $`a_l^{}`$ and $`b_l^+`$ cycles is denoted by $`\mathrm{\Phi }_l`$ and the point of intersection of $`a_1^+`$ and $`b_g^{}`$ cycles is denoted by $`\mathrm{\Phi }_0`$. By $`\left(\mathrm{\Gamma }_g\right)_{\text{cut}}`$ we denote a Riemann surface obtained from $`\mathrm{\Gamma }_g`$ by making a cut $`\gamma `$ from $`\mathrm{\Phi }_0`$ to $`P_1`$, and by $`\left(\widehat{\mathrm{\Gamma }}_g\right)_{\text{cut}}`$ we denote its normal polygon. Note that topologically $`\left(\mathrm{\Gamma }_g\right)_{\text{cut}}\mathrm{\Gamma }_gP_1`$.
The multivalued differentials that we consider are single valued on either $`\widehat{\mathrm{\Gamma }}_g`$ or $`\left(\widehat{\mathrm{\Gamma }}_g\right)_{\text{cut}}`$ and, as differentials on $`\mathrm{\Gamma }_g`$ or $`\left(\mathrm{\Gamma }_g\right)_{\text{cut}}`$, they can have jumps across the $`a`$ and $`b`$-cycles, and a cut $`\gamma `$. Note that for such differentials, the integral over a cycle depends not only on the homology class of the cycle, but also on the side of the cycle and the actual choice of a representative in the homology class. Thus, in order to make our construction independent of a choice of such representative in the homology class, for each differential $`fd\omega `$ we introduce a corresponding cocycle $`[fd\omega ]`$ by the following procedure. First, we define $`[fd\omega ]`$ on basic cycles $`a_k`$, $`b_k`$ by adding certain *correction terms* to $`f𝑑\omega `$, and then extend it to an arbitrary cycle by linearity. We use the following notation.
###### Notation.
For any cycle $`c=\alpha _1a_1+\mathrm{}\beta _gb_g`$, by
$$_{[c]}f𝑑\omega =\left[fd\omega \right](c)=\alpha _1[fd\omega ](a_1)+\mathrm{}+\beta _g[fd\omega ](b_g)$$
we denote the value of $`[fd\omega ]`$ on a cycle $`c`$. Note that $`[fd\omega ](a_k)`$ and $`[fd\omega ](b_k)`$ still depend on a side of the cycle. We always choose a right side for $`a`$-cycles and left side for $`b`$-cycles,
$$_{[a_k]}f𝑑\omega :=_{[a_k^{}]}f𝑑\omega ,_{[b_k]}f𝑑\omega :=_{[b_k^+]}f𝑑\omega .$$
However, it is convenient to indicate the side explicitly in the intermediate calculations. For consistency, we use same notation for usual Abelian differentials on $`\mathrm{\Gamma }_g`$. In addition, for any differential $`df`$, we denote by $`f_l=^{\mathrm{\Phi }_l}𝑑f`$ the value of the corresponding Abelian integral $`f`$ at the point $`\mathrm{\Phi }_l`$ of intersection of the $`a_l^{}`$ and $`b_l^+`$ cycles. We also put
$$f_l^𝔯=\mathrm{}\left[f(\mathrm{\Phi }_l)\right],f_l^𝔦=\mathrm{}\left[f(\mathrm{\Phi }_l)\right].$$
We now define Whitahm times on $`𝒱_{}(n,m)`$ by
$$T_k^𝔯=\mathrm{}\left[\frac{1}{k}\mathrm{res}_{P_1}z^kQdE\right],T_k^𝔦=\mathrm{}\left[\frac{1}{k}\mathrm{res}_{P_1}z^kQdE\right],k=1,\mathrm{},n+m$$
and
$`T_{h,k}^a`$ $`=\mathrm{}\left[{\displaystyle _{[a_k^{}]}}Q𝑑E\right],`$ $`T_{h,k}^b`$ $`=\mathrm{}\left[{\displaystyle _{[b_k^+]}}Q𝑑E\right],`$ $`T_{E,k}^a`$ $`=\mathrm{}\left[{\displaystyle _{[b_k]}}𝑑Q\right],`$
$`T_{E,k}^b`$ $`=\mathrm{}\left[{\displaystyle _{[a_k]}}𝑑Q\right],`$ $`T_{Q,k}^a`$ $`=\mathrm{}\left[{\displaystyle _{[b_k]}}𝑑E\right],`$ $`T_{Q,k}^b`$ $`=\mathrm{}\left[{\displaystyle _{[a_k]}}𝑑E\right].`$
Except for some relabeling, the main change occurs in the definition of $`T_{h,k}^a`$, $`T_{h,k}^b`$:
$$T_{h,k}^a=t_{h,a_k}^𝔦t_{Q,a_k}^𝔦E_k^𝔯,T_{h,k}^b=t_{h,b_k}^𝔦t_{Q,b_k}^𝔦E_k^𝔯.$$
### Prepotential
The prepotential $`dS=QdE`$ of the real leaf $`𝒱_{}(n,m)`$ is described by the following proposition.
###### Proposition 1.
The prepotential $`dS`$ is a meromorphic differential on $`\mathrm{\Gamma }_g`$, holomorphic on $`\mathrm{\Gamma }_gP_1`$, with pole of order $`n+m+1`$ at $`P_1`$, and with jumps on $`a`$ and $`b`$ cycles. Near $`P_1`$,
$$dSd\left(\underset{j=1}{\overset{m+n}{}}(T_j^𝔯+\sqrt{1}T_j^𝔦)𝔨^j+O(z)\right)+R^S\frac{dz}{z},$$
where $`z=𝔨^1`$ is our preferred local coordinate in the neighborhood of $`P_1`$, and
(17)
$$R^S=\mathrm{res}_{P_1}dS=\frac{1}{2\pi \sqrt{1}}\underset{l=1}{\overset{g}{}}\left(T_{E,l}^aT_{Q,l}^bT_{E,l}^bT_{Q,l}^a\right).$$
The jumps of $`dS`$ across the $`a`$ and $`b`$-cycles come from the jumps of $`Q`$ and are given by
(18)
$$dS(P_{a_l}^+P_{a_l}^{})=\sqrt{1}T_{E,l}^adE,dS(P_{b_l}^+P_{b_l}^{})=\sqrt{1}T_{E,l}^bdE.$$
The corresponding cocycle $`[dS]`$ is well-defined only on $`\mathrm{\Gamma }_gP_1`$, and is given by
$$_{[a_l^\pm ]}𝑑S=_{a_l^\pm }𝑑S\sqrt{1}T_{E,l}^bE_l_{[b_l^\pm ]}𝑑S=_{b_l^\pm }𝑑S+\sqrt{1}T_{E,l}^aE_l$$
The proof of this proposition is a direct calculation.
Since $`dS`$ has a residue at $`P_1`$, the corresponding integral $`S(P)=^P𝑑S`$ is well-defined only on $`\left(\widehat{\mathrm{\Gamma }}_g\right)_{\text{cut}}`$. We choose an additive normalization constant in such a way that the regular part $`(S)_{}`$ vanishes at $`P_1`$. We can now consider cocycles corresponding to the differentials $`EdS`$ and $`SdE`$, which are well-defined on $`\mathrm{\Gamma }_gP_1`$.
###### Proposition 2.
The cocycles $`[EdS]`$ and $`[SdE]`$ are given by
$`{\displaystyle _{[a_l^{}]}}E𝑑S`$ $`={\displaystyle _{a_l^{}}}E𝑑S+\sqrt{1}T_{Q,l}^bS_l+T_{Q,l}^bT_{E,l}^bE_l\sqrt{1}T_{E,l}^b{\displaystyle \frac{E_l^2}{2}},`$
$`{\displaystyle _{[b_l^+]}}E𝑑S`$ $`={\displaystyle _{b_l^+}}E𝑑S\sqrt{1}T_{Q,l}^aS_l+T_{Q,l}^aT_{E,l}^aE_l+\sqrt{1}T_{E,l}^a{\displaystyle \frac{E_l^2}{2}},`$
$`{\displaystyle _{[a_l^{}]}}S𝑑E`$ $`={\displaystyle _{a_l^{}}}S𝑑E(\omega _S)_l^a^{}E_l+\sqrt{1}T_{E,l}^b{\displaystyle \frac{E_l^2}{2}},`$
$`{\displaystyle _{[b_l^+]}}S𝑑E`$ $`={\displaystyle _{b_l^+}}S𝑑E(\omega _S)_l^{b^+}E_l\sqrt{1}T_{E,l}^a{\displaystyle \frac{E_l^2}{2}}.`$
Moreover, the following “*integration by parts*” formulas hold:
$`\mathrm{}\left[{\displaystyle _{[a_l^{}]}}S𝑑E\right]`$ $`=\mathrm{}\left[{\displaystyle _{[a_l^{}]}}E𝑑S+T_{h,l}^aT_{Q,l}^b\right],`$
$`\mathrm{}\left[{\displaystyle _{[b_l^+]}}S𝑑E\right]`$ $`=\mathrm{}\left[{\displaystyle _{[b_l^+]}}E𝑑ST_{h,l}^bT_{Q,l}^a\right].`$
### Whitham differentials
Similarly to the complex-normalized case, Whitham differentials $`d\mathrm{\Omega }_A(E|T)=_{T_A}dS(E|T)`$ can be divided into four groups:
* $`d\mathrm{\Omega }_k^𝔯`$ and $`d\mathrm{\Omega }_k^𝔦`$ are real-normalized meromorphic differentials with prescribed singularities at $`P_1`$, i.e., exactly the differentials of the real-normalized Whitham hierarchy,
* $`d\mathrm{\Omega }_{h,k}^a`$ and $`d\mathrm{\Omega }_{h,k}^b`$ form a canonical real basis in the space of real-normalized holomorphic differentials that is dual to our basis $``$,
* $`d\mathrm{\Omega }_{E,k}^a`$ and $`d\mathrm{\Omega }_{E,k}^b`$ are meromorphic differentials with a simple pole at $`P_1`$ and a $`dE`$-jump across $`a`$ and $`b`$-cycles,
* $`d\mathrm{\Omega }_{Q,k}^a`$ and $`d\mathrm{\Omega }_{Q,k}^b`$ are meromorphic differentials with a simple pole at $`P_1`$ and a $`dQ`$-jump across $`a`$ and $`b`$-cycles.
In the sequel we restrict to leaves $`𝒱_{}^{}(n,m)`$ of a foliation defined by the level sets of $`T_{Q,k}^a`$, $`T_{Q,k}^b`$. Then all the differentials but $`d\mathrm{\Omega }_{Q,k}^a`$ and $`d\mathrm{\Omega }_{Q,k}^b`$ generate flows preserving the foliation. For this reason, we do not describe differentials $`d\mathrm{\Omega }_{Q,k}^a`$, $`d\mathrm{\Omega }_{Q,k}^b`$ in detail. We also need to consider corresponding Abelian differentials, which we always normalize by $`(\mathrm{\Omega }_A)_(P_1)=0`$.
###### Proposition 3.
For any Whitham time $`T_A`$, the differential $`d\mathrm{\Omega }_A(E|T)=_{T_A}dS(E|T)`$ is holomorphic on $`\mathrm{\Gamma }_gP_1`$.
###### Proof.
The only possible extra poles can appear when $`E(P)`$ does not define a local coordinate, i.e., at the points $`q_s`$ such that $`dE(q_s)=0`$. Assuming for simplicity that $`dE`$ has a simple pole at $`q_s`$, let $`E_s(T)=E(q_s(T)|T)`$, and choose $`\xi (E|T)=\sqrt{EE_s(T)}`$ to be a local coordinate near $`q_s`$. Then we have
$`_{T_A}dS(E|T)`$ $`=_{T_A}(Q(\xi (E|T)|T))d(\xi ^2+E_s)`$
$`={\displaystyle \frac{dQ}{d\xi }}{\displaystyle \frac{_{T_A}(E_s)}{\sqrt{EE_s}}}2\xi d\xi =(_{T_A}E_s)dQ(E|T),`$
which is holomorphic at $`q_s`$. ∎
Main properties of the Whitham differentials are summarized in the following proposition.
###### Proposition 4.
* Differentials $`d\mathrm{\Omega }_k^𝔯(E|T)`$ and $`d\mathrm{\Omega }_k^𝔦(E|T)`$ are *real-normalized*,
$$\mathrm{}\left[_c𝑑\mathrm{\Omega }_k^𝔯\right]=\mathrm{}\left[_c𝑑\mathrm{\Omega }_k^𝔦\right]=0cH_1(\mathrm{\Gamma }_g,),$$
meromorphic on $`\mathrm{\Gamma }_g`$, with a single pole at $`P_1`$, where
$$d\mathrm{\Omega }_k^𝔯d(𝔨^k+O(1)),d\mathrm{\Omega }_k^𝔦d(\sqrt{1}𝔨^k+O(1))\mathrm{res}_{P_1}\mathrm{\Omega }_k^𝔯=\mathrm{res}_{P_1}\mathrm{\Omega }_k^𝔦=0.$$
* Differentials $`d\mathrm{\Omega }_{h,k}^a(E|T)`$ and $`d\mathrm{\Omega }_{h,k}^b(E|T)`$ are holomorphic on $`\mathrm{\Gamma }_g`$ and form a canonical basis in the space of real-normalized holomorphic differentials dual to our homology basis $``$, i.e.,
$`\mathrm{}\left[{\displaystyle _{a_l}}𝑑\mathrm{\Omega }_{h,k}^a\right]`$ $`=\delta _{kl},`$ $`\mathrm{}\left[{\displaystyle _{b_l}}𝑑\mathrm{\Omega }_{h,k}^a\right]`$ $`=0,`$
$`\mathrm{}\left[{\displaystyle _{a_l}}𝑑\mathrm{\Omega }_{h,k}^b\right]`$ $`=0,`$ $`\mathrm{}\left[{\displaystyle _{b_l}}𝑑\mathrm{\Omega }_{h,k}^a\right]`$ $`=\delta _{kl}`$
* Differentials $`d\mathrm{\Omega }_{E,k}^a(E|T)`$ and $`d\mathrm{\Omega }_{E,k}^b(E|T)`$ are meromorphic on $`\mathrm{\Gamma }_g`$, holomorphic on $`\mathrm{\Gamma }_gP_1`$, have a simple pole at $`P_1`$ which is balanced by a single jump across one of the cycles:
$`\mathrm{res}_{P_1}d\mathrm{\Omega }_{E,k}^a`$ $`={\displaystyle \frac{1}{2\pi \sqrt{1}}}T_{Q,k}^b={\displaystyle \frac{1}{2\pi }}{\displaystyle _{a_k}}𝑑E,`$ $`d\mathrm{\Omega }_{E,k}^a(P_{a_l}^+P_{a_l}^{})`$ $`=\sqrt{1}\delta _{kl}dE(P_{a_l}),`$
$`d\mathrm{\Omega }_{E,k}^a(P_{b_l}^+P_{b_l}^{})`$ $`=0,`$
$`\mathrm{res}_{P_1}d\mathrm{\Omega }_{E,k}^b`$ $`={\displaystyle \frac{1}{2\pi \sqrt{1}}}T_{Q,k}^a={\displaystyle \frac{1}{2\pi }}{\displaystyle _{b_k}}𝑑E,`$ $`d\mathrm{\Omega }_{E,k}^b(P_{a_l}^+P_{a_l}^{})`$ $`=0,`$
$`d\mathrm{\Omega }_{E,k}^b(P_{b_l}^+P_{b_l}^{})`$ $`=\sqrt{1}\delta _{kl}dE(P_{b_l}),`$
differentials themselves are *not* real-normalized, but the corresponding cocycles are,
$$\mathrm{}\left[_{[c]}𝑑\mathrm{\Omega }_{E,k}^a\right]=\mathrm{}\left[_{[c]}𝑑\mathrm{\Omega }_{E,k}^b\right]=0.$$
### Duality and Riemann Bilinear Relations
In the real-normalized case, the dual “integral” operators $`_{T_A}`$ are given by the following formulas:
$`{\displaystyle _{T_i^𝔯}}𝑑\mathrm{\Omega }`$ $`=\mathrm{}\left[\mathrm{res}_{P_1}𝔨^id\mathrm{\Omega }\right],`$ $`{\displaystyle _{T_i^𝔦}}𝑑\mathrm{\Omega }`$ $`=\mathrm{}\left[\mathrm{res}_{P_1}\sqrt{1}𝔨^id\mathrm{\Omega }\right],`$
$`{\displaystyle _{T_{h,k}^a}}𝑑\mathrm{\Omega }`$ $`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}{\displaystyle _{[b_{k^+}]}}𝑑\mathrm{\Omega }\right],`$ $`{\displaystyle _{T_{h,k}^a}}𝑑\mathrm{\Omega }`$ $`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}{\displaystyle _{[a_k^{}]}}𝑑\mathrm{\Omega }\right],`$
$`{\displaystyle _{T_{E,k}^a}}𝑑\mathrm{\Omega }`$ $`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}\left({\displaystyle _{[a_k^{}]}}E𝑑\mathrm{\Omega }+\mathrm{Corr}_k^a(d\mathrm{\Omega })\right)\right],`$
$`{\displaystyle _{T_{E,k}^a}}𝑑\mathrm{\Omega }`$ $`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}\left({\displaystyle _{[b_{k^+}]}}E𝑑\mathrm{\Omega }+\mathrm{Corr}_k^b(d\mathrm{\Omega })\right)\right],`$
where the correction terms $`\mathrm{Corr}(d\mathrm{\Omega })`$ are given by
$`\mathrm{Corr}_k^a(dS)`$ $`=T_{h,k}^aT_{Q,k}^b+T_{h,k}^bT_{Q,k}^b,\mathrm{Corr}_k^b(dS)=T_{h,k}^aT_{Q,k}^b+T_{h,k}^bT_{Q,k}^b,`$
$`\mathrm{Corr}(d\mathrm{\Omega }_A)`$ $`=_{T_A}\mathrm{Corr}(dS).`$
We now have to establish relations $`(\text{14})`$.
###### Proposition 5.
Whitham differentials satisfy the identity
(19)
$$_{T_A}𝑑\mathrm{\Omega }_B=_{T_B}𝑑\mathrm{\Omega }_A$$
###### Proof.
This proposition is proved by a direct calculation. We illustrate it by considering the following cases.
First, let $`T_A=T_i^𝔯`$ and $`T_B=T_j^𝔯`$. Then we have to prove the following identity:
$$_{T_i^𝔯}𝑑\mathrm{\Omega }_j^𝔯=\mathrm{}\left[\mathrm{res}_{P_1}𝔨^id\mathrm{\Omega }_j^𝔯\right]=\mathrm{}\left[\mathrm{res}_{P_1}𝔨^jd\mathrm{\Omega }_i^𝔯\right]=_{T_j^𝔯}𝑑\mathrm{\Omega }_i^𝔯.$$
But
$`\mathrm{res}_{P_1}𝔨^id\mathrm{\Omega }_j^𝔯`$ $`=\mathrm{res}_{P_1}\mathrm{\Omega }_j^𝔯d𝔨^i=\mathrm{res}_{P_1}\mathrm{\Omega }_j^𝔯d\mathrm{\Omega }_i^𝔯\mathrm{res}_{P_1}\mathrm{\Omega }_j^𝔯d(𝔨^i)_{}`$
$`=\mathrm{res}_{P_1}\mathrm{\Omega }_j^𝔯d\mathrm{\Omega }_i^𝔯\mathrm{res}_{P_1}𝔨^jd(𝔨^i)_{}=\mathrm{res}_{P_1}\mathrm{\Omega }_j^𝔯d\mathrm{\Omega }_i^𝔯+\mathrm{res}_{P_1}(𝔨^jd\mathrm{\Omega }_i^𝔯).`$
Taking real parts and observing that
$$\mathrm{}\left[\mathrm{res}_{P_1}\mathrm{\Omega }_j^𝔯d\mathrm{\Omega }_i^𝔯\right]=\mathrm{}\left[\frac{1}{2\pi \sqrt{1}}_{\widehat{\mathrm{\Gamma }}_g}\mathrm{\Omega }_j^𝔯𝑑\mathrm{\Omega }_i^𝔯\right]=0,$$
we obtain the desired identity.
Let $`T_A=T_{h,k}^a`$, $`T_B=T_{h,l}^a`$. Then
$$_{T_{h,k}^a}𝑑\mathrm{\Omega }_{h,l}^a=\mathrm{}\left[_{[b_k^+]}𝑑\mathrm{\Omega }_{h,l}^a\right]=\mathrm{}\left[_{[b_l^+]}𝑑\mathrm{\Omega }_{h,k}^a\right]=_{T_{h,l}^a}𝑑\mathrm{\Omega }_{h,k}^a$$
is just the usual Riemann bilinear identity for a canonical basis of real-normalized holomorphic differentials and it is proved in the regular way:
$$0=\mathrm{}\left[\mathrm{res}_{P_1}\mathrm{\Omega }_{h,k}^ad\mathrm{\Omega }_{h,l}^a\right]=\mathrm{}\left[\frac{1}{2\pi }\left(_{b_l}𝑑\mathrm{\Omega }_{h,k}^a+_{b_k}𝑑\mathrm{\Omega }_{h,l}^a\right)\right].$$
The most difficult identities to establish correspond to $`T_A=T_{E,k}^a`$, $`T_B=T_{E,s}^a`$ and the like, since $`d\mathrm{\Omega }_{E,k}^a`$, $`d\mathrm{\Omega }_{E,k}^b`$ have simple poles at $`P_1`$. In this case, let $`C_\epsilon `$ be a small circle around $`P_1`$, $`\mathrm{\Phi }_\epsilon C_\epsilon `$, and let $`\gamma _\epsilon `$ be a cut from $`\mathrm{\Phi }_0`$ to $`\mathrm{\Phi }_\epsilon `$. We consider the integral $`\mathrm{\Omega }_{E,k}^ad\mathrm{\Omega }_{E,s}^a`$ along the contour
(20)
$$\underset{l=1}{\overset{g}{}}(a_i^++b_i^+a_i^{}b_i^{})+\gamma _\epsilon ^+\gamma _\epsilon ^{}+C_\epsilon .$$
First, we compute that
$`\mathrm{}\left[{\displaystyle \frac{1}{2\pi \sqrt{1}}}{\displaystyle \underset{l=1}{\overset{g}{}}}\left({\displaystyle _{a_{l^+}}}+{\displaystyle _{b_{l^+}}}{\displaystyle _{a_l^{}}}{\displaystyle _{b_l^{}}}\right)\mathrm{\Omega }_{E,k}^ad\mathrm{\Omega }_{E,s}^a\right]`$
$`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}\left({\displaystyle _{a_k^{}}}E𝑑\mathrm{\Omega }_{E,s}^a+{\displaystyle _{a_s^{}}}\mathrm{\Omega }_{E,k}^a𝑑E+\delta _{ks}E_k^𝔯T_{Q,k}^b\left({\displaystyle _{[a_s^{}]}}𝑑\mathrm{\Omega }_{E,k}^a\right)E_s^𝔯\right)\right]`$
$`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}\left({\displaystyle _{[a_k^{}]}}E𝑑\mathrm{\Omega }_{E,s}^a+{\displaystyle _{[a_s^{}]}}\mathrm{\Omega }_{E,k}^a𝑑E+T_{Q,k}^b((\mathrm{\Omega }_{E,s}^a)_k^𝔦+\delta _{ks}E_s^𝔯)\right)\right].`$
Note that
$`(\mathrm{\Omega }_{E,s}^a)_k^𝔦+\delta _{ks}E_k^𝔯`$ $`=(\mathrm{\Omega }_{E,s}^a)_0^𝔦+\mathrm{}\left[{\displaystyle _{\mathrm{\Phi }_0}^{\mathrm{\Phi }_k}}𝑑\mathrm{\Omega }_{E,s}^a\right]+\delta _{ks}E_k^𝔯`$
$`=(\mathrm{\Omega }_{E,s}^a)_0^𝔦=\mathrm{}\left[\mathrm{\Omega }_{E,s}^a(\mathrm{\Phi }_0^{})\right]=\mathrm{}\left[\mathrm{\Omega }_{E,s}^a(\mathrm{\Phi }_0^+)\right],`$
since the only non-trivial imaginary contribution for periods of $`d\mathrm{\Omega }_{E,s}^a`$ comes from $`b`$-periods, and in order for $`b^+`$-contribution not to be canceled by $`b^{}`$-contribution, we need $`k=s`$, in which case $`\mathrm{}\left[_{\mathrm{\Phi }_0}^{\mathrm{\Phi }_k}𝑑\mathrm{\Omega }_{E,s}^a\right]=\delta _{ks}E_k^𝔯.`$
On the cut $`\gamma `$ we have
$$\left(_{\gamma _\epsilon ^+}_{\gamma _\epsilon ^{}}\right)(\mathrm{\Omega }_{E,k}^ad\mathrm{\Omega }_{E,s}^a)=_{\mathrm{\Phi }_0^+}^{\mathrm{\Phi }_\epsilon ^+}T_{Q,k}^b𝑑\mathrm{\Omega }_{E,s}^a=T_{Q,k}^b(\mathrm{\Omega }_{E,s}^a(\mathrm{\Phi }_\epsilon ^+)\mathrm{\Omega }_{E,s}^a(\mathrm{\Phi }_0^+)).$$
To evaluate $`_{C_\epsilon }\mathrm{\Omega }_{E,k}^a𝑑\mathrm{\Omega }_{E,s}^a`$, we rewrite everything in terms of a local coordinate $`z`$ in the neighborhood of $`P_1`$:
$$d\mathrm{\Omega }_{E,k}^a=\frac{T_{Q,k}^b}{2\pi \sqrt{1}}\frac{dz}{z}+(\alpha +O(z))dz,\mathrm{\Omega }_{E,k}^a=\frac{T_{Q,k}^b}{2\pi \sqrt{1}}\mathrm{Log}(z)+O(z).$$
Then
$`{\displaystyle _{\mathrm{\Phi }_\epsilon ^+}^{\mathrm{\Phi }_\epsilon ^{}}}\mathrm{\Omega }_{E,k}^a𝑑\mathrm{\Omega }_{E,s}^a`$ $`={\displaystyle _{\mathrm{\Phi }_\epsilon ^+}^{\mathrm{\Phi }_\epsilon ^{}}}{\displaystyle \frac{T_{Q,k}^bT_{Q,s}^b}{4\pi ^2}}{\displaystyle \frac{\mathrm{Log}(z)}{z}}𝑑z+O(\epsilon )`$
$`={\displaystyle \frac{T_{Q,k}^bT_{Q,s}^b}{8\pi ^2}}\left((\mathrm{Log}(z))^2|_{\mathrm{\Phi }_\epsilon ^{}}^{\mathrm{\Phi }_\epsilon ^+}\right)+O(\epsilon )`$
$`={\displaystyle \frac{T_{Q,k}^bT_{Q,s}^b}{8\pi ^2}}(4\pi \sqrt{1}\mathrm{Log}(z(\mathrm{\Phi }_\epsilon ^+))+4\pi ^2)+O(\epsilon )`$
$`=T_{Q,k}^b(\mathrm{\Omega }_{E,s}^a(\mathrm{\Phi }_\epsilon ^+))+{\displaystyle \frac{T_{Q,k}^bT_{Q,s}^b}{2}}+O(\epsilon ).`$
Collecting all of the above together, we obtain
$`0`$ $`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi \sqrt{1}}}\left({\displaystyle \underset{l=1}{\overset{g}{}}}\left({\displaystyle _{a_{l^+}}}+{\displaystyle _{b_{l^+}}}{\displaystyle _{a_l^{}}}{\displaystyle _{b_l^{}}}\right)+{\displaystyle _{\gamma _\epsilon ^+}}{\displaystyle _{\gamma _\epsilon ^{}}}+{\displaystyle _{C_\epsilon }}\right)\mathrm{\Omega }_{E,k}^ad\mathrm{\Omega }_{E,s}^a\right]`$
$`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}\left({\displaystyle _{[a_k^{}]}}E𝑑\mathrm{\Omega }_{E,s}^a+{\displaystyle _{[a_s^{}]}}\mathrm{\Omega }_{E,k}^a𝑑E+O(\epsilon )\right)\right]`$
$`\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}\left({\displaystyle _{[a_k^{}]}}E𝑑\mathrm{\Omega }_{E,s}^a+{\displaystyle _{[a_s^{}]}}\mathrm{\Omega }_{E,k}^a𝑑E\right)\right]\text{as }\epsilon 0,`$
which proves our identity. All other cases are somewhat intermediate in difficulty to the cases considered above and are proved along the same lines. ∎
The third derivatives of $`F(T)`$ are given by the following theorem.
###### Theorem 1.
$$_{T_AT_BT_C}F(T)=\mathrm{}\left[\underset{q_s|dE(q_s)=0}{}\mathrm{res}_{q_s}\frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_Bd\mathrm{\Omega }_C}{dEdQ}\right]$$
###### Proof.
First we prove the following formula. For any two Whitham differentials $`d\mathrm{\Omega }_B`$, $`d\mathrm{\Omega }_C`$ we have
(21)
$$\begin{array}{cc}\hfill \mathrm{}\left[\frac{1}{2\pi \sqrt{1}}_{\widehat{\mathrm{\Gamma }}}(_{T_A}\mathrm{\Omega }_B)𝑑\mathrm{\Omega }_C\right]& =\mathrm{}\left[\mathrm{res}_{P_1}(_{T_A}\mathrm{\Omega }_B)d\mathrm{\Omega }_C\right]\hfill \\ & +\mathrm{}\left[\underset{q_s|dE(q_s)=0}{}\mathrm{res}_{q_s}\frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_Bd\mathrm{\Omega }_C}{dEdQ}\right].\hfill \end{array}$$
To establish this identity, note that since $`d\mathrm{\Omega }_C`$ is holomorphic outside of $`P_1`$, the right hand side of our equation is a sum of residues at $`P_1`$ and at poles of $`_{T_A}\mathrm{\Omega }_B`$. Using $`E`$ as a coordinate, we see that
$$_{T_A}\mathrm{\Omega }_B(E|T)=_{T_A}_{T_B}S(E|T),$$
is holomorphic. So the only extra poles can appear at $`q_s`$ such that $`dE(q_s)=0`$. Assuming that $`q_s`$ is a simple zero of $`dE`$ and using $`\xi (P|T)=\sqrt{E(P)E_s(T)}`$ as a local coordinate near $`q_s`$, we have
$`\mathrm{\Omega }_B(P|T)`$ $`=\mathrm{\Omega }_B(q_s(T))+{\displaystyle \frac{d\mathrm{\Omega }_B}{d\xi }}(q_s(T))\xi (P|T)+O(\xi (P|T)^2)`$
$`(_{T_A}\mathrm{\Omega }_B)(P|T)`$ $`={\displaystyle \frac{d\mathrm{\Omega }_B}{d\xi }}(q_s(T)){\displaystyle \frac{_{T_A}E_s(T)}{2\xi (P|T)}}+O(1)`$
$`(_{T_A}Q)(P|T)`$ $`={\displaystyle \frac{dQ}{d\xi }}(q_s(T)){\displaystyle \frac{_{T_A}E_s(T)}{2\xi (P|T)}}+O(1).`$
At the same time,
$$(_{T_A}Q)(P|T)=_{T_A}\frac{dS(P|T)}{dE(P)}=\frac{d\mathrm{\Omega }_A}{dE}(P|T).$$
Therefore,
$`(_{T_A}\mathrm{\Omega }_B)(P|T)`$ $`={\displaystyle \frac{d\mathrm{\Omega }_B}{d\xi }}{\displaystyle \frac{d\xi }{dQ}}(q_s(T))\left({\displaystyle \frac{d\mathrm{\Omega }_A}{dE}}(P|T)+O(1)\right)`$
$`={\displaystyle \frac{d\mathrm{\Omega }_B}{dQ}}(q_s(T)){\displaystyle \frac{d\mathrm{\Omega }_A}{dE}}(P|T)+O(1)`$
and we have
$`\mathrm{res}_{q_s}(_{T_A}\mathrm{\Omega }_B)d\mathrm{\Omega }_C`$ $`=\mathrm{res}_{q_s}{\displaystyle \frac{d\mathrm{\Omega }_B}{dQ}}(q_s(T)){\displaystyle \frac{d\mathrm{\Omega }_A}{dE}}d\mathrm{\Omega }_C(P|T)`$
$`=\mathrm{res}_{q_s}{\displaystyle \frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_Bd\mathrm{\Omega }_C}{dEdQ}}(P|T).`$
The rest of the proof is a direct calculation, which we illustrate by three different cases.
$`_{T_A}(_{T_i^𝔯}_{T_j^𝔯}F(T))`$ $`=\mathrm{}\left[\mathrm{res}_{P_1}𝔨^id(_{T_A}\mathrm{\Omega }_j^𝔯)\right]=\mathrm{}\left[\mathrm{res}_{P_1}\mathrm{\Omega }_i^𝔯d(_{T_A}\mathrm{\Omega }_j^𝔯)\right]`$
$`=\mathrm{}\left[\mathrm{res}_{P_1}_{T_A}\mathrm{\Omega }_j^𝔯d\mathrm{\Omega }_i^𝔯\right]=\mathrm{}\left[{\displaystyle \underset{q_s|dE(q_s)=0}{}}\mathrm{res}_{q_s}{\displaystyle \frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_i^𝔯d\mathrm{\Omega }_j^𝔯}{dEdQ}}\right],`$
since $`_{T_A}\mathrm{\Omega }_j^𝔯`$ is holomorphic at $`P_1`$ and $`\mathrm{}\left[\frac{1}{2\pi \sqrt{1}}_{\widehat{\mathrm{\Gamma }}_g}(_{T_A}\mathrm{\Omega }_j^𝔯)𝑑\mathrm{\Omega }_i^𝔯\right]=0.`$
$`_{T_A}(_{T_{h,l}^a}_{T_{h,k}^a}F(T))`$ $`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi \sqrt{1}}}{\displaystyle _{[b_{k^+}]}}d(_{T_A}\mathrm{\Omega }_{h,l}^a)\right]=\mathrm{}\left[{\displaystyle \frac{1}{2\pi \sqrt{1}}}{\displaystyle _{\widehat{\mathrm{\Gamma }}_g}}(_{T_A}\mathrm{\Omega }_{h,l}^a)𝑑\mathrm{\Omega }_{h,k}^a\right]`$
$`=\mathrm{}\left[{\displaystyle \underset{q_s|dE(q_s)=0}{}}\mathrm{res}_{q_s}{\displaystyle \frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_{h,k}^ad\mathrm{\Omega }_{h,l}^a}{dEdQ}}\right],`$
since $`\mathrm{res}_{P_1}(_{T_A}\mathrm{\Omega }_{h,l}^a)d\mathrm{\Omega }_{h,k}^a=0.`$
$`_{T_A}(_{T_{E,l}^a}_{T_{E,k}^a}F(T))`$ $`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}{\displaystyle _{[a_k^{}]}}Ed(_{T_A}\mathrm{\Omega }_{E,k}^a)\right]=\mathrm{}\left[{\displaystyle \frac{1}{2\pi }}{\displaystyle _{[a_l^{}]}}(_{T_A}\mathrm{\Omega }_{E,k}^a)𝑑E\right]`$
$`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi \sqrt{1}}}{\displaystyle _{\widehat{\mathrm{\Gamma }}_g}}(_{T_A}\mathrm{\Omega }_{E,k}^a)𝑑\mathrm{\Omega }_{E,l}^a\right]`$
$`=\mathrm{}\left[{\displaystyle \underset{q_s|dE(q_s)=0}{}}\mathrm{res}_{q_s}{\displaystyle \frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_{E,k}^ad\mathrm{\Omega }_{E,l}^a}{dEdQ}}\right],`$
since $`\mathrm{res}_{P_1}(_{T_A}\mathrm{\Omega }_{E,l}^a)d\mathrm{\Omega }_{E,k}^a=0`$ and $`(_{T_A}\mathrm{\Omega }_{E,k}^a)`$ is well-defined on $`\widehat{\mathrm{\Gamma }}_g`$. ∎
### Reduction to Frobenius Structures and Solutions to the WDVV equations
In the real-normalized case we choose the marked variable to be $`T_1^𝔯`$. It correspond to the direction of unity,
(22)
$$_{T_1^𝔯}=ed\mathrm{\Omega }_1^𝔯.$$
In addition, we choose
(23)
$$dQ=\sqrt{1}d\mathrm{\Omega }_1^𝔯,$$
and truncate the hierarchy by considering $`T_i^𝔯,T_i^𝔦`$ only for $`in`$. Then we have the following theorem.
###### Theorem 2.
Given the above assumptions,
(24)
$$_{T_1^𝔯T_AT_B}F(T)=\{\begin{array}{cc}\frac{ij}{n},\hfill & d\mathrm{\Omega }_A=d\mathrm{\Omega }_i^𝔯,d\mathrm{\Omega }_B=d\mathrm{\Omega }_j^𝔦,\text{ and }i+j=n\hfill \\ \frac{ij}{n},\hfill & d\mathrm{\Omega }_A=d\mathrm{\Omega }_i^𝔦,d\mathrm{\Omega }_B=d\mathrm{\Omega }_j^𝔯,\text{ and }i+j=n\hfill \\ \frac{1}{2\pi },\hfill & d\mathrm{\Omega }_A=d\mathrm{\Omega }_{E,k}^a,d\mathrm{\Omega }_B=d\mathrm{\Omega }_{h,k}^a\hfill \\ \frac{1}{2\pi },\hfill & d\mathrm{\Omega }_A=d\mathrm{\Omega }_{E,k}^b,d\mathrm{\Omega }_B=d\mathrm{\Omega }_{h,k}^a\hfill \\ 0\hfill & \text{ otherwise.}\hfill \end{array}$$
Therefore, the coordinates are flat and $`F(T)`$ is a solution to the WDVV equation.
###### Proof.
Let us rewrite the formula for the third derivative with $`dQ=\sqrt{1}d\mathrm{\Omega }_1^𝔯`$:
$`_{T_1^𝔯T_AT_B}F(T)`$ $`=\mathrm{}\left[{\displaystyle \underset{q_s|dE(q_s)=0}{}}\mathrm{res}_{q_s}{\displaystyle \frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_B}{\sqrt{1}dE}}\right]`$
$`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi \sqrt{1}}}{\displaystyle _{\widehat{\mathrm{\Gamma }}_g}}{\displaystyle \frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_B}{\sqrt{1}dE}}\mathrm{res}_{P_1}{\displaystyle \frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_B}{\sqrt{1}dE}}\right].`$
First let us see what is happening near the puncture $`P_1`$. Since $`dEd(z^n+O(1))`$ near $`P_1`$, $`\frac{dz}{dE}z^{n+1}`$ has a zero of order $`n+1`$ at $`P_1`$. It is clear that for the truncated hierarchy the only differentials that can contribute to the $`\mathrm{res}_{P_1}()`$-term are $`d\mathrm{\Omega }_i^𝔯`$ and $`d\mathrm{\Omega }_i^𝔦`$. Assume that near $`P_1`$,
$$d\mathrm{\Omega }_Ad(\epsilon _Az^i+O(1)),d\mathrm{\Omega }_Bd(\epsilon _Bz^j+O(1)),\epsilon =1\text{ or }\sqrt{1}.$$
Then
$`d\mathrm{\Omega }_Ad\mathrm{\Omega }_B`$ $`=(\epsilon _A(i)z^{i1}+O(1))(\epsilon _B(j)z^{j1}+O(1))(dz)^2`$
$`=\{\epsilon _a\epsilon _B(ij)z^{(i+j)2}+O(z^{i1})+O(z^{j1})+O(1)\}(dz)^2`$
$`\mathrm{}\left[\mathrm{res}_{P_1}{\displaystyle \frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_B}{\sqrt{1}dE}}\right]`$ $`=\mathrm{}\left[{\displaystyle \frac{ij}{n}}\sqrt{1}\epsilon _A\epsilon _B\right]\delta _{i+j,n}={\displaystyle \frac{ij}{n}}\delta _{i+j,n}\delta _{\epsilon _A\epsilon _B,\sqrt{1}}.`$
Now let us consider the boundary term $`{\displaystyle \frac{1}{2\pi \sqrt{1}}}_{\widehat{\mathrm{\Gamma }}_g}\frac{d\mathrm{\Omega }_Ad\mathrm{\Omega }_B}{\sqrt{1}dE}`$. For it to be non-zero, at least one of $`d\mathrm{\Omega }_A,d\mathrm{\Omega }_B`$ should be $`d\mathrm{\Omega }_{E,k}^a`$ or $`d\mathrm{\Omega }_{E,k}^b`$. For example, let $`d\mathrm{\Omega }_A=d\mathrm{\Omega }_{E,k}^a`$. Then, if $`d\mathrm{\Omega }_Bd\mathrm{\Omega }_{E,s}^a,d\mathrm{\Omega }_{E,s}^b`$,
$`{\displaystyle \frac{d\mathrm{\Omega }_{E,k}^a}{dE}}(P_{a_l}^+P_{a_l}^{})`$ $`=\sqrt{1}\delta _{kl}{\displaystyle \frac{d\mathrm{\Omega }_{E,k}^a}{dE}}(P_{b_l}^+P_{b_l}^{})=0`$
$`\mathrm{}\left[{\displaystyle \frac{1}{2\pi \sqrt{1}}}{\displaystyle _{\widehat{\mathrm{\Gamma }}_g}}{\displaystyle \frac{d\mathrm{\Omega }_{E,k}^ad\mathrm{\Omega }_B}{\sqrt{1}dE}}\right]`$ $`=\mathrm{}\left[{\displaystyle \frac{1}{2\pi \sqrt{1}}}{\displaystyle _{a_{k^+}}}𝑑\mathrm{\Omega }_B\right]=\{\begin{array}{cc}\frac{1}{2\pi },\hfill & d\mathrm{\Omega }_B=d\mathrm{\Omega }_{h,k}^a\hfill \\ 0\hfill & \text{otherwise}\hfill \end{array}`$
and if $`d\mathrm{\Omega }_B=d\mathrm{\Omega }_{E,s}^a`$ or $`d\mathrm{\Omega }_B=d\mathrm{\Omega }_{E,s}^b`$,
$$\mathrm{}\left[\frac{1}{2\pi \sqrt{1}}_{\widehat{\mathrm{\Gamma }}_g}\frac{d\mathrm{\Omega }_{E,k}^ad\mathrm{\Omega }_{E,s}^a}{\sqrt{1}dE}\right]=\mathrm{}\left[\frac{1}{2\pi \sqrt{1}}_{\widehat{\mathrm{\Gamma }}_g}\frac{d\mathrm{\Omega }_{E,k}^ad\mathrm{\Omega }_{E,s}^b}{\sqrt{1}dE}\right]=0.$$
All the remaining cases are similar to the ones considered, and this concludes the proof of the theorem. ∎ |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.